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
  41 #include <asm/uaccess.h>
  42 #include "br_private.h"
  43 #ifdef CONFIG_SYSCTL
  44 #include <linux/sysctl.h>
  45 #endif
  46
  47 #ifdef CONFIG_SYSCTL
  48 static struct ctl_table_header *brnf_sysctl_header;
  49 static int brnf_call_iptables __read_mostly = 1;
  50 static int brnf_call_ip6tables __read_mostly = 1;
  51 static int brnf_call_arptables __read_mostly = 1;
  52 static int brnf_filter_vlan_tagged __read_mostly;
  53 static int brnf_filter_pppoe_tagged __read_mostly;
  54 static int brnf_pass_vlan_indev __read_mostly;
  55 #else
  56 #define brnf_call_iptables 1
  57 #define brnf_call_ip6tables 1
  58 #define brnf_call_arptables 1
  59 #define brnf_filter_vlan_tagged 0
  60 #define brnf_filter_pppoe_tagged 0
  61 #define brnf_pass_vlan_indev 0
  62 #endif
  63
  64 #define IS_IP(skb) \
  65         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IP))
  66
  67 #define IS_IPV6(skb) \
  68         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IPV6))
  69
  70 #define IS_ARP(skb) \
  71         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_ARP))
  72
  73 static inline __be16 vlan_proto(const struct sk_buff *skb)
  74 {
  75         if (skb_vlan_tag_present(skb))
  76                 return skb->protocol;
  77         else if (skb->protocol == htons(ETH_P_8021Q))
  78                 return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  79         else
  80                 return 0;
  81 }
  82
  83 #define IS_VLAN_IP(skb) \
  84         (vlan_proto(skb) == htons(ETH_P_IP) && \
  85          brnf_filter_vlan_tagged)
  86
  87 #define IS_VLAN_IPV6(skb) \
  88         (vlan_proto(skb) == htons(ETH_P_IPV6) && \
  89          brnf_filter_vlan_tagged)
  90
  91 #define IS_VLAN_ARP(skb) \
  92         (vlan_proto(skb) == htons(ETH_P_ARP) && \
  93          brnf_filter_vlan_tagged)
  94
  95 static inline __be16 pppoe_proto(const struct sk_buff *skb)
  96 {
  97         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
  98                             sizeof(struct pppoe_hdr)));
  99 }
 100
 101 #define IS_PPPOE_IP(skb) \
 102         (skb->protocol == htons(ETH_P_PPP_SES) && \
 103          pppoe_proto(skb) == htons(PPP_IP) && \
 104          brnf_filter_pppoe_tagged)
 105
 106 #define IS_PPPOE_IPV6(skb) \
 107         (skb->protocol == htons(ETH_P_PPP_SES) && \
 108          pppoe_proto(skb) == htons(PPP_IPV6) && \
 109          brnf_filter_pppoe_tagged)
 110
 111 /* largest possible L2 header, see br_nf_dev_queue_xmit() */
 112 #define NF_BRIDGE_MAX_MAC_HEADER_LENGTH (PPPOE_SES_HLEN + ETH_HLEN)
 113
 114 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4) || IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 115 struct brnf_frag_data {
 116         char mac[NF_BRIDGE_MAX_MAC_HEADER_LENGTH];
 117         u8 encap_size;
 118         u8 size;
 119         u16 vlan_tci;
 120         __be16 vlan_proto;
 121 };
 122
 123 static DEFINE_PER_CPU(struct brnf_frag_data, brnf_frag_data_storage);
 124 #endif
 125
 126 static void nf_bridge_info_free(struct sk_buff *skb)
 127 {
 128         if (skb->nf_bridge) {
 129                 nf_bridge_put(skb->nf_bridge);
 130                 skb->nf_bridge = NULL;
 131         }
 132 }
 133
 134 static inline struct net_device *bridge_parent(const struct net_device *dev)
 135 {
 136         struct net_bridge_port *port;
 137
 138         port = br_port_get_rcu(dev);
 139         return port ? port->br->dev : NULL;
 140 }
 141
 142 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 143 {
 144         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 145
 146         if (atomic_read(&nf_bridge->use) > 1) {
 147                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 148
 149                 if (tmp) {
 150                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 151                         atomic_set(&tmp->use, 1);
 152                 }
 153                 nf_bridge_put(nf_bridge);
 154                 nf_bridge = tmp;
 155         }
 156         return nf_bridge;
 157 }
 158
 159 unsigned int nf_bridge_encap_header_len(const struct sk_buff *skb)
 160 {
 161         switch (skb->protocol) {
 162         case __cpu_to_be16(ETH_P_8021Q):
 163                 return VLAN_HLEN;
 164         case __cpu_to_be16(ETH_P_PPP_SES):
 165                 return PPPOE_SES_HLEN;
 166         default:
 167                 return 0;
 168         }
 169 }
 170
 171 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 172 {
 173         unsigned int len = nf_bridge_encap_header_len(skb);
 174
 175         skb_pull(skb, len);
 176         skb->network_header += len;
 177 }
 178
 179 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 180 {
 181         unsigned int len = nf_bridge_encap_header_len(skb);
 182
 183         skb_pull_rcsum(skb, len);
 184         skb->network_header += len;
 185 }
 186
 187 /* When handing a packet over to the IP layer
 188  * check whether we have a skb that is in the
 189  * expected format
 190  */
 191
 192 static int br_validate_ipv4(struct sk_buff *skb)
 193 {
 194         const struct iphdr *iph;
 195         struct net_device *dev = skb->dev;
 196         u32 len;
 197
 198         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 199                 goto inhdr_error;
 200
 201         iph = ip_hdr(skb);
 202
 203         /* Basic sanity checks */
 204         if (iph->ihl < 5 || iph->version != 4)
 205                 goto inhdr_error;
 206
 207         if (!pskb_may_pull(skb, iph->ihl*4))
 208                 goto inhdr_error;
 209
 210         iph = ip_hdr(skb);
 211         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 212                 goto inhdr_error;
 213
 214         len = ntohs(iph->tot_len);
 215         if (skb->len < len) {
 216                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
 217                 goto drop;
 218         } else if (len < (iph->ihl*4))
 219                 goto inhdr_error;
 220
 221         if (pskb_trim_rcsum(skb, len)) {
 222                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 223                 goto drop;
 224         }
 225
 226         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 227         /* We should really parse IP options here but until
 228          * somebody who actually uses IP options complains to
 229          * us we'll just silently ignore the options because
 230          * we're lazy!
 231          */
 232         return 0;
 233
 234 inhdr_error:
 235         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 236 drop:
 237         return -1;
 238 }
 239
 240 void nf_bridge_update_protocol(struct sk_buff *skb)
 241 {
 242         switch (skb->nf_bridge->orig_proto) {
 243         case BRNF_PROTO_8021Q:
 244                 skb->protocol = htons(ETH_P_8021Q);
 245                 break;
 246         case BRNF_PROTO_PPPOE:
 247                 skb->protocol = htons(ETH_P_PPP_SES);
 248                 break;
 249         case BRNF_PROTO_UNCHANGED:
 250                 break;
 251         }
 252 }
 253
 254 /* Obtain the correct destination MAC address, while preserving the original
 255  * source MAC address. If we already know this address, we just copy it. If we
 256  * don't, we use the neighbour framework to find out. In both cases, we make
 257  * sure that br_handle_frame_finish() is called afterwards.
 258  */
 259 int br_nf_pre_routing_finish_bridge(struct sock *sk, struct sk_buff *skb)
 260 {
 261         struct neighbour *neigh;
 262         struct dst_entry *dst;
 263
 264         skb->dev = bridge_parent(skb->dev);
 265         if (!skb->dev)
 266                 goto free_skb;
 267         dst = skb_dst(skb);
 268         neigh = dst_neigh_lookup_skb(dst, skb);
 269         if (neigh) {
 270                 struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 271                 int ret;
 272
 273                 if (neigh->hh.hh_len) {
 274                         neigh_hh_bridge(&neigh->hh, skb);
 275                         skb->dev = nf_bridge->physindev;
 276                         ret = br_handle_frame_finish(sk, skb);
 277                 } else {
 278                         /* the neighbour function below overwrites the complete
 279                          * MAC header, so we save the Ethernet source address and
 280                          * protocol number.
 281                          */
 282                         skb_copy_from_linear_data_offset(skb,
 283                                                          -(ETH_HLEN-ETH_ALEN),
 284                                                          nf_bridge->neigh_header,
 285                                                          ETH_HLEN-ETH_ALEN);
 286                         /* tell br_dev_xmit to continue with forwarding */
 287                         nf_bridge->bridged_dnat = 1;
 288                         /* FIXME Need to refragment */
 289                         ret = neigh->output(neigh, skb);
 290                 }
 291                 neigh_release(neigh);
 292                 return ret;
 293         }
 294 free_skb:
 295         kfree_skb(skb);
 296         return 0;
 297 }
 298
 299 static inline bool
 300 br_nf_ipv4_daddr_was_changed(const struct sk_buff *skb,
 301                              const struct nf_bridge_info *nf_bridge)
 302 {
 303         return ip_hdr(skb)->daddr != nf_bridge->ipv4_daddr;
 304 }
 305
 306 /* This requires some explaining. If DNAT has taken place,
 307  * we will need to fix up the destination Ethernet address.
 308  * This is also true when SNAT takes place (for the reply direction).
 309  *
 310  * There are two cases to consider:
 311  * 1. The packet was DNAT'ed to a device in the same bridge
 312  *    port group as it was received on. We can still bridge
 313  *    the packet.
 314  * 2. The packet was DNAT'ed to a different device, either
 315  *    a non-bridged device or another bridge port group.
 316  *    The packet will need to be routed.
 317  *
 318  * The correct way of distinguishing between these two cases is to
 319  * call ip_route_input() and to look at skb->dst->dev, which is
 320  * changed to the destination device if ip_route_input() succeeds.
 321  *
 322  * Let's first consider the case that ip_route_input() succeeds:
 323  *
 324  * If the output device equals the logical bridge device the packet
 325  * came in on, we can consider this bridging. The corresponding MAC
 326  * address will be obtained in br_nf_pre_routing_finish_bridge.
 327  * Otherwise, the packet is considered to be routed and we just
 328  * change the destination MAC address so that the packet will
 329  * later be passed up to the IP stack to be routed. For a redirected
 330  * packet, ip_route_input() will give back the localhost as output device,
 331  * which differs from the bridge device.
 332  *
 333  * Let's now consider the case that ip_route_input() fails:
 334  *
 335  * This can be because the destination address is martian, in which case
 336  * the packet will be dropped.
 337  * If IP forwarding is disabled, ip_route_input() will fail, while
 338  * ip_route_output_key() can return success. The source
 339  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 340  * thinks we're handling a locally generated packet and won't care
 341  * if IP forwarding is enabled. If the output device equals the logical bridge
 342  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 343  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 344  */
 345 static int br_nf_pre_routing_finish(struct sock *sk, struct sk_buff *skb)
 346 {
 347         struct net_device *dev = skb->dev;
 348         struct iphdr *iph = ip_hdr(skb);
 349         struct net *net = dev_net(dev);
 350         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 351         struct rtable *rt;
 352         int err;
 353
 354         nf_bridge->frag_max_size = IPCB(skb)->frag_max_size;
 355
 356         if (nf_bridge->pkt_otherhost) {
 357                 skb->pkt_type = PACKET_OTHERHOST;
 358                 nf_bridge->pkt_otherhost = false;
 359         }
 360         nf_bridge->in_prerouting = 0;
 361         if (br_nf_ipv4_daddr_was_changed(skb, nf_bridge)) {
 362                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 363                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 364
 365                         /* If err equals -EHOSTUNREACH the error is due to a
 366                          * martian destination or due to the fact that
 367                          * forwarding is disabled. For most martian packets,
 368                          * ip_route_output_key() will fail. It won't fail for 2 types of
 369                          * martian destinations: loopback destinations and destination
 370                          * 0.0.0.0. In both cases the packet will be dropped because the
 371                          * destination is the loopback device and not the bridge. */
 372                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 373                                 goto free_skb;
 374
 375                         rt = ip_route_output(net, iph->daddr, 0,
 376                                              RT_TOS(iph->tos), 0);
 377                         if (!IS_ERR(rt)) {
 378                                 /* - Bridged-and-DNAT'ed traffic doesn't
 379                                  *   require ip_forwarding. */
 380                                 if (rt->dst.dev == dev) {
 381                                         skb_dst_set(skb, &rt->dst);
 382                                         goto bridged_dnat;
 383                                 }
 384                                 ip_rt_put(rt);
 385                         }
 386 free_skb:
 387                         kfree_skb(skb);
 388                         return 0;
 389                 } else {
 390                         if (skb_dst(skb)->dev == dev) {
 391 bridged_dnat:
 392                                 skb->dev = nf_bridge->physindev;
 393                                 nf_bridge_update_protocol(skb);
 394                                 nf_bridge_push_encap_header(skb);
 395                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
 396                                                NF_BR_PRE_ROUTING,
 397                                                sk, skb, skb->dev, NULL,
 398                                                br_nf_pre_routing_finish_bridge,
 399                                                1);
 400                                 return 0;
 401                         }
 402                         ether_addr_copy(eth_hdr(skb)->h_dest, dev->dev_addr);
 403                         skb->pkt_type = PACKET_HOST;
 404                 }
 405         } else {
 406                 rt = bridge_parent_rtable(nf_bridge->physindev);
 407                 if (!rt) {
 408                         kfree_skb(skb);
 409                         return 0;
 410                 }
 411                 skb_dst_set_noref(skb, &rt->dst);
 412         }
 413
 414         skb->dev = nf_bridge->physindev;
 415         nf_bridge_update_protocol(skb);
 416         nf_bridge_push_encap_header(skb);
 417         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, sk, skb,
 418                        skb->dev, NULL,
 419                        br_handle_frame_finish, 1);
 420
 421         return 0;
 422 }
 423
 424 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb, const struct net_device *dev)
 425 {
 426         struct net_device *vlan, *br;
 427
 428         br = bridge_parent(dev);
 429         if (brnf_pass_vlan_indev == 0 || !skb_vlan_tag_present(skb))
 430                 return br;
 431
 432         vlan = __vlan_find_dev_deep_rcu(br, skb->vlan_proto,
 433                                     skb_vlan_tag_get(skb) & VLAN_VID_MASK);
 434
 435         return vlan ? vlan : br;
 436 }
 437
 438 /* Some common code for IPv4/IPv6 */
 439 struct net_device *setup_pre_routing(struct sk_buff *skb)
 440 {
 441         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 442
 443         if (skb->pkt_type == PACKET_OTHERHOST) {
 444                 skb->pkt_type = PACKET_HOST;
 445                 nf_bridge->pkt_otherhost = true;
 446         }
 447
 448         nf_bridge->in_prerouting = 1;
 449         nf_bridge->physindev = skb->dev;
 450         skb->dev = brnf_get_logical_dev(skb, skb->dev);
 451
 452         if (skb->protocol == htons(ETH_P_8021Q))
 453                 nf_bridge->orig_proto = BRNF_PROTO_8021Q;
 454         else if (skb->protocol == htons(ETH_P_PPP_SES))
 455                 nf_bridge->orig_proto = BRNF_PROTO_PPPOE;
 456
 457         /* Must drop socket now because of tproxy. */
 458         skb_orphan(skb);
 459         return skb->dev;
 460 }
 461
 462 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 463  * Replicate the checks that IPv4 does on packet reception.
 464  * Set skb->dev to the bridge device (i.e. parent of the
 465  * receiving device) to make netfilter happy, the REDIRECT
 466  * target in particular.  Save the original destination IP
 467  * address to be able to detect DNAT afterwards. */
 468 static unsigned int br_nf_pre_routing(const struct nf_hook_ops *ops,
 469                                       struct sk_buff *skb,
 470                                       const struct nf_hook_state *state)
 471 {
 472         struct nf_bridge_info *nf_bridge;
 473         struct net_bridge_port *p;
 474         struct net_bridge *br;
 475         __u32 len = nf_bridge_encap_header_len(skb);
 476
 477         if (unlikely(!pskb_may_pull(skb, len)))
 478                 return NF_DROP;
 479
 480         p = br_port_get_rcu(state->in);
 481         if (p == NULL)
 482                 return NF_DROP;
 483         br = p->br;
 484
 485         if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb)) {
 486                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
 487                         return NF_ACCEPT;
 488
 489                 nf_bridge_pull_encap_header_rcsum(skb);
 490                 return br_nf_pre_routing_ipv6(ops, skb, state);
 491         }
 492
 493         if (!brnf_call_iptables && !br->nf_call_iptables)
 494                 return NF_ACCEPT;
 495
 496         if (!IS_IP(skb) && !IS_VLAN_IP(skb) && !IS_PPPOE_IP(skb))
 497                 return NF_ACCEPT;
 498
 499         nf_bridge_pull_encap_header_rcsum(skb);
 500
 501         if (br_validate_ipv4(skb))
 502                 return NF_DROP;
 503
 504         nf_bridge_put(skb->nf_bridge);
 505         if (!nf_bridge_alloc(skb))
 506                 return NF_DROP;
 507         if (!setup_pre_routing(skb))
 508                 return NF_DROP;
 509
 510         nf_bridge = nf_bridge_info_get(skb);
 511         nf_bridge->ipv4_daddr = ip_hdr(skb)->daddr;
 512
 513         skb->protocol = htons(ETH_P_IP);
 514
 515         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, state->sk, skb,
 516                 skb->dev, NULL,
 517                 br_nf_pre_routing_finish);
 518
 519         return NF_STOLEN;
 520 }
 521
 522
 523 /* PF_BRIDGE/LOCAL_IN ************************************************/
 524 /* The packet is locally destined, which requires a real
 525  * dst_entry, so detach the fake one.  On the way up, the
 526  * packet would pass through PRE_ROUTING again (which already
 527  * took place when the packet entered the bridge), but we
 528  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 529  * prevent this from happening. */
 530 static unsigned int br_nf_local_in(const struct nf_hook_ops *ops,
 531                                    struct sk_buff *skb,
 532                                    const struct nf_hook_state *state)
 533 {
 534         br_drop_fake_rtable(skb);
 535         return NF_ACCEPT;
 536 }
 537
 538 /* PF_BRIDGE/FORWARD *************************************************/
 539 static int br_nf_forward_finish(struct sock *sk, struct sk_buff *skb)
 540 {
 541         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 542         struct net_device *in;
 543
 544         if (!IS_ARP(skb) && !IS_VLAN_ARP(skb)) {
 545
 546                 if (skb->protocol == htons(ETH_P_IP))
 547                         nf_bridge->frag_max_size = IPCB(skb)->frag_max_size;
 548
 549                 if (skb->protocol == htons(ETH_P_IPV6))
 550                         nf_bridge->frag_max_size = IP6CB(skb)->frag_max_size;
 551
 552                 in = nf_bridge->physindev;
 553                 if (nf_bridge->pkt_otherhost) {
 554                         skb->pkt_type = PACKET_OTHERHOST;
 555                         nf_bridge->pkt_otherhost = false;
 556                 }
 557                 nf_bridge_update_protocol(skb);
 558         } else {
 559                 in = *((struct net_device **)(skb->cb));
 560         }
 561         nf_bridge_push_encap_header(skb);
 562
 563         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, sk, skb,
 564                        in, skb->dev, br_forward_finish, 1);
 565         return 0;
 566 }
 567
 568
 569 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 570  * netfilter with indev and outdev set to the bridge device,
 571  * but we are still able to filter on the 'real' indev/outdev
 572  * because of the physdev module. For ARP, indev and outdev are the
 573  * bridge ports. */
 574 static unsigned int br_nf_forward_ip(const struct nf_hook_ops *ops,
 575                                      struct sk_buff *skb,
 576                                      const struct nf_hook_state *state)
 577 {
 578         struct nf_bridge_info *nf_bridge;
 579         struct net_device *parent;
 580         u_int8_t pf;
 581
 582         if (!skb->nf_bridge)
 583                 return NF_ACCEPT;
 584
 585         /* Need exclusive nf_bridge_info since we might have multiple
 586          * different physoutdevs. */
 587         if (!nf_bridge_unshare(skb))
 588                 return NF_DROP;
 589
 590         nf_bridge = nf_bridge_info_get(skb);
 591         if (!nf_bridge)
 592                 return NF_DROP;
 593
 594         parent = bridge_parent(state->out);
 595         if (!parent)
 596                 return NF_DROP;
 597
 598         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 599                 pf = NFPROTO_IPV4;
 600         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 601                 pf = NFPROTO_IPV6;
 602         else
 603                 return NF_ACCEPT;
 604
 605         nf_bridge_pull_encap_header(skb);
 606
 607         if (skb->pkt_type == PACKET_OTHERHOST) {
 608                 skb->pkt_type = PACKET_HOST;
 609                 nf_bridge->pkt_otherhost = true;
 610         }
 611
 612         if (pf == NFPROTO_IPV4) {
 613                 if (br_validate_ipv4(skb))
 614                         return NF_DROP;
 615                 IPCB(skb)->frag_max_size = nf_bridge->frag_max_size;
 616         }
 617
 618         if (pf == NFPROTO_IPV6) {
 619                 if (br_validate_ipv6(skb))
 620                         return NF_DROP;
 621                 IP6CB(skb)->frag_max_size = nf_bridge->frag_max_size;
 622         }
 623
 624         nf_bridge->physoutdev = skb->dev;
 625         if (pf == NFPROTO_IPV4)
 626                 skb->protocol = htons(ETH_P_IP);
 627         else
 628                 skb->protocol = htons(ETH_P_IPV6);
 629
 630         NF_HOOK(pf, NF_INET_FORWARD, NULL, skb,
 631                 brnf_get_logical_dev(skb, state->in),
 632                 parent, br_nf_forward_finish);
 633
 634         return NF_STOLEN;
 635 }
 636
 637 static unsigned int br_nf_forward_arp(const struct nf_hook_ops *ops,
 638                                       struct sk_buff *skb,
 639                                       const struct nf_hook_state *state)
 640 {
 641         struct net_bridge_port *p;
 642         struct net_bridge *br;
 643         struct net_device **d = (struct net_device **)(skb->cb);
 644
 645         p = br_port_get_rcu(state->out);
 646         if (p == NULL)
 647                 return NF_ACCEPT;
 648         br = p->br;
 649
 650         if (!brnf_call_arptables && !br->nf_call_arptables)
 651                 return NF_ACCEPT;
 652
 653         if (!IS_ARP(skb)) {
 654                 if (!IS_VLAN_ARP(skb))
 655                         return NF_ACCEPT;
 656                 nf_bridge_pull_encap_header(skb);
 657         }
 658
 659         if (arp_hdr(skb)->ar_pln != 4) {
 660                 if (IS_VLAN_ARP(skb))
 661                         nf_bridge_push_encap_header(skb);
 662                 return NF_ACCEPT;
 663         }
 664         *d = state->in;
 665         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, state->sk, skb,
 666                 state->in, state->out, br_nf_forward_finish);
 667
 668         return NF_STOLEN;
 669 }
 670
 671 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4) || IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 672 static int br_nf_push_frag_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
 673 {
 674         struct brnf_frag_data *data;
 675         int err;
 676
 677         data = this_cpu_ptr(&brnf_frag_data_storage);
 678         err = skb_cow_head(skb, data->size);
 679
 680         if (err) {
 681                 kfree_skb(skb);
 682                 return 0;
 683         }
 684
 685         if (data->vlan_tci) {
 686                 skb->vlan_tci = data->vlan_tci;
 687                 skb->vlan_proto = data->vlan_proto;
 688         }
 689
 690         skb_copy_to_linear_data_offset(skb, -data->size, data->mac, data->size);
 691         __skb_push(skb, data->encap_size);
 692
 693         nf_bridge_info_free(skb);
 694         return br_dev_queue_push_xmit(sk, skb);
 695 }
 696 static int br_nf_push_frag_xmit_sk(struct sock *sk, struct sk_buff *skb)
 697 {
 698         struct net *net = dev_net(skb_dst(skb)->dev);
 699         return br_nf_push_frag_xmit(net, sk, skb);
 700 }
 701 #endif
 702
 703 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4)
 704 static int
 705 br_nf_ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
 706                   int (*output)(struct sock *, struct sk_buff *))
 707 {
 708         unsigned int mtu = ip_skb_dst_mtu(skb);
 709         struct iphdr *iph = ip_hdr(skb);
 710
 711         if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->ignore_df) ||
 712                      (IPCB(skb)->frag_max_size &&
 713                       IPCB(skb)->frag_max_size > mtu))) {
 714                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
 715                 kfree_skb(skb);
 716                 return -EMSGSIZE;
 717         }
 718
 719         return ip_do_fragment(sk, skb, output);
 720 }
 721 #endif
 722
 723 static unsigned int nf_bridge_mtu_reduction(const struct sk_buff *skb)
 724 {
 725         if (skb->nf_bridge->orig_proto == BRNF_PROTO_PPPOE)
 726                 return PPPOE_SES_HLEN;
 727         return 0;
 728 }
 729
 730 static int br_nf_dev_queue_xmit(struct sock *sk, struct sk_buff *skb)
 731 {
 732         struct nf_bridge_info *nf_bridge;
 733         unsigned int mtu_reserved;
 734         struct net *net = dev_net(skb_dst(skb)->dev);
 735
 736         mtu_reserved = nf_bridge_mtu_reduction(skb);
 737
 738         if (skb_is_gso(skb) || skb->len + mtu_reserved <= skb->dev->mtu) {
 739                 nf_bridge_info_free(skb);
 740                 return br_dev_queue_push_xmit(sk, skb);
 741         }
 742
 743         nf_bridge = nf_bridge_info_get(skb);
 744
 745 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4)
 746         /* This is wrong! We should preserve the original fragment
 747          * boundaries by preserving frag_list rather than refragmenting.
 748          */
 749         if (skb->protocol == htons(ETH_P_IP)) {
 750                 struct brnf_frag_data *data;
 751
 752                 if (br_validate_ipv4(skb))
 753                         goto drop;
 754
 755                 IPCB(skb)->frag_max_size = nf_bridge->frag_max_size;
 756
 757                 nf_bridge_update_protocol(skb);
 758
 759                 data = this_cpu_ptr(&brnf_frag_data_storage);
 760
 761                 data->vlan_tci = skb->vlan_tci;
 762                 data->vlan_proto = skb->vlan_proto;
 763                 data->encap_size = nf_bridge_encap_header_len(skb);
 764                 data->size = ETH_HLEN + data->encap_size;
 765
 766                 skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
 767                                                  data->size);
 768
 769                 return br_nf_ip_fragment(net, sk, skb, br_nf_push_frag_xmit_sk);
 770         }
 771 #endif
 772 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 773         if (skb->protocol == htons(ETH_P_IPV6)) {
 774                 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
 775                 struct brnf_frag_data *data;
 776
 777                 if (br_validate_ipv6(skb))
 778                         goto drop;
 779
 780                 IP6CB(skb)->frag_max_size = nf_bridge->frag_max_size;
 781
 782                 nf_bridge_update_protocol(skb);
 783
 784                 data = this_cpu_ptr(&brnf_frag_data_storage);
 785                 data->encap_size = nf_bridge_encap_header_len(skb);
 786                 data->size = ETH_HLEN + data->encap_size;
 787
 788                 skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
 789                                                  data->size);
 790
 791                 if (v6ops)
 792                         return v6ops->fragment(sk, skb, br_nf_push_frag_xmit_sk);
 793
 794                 kfree_skb(skb);
 795                 return -EMSGSIZE;
 796         }
 797 #endif
 798         nf_bridge_info_free(skb);
 799         return br_dev_queue_push_xmit(sk, skb);
 800  drop:
 801         kfree_skb(skb);
 802         return 0;
 803 }
 804
 805 /* PF_BRIDGE/POST_ROUTING ********************************************/
 806 static unsigned int br_nf_post_routing(const struct nf_hook_ops *ops,
 807                                        struct sk_buff *skb,
 808                                        const struct nf_hook_state *state)
 809 {
 810         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 811         struct net_device *realoutdev = bridge_parent(skb->dev);
 812         u_int8_t pf;
 813
 814         /* if nf_bridge is set, but ->physoutdev is NULL, this packet came in
 815          * on a bridge, but was delivered locally and is now being routed:
 816          *
 817          * POST_ROUTING was already invoked from the ip stack.
 818          */
 819         if (!nf_bridge || !nf_bridge->physoutdev)
 820                 return NF_ACCEPT;
 821
 822         if (!realoutdev)
 823                 return NF_DROP;
 824
 825         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 826                 pf = NFPROTO_IPV4;
 827         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 828                 pf = NFPROTO_IPV6;
 829         else
 830                 return NF_ACCEPT;
 831
 832         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 833          * about the value of skb->pkt_type. */
 834         if (skb->pkt_type == PACKET_OTHERHOST) {
 835                 skb->pkt_type = PACKET_HOST;
 836                 nf_bridge->pkt_otherhost = true;
 837         }
 838
 839         nf_bridge_pull_encap_header(skb);
 840         if (pf == NFPROTO_IPV4)
 841                 skb->protocol = htons(ETH_P_IP);
 842         else
 843                 skb->protocol = htons(ETH_P_IPV6);
 844
 845         NF_HOOK(pf, NF_INET_POST_ROUTING, state->sk, skb,
 846                 NULL, realoutdev,
 847                 br_nf_dev_queue_xmit);
 848
 849         return NF_STOLEN;
 850 }
 851
 852 /* IP/SABOTAGE *****************************************************/
 853 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 854  * for the second time. */
 855 static unsigned int ip_sabotage_in(const struct nf_hook_ops *ops,
 856                                    struct sk_buff *skb,
 857                                    const struct nf_hook_state *state)
 858 {
 859         if (skb->nf_bridge && !skb->nf_bridge->in_prerouting)
 860                 return NF_STOP;
 861
 862         return NF_ACCEPT;
 863 }
 864
 865 /* This is called when br_netfilter has called into iptables/netfilter,
 866  * and DNAT has taken place on a bridge-forwarded packet.
 867  *
 868  * neigh->output has created a new MAC header, with local br0 MAC
 869  * as saddr.
 870  *
 871  * This restores the original MAC saddr of the bridged packet
 872  * before invoking bridge forward logic to transmit the packet.
 873  */
 874 static void br_nf_pre_routing_finish_bridge_slow(struct sk_buff *skb)
 875 {
 876         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 877
 878         skb_pull(skb, ETH_HLEN);
 879         nf_bridge->bridged_dnat = 0;
 880
 881         BUILD_BUG_ON(sizeof(nf_bridge->neigh_header) != (ETH_HLEN - ETH_ALEN));
 882
 883         skb_copy_to_linear_data_offset(skb, -(ETH_HLEN - ETH_ALEN),
 884                                        nf_bridge->neigh_header,
 885                                        ETH_HLEN - ETH_ALEN);
 886         skb->dev = nf_bridge->physindev;
 887
 888         nf_bridge->physoutdev = NULL;
 889         br_handle_frame_finish(NULL, skb);
 890 }
 891
 892 static int br_nf_dev_xmit(struct sk_buff *skb)
 893 {
 894         if (skb->nf_bridge && skb->nf_bridge->bridged_dnat) {
 895                 br_nf_pre_routing_finish_bridge_slow(skb);
 896                 return 1;
 897         }
 898         return 0;
 899 }
 900
 901 static const struct nf_br_ops br_ops = {
 902         .br_dev_xmit_hook =     br_nf_dev_xmit,
 903 };
 904
 905 void br_netfilter_enable(void)
 906 {
 907 }
 908 EXPORT_SYMBOL_GPL(br_netfilter_enable);
 909
 910 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 911  * br_dev_queue_push_xmit is called afterwards */
 912 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
 913         {
 914                 .hook = br_nf_pre_routing,
 915                 .owner = THIS_MODULE,
 916                 .pf = NFPROTO_BRIDGE,
 917                 .hooknum = NF_BR_PRE_ROUTING,
 918                 .priority = NF_BR_PRI_BRNF,
 919         },
 920         {
 921                 .hook = br_nf_local_in,
 922                 .owner = THIS_MODULE,
 923                 .pf = NFPROTO_BRIDGE,
 924                 .hooknum = NF_BR_LOCAL_IN,
 925                 .priority = NF_BR_PRI_BRNF,
 926         },
 927         {
 928                 .hook = br_nf_forward_ip,
 929                 .owner = THIS_MODULE,
 930                 .pf = NFPROTO_BRIDGE,
 931                 .hooknum = NF_BR_FORWARD,
 932                 .priority = NF_BR_PRI_BRNF - 1,
 933         },
 934         {
 935                 .hook = br_nf_forward_arp,
 936                 .owner = THIS_MODULE,
 937                 .pf = NFPROTO_BRIDGE,
 938                 .hooknum = NF_BR_FORWARD,
 939                 .priority = NF_BR_PRI_BRNF,
 940         },
 941         {
 942                 .hook = br_nf_post_routing,
 943                 .owner = THIS_MODULE,
 944                 .pf = NFPROTO_BRIDGE,
 945                 .hooknum = NF_BR_POST_ROUTING,
 946                 .priority = NF_BR_PRI_LAST,
 947         },
 948         {
 949                 .hook = ip_sabotage_in,
 950                 .owner = THIS_MODULE,
 951                 .pf = NFPROTO_IPV4,
 952                 .hooknum = NF_INET_PRE_ROUTING,
 953                 .priority = NF_IP_PRI_FIRST,
 954         },
 955         {
 956                 .hook = ip_sabotage_in,
 957                 .owner = THIS_MODULE,
 958                 .pf = NFPROTO_IPV6,
 959                 .hooknum = NF_INET_PRE_ROUTING,
 960                 .priority = NF_IP6_PRI_FIRST,
 961         },
 962 };
 963
 964 #ifdef CONFIG_SYSCTL
 965 static
 966 int brnf_sysctl_call_tables(struct ctl_table *ctl, int write,
 967                             void __user *buffer, size_t *lenp, loff_t *ppos)
 968 {
 969         int ret;
 970
 971         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
 972
 973         if (write && *(int *)(ctl->data))
 974                 *(int *)(ctl->data) = 1;
 975         return ret;
 976 }
 977
 978 static struct ctl_table brnf_table[] = {
 979         {
 980                 .procname       = "bridge-nf-call-arptables",
 981                 .data           = &brnf_call_arptables,
 982                 .maxlen         = sizeof(int),
 983                 .mode           = 0644,
 984                 .proc_handler   = brnf_sysctl_call_tables,
 985         },
 986         {
 987                 .procname       = "bridge-nf-call-iptables",
 988                 .data           = &brnf_call_iptables,
 989                 .maxlen         = sizeof(int),
 990                 .mode           = 0644,
 991                 .proc_handler   = brnf_sysctl_call_tables,
 992         },
 993         {
 994                 .procname       = "bridge-nf-call-ip6tables",
 995                 .data           = &brnf_call_ip6tables,
 996                 .maxlen         = sizeof(int),
 997                 .mode           = 0644,
 998                 .proc_handler   = brnf_sysctl_call_tables,
 999         },
1000         {
1001                 .procname       = "bridge-nf-filter-vlan-tagged",
1002                 .data           = &brnf_filter_vlan_tagged,
1003                 .maxlen         = sizeof(int),
1004                 .mode           = 0644,
1005                 .proc_handler   = brnf_sysctl_call_tables,
1006         },
1007         {
1008                 .procname       = "bridge-nf-filter-pppoe-tagged",
1009                 .data           = &brnf_filter_pppoe_tagged,
1010                 .maxlen         = sizeof(int),
1011                 .mode           = 0644,
1012                 .proc_handler   = brnf_sysctl_call_tables,
1013         },
1014         {
1015                 .procname       = "bridge-nf-pass-vlan-input-dev",
1016                 .data           = &brnf_pass_vlan_indev,
1017                 .maxlen         = sizeof(int),
1018                 .mode           = 0644,
1019                 .proc_handler   = brnf_sysctl_call_tables,
1020         },
1021         { }
1022 };
1023 #endif
1024
1025 static int __init br_netfilter_init(void)
1026 {
1027         int ret;
1028
1029         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1030         if (ret < 0)
1031                 return ret;
1032
1033 #ifdef CONFIG_SYSCTL
1034         brnf_sysctl_header = register_net_sysctl(&init_net, "net/bridge", brnf_table);
1035         if (brnf_sysctl_header == NULL) {
1036                 printk(KERN_WARNING
1037                        "br_netfilter: can't register to sysctl.\n");
1038                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1039                 return -ENOMEM;
1040         }
1041 #endif
1042         RCU_INIT_POINTER(nf_br_ops, &br_ops);
1043         printk(KERN_NOTICE "Bridge firewalling registered\n");
1044         return 0;
1045 }
1046
1047 static void __exit br_netfilter_fini(void)
1048 {
1049         RCU_INIT_POINTER(nf_br_ops, NULL);
1050         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1051 #ifdef CONFIG_SYSCTL
1052         unregister_net_sysctl_table(brnf_sysctl_header);
1053 #endif
1054 }
1055
1056 module_init(br_netfilter_init);
1057 module_exit(br_netfilter_fini);
1058
1059 MODULE_LICENSE("GPL");
1060 MODULE_AUTHOR("Lennert Buytenhek <buytenh@gnu.org>");
1061 MODULE_AUTHOR("Bart De Schuymer <bdschuym@pandora.be>");
1062 MODULE_DESCRIPTION("Linux ethernet netfilter firewall bridge");