net/bridge/br_netfilter_hooks.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/addrconf.h>
  38 #include <net/route.h>
  39 #include <net/netfilter/br_netfilter.h>
  40 #include <net/netns/generic.h>
  41
  42 #include <asm/uaccess.h>
  43 #include "br_private.h"
  44 #ifdef CONFIG_SYSCTL
  45 #include <linux/sysctl.h>
  46 #endif
  47
  48 static int brnf_net_id __read_mostly;
  49
  50 struct brnf_net {
  51         bool enabled;
  52 };
  53
  54 #ifdef CONFIG_SYSCTL
  55 static struct ctl_table_header *brnf_sysctl_header;
  56 static int brnf_call_iptables __read_mostly = 1;
  57 static int brnf_call_ip6tables __read_mostly = 1;
  58 static int brnf_call_arptables __read_mostly = 1;
  59 static int brnf_filter_vlan_tagged __read_mostly;
  60 static int brnf_filter_pppoe_tagged __read_mostly;
  61 static int brnf_pass_vlan_indev __read_mostly;
  62 #else
  63 #define brnf_call_iptables 1
  64 #define brnf_call_ip6tables 1
  65 #define brnf_call_arptables 1
  66 #define brnf_filter_vlan_tagged 0
  67 #define brnf_filter_pppoe_tagged 0
  68 #define brnf_pass_vlan_indev 0
  69 #endif
  70
  71 #define IS_IP(skb) \
  72         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IP))
  73
  74 #define IS_IPV6(skb) \
  75         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IPV6))
  76
  77 #define IS_ARP(skb) \
  78         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_ARP))
  79
  80 static inline __be16 vlan_proto(const struct sk_buff *skb)
  81 {
  82         if (skb_vlan_tag_present(skb))
  83                 return skb->protocol;
  84         else if (skb->protocol == htons(ETH_P_8021Q))
  85                 return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  86         else
  87                 return 0;
  88 }
  89
  90 #define IS_VLAN_IP(skb) \
  91         (vlan_proto(skb) == htons(ETH_P_IP) && \
  92          brnf_filter_vlan_tagged)
  93
  94 #define IS_VLAN_IPV6(skb) \
  95         (vlan_proto(skb) == htons(ETH_P_IPV6) && \
  96          brnf_filter_vlan_tagged)
  97
  98 #define IS_VLAN_ARP(skb) \
  99         (vlan_proto(skb) == htons(ETH_P_ARP) && \
 100          brnf_filter_vlan_tagged)
 101
 102 static inline __be16 pppoe_proto(const struct sk_buff *skb)
 103 {
 104         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
 105                             sizeof(struct pppoe_hdr)));
 106 }
 107
 108 #define IS_PPPOE_IP(skb) \
 109         (skb->protocol == htons(ETH_P_PPP_SES) && \
 110          pppoe_proto(skb) == htons(PPP_IP) && \
 111          brnf_filter_pppoe_tagged)
 112
 113 #define IS_PPPOE_IPV6(skb) \
 114         (skb->protocol == htons(ETH_P_PPP_SES) && \
 115          pppoe_proto(skb) == htons(PPP_IPV6) && \
 116          brnf_filter_pppoe_tagged)
 117
 118 /* largest possible L2 header, see br_nf_dev_queue_xmit() */
 119 #define NF_BRIDGE_MAX_MAC_HEADER_LENGTH (PPPOE_SES_HLEN + ETH_HLEN)
 120
 121 struct brnf_frag_data {
 122         char mac[NF_BRIDGE_MAX_MAC_HEADER_LENGTH];
 123         u8 encap_size;
 124         u8 size;
 125         u16 vlan_tci;
 126         __be16 vlan_proto;
 127 };
 128
 129 static DEFINE_PER_CPU(struct brnf_frag_data, brnf_frag_data_storage);
 130
 131 static void nf_bridge_info_free(struct sk_buff *skb)
 132 {
 133         if (skb->nf_bridge) {
 134                 nf_bridge_put(skb->nf_bridge);
 135                 skb->nf_bridge = NULL;
 136         }
 137 }
 138
 139 static inline struct net_device *bridge_parent(const struct net_device *dev)
 140 {
 141         struct net_bridge_port *port;
 142
 143         port = br_port_get_rcu(dev);
 144         return port ? port->br->dev : NULL;
 145 }
 146
 147 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 148 {
 149         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 150
 151         if (atomic_read(&nf_bridge->use) > 1) {
 152                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 153
 154                 if (tmp) {
 155                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 156                         atomic_set(&tmp->use, 1);
 157                 }
 158                 nf_bridge_put(nf_bridge);
 159                 nf_bridge = tmp;
 160         }
 161         return nf_bridge;
 162 }
 163
 164 unsigned int nf_bridge_encap_header_len(const struct sk_buff *skb)
 165 {
 166         switch (skb->protocol) {
 167         case __cpu_to_be16(ETH_P_8021Q):
 168                 return VLAN_HLEN;
 169         case __cpu_to_be16(ETH_P_PPP_SES):
 170                 return PPPOE_SES_HLEN;
 171         default:
 172                 return 0;
 173         }
 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 /* When handing a packet over to the IP layer
 193  * check whether we have a skb that is in the
 194  * expected format
 195  */
 196
 197 static int br_validate_ipv4(struct net *net, struct sk_buff *skb)
 198 {
 199         const struct iphdr *iph;
 200         u32 len;
 201
 202         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 203                 goto inhdr_error;
 204
 205         iph = ip_hdr(skb);
 206
 207         /* Basic sanity checks */
 208         if (iph->ihl < 5 || iph->version != 4)
 209                 goto inhdr_error;
 210
 211         if (!pskb_may_pull(skb, iph->ihl*4))
 212                 goto inhdr_error;
 213
 214         iph = ip_hdr(skb);
 215         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 216                 goto inhdr_error;
 217
 218         len = ntohs(iph->tot_len);
 219         if (skb->len < len) {
 220                 __IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS);
 221                 goto drop;
 222         } else if (len < (iph->ihl*4))
 223                 goto inhdr_error;
 224
 225         if (pskb_trim_rcsum(skb, len)) {
 226                 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
 227                 goto drop;
 228         }
 229
 230         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 231         /* We should really parse IP options here but until
 232          * somebody who actually uses IP options complains to
 233          * us we'll just silently ignore the options because
 234          * we're lazy!
 235          */
 236         return 0;
 237
 238 inhdr_error:
 239         __IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
 240 drop:
 241         return -1;
 242 }
 243
 244 void nf_bridge_update_protocol(struct sk_buff *skb)
 245 {
 246         switch (skb->nf_bridge->orig_proto) {
 247         case BRNF_PROTO_8021Q:
 248                 skb->protocol = htons(ETH_P_8021Q);
 249                 break;
 250         case BRNF_PROTO_PPPOE:
 251                 skb->protocol = htons(ETH_P_PPP_SES);
 252                 break;
 253         case BRNF_PROTO_UNCHANGED:
 254                 break;
 255         }
 256 }
 257
 258 /* Obtain the correct destination MAC address, while preserving the original
 259  * source MAC address. If we already know this address, we just copy it. If we
 260  * don't, we use the neighbour framework to find out. In both cases, we make
 261  * sure that br_handle_frame_finish() is called afterwards.
 262  */
 263 int br_nf_pre_routing_finish_bridge(struct net *net, struct sock *sk, struct sk_buff *skb)
 264 {
 265         struct neighbour *neigh;
 266         struct dst_entry *dst;
 267
 268         skb->dev = bridge_parent(skb->dev);
 269         if (!skb->dev)
 270                 goto free_skb;
 271         dst = skb_dst(skb);
 272         neigh = dst_neigh_lookup_skb(dst, skb);
 273         if (neigh) {
 274                 struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 275                 int ret;
 276
 277                 if (neigh->hh.hh_len) {
 278                         neigh_hh_bridge(&neigh->hh, skb);
 279                         skb->dev = nf_bridge->physindev;
 280                         ret = br_handle_frame_finish(net, sk, skb);
 281                 } else {
 282                         /* the neighbour function below overwrites the complete
 283                          * MAC header, so we save the Ethernet source address and
 284                          * protocol number.
 285                          */
 286                         skb_copy_from_linear_data_offset(skb,
 287                                                          -(ETH_HLEN-ETH_ALEN),
 288                                                          nf_bridge->neigh_header,
 289                                                          ETH_HLEN-ETH_ALEN);
 290                         /* tell br_dev_xmit to continue with forwarding */
 291                         nf_bridge->bridged_dnat = 1;
 292                         /* FIXME Need to refragment */
 293                         ret = neigh->output(neigh, skb);
 294                 }
 295                 neigh_release(neigh);
 296                 return ret;
 297         }
 298 free_skb:
 299         kfree_skb(skb);
 300         return 0;
 301 }
 302
 303 static inline bool
 304 br_nf_ipv4_daddr_was_changed(const struct sk_buff *skb,
 305                              const struct nf_bridge_info *nf_bridge)
 306 {
 307         return ip_hdr(skb)->daddr != nf_bridge->ipv4_daddr;
 308 }
 309
 310 /* This requires some explaining. If DNAT has taken place,
 311  * we will need to fix up the destination Ethernet address.
 312  * This is also true when SNAT takes place (for the reply direction).
 313  *
 314  * There are two cases to consider:
 315  * 1. The packet was DNAT'ed to a device in the same bridge
 316  *    port group as it was received on. We can still bridge
 317  *    the packet.
 318  * 2. The packet was DNAT'ed to a different device, either
 319  *    a non-bridged device or another bridge port group.
 320  *    The packet will need to be routed.
 321  *
 322  * The correct way of distinguishing between these two cases is to
 323  * call ip_route_input() and to look at skb->dst->dev, which is
 324  * changed to the destination device if ip_route_input() succeeds.
 325  *
 326  * Let's first consider the case that ip_route_input() succeeds:
 327  *
 328  * If the output device equals the logical bridge device the packet
 329  * came in on, we can consider this bridging. The corresponding MAC
 330  * address will be obtained in br_nf_pre_routing_finish_bridge.
 331  * Otherwise, the packet is considered to be routed and we just
 332  * change the destination MAC address so that the packet will
 333  * later be passed up to the IP stack to be routed. For a redirected
 334  * packet, ip_route_input() will give back the localhost as output device,
 335  * which differs from the bridge device.
 336  *
 337  * Let's now consider the case that ip_route_input() fails:
 338  *
 339  * This can be because the destination address is martian, in which case
 340  * the packet will be dropped.
 341  * If IP forwarding is disabled, ip_route_input() will fail, while
 342  * ip_route_output_key() can return success. The source
 343  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 344  * thinks we're handling a locally generated packet and won't care
 345  * if IP forwarding is enabled. If the output device equals the logical bridge
 346  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 347  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 348  */
 349 static int br_nf_pre_routing_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
 350 {
 351         struct net_device *dev = skb->dev;
 352         struct iphdr *iph = ip_hdr(skb);
 353         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 354         struct rtable *rt;
 355         int err;
 356
 357         nf_bridge->frag_max_size = IPCB(skb)->frag_max_size;
 358
 359         if (nf_bridge->pkt_otherhost) {
 360                 skb->pkt_type = PACKET_OTHERHOST;
 361                 nf_bridge->pkt_otherhost = false;
 362         }
 363         nf_bridge->in_prerouting = 0;
 364         if (br_nf_ipv4_daddr_was_changed(skb, nf_bridge)) {
 365                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 366                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 367
 368                         /* If err equals -EHOSTUNREACH the error is due to a
 369                          * martian destination or due to the fact that
 370                          * forwarding is disabled. For most martian packets,
 371                          * ip_route_output_key() will fail. It won't fail for 2 types of
 372                          * martian destinations: loopback destinations and destination
 373                          * 0.0.0.0. In both cases the packet will be dropped because the
 374                          * destination is the loopback device and not the bridge. */
 375                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 376                                 goto free_skb;
 377
 378                         rt = ip_route_output(net, iph->daddr, 0,
 379                                              RT_TOS(iph->tos), 0);
 380                         if (!IS_ERR(rt)) {
 381                                 /* - Bridged-and-DNAT'ed traffic doesn't
 382                                  *   require ip_forwarding. */
 383                                 if (rt->dst.dev == dev) {
 384                                         skb_dst_set(skb, &rt->dst);
 385                                         goto bridged_dnat;
 386                                 }
 387                                 ip_rt_put(rt);
 388                         }
 389 free_skb:
 390                         kfree_skb(skb);
 391                         return 0;
 392                 } else {
 393                         if (skb_dst(skb)->dev == dev) {
 394 bridged_dnat:
 395                                 skb->dev = nf_bridge->physindev;
 396                                 nf_bridge_update_protocol(skb);
 397                                 nf_bridge_push_encap_header(skb);
 398                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
 399                                                NF_BR_PRE_ROUTING,
 400                                                net, sk, skb, skb->dev, NULL,
 401                                                br_nf_pre_routing_finish_bridge,
 402                                                1);
 403                                 return 0;
 404                         }
 405                         ether_addr_copy(eth_hdr(skb)->h_dest, dev->dev_addr);
 406                         skb->pkt_type = PACKET_HOST;
 407                 }
 408         } else {
 409                 rt = bridge_parent_rtable(nf_bridge->physindev);
 410                 if (!rt) {
 411                         kfree_skb(skb);
 412                         return 0;
 413                 }
 414                 skb_dst_set_noref(skb, &rt->dst);
 415         }
 416
 417         skb->dev = nf_bridge->physindev;
 418         nf_bridge_update_protocol(skb);
 419         nf_bridge_push_encap_header(skb);
 420         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, net, sk, skb,
 421                        skb->dev, NULL,
 422                        br_handle_frame_finish, 1);
 423
 424         return 0;
 425 }
 426
 427 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb, const struct net_device *dev)
 428 {
 429         struct net_device *vlan, *br;
 430
 431         br = bridge_parent(dev);
 432         if (brnf_pass_vlan_indev == 0 || !skb_vlan_tag_present(skb))
 433                 return br;
 434
 435         vlan = __vlan_find_dev_deep_rcu(br, skb->vlan_proto,
 436                                     skb_vlan_tag_get(skb) & VLAN_VID_MASK);
 437
 438         return vlan ? vlan : br;
 439 }
 440
 441 /* Some common code for IPv4/IPv6 */
 442 struct net_device *setup_pre_routing(struct sk_buff *skb)
 443 {
 444         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 445
 446         if (skb->pkt_type == PACKET_OTHERHOST) {
 447                 skb->pkt_type = PACKET_HOST;
 448                 nf_bridge->pkt_otherhost = true;
 449         }
 450
 451         nf_bridge->in_prerouting = 1;
 452         nf_bridge->physindev = skb->dev;
 453         skb->dev = brnf_get_logical_dev(skb, skb->dev);
 454
 455         if (skb->protocol == htons(ETH_P_8021Q))
 456                 nf_bridge->orig_proto = BRNF_PROTO_8021Q;
 457         else if (skb->protocol == htons(ETH_P_PPP_SES))
 458                 nf_bridge->orig_proto = BRNF_PROTO_PPPOE;
 459
 460         /* Must drop socket now because of tproxy. */
 461         skb_orphan(skb);
 462         return skb->dev;
 463 }
 464
 465 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 466  * Replicate the checks that IPv4 does on packet reception.
 467  * Set skb->dev to the bridge device (i.e. parent of the
 468  * receiving device) to make netfilter happy, the REDIRECT
 469  * target in particular.  Save the original destination IP
 470  * address to be able to detect DNAT afterwards. */
 471 static unsigned int br_nf_pre_routing(void *priv,
 472                                       struct sk_buff *skb,
 473                                       const struct nf_hook_state *state)
 474 {
 475         struct nf_bridge_info *nf_bridge;
 476         struct net_bridge_port *p;
 477         struct net_bridge *br;
 478         __u32 len = nf_bridge_encap_header_len(skb);
 479
 480         if (unlikely(!pskb_may_pull(skb, len)))
 481                 return NF_DROP;
 482
 483         p = br_port_get_rcu(state->in);
 484         if (p == NULL)
 485                 return NF_DROP;
 486         br = p->br;
 487
 488         if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb)) {
 489                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
 490                         return NF_ACCEPT;
 491
 492                 nf_bridge_pull_encap_header_rcsum(skb);
 493                 return br_nf_pre_routing_ipv6(priv, skb, state);
 494         }
 495
 496         if (!brnf_call_iptables && !br->nf_call_iptables)
 497                 return NF_ACCEPT;
 498
 499         if (!IS_IP(skb) && !IS_VLAN_IP(skb) && !IS_PPPOE_IP(skb))
 500                 return NF_ACCEPT;
 501
 502         nf_bridge_pull_encap_header_rcsum(skb);
 503
 504         if (br_validate_ipv4(state->net, skb))
 505                 return NF_DROP;
 506
 507         nf_bridge_put(skb->nf_bridge);
 508         if (!nf_bridge_alloc(skb))
 509                 return NF_DROP;
 510         if (!setup_pre_routing(skb))
 511                 return NF_DROP;
 512
 513         nf_bridge = nf_bridge_info_get(skb);
 514         nf_bridge->ipv4_daddr = ip_hdr(skb)->daddr;
 515
 516         skb->protocol = htons(ETH_P_IP);
 517
 518         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, state->net, state->sk, skb,
 519                 skb->dev, NULL,
 520                 br_nf_pre_routing_finish);
 521
 522         return NF_STOLEN;
 523 }
 524
 525
 526 /* PF_BRIDGE/LOCAL_IN ************************************************/
 527 /* The packet is locally destined, which requires a real
 528  * dst_entry, so detach the fake one.  On the way up, the
 529  * packet would pass through PRE_ROUTING again (which already
 530  * took place when the packet entered the bridge), but we
 531  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 532  * prevent this from happening. */
 533 static unsigned int br_nf_local_in(void *priv,
 534                                    struct sk_buff *skb,
 535                                    const struct nf_hook_state *state)
 536 {
 537         br_drop_fake_rtable(skb);
 538         return NF_ACCEPT;
 539 }
 540
 541 /* PF_BRIDGE/FORWARD *************************************************/
 542 static int br_nf_forward_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
 543 {
 544         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 545         struct net_device *in;
 546
 547         if (!IS_ARP(skb) && !IS_VLAN_ARP(skb)) {
 548
 549                 if (skb->protocol == htons(ETH_P_IP))
 550                         nf_bridge->frag_max_size = IPCB(skb)->frag_max_size;
 551
 552                 if (skb->protocol == htons(ETH_P_IPV6))
 553                         nf_bridge->frag_max_size = IP6CB(skb)->frag_max_size;
 554
 555                 in = nf_bridge->physindev;
 556                 if (nf_bridge->pkt_otherhost) {
 557                         skb->pkt_type = PACKET_OTHERHOST;
 558                         nf_bridge->pkt_otherhost = false;
 559                 }
 560                 nf_bridge_update_protocol(skb);
 561         } else {
 562                 in = *((struct net_device **)(skb->cb));
 563         }
 564         nf_bridge_push_encap_header(skb);
 565
 566         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, net, sk, skb,
 567                        in, skb->dev, br_forward_finish, 1);
 568         return 0;
 569 }
 570
 571
 572 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 573  * netfilter with indev and outdev set to the bridge device,
 574  * but we are still able to filter on the 'real' indev/outdev
 575  * because of the physdev module. For ARP, indev and outdev are the
 576  * bridge ports. */
 577 static unsigned int br_nf_forward_ip(void *priv,
 578                                      struct sk_buff *skb,
 579                                      const struct nf_hook_state *state)
 580 {
 581         struct nf_bridge_info *nf_bridge;
 582         struct net_device *parent;
 583         u_int8_t pf;
 584
 585         if (!skb->nf_bridge)
 586                 return NF_ACCEPT;
 587
 588         /* Need exclusive nf_bridge_info since we might have multiple
 589          * different physoutdevs. */
 590         if (!nf_bridge_unshare(skb))
 591                 return NF_DROP;
 592
 593         nf_bridge = nf_bridge_info_get(skb);
 594         if (!nf_bridge)
 595                 return NF_DROP;
 596
 597         parent = bridge_parent(state->out);
 598         if (!parent)
 599                 return NF_DROP;
 600
 601         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 602                 pf = NFPROTO_IPV4;
 603         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 604                 pf = NFPROTO_IPV6;
 605         else
 606                 return NF_ACCEPT;
 607
 608         nf_bridge_pull_encap_header(skb);
 609
 610         if (skb->pkt_type == PACKET_OTHERHOST) {
 611                 skb->pkt_type = PACKET_HOST;
 612                 nf_bridge->pkt_otherhost = true;
 613         }
 614
 615         if (pf == NFPROTO_IPV4) {
 616                 if (br_validate_ipv4(state->net, skb))
 617                         return NF_DROP;
 618                 IPCB(skb)->frag_max_size = nf_bridge->frag_max_size;
 619         }
 620
 621         if (pf == NFPROTO_IPV6) {
 622                 if (br_validate_ipv6(state->net, skb))
 623                         return NF_DROP;
 624                 IP6CB(skb)->frag_max_size = nf_bridge->frag_max_size;
 625         }
 626
 627         nf_bridge->physoutdev = skb->dev;
 628         if (pf == NFPROTO_IPV4)
 629                 skb->protocol = htons(ETH_P_IP);
 630         else
 631                 skb->protocol = htons(ETH_P_IPV6);
 632
 633         NF_HOOK(pf, NF_INET_FORWARD, state->net, NULL, skb,
 634                 brnf_get_logical_dev(skb, state->in),
 635                 parent, br_nf_forward_finish);
 636
 637         return NF_STOLEN;
 638 }
 639
 640 static unsigned int br_nf_forward_arp(void *priv,
 641                                       struct sk_buff *skb,
 642                                       const struct nf_hook_state *state)
 643 {
 644         struct net_bridge_port *p;
 645         struct net_bridge *br;
 646         struct net_device **d = (struct net_device **)(skb->cb);
 647
 648         p = br_port_get_rcu(state->out);
 649         if (p == NULL)
 650                 return NF_ACCEPT;
 651         br = p->br;
 652
 653         if (!brnf_call_arptables && !br->nf_call_arptables)
 654                 return NF_ACCEPT;
 655
 656         if (!IS_ARP(skb)) {
 657                 if (!IS_VLAN_ARP(skb))
 658                         return NF_ACCEPT;
 659                 nf_bridge_pull_encap_header(skb);
 660         }
 661
 662         if (arp_hdr(skb)->ar_pln != 4) {
 663                 if (IS_VLAN_ARP(skb))
 664                         nf_bridge_push_encap_header(skb);
 665                 return NF_ACCEPT;
 666         }
 667         *d = state->in;
 668         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, state->net, state->sk, skb,
 669                 state->in, state->out, br_nf_forward_finish);
 670
 671         return NF_STOLEN;
 672 }
 673
 674 static int br_nf_push_frag_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
 675 {
 676         struct brnf_frag_data *data;
 677         int err;
 678
 679         data = this_cpu_ptr(&brnf_frag_data_storage);
 680         err = skb_cow_head(skb, data->size);
 681
 682         if (err) {
 683                 kfree_skb(skb);
 684                 return 0;
 685         }
 686
 687         if (data->vlan_tci) {
 688                 skb->vlan_tci = data->vlan_tci;
 689                 skb->vlan_proto = data->vlan_proto;
 690         }
 691
 692         skb_copy_to_linear_data_offset(skb, -data->size, data->mac, data->size);
 693         __skb_push(skb, data->encap_size);
 694
 695         nf_bridge_info_free(skb);
 696         return br_dev_queue_push_xmit(net, sk, skb);
 697 }
 698
 699 static int
 700 br_nf_ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
 701                   int (*output)(struct net *, struct sock *, struct sk_buff *))
 702 {
 703         unsigned int mtu = ip_skb_dst_mtu(skb);
 704         struct iphdr *iph = ip_hdr(skb);
 705
 706         if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->ignore_df) ||
 707                      (IPCB(skb)->frag_max_size &&
 708                       IPCB(skb)->frag_max_size > mtu))) {
 709                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
 710                 kfree_skb(skb);
 711                 return -EMSGSIZE;
 712         }
 713
 714         return ip_do_fragment(net, sk, skb, output);
 715 }
 716
 717 static unsigned int nf_bridge_mtu_reduction(const struct sk_buff *skb)
 718 {
 719         if (skb->nf_bridge->orig_proto == BRNF_PROTO_PPPOE)
 720                 return PPPOE_SES_HLEN;
 721         return 0;
 722 }
 723
 724 static int br_nf_dev_queue_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
 725 {
 726         struct nf_bridge_info *nf_bridge;
 727         unsigned int mtu_reserved;
 728
 729         mtu_reserved = nf_bridge_mtu_reduction(skb);
 730
 731         if (skb_is_gso(skb) || skb->len + mtu_reserved <= skb->dev->mtu) {
 732                 nf_bridge_info_free(skb);
 733                 return br_dev_queue_push_xmit(net, sk, skb);
 734         }
 735
 736         nf_bridge = nf_bridge_info_get(skb);
 737
 738         /* This is wrong! We should preserve the original fragment
 739          * boundaries by preserving frag_list rather than refragmenting.
 740          */
 741         if (IS_ENABLED(CONFIG_NF_DEFRAG_IPV4) &&
 742             skb->protocol == htons(ETH_P_IP)) {
 743                 struct brnf_frag_data *data;
 744
 745                 if (br_validate_ipv4(net, skb))
 746                         goto drop;
 747
 748                 IPCB(skb)->frag_max_size = nf_bridge->frag_max_size;
 749
 750                 nf_bridge_update_protocol(skb);
 751
 752                 data = this_cpu_ptr(&brnf_frag_data_storage);
 753
 754                 data->vlan_tci = skb->vlan_tci;
 755                 data->vlan_proto = skb->vlan_proto;
 756                 data->encap_size = nf_bridge_encap_header_len(skb);
 757                 data->size = ETH_HLEN + data->encap_size;
 758
 759                 skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
 760                                                  data->size);
 761
 762                 return br_nf_ip_fragment(net, sk, skb, br_nf_push_frag_xmit);
 763         }
 764         if (IS_ENABLED(CONFIG_NF_DEFRAG_IPV6) &&
 765             skb->protocol == htons(ETH_P_IPV6)) {
 766                 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
 767                 struct brnf_frag_data *data;
 768
 769                 if (br_validate_ipv6(net, skb))
 770                         goto drop;
 771
 772                 IP6CB(skb)->frag_max_size = nf_bridge->frag_max_size;
 773
 774                 nf_bridge_update_protocol(skb);
 775
 776                 data = this_cpu_ptr(&brnf_frag_data_storage);
 777                 data->encap_size = nf_bridge_encap_header_len(skb);
 778                 data->size = ETH_HLEN + data->encap_size;
 779
 780                 skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
 781                                                  data->size);
 782
 783                 if (v6ops)
 784                         return v6ops->fragment(net, sk, skb, br_nf_push_frag_xmit);
 785
 786                 kfree_skb(skb);
 787                 return -EMSGSIZE;
 788         }
 789         nf_bridge_info_free(skb);
 790         return br_dev_queue_push_xmit(net, sk, skb);
 791  drop:
 792         kfree_skb(skb);
 793         return 0;
 794 }
 795
 796 /* PF_BRIDGE/POST_ROUTING ********************************************/
 797 static unsigned int br_nf_post_routing(void *priv,
 798                                        struct sk_buff *skb,
 799                                        const struct nf_hook_state *state)
 800 {
 801         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 802         struct net_device *realoutdev = bridge_parent(skb->dev);
 803         u_int8_t pf;
 804
 805         /* if nf_bridge is set, but ->physoutdev is NULL, this packet came in
 806          * on a bridge, but was delivered locally and is now being routed:
 807          *
 808          * POST_ROUTING was already invoked from the ip stack.
 809          */
 810         if (!nf_bridge || !nf_bridge->physoutdev)
 811                 return NF_ACCEPT;
 812
 813         if (!realoutdev)
 814                 return NF_DROP;
 815
 816         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 817                 pf = NFPROTO_IPV4;
 818         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 819                 pf = NFPROTO_IPV6;
 820         else
 821                 return NF_ACCEPT;
 822
 823         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 824          * about the value of skb->pkt_type. */
 825         if (skb->pkt_type == PACKET_OTHERHOST) {
 826                 skb->pkt_type = PACKET_HOST;
 827                 nf_bridge->pkt_otherhost = true;
 828         }
 829
 830         nf_bridge_pull_encap_header(skb);
 831         if (pf == NFPROTO_IPV4)
 832                 skb->protocol = htons(ETH_P_IP);
 833         else
 834                 skb->protocol = htons(ETH_P_IPV6);
 835
 836         NF_HOOK(pf, NF_INET_POST_ROUTING, state->net, state->sk, skb,
 837                 NULL, realoutdev,
 838                 br_nf_dev_queue_xmit);
 839
 840         return NF_STOLEN;
 841 }
 842
 843 /* IP/SABOTAGE *****************************************************/
 844 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 845  * for the second time. */
 846 static unsigned int ip_sabotage_in(void *priv,
 847                                    struct sk_buff *skb,
 848                                    const struct nf_hook_state *state)
 849 {
 850         if (skb->nf_bridge && !skb->nf_bridge->in_prerouting)
 851                 return NF_STOP;
 852
 853         return NF_ACCEPT;
 854 }
 855
 856 /* This is called when br_netfilter has called into iptables/netfilter,
 857  * and DNAT has taken place on a bridge-forwarded packet.
 858  *
 859  * neigh->output has created a new MAC header, with local br0 MAC
 860  * as saddr.
 861  *
 862  * This restores the original MAC saddr of the bridged packet
 863  * before invoking bridge forward logic to transmit the packet.
 864  */
 865 static void br_nf_pre_routing_finish_bridge_slow(struct sk_buff *skb)
 866 {
 867         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 868
 869         skb_pull(skb, ETH_HLEN);
 870         nf_bridge->bridged_dnat = 0;
 871
 872         BUILD_BUG_ON(sizeof(nf_bridge->neigh_header) != (ETH_HLEN - ETH_ALEN));
 873
 874         skb_copy_to_linear_data_offset(skb, -(ETH_HLEN - ETH_ALEN),
 875                                        nf_bridge->neigh_header,
 876                                        ETH_HLEN - ETH_ALEN);
 877         skb->dev = nf_bridge->physindev;
 878
 879         nf_bridge->physoutdev = NULL;
 880         br_handle_frame_finish(dev_net(skb->dev), NULL, skb);
 881 }
 882
 883 static int br_nf_dev_xmit(struct sk_buff *skb)
 884 {
 885         if (skb->nf_bridge && skb->nf_bridge->bridged_dnat) {
 886                 br_nf_pre_routing_finish_bridge_slow(skb);
 887                 return 1;
 888         }
 889         return 0;
 890 }
 891
 892 static const struct nf_br_ops br_ops = {
 893         .br_dev_xmit_hook =     br_nf_dev_xmit,
 894 };
 895
 896 void br_netfilter_enable(void)
 897 {
 898 }
 899 EXPORT_SYMBOL_GPL(br_netfilter_enable);
 900
 901 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 902  * br_dev_queue_push_xmit is called afterwards */
 903 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
 904         {
 905                 .hook = br_nf_pre_routing,
 906                 .pf = NFPROTO_BRIDGE,
 907                 .hooknum = NF_BR_PRE_ROUTING,
 908                 .priority = NF_BR_PRI_BRNF,
 909         },
 910         {
 911                 .hook = br_nf_local_in,
 912                 .pf = NFPROTO_BRIDGE,
 913                 .hooknum = NF_BR_LOCAL_IN,
 914                 .priority = NF_BR_PRI_BRNF,
 915         },
 916         {
 917                 .hook = br_nf_forward_ip,
 918                 .pf = NFPROTO_BRIDGE,
 919                 .hooknum = NF_BR_FORWARD,
 920                 .priority = NF_BR_PRI_BRNF - 1,
 921         },
 922         {
 923                 .hook = br_nf_forward_arp,
 924                 .pf = NFPROTO_BRIDGE,
 925                 .hooknum = NF_BR_FORWARD,
 926                 .priority = NF_BR_PRI_BRNF,
 927         },
 928         {
 929                 .hook = br_nf_post_routing,
 930                 .pf = NFPROTO_BRIDGE,
 931                 .hooknum = NF_BR_POST_ROUTING,
 932                 .priority = NF_BR_PRI_LAST,
 933         },
 934         {
 935                 .hook = ip_sabotage_in,
 936                 .pf = NFPROTO_IPV4,
 937                 .hooknum = NF_INET_PRE_ROUTING,
 938                 .priority = NF_IP_PRI_FIRST,
 939         },
 940         {
 941                 .hook = ip_sabotage_in,
 942                 .pf = NFPROTO_IPV6,
 943                 .hooknum = NF_INET_PRE_ROUTING,
 944                 .priority = NF_IP6_PRI_FIRST,
 945         },
 946 };
 947
 948 static int brnf_device_event(struct notifier_block *unused, unsigned long event,
 949                              void *ptr)
 950 {
 951         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 952         struct brnf_net *brnet;
 953         struct net *net;
 954         int ret;
 955
 956         if (event != NETDEV_REGISTER || !(dev->priv_flags & IFF_EBRIDGE))
 957                 return NOTIFY_DONE;
 958
 959         ASSERT_RTNL();
 960
 961         net = dev_net(dev);
 962         brnet = net_generic(net, brnf_net_id);
 963         if (brnet->enabled)
 964                 return NOTIFY_OK;
 965
 966         ret = nf_register_net_hooks(net, br_nf_ops, ARRAY_SIZE(br_nf_ops));
 967         if (ret)
 968                 return NOTIFY_BAD;
 969
 970         brnet->enabled = true;
 971         return NOTIFY_OK;
 972 }
 973
 974 static void __net_exit brnf_exit_net(struct net *net)
 975 {
 976         struct brnf_net *brnet = net_generic(net, brnf_net_id);
 977
 978         if (!brnet->enabled)
 979                 return;
 980
 981         nf_unregister_net_hooks(net, br_nf_ops, ARRAY_SIZE(br_nf_ops));
 982         brnet->enabled = false;
 983 }
 984
 985 static struct pernet_operations brnf_net_ops __read_mostly = {
 986         .exit = brnf_exit_net,
 987         .id   = &brnf_net_id,
 988         .size = sizeof(struct brnf_net),
 989 };
 990
 991 static struct notifier_block brnf_notifier __read_mostly = {
 992         .notifier_call = brnf_device_event,
 993 };
 994
 995 #ifdef CONFIG_SYSCTL
 996 static
 997 int brnf_sysctl_call_tables(struct ctl_table *ctl, int write,
 998                             void __user *buffer, size_t *lenp, loff_t *ppos)
 999 {
1000         int ret;
1001
1002         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1003
1004         if (write && *(int *)(ctl->data))
1005                 *(int *)(ctl->data) = 1;
1006         return ret;
1007 }
1008
1009 static struct ctl_table brnf_table[] = {
1010         {
1011                 .procname       = "bridge-nf-call-arptables",
1012                 .data           = &brnf_call_arptables,
1013                 .maxlen         = sizeof(int),
1014                 .mode           = 0644,
1015                 .proc_handler   = brnf_sysctl_call_tables,
1016         },
1017         {
1018                 .procname       = "bridge-nf-call-iptables",
1019                 .data           = &brnf_call_iptables,
1020                 .maxlen         = sizeof(int),
1021                 .mode           = 0644,
1022                 .proc_handler   = brnf_sysctl_call_tables,
1023         },
1024         {
1025                 .procname       = "bridge-nf-call-ip6tables",
1026                 .data           = &brnf_call_ip6tables,
1027                 .maxlen         = sizeof(int),
1028                 .mode           = 0644,
1029                 .proc_handler   = brnf_sysctl_call_tables,
1030         },
1031         {
1032                 .procname       = "bridge-nf-filter-vlan-tagged",
1033                 .data           = &brnf_filter_vlan_tagged,
1034                 .maxlen         = sizeof(int),
1035                 .mode           = 0644,
1036                 .proc_handler   = brnf_sysctl_call_tables,
1037         },
1038         {
1039                 .procname       = "bridge-nf-filter-pppoe-tagged",
1040                 .data           = &brnf_filter_pppoe_tagged,
1041                 .maxlen         = sizeof(int),
1042                 .mode           = 0644,
1043                 .proc_handler   = brnf_sysctl_call_tables,
1044         },
1045         {
1046                 .procname       = "bridge-nf-pass-vlan-input-dev",
1047                 .data           = &brnf_pass_vlan_indev,
1048                 .maxlen         = sizeof(int),
1049                 .mode           = 0644,
1050                 .proc_handler   = brnf_sysctl_call_tables,
1051         },
1052         { }
1053 };
1054 #endif
1055
1056 static int __init br_netfilter_init(void)
1057 {
1058         int ret;
1059
1060         ret = register_pernet_subsys(&brnf_net_ops);
1061         if (ret < 0)
1062                 return ret;
1063
1064         ret = register_netdevice_notifier(&brnf_notifier);
1065         if (ret < 0) {
1066                 unregister_pernet_subsys(&brnf_net_ops);
1067                 return ret;
1068         }
1069
1070 #ifdef CONFIG_SYSCTL
1071         brnf_sysctl_header = register_net_sysctl(&init_net, "net/bridge", brnf_table);
1072         if (brnf_sysctl_header == NULL) {
1073                 printk(KERN_WARNING
1074                        "br_netfilter: can't register to sysctl.\n");
1075                 unregister_netdevice_notifier(&brnf_notifier);
1076                 unregister_pernet_subsys(&brnf_net_ops);
1077                 return -ENOMEM;
1078         }
1079 #endif
1080         RCU_INIT_POINTER(nf_br_ops, &br_ops);
1081         printk(KERN_NOTICE "Bridge firewalling registered\n");
1082         return 0;
1083 }
1084
1085 static void __exit br_netfilter_fini(void)
1086 {
1087         RCU_INIT_POINTER(nf_br_ops, NULL);
1088         unregister_netdevice_notifier(&brnf_notifier);
1089         unregister_pernet_subsys(&brnf_net_ops);
1090 #ifdef CONFIG_SYSCTL
1091         unregister_net_sysctl_table(brnf_sysctl_header);
1092 #endif
1093 }
1094
1095 module_init(br_netfilter_init);
1096 module_exit(br_netfilter_fini);
1097
1098 MODULE_LICENSE("GPL");
1099 MODULE_AUTHOR("Lennert Buytenhek <buytenh@gnu.org>");
1100 MODULE_AUTHOR("Bart De Schuymer <bdschuym@pandora.be>");
1101 MODULE_DESCRIPTION("Linux ethernet netfilter firewall bridge");