net/bridge/br_netfilter.c

   1 /*
   2  *      Handle firewalling
   3  *      Linux ethernet bridge
   4  *
   5  *      Authors:
   6  *      Lennert Buytenhek               <buytenh@gnu.org>
   7  *      Bart De Schuymer                <bdschuym@pandora.be>
   8  *
   9  *      This program is free software; you can redistribute it and/or
  10  *      modify it under the terms of the GNU General Public License
  11  *      as published by the Free Software Foundation; either version
  12  *      2 of the License, or (at your option) any later version.
  13  *
  14  *      Lennert dedicates this file to Kerstin Wurdinger.
  15  */
  16
  17 #include <linux/module.h>
  18 #include <linux/kernel.h>
  19 #include <linux/slab.h>
  20 #include <linux/ip.h>
  21 #include <linux/netdevice.h>
  22 #include <linux/skbuff.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_ether.h>
  25 #include <linux/if_vlan.h>
  26 #include <linux/if_pppox.h>
  27 #include <linux/ppp_defs.h>
  28 #include <linux/netfilter_bridge.h>
  29 #include <linux/netfilter_ipv4.h>
  30 #include <linux/netfilter_ipv6.h>
  31 #include <linux/netfilter_arp.h>
  32 #include <linux/in_route.h>
  33 #include <linux/inetdevice.h>
  34
  35 #include <net/ip.h>
  36 #include <net/ipv6.h>
  37 #include <net/route.h>
  38 #include <net/netfilter/br_netfilter.h>
  39
  40 #include <asm/uaccess.h>
  41 #include "br_private.h"
  42 #ifdef CONFIG_SYSCTL
  43 #include <linux/sysctl.h>
  44 #endif
  45
  46 #ifdef CONFIG_SYSCTL
  47 static struct ctl_table_header *brnf_sysctl_header;
  48 static int brnf_call_iptables __read_mostly = 1;
  49 static int brnf_call_ip6tables __read_mostly = 1;
  50 static int brnf_call_arptables __read_mostly = 1;
  51 static int brnf_filter_vlan_tagged __read_mostly = 0;
  52 static int brnf_filter_pppoe_tagged __read_mostly = 0;
  53 static int brnf_pass_vlan_indev __read_mostly = 0;
  54 #else
  55 #define brnf_call_iptables 1
  56 #define brnf_call_ip6tables 1
  57 #define brnf_call_arptables 1
  58 #define brnf_filter_vlan_tagged 0
  59 #define brnf_filter_pppoe_tagged 0
  60 #define brnf_pass_vlan_indev 0
  61 #endif
  62
  63 #define IS_IP(skb) \
  64         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IP))
  65
  66 #define IS_IPV6(skb) \
  67         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IPV6))
  68
  69 #define IS_ARP(skb) \
  70         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_ARP))
  71
  72 static inline __be16 vlan_proto(const struct sk_buff *skb)
  73 {
  74         if (skb_vlan_tag_present(skb))
  75                 return skb->protocol;
  76         else if (skb->protocol == htons(ETH_P_8021Q))
  77                 return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  78         else
  79                 return 0;
  80 }
  81
  82 #define IS_VLAN_IP(skb) \
  83         (vlan_proto(skb) == htons(ETH_P_IP) && \
  84          brnf_filter_vlan_tagged)
  85
  86 #define IS_VLAN_IPV6(skb) \
  87         (vlan_proto(skb) == htons(ETH_P_IPV6) && \
  88          brnf_filter_vlan_tagged)
  89
  90 #define IS_VLAN_ARP(skb) \
  91         (vlan_proto(skb) == htons(ETH_P_ARP) && \
  92          brnf_filter_vlan_tagged)
  93
  94 static inline __be16 pppoe_proto(const struct sk_buff *skb)
  95 {
  96         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
  97                             sizeof(struct pppoe_hdr)));
  98 }
  99
 100 #define IS_PPPOE_IP(skb) \
 101         (skb->protocol == htons(ETH_P_PPP_SES) && \
 102          pppoe_proto(skb) == htons(PPP_IP) && \
 103          brnf_filter_pppoe_tagged)
 104
 105 #define IS_PPPOE_IPV6(skb) \
 106         (skb->protocol == htons(ETH_P_PPP_SES) && \
 107          pppoe_proto(skb) == htons(PPP_IPV6) && \
 108          brnf_filter_pppoe_tagged)
 109
 110 /* largest possible L2 header, see br_nf_dev_queue_xmit() */
 111 #define NF_BRIDGE_MAX_MAC_HEADER_LENGTH (PPPOE_SES_HLEN + ETH_HLEN)
 112
 113 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4)
 114 struct brnf_frag_data {
 115         char mac[NF_BRIDGE_MAX_MAC_HEADER_LENGTH];
 116         u8 encap_size;
 117         u8 size;
 118 };
 119
 120 static DEFINE_PER_CPU(struct brnf_frag_data, brnf_frag_data_storage);
 121 #endif
 122
 123 static struct nf_bridge_info *nf_bridge_info_get(const struct sk_buff *skb)
 124 {
 125         return skb->nf_bridge;
 126 }
 127
 128 static void nf_bridge_info_free(struct sk_buff *skb)
 129 {
 130         if (skb->nf_bridge) {
 131                 nf_bridge_put(skb->nf_bridge);
 132                 skb->nf_bridge = NULL;
 133         }
 134 }
 135
 136 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
 137 {
 138         struct net_bridge_port *port;
 139
 140         port = br_port_get_rcu(dev);
 141         return port ? &port->br->fake_rtable : NULL;
 142 }
 143
 144 static inline struct net_device *bridge_parent(const struct net_device *dev)
 145 {
 146         struct net_bridge_port *port;
 147
 148         port = br_port_get_rcu(dev);
 149         return port ? port->br->dev : NULL;
 150 }
 151
 152 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 153 {
 154         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 155         if (likely(skb->nf_bridge))
 156                 atomic_set(&(skb->nf_bridge->use), 1);
 157
 158         return skb->nf_bridge;
 159 }
 160
 161 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 162 {
 163         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 164
 165         if (atomic_read(&nf_bridge->use) > 1) {
 166                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 167
 168                 if (tmp) {
 169                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 170                         atomic_set(&tmp->use, 1);
 171                 }
 172                 nf_bridge_put(nf_bridge);
 173                 nf_bridge = tmp;
 174         }
 175         return nf_bridge;
 176 }
 177
 178 static unsigned int nf_bridge_encap_header_len(const struct sk_buff *skb)
 179 {
 180         switch (skb->protocol) {
 181         case __cpu_to_be16(ETH_P_8021Q):
 182                 return VLAN_HLEN;
 183         case __cpu_to_be16(ETH_P_PPP_SES):
 184                 return PPPOE_SES_HLEN;
 185         default:
 186                 return 0;
 187         }
 188 }
 189
 190 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 191 {
 192         unsigned int len = nf_bridge_encap_header_len(skb);
 193
 194         skb_push(skb, len);
 195         skb->network_header -= len;
 196 }
 197
 198 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 199 {
 200         unsigned int len = nf_bridge_encap_header_len(skb);
 201
 202         skb_pull(skb, len);
 203         skb->network_header += len;
 204 }
 205
 206 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 207 {
 208         unsigned int len = nf_bridge_encap_header_len(skb);
 209
 210         skb_pull_rcsum(skb, len);
 211         skb->network_header += len;
 212 }
 213
 214 /* When handing a packet over to the IP layer
 215  * check whether we have a skb that is in the
 216  * expected format
 217  */
 218
 219 static int br_parse_ip_options(struct sk_buff *skb)
 220 {
 221         const struct iphdr *iph;
 222         struct net_device *dev = skb->dev;
 223         u32 len;
 224
 225         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 226                 goto inhdr_error;
 227
 228         iph = ip_hdr(skb);
 229
 230         /* Basic sanity checks */
 231         if (iph->ihl < 5 || iph->version != 4)
 232                 goto inhdr_error;
 233
 234         if (!pskb_may_pull(skb, iph->ihl*4))
 235                 goto inhdr_error;
 236
 237         iph = ip_hdr(skb);
 238         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 239                 goto inhdr_error;
 240
 241         len = ntohs(iph->tot_len);
 242         if (skb->len < len) {
 243                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
 244                 goto drop;
 245         } else if (len < (iph->ihl*4))
 246                 goto inhdr_error;
 247
 248         if (pskb_trim_rcsum(skb, len)) {
 249                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 250                 goto drop;
 251         }
 252
 253         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 254         /* We should really parse IP options here but until
 255          * somebody who actually uses IP options complains to
 256          * us we'll just silently ignore the options because
 257          * we're lazy!
 258          */
 259         return 0;
 260
 261 inhdr_error:
 262         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 263 drop:
 264         return -1;
 265 }
 266
 267 static void nf_bridge_update_protocol(struct sk_buff *skb)
 268 {
 269         switch (skb->nf_bridge->orig_proto) {
 270         case BRNF_PROTO_8021Q:
 271                 skb->protocol = htons(ETH_P_8021Q);
 272                 break;
 273         case BRNF_PROTO_PPPOE:
 274                 skb->protocol = htons(ETH_P_PPP_SES);
 275                 break;
 276         case BRNF_PROTO_UNCHANGED:
 277                 break;
 278         }
 279 }
 280
 281 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 282 /* Undo the changes made for ip6tables PREROUTING and continue the
 283  * bridge PRE_ROUTING hook. */
 284 static int br_nf_pre_routing_finish_ipv6(struct sock *sk, struct sk_buff *skb)
 285 {
 286         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 287         struct rtable *rt;
 288
 289         if (nf_bridge->pkt_otherhost) {
 290                 skb->pkt_type = PACKET_OTHERHOST;
 291                 nf_bridge->pkt_otherhost = false;
 292         }
 293         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 294
 295         rt = bridge_parent_rtable(nf_bridge->physindev);
 296         if (!rt) {
 297                 kfree_skb(skb);
 298                 return 0;
 299         }
 300         skb_dst_set_noref(skb, &rt->dst);
 301
 302         skb->dev = nf_bridge->physindev;
 303         nf_bridge_update_protocol(skb);
 304         nf_bridge_push_encap_header(skb);
 305         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, sk, skb,
 306                        skb->dev, NULL,
 307                        br_handle_frame_finish, 1);
 308
 309         return 0;
 310 }
 311
 312 /* Obtain the correct destination MAC address, while preserving the original
 313  * source MAC address. If we already know this address, we just copy it. If we
 314  * don't, we use the neighbour framework to find out. In both cases, we make
 315  * sure that br_handle_frame_finish() is called afterwards.
 316  */
 317 static int br_nf_pre_routing_finish_bridge(struct sock *sk, struct sk_buff *skb)
 318 {
 319         struct neighbour *neigh;
 320         struct dst_entry *dst;
 321
 322         skb->dev = bridge_parent(skb->dev);
 323         if (!skb->dev)
 324                 goto free_skb;
 325         dst = skb_dst(skb);
 326         neigh = dst_neigh_lookup_skb(dst, skb);
 327         if (neigh) {
 328                 struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 329                 int ret;
 330
 331                 if (neigh->hh.hh_len) {
 332                         neigh_hh_bridge(&neigh->hh, skb);
 333                         skb->dev = nf_bridge->physindev;
 334                         ret = br_handle_frame_finish(sk, skb);
 335                 } else {
 336                         /* the neighbour function below overwrites the complete
 337                          * MAC header, so we save the Ethernet source address and
 338                          * protocol number.
 339                          */
 340                         skb_copy_from_linear_data_offset(skb,
 341                                                          -(ETH_HLEN-ETH_ALEN),
 342                                                          nf_bridge->neigh_header,
 343                                                          ETH_HLEN-ETH_ALEN);
 344                         /* tell br_dev_xmit to continue with forwarding */
 345                         nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 346                         /* FIXME Need to refragment */
 347                         ret = neigh->output(neigh, skb);
 348                 }
 349                 neigh_release(neigh);
 350                 return ret;
 351         }
 352 free_skb:
 353         kfree_skb(skb);
 354         return 0;
 355 }
 356
 357 static bool daddr_was_changed(const struct sk_buff *skb,
 358                               const struct nf_bridge_info *nf_bridge)
 359 {
 360         return ip_hdr(skb)->daddr != nf_bridge->ipv4_daddr;
 361 }
 362
 363 /* This requires some explaining. If DNAT has taken place,
 364  * we will need to fix up the destination Ethernet address.
 365  * This is also true when SNAT takes place (for the reply direction).
 366  *
 367  * There are two cases to consider:
 368  * 1. The packet was DNAT'ed to a device in the same bridge
 369  *    port group as it was received on. We can still bridge
 370  *    the packet.
 371  * 2. The packet was DNAT'ed to a different device, either
 372  *    a non-bridged device or another bridge port group.
 373  *    The packet will need to be routed.
 374  *
 375  * The correct way of distinguishing between these two cases is to
 376  * call ip_route_input() and to look at skb->dst->dev, which is
 377  * changed to the destination device if ip_route_input() succeeds.
 378  *
 379  * Let's first consider the case that ip_route_input() succeeds:
 380  *
 381  * If the output device equals the logical bridge device the packet
 382  * came in on, we can consider this bridging. The corresponding MAC
 383  * address will be obtained in br_nf_pre_routing_finish_bridge.
 384  * Otherwise, the packet is considered to be routed and we just
 385  * change the destination MAC address so that the packet will
 386  * later be passed up to the IP stack to be routed. For a redirected
 387  * packet, ip_route_input() will give back the localhost as output device,
 388  * which differs from the bridge device.
 389  *
 390  * Let's now consider the case that ip_route_input() fails:
 391  *
 392  * This can be because the destination address is martian, in which case
 393  * the packet will be dropped.
 394  * If IP forwarding is disabled, ip_route_input() will fail, while
 395  * ip_route_output_key() can return success. The source
 396  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 397  * thinks we're handling a locally generated packet and won't care
 398  * if IP forwarding is enabled. If the output device equals the logical bridge
 399  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 400  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 401  */
 402 static int br_nf_pre_routing_finish(struct sock *sk, struct sk_buff *skb)
 403 {
 404         struct net_device *dev = skb->dev;
 405         struct iphdr *iph = ip_hdr(skb);
 406         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 407         struct rtable *rt;
 408         int err;
 409         int frag_max_size;
 410
 411         frag_max_size = IPCB(skb)->frag_max_size;
 412         BR_INPUT_SKB_CB(skb)->frag_max_size = frag_max_size;
 413
 414         if (nf_bridge->pkt_otherhost) {
 415                 skb->pkt_type = PACKET_OTHERHOST;
 416                 nf_bridge->pkt_otherhost = false;
 417         }
 418         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 419         if (daddr_was_changed(skb, nf_bridge)) {
 420                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 421                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 422
 423                         /* If err equals -EHOSTUNREACH the error is due to a
 424                          * martian destination or due to the fact that
 425                          * forwarding is disabled. For most martian packets,
 426                          * ip_route_output_key() will fail. It won't fail for 2 types of
 427                          * martian destinations: loopback destinations and destination
 428                          * 0.0.0.0. In both cases the packet will be dropped because the
 429                          * destination is the loopback device and not the bridge. */
 430                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 431                                 goto free_skb;
 432
 433                         rt = ip_route_output(dev_net(dev), iph->daddr, 0,
 434                                              RT_TOS(iph->tos), 0);
 435                         if (!IS_ERR(rt)) {
 436                                 /* - Bridged-and-DNAT'ed traffic doesn't
 437                                  *   require ip_forwarding. */
 438                                 if (rt->dst.dev == dev) {
 439                                         skb_dst_set(skb, &rt->dst);
 440                                         goto bridged_dnat;
 441                                 }
 442                                 ip_rt_put(rt);
 443                         }
 444 free_skb:
 445                         kfree_skb(skb);
 446                         return 0;
 447                 } else {
 448                         if (skb_dst(skb)->dev == dev) {
 449 bridged_dnat:
 450                                 skb->dev = nf_bridge->physindev;
 451                                 nf_bridge_update_protocol(skb);
 452                                 nf_bridge_push_encap_header(skb);
 453                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
 454                                                NF_BR_PRE_ROUTING,
 455                                                sk, skb, skb->dev, NULL,
 456                                                br_nf_pre_routing_finish_bridge,
 457                                                1);
 458                                 return 0;
 459                         }
 460                         ether_addr_copy(eth_hdr(skb)->h_dest, dev->dev_addr);
 461                         skb->pkt_type = PACKET_HOST;
 462                 }
 463         } else {
 464                 rt = bridge_parent_rtable(nf_bridge->physindev);
 465                 if (!rt) {
 466                         kfree_skb(skb);
 467                         return 0;
 468                 }
 469                 skb_dst_set_noref(skb, &rt->dst);
 470         }
 471
 472         skb->dev = nf_bridge->physindev;
 473         nf_bridge_update_protocol(skb);
 474         nf_bridge_push_encap_header(skb);
 475         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, sk, skb,
 476                        skb->dev, NULL,
 477                        br_handle_frame_finish, 1);
 478
 479         return 0;
 480 }
 481
 482 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb, const struct net_device *dev)
 483 {
 484         struct net_device *vlan, *br;
 485
 486         br = bridge_parent(dev);
 487         if (brnf_pass_vlan_indev == 0 || !skb_vlan_tag_present(skb))
 488                 return br;
 489
 490         vlan = __vlan_find_dev_deep_rcu(br, skb->vlan_proto,
 491                                     skb_vlan_tag_get(skb) & VLAN_VID_MASK);
 492
 493         return vlan ? vlan : br;
 494 }
 495
 496 /* Some common code for IPv4/IPv6 */
 497 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 498 {
 499         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 500
 501         if (skb->pkt_type == PACKET_OTHERHOST) {
 502                 skb->pkt_type = PACKET_HOST;
 503                 nf_bridge->pkt_otherhost = true;
 504         }
 505
 506         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 507         nf_bridge->physindev = skb->dev;
 508         skb->dev = brnf_get_logical_dev(skb, skb->dev);
 509
 510         if (skb->protocol == htons(ETH_P_8021Q))
 511                 nf_bridge->orig_proto = BRNF_PROTO_8021Q;
 512         else if (skb->protocol == htons(ETH_P_PPP_SES))
 513                 nf_bridge->orig_proto = BRNF_PROTO_PPPOE;
 514
 515         /* Must drop socket now because of tproxy. */
 516         skb_orphan(skb);
 517         return skb->dev;
 518 }
 519
 520 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 521 static int check_hbh_len(struct sk_buff *skb)
 522 {
 523         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 524         u32 pkt_len;
 525         const unsigned char *nh = skb_network_header(skb);
 526         int off = raw - nh;
 527         int len = (raw[1] + 1) << 3;
 528
 529         if ((raw + len) - skb->data > skb_headlen(skb))
 530                 goto bad;
 531
 532         off += 2;
 533         len -= 2;
 534
 535         while (len > 0) {
 536                 int optlen = nh[off + 1] + 2;
 537
 538                 switch (nh[off]) {
 539                 case IPV6_TLV_PAD1:
 540                         optlen = 1;
 541                         break;
 542
 543                 case IPV6_TLV_PADN:
 544                         break;
 545
 546                 case IPV6_TLV_JUMBO:
 547                         if (nh[off + 1] != 4 || (off & 3) != 2)
 548                                 goto bad;
 549                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 550                         if (pkt_len <= IPV6_MAXPLEN ||
 551                             ipv6_hdr(skb)->payload_len)
 552                                 goto bad;
 553                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 554                                 goto bad;
 555                         if (pskb_trim_rcsum(skb,
 556                                             pkt_len + sizeof(struct ipv6hdr)))
 557                                 goto bad;
 558                         nh = skb_network_header(skb);
 559                         break;
 560                 default:
 561                         if (optlen > len)
 562                                 goto bad;
 563                         break;
 564                 }
 565                 off += optlen;
 566                 len -= optlen;
 567         }
 568         if (len == 0)
 569                 return 0;
 570 bad:
 571         return -1;
 572
 573 }
 574
 575 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 576  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 577 static unsigned int br_nf_pre_routing_ipv6(const struct nf_hook_ops *ops,
 578                                            struct sk_buff *skb,
 579                                            const struct nf_hook_state *state)
 580 {
 581         const struct ipv6hdr *hdr;
 582         u32 pkt_len;
 583
 584         if (skb->len < sizeof(struct ipv6hdr))
 585                 return NF_DROP;
 586
 587         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 588                 return NF_DROP;
 589
 590         hdr = ipv6_hdr(skb);
 591
 592         if (hdr->version != 6)
 593                 return NF_DROP;
 594
 595         pkt_len = ntohs(hdr->payload_len);
 596
 597         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 598                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 599                         return NF_DROP;
 600                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 601                         return NF_DROP;
 602         }
 603         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 604                 return NF_DROP;
 605
 606         nf_bridge_put(skb->nf_bridge);
 607         if (!nf_bridge_alloc(skb))
 608                 return NF_DROP;
 609         if (!setup_pre_routing(skb))
 610                 return NF_DROP;
 611
 612         skb->protocol = htons(ETH_P_IPV6);
 613         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, state->sk, skb,
 614                 skb->dev, NULL,
 615                 br_nf_pre_routing_finish_ipv6);
 616
 617         return NF_STOLEN;
 618 }
 619
 620 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 621  * Replicate the checks that IPv4 does on packet reception.
 622  * Set skb->dev to the bridge device (i.e. parent of the
 623  * receiving device) to make netfilter happy, the REDIRECT
 624  * target in particular.  Save the original destination IP
 625  * address to be able to detect DNAT afterwards. */
 626 static unsigned int br_nf_pre_routing(const struct nf_hook_ops *ops,
 627                                       struct sk_buff *skb,
 628                                       const struct nf_hook_state *state)
 629 {
 630         struct nf_bridge_info *nf_bridge;
 631         struct net_bridge_port *p;
 632         struct net_bridge *br;
 633         __u32 len = nf_bridge_encap_header_len(skb);
 634
 635         if (unlikely(!pskb_may_pull(skb, len)))
 636                 return NF_DROP;
 637
 638         p = br_port_get_rcu(state->in);
 639         if (p == NULL)
 640                 return NF_DROP;
 641         br = p->br;
 642
 643         if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb)) {
 644                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
 645                         return NF_ACCEPT;
 646
 647                 nf_bridge_pull_encap_header_rcsum(skb);
 648                 return br_nf_pre_routing_ipv6(ops, skb, state);
 649         }
 650
 651         if (!brnf_call_iptables && !br->nf_call_iptables)
 652                 return NF_ACCEPT;
 653
 654         if (!IS_IP(skb) && !IS_VLAN_IP(skb) && !IS_PPPOE_IP(skb))
 655                 return NF_ACCEPT;
 656
 657         nf_bridge_pull_encap_header_rcsum(skb);
 658
 659         if (br_parse_ip_options(skb))
 660                 return NF_DROP;
 661
 662         nf_bridge_put(skb->nf_bridge);
 663         if (!nf_bridge_alloc(skb))
 664                 return NF_DROP;
 665         if (!setup_pre_routing(skb))
 666                 return NF_DROP;
 667
 668         nf_bridge = nf_bridge_info_get(skb);
 669         nf_bridge->ipv4_daddr = ip_hdr(skb)->daddr;
 670
 671         skb->protocol = htons(ETH_P_IP);
 672
 673         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, state->sk, skb,
 674                 skb->dev, NULL,
 675                 br_nf_pre_routing_finish);
 676
 677         return NF_STOLEN;
 678 }
 679
 680
 681 /* PF_BRIDGE/LOCAL_IN ************************************************/
 682 /* The packet is locally destined, which requires a real
 683  * dst_entry, so detach the fake one.  On the way up, the
 684  * packet would pass through PRE_ROUTING again (which already
 685  * took place when the packet entered the bridge), but we
 686  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 687  * prevent this from happening. */
 688 static unsigned int br_nf_local_in(const struct nf_hook_ops *ops,
 689                                    struct sk_buff *skb,
 690                                    const struct nf_hook_state *state)
 691 {
 692         br_drop_fake_rtable(skb);
 693         return NF_ACCEPT;
 694 }
 695
 696 /* PF_BRIDGE/FORWARD *************************************************/
 697 static int br_nf_forward_finish(struct sock *sk, struct sk_buff *skb)
 698 {
 699         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 700         struct net_device *in;
 701
 702         if (!IS_ARP(skb) && !IS_VLAN_ARP(skb)) {
 703                 int frag_max_size;
 704
 705                 if (skb->protocol == htons(ETH_P_IP)) {
 706                         frag_max_size = IPCB(skb)->frag_max_size;
 707                         BR_INPUT_SKB_CB(skb)->frag_max_size = frag_max_size;
 708                 }
 709
 710                 in = nf_bridge->physindev;
 711                 if (nf_bridge->pkt_otherhost) {
 712                         skb->pkt_type = PACKET_OTHERHOST;
 713                         nf_bridge->pkt_otherhost = false;
 714                 }
 715                 nf_bridge_update_protocol(skb);
 716         } else {
 717                 in = *((struct net_device **)(skb->cb));
 718         }
 719         nf_bridge_push_encap_header(skb);
 720
 721         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, sk, skb,
 722                        in, skb->dev, br_forward_finish, 1);
 723         return 0;
 724 }
 725
 726
 727 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 728  * netfilter with indev and outdev set to the bridge device,
 729  * but we are still able to filter on the 'real' indev/outdev
 730  * because of the physdev module. For ARP, indev and outdev are the
 731  * bridge ports. */
 732 static unsigned int br_nf_forward_ip(const struct nf_hook_ops *ops,
 733                                      struct sk_buff *skb,
 734                                      const struct nf_hook_state *state)
 735 {
 736         struct nf_bridge_info *nf_bridge;
 737         struct net_device *parent;
 738         u_int8_t pf;
 739
 740         if (!skb->nf_bridge)
 741                 return NF_ACCEPT;
 742
 743         /* Need exclusive nf_bridge_info since we might have multiple
 744          * different physoutdevs. */
 745         if (!nf_bridge_unshare(skb))
 746                 return NF_DROP;
 747
 748         nf_bridge = nf_bridge_info_get(skb);
 749         if (!nf_bridge)
 750                 return NF_DROP;
 751
 752         parent = bridge_parent(state->out);
 753         if (!parent)
 754                 return NF_DROP;
 755
 756         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 757                 pf = NFPROTO_IPV4;
 758         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 759                 pf = NFPROTO_IPV6;
 760         else
 761                 return NF_ACCEPT;
 762
 763         nf_bridge_pull_encap_header(skb);
 764
 765         if (skb->pkt_type == PACKET_OTHERHOST) {
 766                 skb->pkt_type = PACKET_HOST;
 767                 nf_bridge->pkt_otherhost = true;
 768         }
 769
 770         if (pf == NFPROTO_IPV4) {
 771                 int frag_max = BR_INPUT_SKB_CB(skb)->frag_max_size;
 772
 773                 if (br_parse_ip_options(skb))
 774                         return NF_DROP;
 775
 776                 IPCB(skb)->frag_max_size = frag_max;
 777         }
 778
 779         nf_bridge->physoutdev = skb->dev;
 780         if (pf == NFPROTO_IPV4)
 781                 skb->protocol = htons(ETH_P_IP);
 782         else
 783                 skb->protocol = htons(ETH_P_IPV6);
 784
 785         NF_HOOK(pf, NF_INET_FORWARD, NULL, skb,
 786                 brnf_get_logical_dev(skb, state->in),
 787                 parent, br_nf_forward_finish);
 788
 789         return NF_STOLEN;
 790 }
 791
 792 static unsigned int br_nf_forward_arp(const struct nf_hook_ops *ops,
 793                                       struct sk_buff *skb,
 794                                       const struct nf_hook_state *state)
 795 {
 796         struct net_bridge_port *p;
 797         struct net_bridge *br;
 798         struct net_device **d = (struct net_device **)(skb->cb);
 799
 800         p = br_port_get_rcu(state->out);
 801         if (p == NULL)
 802                 return NF_ACCEPT;
 803         br = p->br;
 804
 805         if (!brnf_call_arptables && !br->nf_call_arptables)
 806                 return NF_ACCEPT;
 807
 808         if (!IS_ARP(skb)) {
 809                 if (!IS_VLAN_ARP(skb))
 810                         return NF_ACCEPT;
 811                 nf_bridge_pull_encap_header(skb);
 812         }
 813
 814         if (arp_hdr(skb)->ar_pln != 4) {
 815                 if (IS_VLAN_ARP(skb))
 816                         nf_bridge_push_encap_header(skb);
 817                 return NF_ACCEPT;
 818         }
 819         *d = state->in;
 820         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, state->sk, skb,
 821                 state->in, state->out, br_nf_forward_finish);
 822
 823         return NF_STOLEN;
 824 }
 825
 826 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4)
 827 static int br_nf_push_frag_xmit(struct sock *sk, struct sk_buff *skb)
 828 {
 829         struct brnf_frag_data *data;
 830         int err;
 831
 832         data = this_cpu_ptr(&brnf_frag_data_storage);
 833         err = skb_cow_head(skb, data->size);
 834
 835         if (err) {
 836                 kfree_skb(skb);
 837                 return 0;
 838         }
 839
 840         skb_copy_to_linear_data_offset(skb, -data->size, data->mac, data->size);
 841         __skb_push(skb, data->encap_size);
 842
 843         nf_bridge_info_free(skb);
 844         return br_dev_queue_push_xmit(sk, skb);
 845 }
 846
 847 static int br_nf_ip_fragment(struct sock *sk, struct sk_buff *skb,
 848                              int (*output)(struct sock *, struct sk_buff *))
 849 {
 850         unsigned int mtu = ip_skb_dst_mtu(skb);
 851         struct iphdr *iph = ip_hdr(skb);
 852         struct rtable *rt = skb_rtable(skb);
 853         struct net_device *dev = rt->dst.dev;
 854
 855         if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->ignore_df) ||
 856                      (IPCB(skb)->frag_max_size &&
 857                       IPCB(skb)->frag_max_size > mtu))) {
 858                 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
 859                 kfree_skb(skb);
 860                 return -EMSGSIZE;
 861         }
 862
 863         return ip_do_fragment(sk, skb, output);
 864 }
 865
 866 static int br_nf_dev_queue_xmit(struct sock *sk, struct sk_buff *skb)
 867 {
 868         int ret;
 869         int frag_max_size;
 870         unsigned int mtu_reserved;
 871
 872         if (skb_is_gso(skb) || skb->protocol != htons(ETH_P_IP)) {
 873                 nf_bridge_info_free(skb);
 874                 return br_dev_queue_push_xmit(sk, skb);
 875         }
 876
 877         mtu_reserved = nf_bridge_mtu_reduction(skb);
 878         /* This is wrong! We should preserve the original fragment
 879          * boundaries by preserving frag_list rather than refragmenting.
 880          */
 881         if (skb->len + mtu_reserved > skb->dev->mtu) {
 882                 struct brnf_frag_data *data;
 883
 884                 frag_max_size = BR_INPUT_SKB_CB(skb)->frag_max_size;
 885                 if (br_parse_ip_options(skb))
 886                         /* Drop invalid packet */
 887                         return NF_DROP;
 888                 IPCB(skb)->frag_max_size = frag_max_size;
 889
 890                 nf_bridge_update_protocol(skb);
 891
 892                 data = this_cpu_ptr(&brnf_frag_data_storage);
 893                 data->encap_size = nf_bridge_encap_header_len(skb);
 894                 data->size = ETH_HLEN + data->encap_size;
 895
 896                 skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
 897                                                  data->size);
 898
 899                 ret = br_nf_ip_fragment(sk, skb, br_nf_push_frag_xmit);
 900         } else {
 901                 nf_bridge_info_free(skb);
 902                 ret = br_dev_queue_push_xmit(sk, skb);
 903         }
 904
 905         return ret;
 906 }
 907 #else
 908 static int br_nf_dev_queue_xmit(struct sock *sk, struct sk_buff *skb)
 909 {
 910         nf_bridge_info_free(skb);
 911         return br_dev_queue_push_xmit(sk, skb);
 912 }
 913 #endif
 914
 915 /* PF_BRIDGE/POST_ROUTING ********************************************/
 916 static unsigned int br_nf_post_routing(const struct nf_hook_ops *ops,
 917                                        struct sk_buff *skb,
 918                                        const struct nf_hook_state *state)
 919 {
 920         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 921         struct net_device *realoutdev = bridge_parent(skb->dev);
 922         u_int8_t pf;
 923
 924         /* if nf_bridge is set, but ->physoutdev is NULL, this packet came in
 925          * on a bridge, but was delivered locally and is now being routed:
 926          *
 927          * POST_ROUTING was already invoked from the ip stack.
 928          */
 929         if (!nf_bridge || !nf_bridge->physoutdev)
 930                 return NF_ACCEPT;
 931
 932         if (!realoutdev)
 933                 return NF_DROP;
 934
 935         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 936                 pf = NFPROTO_IPV4;
 937         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 938                 pf = NFPROTO_IPV6;
 939         else
 940                 return NF_ACCEPT;
 941
 942         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 943          * about the value of skb->pkt_type. */
 944         if (skb->pkt_type == PACKET_OTHERHOST) {
 945                 skb->pkt_type = PACKET_HOST;
 946                 nf_bridge->pkt_otherhost = true;
 947         }
 948
 949         nf_bridge_pull_encap_header(skb);
 950         if (pf == NFPROTO_IPV4)
 951                 skb->protocol = htons(ETH_P_IP);
 952         else
 953                 skb->protocol = htons(ETH_P_IPV6);
 954
 955         NF_HOOK(pf, NF_INET_POST_ROUTING, state->sk, skb,
 956                 NULL, realoutdev,
 957                 br_nf_dev_queue_xmit);
 958
 959         return NF_STOLEN;
 960 }
 961
 962 /* IP/SABOTAGE *****************************************************/
 963 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 964  * for the second time. */
 965 static unsigned int ip_sabotage_in(const struct nf_hook_ops *ops,
 966                                    struct sk_buff *skb,
 967                                    const struct nf_hook_state *state)
 968 {
 969         if (skb->nf_bridge &&
 970             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 971                 return NF_STOP;
 972         }
 973
 974         return NF_ACCEPT;
 975 }
 976
 977 /* This is called when br_netfilter has called into iptables/netfilter,
 978  * and DNAT has taken place on a bridge-forwarded packet.
 979  *
 980  * neigh->output has created a new MAC header, with local br0 MAC
 981  * as saddr.
 982  *
 983  * This restores the original MAC saddr of the bridged packet
 984  * before invoking bridge forward logic to transmit the packet.
 985  */
 986 static void br_nf_pre_routing_finish_bridge_slow(struct sk_buff *skb)
 987 {
 988         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 989
 990         skb_pull(skb, ETH_HLEN);
 991         nf_bridge->mask &= ~BRNF_BRIDGED_DNAT;
 992
 993         BUILD_BUG_ON(sizeof(nf_bridge->neigh_header) != (ETH_HLEN - ETH_ALEN));
 994
 995         skb_copy_to_linear_data_offset(skb, -(ETH_HLEN - ETH_ALEN),
 996                                        nf_bridge->neigh_header,
 997                                        ETH_HLEN - ETH_ALEN);
 998         skb->dev = nf_bridge->physindev;
 999
1000         nf_bridge->physoutdev = NULL;
1001         br_handle_frame_finish(NULL, skb);
1002 }
1003
1004 static int br_nf_dev_xmit(struct sk_buff *skb)
1005 {
1006         if (skb->nf_bridge && (skb->nf_bridge->mask & BRNF_BRIDGED_DNAT)) {
1007                 br_nf_pre_routing_finish_bridge_slow(skb);
1008                 return 1;
1009         }
1010         return 0;
1011 }
1012
1013 static const struct nf_br_ops br_ops = {
1014         .br_dev_xmit_hook =     br_nf_dev_xmit,
1015 };
1016
1017 void br_netfilter_enable(void)
1018 {
1019 }
1020 EXPORT_SYMBOL_GPL(br_netfilter_enable);
1021
1022 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
1023  * br_dev_queue_push_xmit is called afterwards */
1024 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
1025         {
1026                 .hook = br_nf_pre_routing,
1027                 .owner = THIS_MODULE,
1028                 .pf = NFPROTO_BRIDGE,
1029                 .hooknum = NF_BR_PRE_ROUTING,
1030                 .priority = NF_BR_PRI_BRNF,
1031         },
1032         {
1033                 .hook = br_nf_local_in,
1034                 .owner = THIS_MODULE,
1035                 .pf = NFPROTO_BRIDGE,
1036                 .hooknum = NF_BR_LOCAL_IN,
1037                 .priority = NF_BR_PRI_BRNF,
1038         },
1039         {
1040                 .hook = br_nf_forward_ip,
1041                 .owner = THIS_MODULE,
1042                 .pf = NFPROTO_BRIDGE,
1043                 .hooknum = NF_BR_FORWARD,
1044                 .priority = NF_BR_PRI_BRNF - 1,
1045         },
1046         {
1047                 .hook = br_nf_forward_arp,
1048                 .owner = THIS_MODULE,
1049                 .pf = NFPROTO_BRIDGE,
1050                 .hooknum = NF_BR_FORWARD,
1051                 .priority = NF_BR_PRI_BRNF,
1052         },
1053         {
1054                 .hook = br_nf_post_routing,
1055                 .owner = THIS_MODULE,
1056                 .pf = NFPROTO_BRIDGE,
1057                 .hooknum = NF_BR_POST_ROUTING,
1058                 .priority = NF_BR_PRI_LAST,
1059         },
1060         {
1061                 .hook = ip_sabotage_in,
1062                 .owner = THIS_MODULE,
1063                 .pf = NFPROTO_IPV4,
1064                 .hooknum = NF_INET_PRE_ROUTING,
1065                 .priority = NF_IP_PRI_FIRST,
1066         },
1067         {
1068                 .hook = ip_sabotage_in,
1069                 .owner = THIS_MODULE,
1070                 .pf = NFPROTO_IPV6,
1071                 .hooknum = NF_INET_PRE_ROUTING,
1072                 .priority = NF_IP6_PRI_FIRST,
1073         },
1074 };
1075
1076 #ifdef CONFIG_SYSCTL
1077 static
1078 int brnf_sysctl_call_tables(struct ctl_table *ctl, int write,
1079                             void __user *buffer, size_t *lenp, loff_t *ppos)
1080 {
1081         int ret;
1082
1083         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1084
1085         if (write && *(int *)(ctl->data))
1086                 *(int *)(ctl->data) = 1;
1087         return ret;
1088 }
1089
1090 static struct ctl_table brnf_table[] = {
1091         {
1092                 .procname       = "bridge-nf-call-arptables",
1093                 .data           = &brnf_call_arptables,
1094                 .maxlen         = sizeof(int),
1095                 .mode           = 0644,
1096                 .proc_handler   = brnf_sysctl_call_tables,
1097         },
1098         {
1099                 .procname       = "bridge-nf-call-iptables",
1100                 .data           = &brnf_call_iptables,
1101                 .maxlen         = sizeof(int),
1102                 .mode           = 0644,
1103                 .proc_handler   = brnf_sysctl_call_tables,
1104         },
1105         {
1106                 .procname       = "bridge-nf-call-ip6tables",
1107                 .data           = &brnf_call_ip6tables,
1108                 .maxlen         = sizeof(int),
1109                 .mode           = 0644,
1110                 .proc_handler   = brnf_sysctl_call_tables,
1111         },
1112         {
1113                 .procname       = "bridge-nf-filter-vlan-tagged",
1114                 .data           = &brnf_filter_vlan_tagged,
1115                 .maxlen         = sizeof(int),
1116                 .mode           = 0644,
1117                 .proc_handler   = brnf_sysctl_call_tables,
1118         },
1119         {
1120                 .procname       = "bridge-nf-filter-pppoe-tagged",
1121                 .data           = &brnf_filter_pppoe_tagged,
1122                 .maxlen         = sizeof(int),
1123                 .mode           = 0644,
1124                 .proc_handler   = brnf_sysctl_call_tables,
1125         },
1126         {
1127                 .procname       = "bridge-nf-pass-vlan-input-dev",
1128                 .data           = &brnf_pass_vlan_indev,
1129                 .maxlen         = sizeof(int),
1130                 .mode           = 0644,
1131                 .proc_handler   = brnf_sysctl_call_tables,
1132         },
1133         { }
1134 };
1135 #endif
1136
1137 static int __init br_netfilter_init(void)
1138 {
1139         int ret;
1140
1141         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1142         if (ret < 0)
1143                 return ret;
1144
1145 #ifdef CONFIG_SYSCTL
1146         brnf_sysctl_header = register_net_sysctl(&init_net, "net/bridge", brnf_table);
1147         if (brnf_sysctl_header == NULL) {
1148                 printk(KERN_WARNING
1149                        "br_netfilter: can't register to sysctl.\n");
1150                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1151                 return -ENOMEM;
1152         }
1153 #endif
1154         RCU_INIT_POINTER(nf_br_ops, &br_ops);
1155         printk(KERN_NOTICE "Bridge firewalling registered\n");
1156         return 0;
1157 }
1158
1159 static void __exit br_netfilter_fini(void)
1160 {
1161         RCU_INIT_POINTER(nf_br_ops, NULL);
1162         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1163 #ifdef CONFIG_SYSCTL
1164         unregister_net_sysctl_table(brnf_sysctl_header);
1165 #endif
1166 }
1167
1168 module_init(br_netfilter_init);
1169 module_exit(br_netfilter_fini);
1170
1171 MODULE_LICENSE("GPL");
1172 MODULE_AUTHOR("Lennert Buytenhek <buytenh@gnu.org>");
1173 MODULE_AUTHOR("Bart De Schuymer <bdschuym@pandora.be>");
1174 MODULE_DESCRIPTION("Linux ethernet netfilter firewall bridge");