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 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 unsigned int nf_bridge_encap_header_len(const struct sk_buff *skb)
 157 {
 158         switch (skb->protocol) {
 159         case __cpu_to_be16(ETH_P_8021Q):
 160                 return VLAN_HLEN;
 161         case __cpu_to_be16(ETH_P_PPP_SES):
 162                 return PPPOE_SES_HLEN;
 163         default:
 164                 return 0;
 165         }
 166 }
 167
 168 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 169 {
 170         unsigned int len = nf_bridge_encap_header_len(skb);
 171
 172         skb_push(skb, len);
 173         skb->network_header -= len;
 174 }
 175
 176 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 177 {
 178         unsigned int len = nf_bridge_encap_header_len(skb);
 179
 180         skb_pull(skb, len);
 181         skb->network_header += len;
 182 }
 183
 184 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 185 {
 186         unsigned int len = nf_bridge_encap_header_len(skb);
 187
 188         skb_pull_rcsum(skb, len);
 189         skb->network_header += len;
 190 }
 191
 192 static inline void nf_bridge_save_header(struct sk_buff *skb)
 193 {
 194         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 195
 196         skb_copy_from_linear_data_offset(skb, -header_size,
 197                                          skb->nf_bridge->data, header_size);
 198 }
 199
 200 /* When handing a packet over to the IP layer
 201  * check whether we have a skb that is in the
 202  * expected format
 203  */
 204
 205 static int br_parse_ip_options(struct sk_buff *skb)
 206 {
 207         const struct iphdr *iph;
 208         struct net_device *dev = skb->dev;
 209         u32 len;
 210
 211         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 212                 goto inhdr_error;
 213
 214         iph = ip_hdr(skb);
 215
 216         /* Basic sanity checks */
 217         if (iph->ihl < 5 || iph->version != 4)
 218                 goto inhdr_error;
 219
 220         if (!pskb_may_pull(skb, iph->ihl*4))
 221                 goto inhdr_error;
 222
 223         iph = ip_hdr(skb);
 224         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 225                 goto inhdr_error;
 226
 227         len = ntohs(iph->tot_len);
 228         if (skb->len < len) {
 229                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
 230                 goto drop;
 231         } else if (len < (iph->ihl*4))
 232                 goto inhdr_error;
 233
 234         if (pskb_trim_rcsum(skb, len)) {
 235                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 236                 goto drop;
 237         }
 238
 239         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 240         /* We should really parse IP options here but until
 241          * somebody who actually uses IP options complains to
 242          * us we'll just silently ignore the options because
 243          * we're lazy!
 244          */
 245         return 0;
 246
 247 inhdr_error:
 248         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 249 drop:
 250         return -1;
 251 }
 252
 253 static void nf_bridge_update_protocol(struct sk_buff *skb)
 254 {
 255         if (skb->nf_bridge->mask & BRNF_8021Q)
 256                 skb->protocol = htons(ETH_P_8021Q);
 257         else if (skb->nf_bridge->mask & BRNF_PPPoE)
 258                 skb->protocol = htons(ETH_P_PPP_SES);
 259 }
 260
 261 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 262 /* Undo the changes made for ip6tables PREROUTING and continue the
 263  * bridge PRE_ROUTING hook. */
 264 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 265 {
 266         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 267         struct rtable *rt;
 268
 269         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 270                 skb->pkt_type = PACKET_OTHERHOST;
 271                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 272         }
 273         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 274
 275         rt = bridge_parent_rtable(nf_bridge->physindev);
 276         if (!rt) {
 277                 kfree_skb(skb);
 278                 return 0;
 279         }
 280         skb_dst_set_noref(skb, &rt->dst);
 281
 282         skb->dev = nf_bridge->physindev;
 283         nf_bridge_update_protocol(skb);
 284         nf_bridge_push_encap_header(skb);
 285         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 286                        br_handle_frame_finish, 1);
 287
 288         return 0;
 289 }
 290
 291 /* Obtain the correct destination MAC address, while preserving the original
 292  * source MAC address. If we already know this address, we just copy it. If we
 293  * don't, we use the neighbour framework to find out. In both cases, we make
 294  * sure that br_handle_frame_finish() is called afterwards.
 295  */
 296 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 297 {
 298         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 299         struct neighbour *neigh;
 300         struct dst_entry *dst;
 301
 302         skb->dev = bridge_parent(skb->dev);
 303         if (!skb->dev)
 304                 goto free_skb;
 305         dst = skb_dst(skb);
 306         neigh = dst_neigh_lookup_skb(dst, skb);
 307         if (neigh) {
 308                 int ret;
 309
 310                 if (neigh->hh.hh_len) {
 311                         neigh_hh_bridge(&neigh->hh, skb);
 312                         skb->dev = nf_bridge->physindev;
 313                         ret = br_handle_frame_finish(skb);
 314                 } else {
 315                         /* the neighbour function below overwrites the complete
 316                          * MAC header, so we save the Ethernet source address and
 317                          * protocol number.
 318                          */
 319                         skb_copy_from_linear_data_offset(skb,
 320                                                          -(ETH_HLEN-ETH_ALEN),
 321                                                          skb->nf_bridge->data,
 322                                                          ETH_HLEN-ETH_ALEN);
 323                         /* tell br_dev_xmit to continue with forwarding */
 324                         nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 325                         /* FIXME Need to refragment */
 326                         ret = neigh->output(neigh, skb);
 327                 }
 328                 neigh_release(neigh);
 329                 return ret;
 330         }
 331 free_skb:
 332         kfree_skb(skb);
 333         return 0;
 334 }
 335
 336 static bool dnat_took_place(const struct sk_buff *skb)
 337 {
 338 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 339         enum ip_conntrack_info ctinfo;
 340         struct nf_conn *ct;
 341
 342         ct = nf_ct_get(skb, &ctinfo);
 343         if (!ct || nf_ct_is_untracked(ct))
 344                 return false;
 345
 346         return test_bit(IPS_DST_NAT_BIT, &ct->status);
 347 #else
 348         return false;
 349 #endif
 350 }
 351
 352 /* This requires some explaining. If DNAT has taken place,
 353  * we will need to fix up the destination Ethernet address.
 354  *
 355  * There are two cases to consider:
 356  * 1. The packet was DNAT'ed to a device in the same bridge
 357  *    port group as it was received on. We can still bridge
 358  *    the packet.
 359  * 2. The packet was DNAT'ed to a different device, either
 360  *    a non-bridged device or another bridge port group.
 361  *    The packet will need to be routed.
 362  *
 363  * The correct way of distinguishing between these two cases is to
 364  * call ip_route_input() and to look at skb->dst->dev, which is
 365  * changed to the destination device if ip_route_input() succeeds.
 366  *
 367  * Let's first consider the case that ip_route_input() succeeds:
 368  *
 369  * If the output device equals the logical bridge device the packet
 370  * came in on, we can consider this bridging. The corresponding MAC
 371  * address will be obtained in br_nf_pre_routing_finish_bridge.
 372  * Otherwise, the packet is considered to be routed and we just
 373  * change the destination MAC address so that the packet will
 374  * later be passed up to the IP stack to be routed. For a redirected
 375  * packet, ip_route_input() will give back the localhost as output device,
 376  * which differs from the bridge device.
 377  *
 378  * Let's now consider the case that ip_route_input() fails:
 379  *
 380  * This can be because the destination address is martian, in which case
 381  * the packet will be dropped.
 382  * If IP forwarding is disabled, ip_route_input() will fail, while
 383  * ip_route_output_key() can return success. The source
 384  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 385  * thinks we're handling a locally generated packet and won't care
 386  * if IP forwarding is enabled. If the output device equals the logical bridge
 387  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 388  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 389  */
 390 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 391 {
 392         struct net_device *dev = skb->dev;
 393         struct iphdr *iph = ip_hdr(skb);
 394         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 395         struct rtable *rt;
 396         int err;
 397         int frag_max_size;
 398
 399         frag_max_size = IPCB(skb)->frag_max_size;
 400         BR_INPUT_SKB_CB(skb)->frag_max_size = frag_max_size;
 401
 402         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 403                 skb->pkt_type = PACKET_OTHERHOST;
 404                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 405         }
 406         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 407         if (dnat_took_place(skb)) {
 408                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 409                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 410
 411                         /* If err equals -EHOSTUNREACH the error is due to a
 412                          * martian destination or due to the fact that
 413                          * forwarding is disabled. For most martian packets,
 414                          * ip_route_output_key() will fail. It won't fail for 2 types of
 415                          * martian destinations: loopback destinations and destination
 416                          * 0.0.0.0. In both cases the packet will be dropped because the
 417                          * destination is the loopback device and not the bridge. */
 418                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 419                                 goto free_skb;
 420
 421                         rt = ip_route_output(dev_net(dev), iph->daddr, 0,
 422                                              RT_TOS(iph->tos), 0);
 423                         if (!IS_ERR(rt)) {
 424                                 /* - Bridged-and-DNAT'ed traffic doesn't
 425                                  *   require ip_forwarding. */
 426                                 if (rt->dst.dev == dev) {
 427                                         skb_dst_set(skb, &rt->dst);
 428                                         goto bridged_dnat;
 429                                 }
 430                                 ip_rt_put(rt);
 431                         }
 432 free_skb:
 433                         kfree_skb(skb);
 434                         return 0;
 435                 } else {
 436                         if (skb_dst(skb)->dev == dev) {
 437 bridged_dnat:
 438                                 skb->dev = nf_bridge->physindev;
 439                                 nf_bridge_update_protocol(skb);
 440                                 nf_bridge_push_encap_header(skb);
 441                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
 442                                                NF_BR_PRE_ROUTING,
 443                                                skb, skb->dev, NULL,
 444                                                br_nf_pre_routing_finish_bridge,
 445                                                1);
 446                                 return 0;
 447                         }
 448                         ether_addr_copy(eth_hdr(skb)->h_dest, dev->dev_addr);
 449                         skb->pkt_type = PACKET_HOST;
 450                 }
 451         } else {
 452                 rt = bridge_parent_rtable(nf_bridge->physindev);
 453                 if (!rt) {
 454                         kfree_skb(skb);
 455                         return 0;
 456                 }
 457                 skb_dst_set_noref(skb, &rt->dst);
 458         }
 459
 460         skb->dev = nf_bridge->physindev;
 461         nf_bridge_update_protocol(skb);
 462         nf_bridge_push_encap_header(skb);
 463         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 464                        br_handle_frame_finish, 1);
 465
 466         return 0;
 467 }
 468
 469 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb, const struct net_device *dev)
 470 {
 471         struct net_device *vlan, *br;
 472
 473         br = bridge_parent(dev);
 474         if (brnf_pass_vlan_indev == 0 || !skb_vlan_tag_present(skb))
 475                 return br;
 476
 477         vlan = __vlan_find_dev_deep_rcu(br, skb->vlan_proto,
 478                                     skb_vlan_tag_get(skb) & VLAN_VID_MASK);
 479
 480         return vlan ? vlan : br;
 481 }
 482
 483 /* Some common code for IPv4/IPv6 */
 484 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 485 {
 486         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 487
 488         if (skb->pkt_type == PACKET_OTHERHOST) {
 489                 skb->pkt_type = PACKET_HOST;
 490                 nf_bridge->mask |= BRNF_PKT_TYPE;
 491         }
 492
 493         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 494         nf_bridge->physindev = skb->dev;
 495         skb->dev = brnf_get_logical_dev(skb, skb->dev);
 496         if (skb->protocol == htons(ETH_P_8021Q))
 497                 nf_bridge->mask |= BRNF_8021Q;
 498         else if (skb->protocol == htons(ETH_P_PPP_SES))
 499                 nf_bridge->mask |= BRNF_PPPoE;
 500
 501         /* Must drop socket now because of tproxy. */
 502         skb_orphan(skb);
 503         return skb->dev;
 504 }
 505
 506 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 507 static int check_hbh_len(struct sk_buff *skb)
 508 {
 509         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 510         u32 pkt_len;
 511         const unsigned char *nh = skb_network_header(skb);
 512         int off = raw - nh;
 513         int len = (raw[1] + 1) << 3;
 514
 515         if ((raw + len) - skb->data > skb_headlen(skb))
 516                 goto bad;
 517
 518         off += 2;
 519         len -= 2;
 520
 521         while (len > 0) {
 522                 int optlen = nh[off + 1] + 2;
 523
 524                 switch (nh[off]) {
 525                 case IPV6_TLV_PAD1:
 526                         optlen = 1;
 527                         break;
 528
 529                 case IPV6_TLV_PADN:
 530                         break;
 531
 532                 case IPV6_TLV_JUMBO:
 533                         if (nh[off + 1] != 4 || (off & 3) != 2)
 534                                 goto bad;
 535                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 536                         if (pkt_len <= IPV6_MAXPLEN ||
 537                             ipv6_hdr(skb)->payload_len)
 538                                 goto bad;
 539                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 540                                 goto bad;
 541                         if (pskb_trim_rcsum(skb,
 542                                             pkt_len + sizeof(struct ipv6hdr)))
 543                                 goto bad;
 544                         nh = skb_network_header(skb);
 545                         break;
 546                 default:
 547                         if (optlen > len)
 548                                 goto bad;
 549                         break;
 550                 }
 551                 off += optlen;
 552                 len -= optlen;
 553         }
 554         if (len == 0)
 555                 return 0;
 556 bad:
 557         return -1;
 558
 559 }
 560
 561 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 562  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 563 static unsigned int br_nf_pre_routing_ipv6(const struct nf_hook_ops *ops,
 564                                            struct sk_buff *skb,
 565                                            const struct nf_hook_state *state)
 566 {
 567         const struct ipv6hdr *hdr;
 568         u32 pkt_len;
 569
 570         if (skb->len < sizeof(struct ipv6hdr))
 571                 return NF_DROP;
 572
 573         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 574                 return NF_DROP;
 575
 576         hdr = ipv6_hdr(skb);
 577
 578         if (hdr->version != 6)
 579                 return NF_DROP;
 580
 581         pkt_len = ntohs(hdr->payload_len);
 582
 583         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 584                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 585                         return NF_DROP;
 586                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 587                         return NF_DROP;
 588         }
 589         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 590                 return NF_DROP;
 591
 592         nf_bridge_put(skb->nf_bridge);
 593         if (!nf_bridge_alloc(skb))
 594                 return NF_DROP;
 595         if (!setup_pre_routing(skb))
 596                 return NF_DROP;
 597
 598         skb->protocol = htons(ETH_P_IPV6);
 599         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 600                 br_nf_pre_routing_finish_ipv6);
 601
 602         return NF_STOLEN;
 603 }
 604
 605 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 606  * Replicate the checks that IPv4 does on packet reception.
 607  * Set skb->dev to the bridge device (i.e. parent of the
 608  * receiving device) to make netfilter happy, the REDIRECT
 609  * target in particular.  Save the original destination IP
 610  * address to be able to detect DNAT afterwards. */
 611 static unsigned int br_nf_pre_routing(const struct nf_hook_ops *ops,
 612                                       struct sk_buff *skb,
 613                                       const struct nf_hook_state *state)
 614 {
 615         struct net_bridge_port *p;
 616         struct net_bridge *br;
 617         __u32 len = nf_bridge_encap_header_len(skb);
 618
 619         if (unlikely(!pskb_may_pull(skb, len)))
 620                 return NF_DROP;
 621
 622         p = br_port_get_rcu(state->in);
 623         if (p == NULL)
 624                 return NF_DROP;
 625         br = p->br;
 626
 627         if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb)) {
 628                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
 629                         return NF_ACCEPT;
 630
 631                 nf_bridge_pull_encap_header_rcsum(skb);
 632                 return br_nf_pre_routing_ipv6(ops, skb, state);
 633         }
 634
 635         if (!brnf_call_iptables && !br->nf_call_iptables)
 636                 return NF_ACCEPT;
 637
 638         if (!IS_IP(skb) && !IS_VLAN_IP(skb) && !IS_PPPOE_IP(skb))
 639                 return NF_ACCEPT;
 640
 641         nf_bridge_pull_encap_header_rcsum(skb);
 642
 643         if (br_parse_ip_options(skb))
 644                 return NF_DROP;
 645
 646         nf_bridge_put(skb->nf_bridge);
 647         if (!nf_bridge_alloc(skb))
 648                 return NF_DROP;
 649         if (!setup_pre_routing(skb))
 650                 return NF_DROP;
 651
 652         skb->protocol = htons(ETH_P_IP);
 653
 654         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 655                 br_nf_pre_routing_finish);
 656
 657         return NF_STOLEN;
 658 }
 659
 660
 661 /* PF_BRIDGE/LOCAL_IN ************************************************/
 662 /* The packet is locally destined, which requires a real
 663  * dst_entry, so detach the fake one.  On the way up, the
 664  * packet would pass through PRE_ROUTING again (which already
 665  * took place when the packet entered the bridge), but we
 666  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 667  * prevent this from happening. */
 668 static unsigned int br_nf_local_in(const struct nf_hook_ops *ops,
 669                                    struct sk_buff *skb,
 670                                    const struct nf_hook_state *state)
 671 {
 672         br_drop_fake_rtable(skb);
 673         return NF_ACCEPT;
 674 }
 675
 676 /* PF_BRIDGE/FORWARD *************************************************/
 677 static int br_nf_forward_finish(struct sk_buff *skb)
 678 {
 679         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 680         struct net_device *in;
 681
 682         if (!IS_ARP(skb) && !IS_VLAN_ARP(skb)) {
 683                 in = nf_bridge->physindev;
 684                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 685                         skb->pkt_type = PACKET_OTHERHOST;
 686                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 687                 }
 688                 nf_bridge_update_protocol(skb);
 689         } else {
 690                 in = *((struct net_device **)(skb->cb));
 691         }
 692         nf_bridge_push_encap_header(skb);
 693
 694         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
 695                        skb->dev, br_forward_finish, 1);
 696         return 0;
 697 }
 698
 699
 700 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 701  * netfilter with indev and outdev set to the bridge device,
 702  * but we are still able to filter on the 'real' indev/outdev
 703  * because of the physdev module. For ARP, indev and outdev are the
 704  * bridge ports. */
 705 static unsigned int br_nf_forward_ip(const struct nf_hook_ops *ops,
 706                                      struct sk_buff *skb,
 707                                      const struct nf_hook_state *state)
 708 {
 709         struct nf_bridge_info *nf_bridge;
 710         struct net_device *parent;
 711         u_int8_t pf;
 712
 713         if (!skb->nf_bridge)
 714                 return NF_ACCEPT;
 715
 716         /* Need exclusive nf_bridge_info since we might have multiple
 717          * different physoutdevs. */
 718         if (!nf_bridge_unshare(skb))
 719                 return NF_DROP;
 720
 721         parent = bridge_parent(state->out);
 722         if (!parent)
 723                 return NF_DROP;
 724
 725         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 726                 pf = NFPROTO_IPV4;
 727         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 728                 pf = NFPROTO_IPV6;
 729         else
 730                 return NF_ACCEPT;
 731
 732         nf_bridge_pull_encap_header(skb);
 733
 734         nf_bridge = skb->nf_bridge;
 735         if (skb->pkt_type == PACKET_OTHERHOST) {
 736                 skb->pkt_type = PACKET_HOST;
 737                 nf_bridge->mask |= BRNF_PKT_TYPE;
 738         }
 739
 740         if (pf == NFPROTO_IPV4 && br_parse_ip_options(skb))
 741                 return NF_DROP;
 742
 743         nf_bridge->physoutdev = skb->dev;
 744         if (pf == NFPROTO_IPV4)
 745                 skb->protocol = htons(ETH_P_IP);
 746         else
 747                 skb->protocol = htons(ETH_P_IPV6);
 748
 749         NF_HOOK(pf, NF_INET_FORWARD, skb, brnf_get_logical_dev(skb, state->in),
 750                 parent, br_nf_forward_finish);
 751
 752         return NF_STOLEN;
 753 }
 754
 755 static unsigned int br_nf_forward_arp(const struct nf_hook_ops *ops,
 756                                       struct sk_buff *skb,
 757                                       const struct nf_hook_state *state)
 758 {
 759         struct net_bridge_port *p;
 760         struct net_bridge *br;
 761         struct net_device **d = (struct net_device **)(skb->cb);
 762
 763         p = br_port_get_rcu(state->out);
 764         if (p == NULL)
 765                 return NF_ACCEPT;
 766         br = p->br;
 767
 768         if (!brnf_call_arptables && !br->nf_call_arptables)
 769                 return NF_ACCEPT;
 770
 771         if (!IS_ARP(skb)) {
 772                 if (!IS_VLAN_ARP(skb))
 773                         return NF_ACCEPT;
 774                 nf_bridge_pull_encap_header(skb);
 775         }
 776
 777         if (arp_hdr(skb)->ar_pln != 4) {
 778                 if (IS_VLAN_ARP(skb))
 779                         nf_bridge_push_encap_header(skb);
 780                 return NF_ACCEPT;
 781         }
 782         *d = state->in;
 783         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, state->in,
 784                 state->out, br_nf_forward_finish);
 785
 786         return NF_STOLEN;
 787 }
 788
 789 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4)
 790 static bool nf_bridge_copy_header(struct sk_buff *skb)
 791 {
 792         int err;
 793         unsigned int header_size;
 794
 795         nf_bridge_update_protocol(skb);
 796         header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 797         err = skb_cow_head(skb, header_size);
 798         if (err)
 799                 return false;
 800
 801         skb_copy_to_linear_data_offset(skb, -header_size,
 802                                        skb->nf_bridge->data, header_size);
 803         __skb_push(skb, nf_bridge_encap_header_len(skb));
 804         return true;
 805 }
 806
 807 static int br_nf_push_frag_xmit(struct sk_buff *skb)
 808 {
 809         if (!nf_bridge_copy_header(skb)) {
 810                 kfree_skb(skb);
 811                 return 0;
 812         }
 813
 814         return br_dev_queue_push_xmit(skb);
 815 }
 816
 817 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 818 {
 819         int ret;
 820         int frag_max_size;
 821         unsigned int mtu_reserved;
 822
 823         if (skb_is_gso(skb) || skb->protocol != htons(ETH_P_IP))
 824                 return br_dev_queue_push_xmit(skb);
 825
 826         mtu_reserved = nf_bridge_mtu_reduction(skb);
 827         /* This is wrong! We should preserve the original fragment
 828          * boundaries by preserving frag_list rather than refragmenting.
 829          */
 830         if (skb->len + mtu_reserved > skb->dev->mtu) {
 831                 frag_max_size = BR_INPUT_SKB_CB(skb)->frag_max_size;
 832                 if (br_parse_ip_options(skb))
 833                         /* Drop invalid packet */
 834                         return NF_DROP;
 835                 IPCB(skb)->frag_max_size = frag_max_size;
 836                 ret = ip_fragment(skb, br_nf_push_frag_xmit);
 837         } else
 838                 ret = br_dev_queue_push_xmit(skb);
 839
 840         return ret;
 841 }
 842 #else
 843 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 844 {
 845         return br_dev_queue_push_xmit(skb);
 846 }
 847 #endif
 848
 849 /* PF_BRIDGE/POST_ROUTING ********************************************/
 850 static unsigned int br_nf_post_routing(const struct nf_hook_ops *ops,
 851                                        struct sk_buff *skb,
 852                                        const struct nf_hook_state *state)
 853 {
 854         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 855         struct net_device *realoutdev = bridge_parent(skb->dev);
 856         u_int8_t pf;
 857
 858         /* if nf_bridge is set, but ->physoutdev is NULL, this packet came in
 859          * on a bridge, but was delivered locally and is now being routed:
 860          *
 861          * POST_ROUTING was already invoked from the ip stack.
 862          */
 863         if (!nf_bridge || !nf_bridge->physoutdev)
 864                 return NF_ACCEPT;
 865
 866         if (!realoutdev)
 867                 return NF_DROP;
 868
 869         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 870                 pf = NFPROTO_IPV4;
 871         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 872                 pf = NFPROTO_IPV6;
 873         else
 874                 return NF_ACCEPT;
 875
 876         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 877          * about the value of skb->pkt_type. */
 878         if (skb->pkt_type == PACKET_OTHERHOST) {
 879                 skb->pkt_type = PACKET_HOST;
 880                 nf_bridge->mask |= BRNF_PKT_TYPE;
 881         }
 882
 883         nf_bridge_pull_encap_header(skb);
 884         nf_bridge_save_header(skb);
 885         if (pf == NFPROTO_IPV4)
 886                 skb->protocol = htons(ETH_P_IP);
 887         else
 888                 skb->protocol = htons(ETH_P_IPV6);
 889
 890         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
 891                 br_nf_dev_queue_xmit);
 892
 893         return NF_STOLEN;
 894 }
 895
 896 /* IP/SABOTAGE *****************************************************/
 897 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 898  * for the second time. */
 899 static unsigned int ip_sabotage_in(const struct nf_hook_ops *ops,
 900                                    struct sk_buff *skb,
 901                                    const struct nf_hook_state *state)
 902 {
 903         if (skb->nf_bridge &&
 904             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 905                 return NF_STOP;
 906         }
 907
 908         return NF_ACCEPT;
 909 }
 910
 911 /* This is called when br_netfilter has called into iptables/netfilter,
 912  * and DNAT has taken place on a bridge-forwarded packet.
 913  *
 914  * neigh->output has created a new MAC header, with local br0 MAC
 915  * as saddr.
 916  *
 917  * This restores the original MAC saddr of the bridged packet
 918  * before invoking bridge forward logic to transmit the packet.
 919  */
 920 static void br_nf_pre_routing_finish_bridge_slow(struct sk_buff *skb)
 921 {
 922         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 923
 924         skb_pull(skb, ETH_HLEN);
 925         nf_bridge->mask &= ~BRNF_BRIDGED_DNAT;
 926
 927         skb_copy_to_linear_data_offset(skb, -(ETH_HLEN-ETH_ALEN),
 928                                        skb->nf_bridge->data, ETH_HLEN-ETH_ALEN);
 929         skb->dev = nf_bridge->physindev;
 930         br_handle_frame_finish(skb);
 931 }
 932
 933 static int br_nf_dev_xmit(struct sk_buff *skb)
 934 {
 935         if (skb->nf_bridge && (skb->nf_bridge->mask & BRNF_BRIDGED_DNAT)) {
 936                 br_nf_pre_routing_finish_bridge_slow(skb);
 937                 return 1;
 938         }
 939         return 0;
 940 }
 941
 942 static const struct nf_br_ops br_ops = {
 943         .br_dev_xmit_hook =     br_nf_dev_xmit,
 944 };
 945
 946 void br_netfilter_enable(void)
 947 {
 948 }
 949 EXPORT_SYMBOL_GPL(br_netfilter_enable);
 950
 951 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 952  * br_dev_queue_push_xmit is called afterwards */
 953 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
 954         {
 955                 .hook = br_nf_pre_routing,
 956                 .owner = THIS_MODULE,
 957                 .pf = NFPROTO_BRIDGE,
 958                 .hooknum = NF_BR_PRE_ROUTING,
 959                 .priority = NF_BR_PRI_BRNF,
 960         },
 961         {
 962                 .hook = br_nf_local_in,
 963                 .owner = THIS_MODULE,
 964                 .pf = NFPROTO_BRIDGE,
 965                 .hooknum = NF_BR_LOCAL_IN,
 966                 .priority = NF_BR_PRI_BRNF,
 967         },
 968         {
 969                 .hook = br_nf_forward_ip,
 970                 .owner = THIS_MODULE,
 971                 .pf = NFPROTO_BRIDGE,
 972                 .hooknum = NF_BR_FORWARD,
 973                 .priority = NF_BR_PRI_BRNF - 1,
 974         },
 975         {
 976                 .hook = br_nf_forward_arp,
 977                 .owner = THIS_MODULE,
 978                 .pf = NFPROTO_BRIDGE,
 979                 .hooknum = NF_BR_FORWARD,
 980                 .priority = NF_BR_PRI_BRNF,
 981         },
 982         {
 983                 .hook = br_nf_post_routing,
 984                 .owner = THIS_MODULE,
 985                 .pf = NFPROTO_BRIDGE,
 986                 .hooknum = NF_BR_POST_ROUTING,
 987                 .priority = NF_BR_PRI_LAST,
 988         },
 989         {
 990                 .hook = ip_sabotage_in,
 991                 .owner = THIS_MODULE,
 992                 .pf = NFPROTO_IPV4,
 993                 .hooknum = NF_INET_PRE_ROUTING,
 994                 .priority = NF_IP_PRI_FIRST,
 995         },
 996         {
 997                 .hook = ip_sabotage_in,
 998                 .owner = THIS_MODULE,
 999                 .pf = NFPROTO_IPV6,
1000                 .hooknum = NF_INET_PRE_ROUTING,
1001                 .priority = NF_IP6_PRI_FIRST,
1002         },
1003 };
1004
1005 #ifdef CONFIG_SYSCTL
1006 static
1007 int brnf_sysctl_call_tables(struct ctl_table *ctl, int write,
1008                             void __user *buffer, size_t *lenp, loff_t *ppos)
1009 {
1010         int ret;
1011
1012         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1013
1014         if (write && *(int *)(ctl->data))
1015                 *(int *)(ctl->data) = 1;
1016         return ret;
1017 }
1018
1019 static struct ctl_table brnf_table[] = {
1020         {
1021                 .procname       = "bridge-nf-call-arptables",
1022                 .data           = &brnf_call_arptables,
1023                 .maxlen         = sizeof(int),
1024                 .mode           = 0644,
1025                 .proc_handler   = brnf_sysctl_call_tables,
1026         },
1027         {
1028                 .procname       = "bridge-nf-call-iptables",
1029                 .data           = &brnf_call_iptables,
1030                 .maxlen         = sizeof(int),
1031                 .mode           = 0644,
1032                 .proc_handler   = brnf_sysctl_call_tables,
1033         },
1034         {
1035                 .procname       = "bridge-nf-call-ip6tables",
1036                 .data           = &brnf_call_ip6tables,
1037                 .maxlen         = sizeof(int),
1038                 .mode           = 0644,
1039                 .proc_handler   = brnf_sysctl_call_tables,
1040         },
1041         {
1042                 .procname       = "bridge-nf-filter-vlan-tagged",
1043                 .data           = &brnf_filter_vlan_tagged,
1044                 .maxlen         = sizeof(int),
1045                 .mode           = 0644,
1046                 .proc_handler   = brnf_sysctl_call_tables,
1047         },
1048         {
1049                 .procname       = "bridge-nf-filter-pppoe-tagged",
1050                 .data           = &brnf_filter_pppoe_tagged,
1051                 .maxlen         = sizeof(int),
1052                 .mode           = 0644,
1053                 .proc_handler   = brnf_sysctl_call_tables,
1054         },
1055         {
1056                 .procname       = "bridge-nf-pass-vlan-input-dev",
1057                 .data           = &brnf_pass_vlan_indev,
1058                 .maxlen         = sizeof(int),
1059                 .mode           = 0644,
1060                 .proc_handler   = brnf_sysctl_call_tables,
1061         },
1062         { }
1063 };
1064 #endif
1065
1066 static int __init br_netfilter_init(void)
1067 {
1068         int ret;
1069
1070         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1071         if (ret < 0)
1072                 return ret;
1073
1074 #ifdef CONFIG_SYSCTL
1075         brnf_sysctl_header = register_net_sysctl(&init_net, "net/bridge", brnf_table);
1076         if (brnf_sysctl_header == NULL) {
1077                 printk(KERN_WARNING
1078                        "br_netfilter: can't register to sysctl.\n");
1079                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1080                 return -ENOMEM;
1081         }
1082 #endif
1083         RCU_INIT_POINTER(nf_br_ops, &br_ops);
1084         printk(KERN_NOTICE "Bridge firewalling registered\n");
1085         return 0;
1086 }
1087
1088 static void __exit br_netfilter_fini(void)
1089 {
1090         RCU_INIT_POINTER(nf_br_ops, NULL);
1091         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1092 #ifdef CONFIG_SYSCTL
1093         unregister_net_sysctl_table(brnf_sysctl_header);
1094 #endif
1095 }
1096
1097 module_init(br_netfilter_init);
1098 module_exit(br_netfilter_fini);
1099
1100 MODULE_LICENSE("GPL");
1101 MODULE_AUTHOR("Lennert Buytenhek <buytenh@gnu.org>");
1102 MODULE_AUTHOR("Bart De Schuymer <bdschuym@pandora.be>");
1103 MODULE_DESCRIPTION("Linux ethernet netfilter firewall bridge");