net/bridge/br_netfilter_hooks.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Handle firewalling
   4  *      Linux ethernet bridge
   5  *
   6  *      Authors:
   7  *      Lennert Buytenhek               <buytenh@gnu.org>
   8  *      Bart De Schuymer                <bdschuym@pandora.be>
   9  *
  10  *      Lennert dedicates this file to Kerstin Wurdinger.
  11  */
  12
  13 #include <linux/module.h>
  14 #include <linux/kernel.h>
  15 #include <linux/slab.h>
  16 #include <linux/ip.h>
  17 #include <linux/netdevice.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/if_arp.h>
  20 #include <linux/if_ether.h>
  21 #include <linux/if_vlan.h>
  22 #include <linux/if_pppox.h>
  23 #include <linux/ppp_defs.h>
  24 #include <linux/netfilter_bridge.h>
  25 #include <uapi/linux/netfilter_bridge.h>
  26 #include <linux/netfilter_ipv4.h>
  27 #include <linux/netfilter_ipv6.h>
  28 #include <linux/netfilter_arp.h>
  29 #include <linux/in_route.h>
  30 #include <linux/rculist.h>
  31 #include <linux/inetdevice.h>
  32
  33 #include <net/ip.h>
  34 #include <net/ipv6.h>
  35 #include <net/addrconf.h>
  36 #include <net/route.h>
  37 #include <net/netfilter/br_netfilter.h>
  38 #include <net/netns/generic.h>
  39
  40 #include <linux/uaccess.h>
  41 #include "br_private.h"
  42 #ifdef CONFIG_SYSCTL
  43 #include <linux/sysctl.h>
  44 #endif
  45
  46 static unsigned int brnf_net_id __read_mostly;
  47
  48 struct brnf_net {
  49         bool enabled;
  50
  51 #ifdef CONFIG_SYSCTL
  52         struct ctl_table_header *ctl_hdr;
  53 #endif
  54
  55         /* default value is 1 */
  56         int call_iptables;
  57         int call_ip6tables;
  58         int call_arptables;
  59
  60         /* default value is 0 */
  61         int filter_vlan_tagged;
  62         int filter_pppoe_tagged;
  63         int pass_vlan_indev;
  64 };
  65
  66 #define IS_IP(skb) \
  67         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IP))
  68
  69 #define IS_IPV6(skb) \
  70         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IPV6))
  71
  72 #define IS_ARP(skb) \
  73         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_ARP))
  74
  75 static inline __be16 vlan_proto(const struct sk_buff *skb)
  76 {
  77         if (skb_vlan_tag_present(skb))
  78                 return skb->protocol;
  79         else if (skb->protocol == htons(ETH_P_8021Q))
  80                 return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  81         else
  82                 return 0;
  83 }
  84
  85 static inline bool is_vlan_ip(const struct sk_buff *skb, const struct net *net)
  86 {
  87         struct brnf_net *brnet = net_generic(net, brnf_net_id);
  88
  89         return vlan_proto(skb) == htons(ETH_P_IP) && brnet->filter_vlan_tagged;
  90 }
  91
  92 static inline bool is_vlan_ipv6(const struct sk_buff *skb,
  93                                 const struct net *net)
  94 {
  95         struct brnf_net *brnet = net_generic(net, brnf_net_id);
  96
  97         return vlan_proto(skb) == htons(ETH_P_IPV6) &&
  98                brnet->filter_vlan_tagged;
  99 }
 100
 101 static inline bool is_vlan_arp(const struct sk_buff *skb, const struct net *net)
 102 {
 103         struct brnf_net *brnet = net_generic(net, brnf_net_id);
 104
 105         return vlan_proto(skb) == htons(ETH_P_ARP) && brnet->filter_vlan_tagged;
 106 }
 107
 108 static inline __be16 pppoe_proto(const struct sk_buff *skb)
 109 {
 110         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
 111                             sizeof(struct pppoe_hdr)));
 112 }
 113
 114 static inline bool is_pppoe_ip(const struct sk_buff *skb, const struct net *net)
 115 {
 116         struct brnf_net *brnet = net_generic(net, brnf_net_id);
 117
 118         return skb->protocol == htons(ETH_P_PPP_SES) &&
 119                pppoe_proto(skb) == htons(PPP_IP) && brnet->filter_pppoe_tagged;
 120 }
 121
 122 static inline bool is_pppoe_ipv6(const struct sk_buff *skb,
 123                                  const struct net *net)
 124 {
 125         struct brnf_net *brnet = net_generic(net, brnf_net_id);
 126
 127         return skb->protocol == htons(ETH_P_PPP_SES) &&
 128                pppoe_proto(skb) == htons(PPP_IPV6) &&
 129                brnet->filter_pppoe_tagged;
 130 }
 131
 132 /* largest possible L2 header, see br_nf_dev_queue_xmit() */
 133 #define NF_BRIDGE_MAX_MAC_HEADER_LENGTH (PPPOE_SES_HLEN + ETH_HLEN)
 134
 135 struct brnf_frag_data {
 136         char mac[NF_BRIDGE_MAX_MAC_HEADER_LENGTH];
 137         u8 encap_size;
 138         u8 size;
 139         u16 vlan_tci;
 140         __be16 vlan_proto;
 141 };
 142
 143 static DEFINE_PER_CPU(struct brnf_frag_data, brnf_frag_data_storage);
 144
 145 static void nf_bridge_info_free(struct sk_buff *skb)
 146 {
 147         skb_ext_del(skb, SKB_EXT_BRIDGE_NF);
 148 }
 149
 150 static inline struct net_device *bridge_parent(const struct net_device *dev)
 151 {
 152         struct net_bridge_port *port;
 153
 154         port = br_port_get_rcu(dev);
 155         return port ? port->br->dev : NULL;
 156 }
 157
 158 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 159 {
 160         return skb_ext_add(skb, SKB_EXT_BRIDGE_NF);
 161 }
 162
 163 unsigned int nf_bridge_encap_header_len(const struct sk_buff *skb)
 164 {
 165         switch (skb->protocol) {
 166         case __cpu_to_be16(ETH_P_8021Q):
 167                 return VLAN_HLEN;
 168         case __cpu_to_be16(ETH_P_PPP_SES):
 169                 return PPPOE_SES_HLEN;
 170         default:
 171                 return 0;
 172         }
 173 }
 174
 175 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 176 {
 177         unsigned int len = nf_bridge_encap_header_len(skb);
 178
 179         skb_pull(skb, len);
 180         skb->network_header += len;
 181 }
 182
 183 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 184 {
 185         unsigned int len = nf_bridge_encap_header_len(skb);
 186
 187         skb_pull_rcsum(skb, len);
 188         skb->network_header += len;
 189 }
 190
 191 /* When handing a packet over to the IP layer
 192  * check whether we have a skb that is in the
 193  * expected format
 194  */
 195
 196 static int br_validate_ipv4(struct net *net, struct sk_buff *skb)
 197 {
 198         const struct iphdr *iph;
 199         u32 len;
 200
 201         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 202                 goto inhdr_error;
 203
 204         iph = ip_hdr(skb);
 205
 206         /* Basic sanity checks */
 207         if (iph->ihl < 5 || iph->version != 4)
 208                 goto inhdr_error;
 209
 210         if (!pskb_may_pull(skb, iph->ihl*4))
 211                 goto inhdr_error;
 212
 213         iph = ip_hdr(skb);
 214         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 215                 goto csum_error;
 216
 217         len = skb_ip_totlen(skb);
 218         if (skb->len < len) {
 219                 __IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS);
 220                 goto drop;
 221         } else if (len < (iph->ihl*4))
 222                 goto inhdr_error;
 223
 224         if (pskb_trim_rcsum(skb, len)) {
 225                 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
 226                 goto drop;
 227         }
 228
 229         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 230         /* We should really parse IP options here but until
 231          * somebody who actually uses IP options complains to
 232          * us we'll just silently ignore the options because
 233          * we're lazy!
 234          */
 235         return 0;
 236
 237 csum_error:
 238         __IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS);
 239 inhdr_error:
 240         __IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
 241 drop:
 242         return -1;
 243 }
 244
 245 void nf_bridge_update_protocol(struct sk_buff *skb)
 246 {
 247         const struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 248
 249         switch (nf_bridge->orig_proto) {
 250         case BRNF_PROTO_8021Q:
 251                 skb->protocol = htons(ETH_P_8021Q);
 252                 break;
 253         case BRNF_PROTO_PPPOE:
 254                 skb->protocol = htons(ETH_P_PPP_SES);
 255                 break;
 256         case BRNF_PROTO_UNCHANGED:
 257                 break;
 258         }
 259 }
 260
 261 /* Obtain the correct destination MAC address, while preserving the original
 262  * source MAC address. If we already know this address, we just copy it. If we
 263  * don't, we use the neighbour framework to find out. In both cases, we make
 264  * sure that br_handle_frame_finish() is called afterwards.
 265  */
 266 int br_nf_pre_routing_finish_bridge(struct net *net, struct sock *sk, struct sk_buff *skb)
 267 {
 268         struct neighbour *neigh;
 269         struct dst_entry *dst;
 270
 271         skb->dev = bridge_parent(skb->dev);
 272         if (!skb->dev)
 273                 goto free_skb;
 274         dst = skb_dst(skb);
 275         neigh = dst_neigh_lookup_skb(dst, skb);
 276         if (neigh) {
 277                 struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 278                 int ret;
 279
 280                 if ((READ_ONCE(neigh->nud_state) & NUD_CONNECTED) &&
 281                     READ_ONCE(neigh->hh.hh_len)) {
 282                         neigh_hh_bridge(&neigh->hh, skb);
 283                         skb->dev = nf_bridge->physindev;
 284                         ret = br_handle_frame_finish(net, sk, skb);
 285                 } else {
 286                         /* the neighbour function below overwrites the complete
 287                          * MAC header, so we save the Ethernet source address and
 288                          * protocol number.
 289                          */
 290                         skb_copy_from_linear_data_offset(skb,
 291                                                          -(ETH_HLEN-ETH_ALEN),
 292                                                          nf_bridge->neigh_header,
 293                                                          ETH_HLEN-ETH_ALEN);
 294                         /* tell br_dev_xmit to continue with forwarding */
 295                         nf_bridge->bridged_dnat = 1;
 296                         /* FIXME Need to refragment */
 297                         ret = neigh->output(neigh, skb);
 298                 }
 299                 neigh_release(neigh);
 300                 return ret;
 301         }
 302 free_skb:
 303         kfree_skb(skb);
 304         return 0;
 305 }
 306
 307 static inline bool
 308 br_nf_ipv4_daddr_was_changed(const struct sk_buff *skb,
 309                              const struct nf_bridge_info *nf_bridge)
 310 {
 311         return ip_hdr(skb)->daddr != nf_bridge->ipv4_daddr;
 312 }
 313
 314 /* This requires some explaining. If DNAT has taken place,
 315  * we will need to fix up the destination Ethernet address.
 316  * This is also true when SNAT takes place (for the reply direction).
 317  *
 318  * There are two cases to consider:
 319  * 1. The packet was DNAT'ed to a device in the same bridge
 320  *    port group as it was received on. We can still bridge
 321  *    the packet.
 322  * 2. The packet was DNAT'ed to a different device, either
 323  *    a non-bridged device or another bridge port group.
 324  *    The packet will need to be routed.
 325  *
 326  * The correct way of distinguishing between these two cases is to
 327  * call ip_route_input() and to look at skb->dst->dev, which is
 328  * changed to the destination device if ip_route_input() succeeds.
 329  *
 330  * Let's first consider the case that ip_route_input() succeeds:
 331  *
 332  * If the output device equals the logical bridge device the packet
 333  * came in on, we can consider this bridging. The corresponding MAC
 334  * address will be obtained in br_nf_pre_routing_finish_bridge.
 335  * Otherwise, the packet is considered to be routed and we just
 336  * change the destination MAC address so that the packet will
 337  * later be passed up to the IP stack to be routed. For a redirected
 338  * packet, ip_route_input() will give back the localhost as output device,
 339  * which differs from the bridge device.
 340  *
 341  * Let's now consider the case that ip_route_input() fails:
 342  *
 343  * This can be because the destination address is martian, in which case
 344  * the packet will be dropped.
 345  * If IP forwarding is disabled, ip_route_input() will fail, while
 346  * ip_route_output_key() can return success. The source
 347  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 348  * thinks we're handling a locally generated packet and won't care
 349  * if IP forwarding is enabled. If the output device equals the logical bridge
 350  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 351  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 352  */
 353 static int br_nf_pre_routing_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
 354 {
 355         struct net_device *dev = skb->dev;
 356         struct iphdr *iph = ip_hdr(skb);
 357         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 358         struct rtable *rt;
 359         int err;
 360
 361         nf_bridge->frag_max_size = IPCB(skb)->frag_max_size;
 362
 363         if (nf_bridge->pkt_otherhost) {
 364                 skb->pkt_type = PACKET_OTHERHOST;
 365                 nf_bridge->pkt_otherhost = false;
 366         }
 367         nf_bridge->in_prerouting = 0;
 368         if (br_nf_ipv4_daddr_was_changed(skb, nf_bridge)) {
 369                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 370                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 371
 372                         /* If err equals -EHOSTUNREACH the error is due to a
 373                          * martian destination or due to the fact that
 374                          * forwarding is disabled. For most martian packets,
 375                          * ip_route_output_key() will fail. It won't fail for 2 types of
 376                          * martian destinations: loopback destinations and destination
 377                          * 0.0.0.0. In both cases the packet will be dropped because the
 378                          * destination is the loopback device and not the bridge. */
 379                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 380                                 goto free_skb;
 381
 382                         rt = ip_route_output(net, iph->daddr, 0,
 383                                              RT_TOS(iph->tos), 0);
 384                         if (!IS_ERR(rt)) {
 385                                 /* - Bridged-and-DNAT'ed traffic doesn't
 386                                  *   require ip_forwarding. */
 387                                 if (rt->dst.dev == dev) {
 388                                         skb_dst_drop(skb);
 389                                         skb_dst_set(skb, &rt->dst);
 390                                         goto bridged_dnat;
 391                                 }
 392                                 ip_rt_put(rt);
 393                         }
 394 free_skb:
 395                         kfree_skb(skb);
 396                         return 0;
 397                 } else {
 398                         if (skb_dst(skb)->dev == dev) {
 399 bridged_dnat:
 400                                 skb->dev = nf_bridge->physindev;
 401                                 nf_bridge_update_protocol(skb);
 402                                 nf_bridge_push_encap_header(skb);
 403                                 br_nf_hook_thresh(NF_BR_PRE_ROUTING,
 404                                                   net, sk, skb, skb->dev,
 405                                                   NULL,
 406                                                   br_nf_pre_routing_finish_bridge);
 407                                 return 0;
 408                         }
 409                         ether_addr_copy(eth_hdr(skb)->h_dest, dev->dev_addr);
 410                         skb->pkt_type = PACKET_HOST;
 411                 }
 412         } else {
 413                 rt = bridge_parent_rtable(nf_bridge->physindev);
 414                 if (!rt) {
 415                         kfree_skb(skb);
 416                         return 0;
 417                 }
 418                 skb_dst_drop(skb);
 419                 skb_dst_set_noref(skb, &rt->dst);
 420         }
 421
 422         skb->dev = nf_bridge->physindev;
 423         nf_bridge_update_protocol(skb);
 424         nf_bridge_push_encap_header(skb);
 425         br_nf_hook_thresh(NF_BR_PRE_ROUTING, net, sk, skb, skb->dev, NULL,
 426                           br_handle_frame_finish);
 427         return 0;
 428 }
 429
 430 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb,
 431                                                const struct net_device *dev,
 432                                                const struct net *net)
 433 {
 434         struct net_device *vlan, *br;
 435         struct brnf_net *brnet = net_generic(net, brnf_net_id);
 436
 437         br = bridge_parent(dev);
 438
 439         if (brnet->pass_vlan_indev == 0 || !skb_vlan_tag_present(skb))
 440                 return br;
 441
 442         vlan = __vlan_find_dev_deep_rcu(br, skb->vlan_proto,
 443                                     skb_vlan_tag_get(skb) & VLAN_VID_MASK);
 444
 445         return vlan ? vlan : br;
 446 }
 447
 448 /* Some common code for IPv4/IPv6 */
 449 struct net_device *setup_pre_routing(struct sk_buff *skb, const struct net *net)
 450 {
 451         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 452
 453         if (skb->pkt_type == PACKET_OTHERHOST) {
 454                 skb->pkt_type = PACKET_HOST;
 455                 nf_bridge->pkt_otherhost = true;
 456         }
 457
 458         nf_bridge->in_prerouting = 1;
 459         nf_bridge->physindev = skb->dev;
 460         skb->dev = brnf_get_logical_dev(skb, skb->dev, net);
 461
 462         if (skb->protocol == htons(ETH_P_8021Q))
 463                 nf_bridge->orig_proto = BRNF_PROTO_8021Q;
 464         else if (skb->protocol == htons(ETH_P_PPP_SES))
 465                 nf_bridge->orig_proto = BRNF_PROTO_PPPOE;
 466
 467         /* Must drop socket now because of tproxy. */
 468         skb_orphan(skb);
 469         return skb->dev;
 470 }
 471
 472 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 473  * Replicate the checks that IPv4 does on packet reception.
 474  * Set skb->dev to the bridge device (i.e. parent of the
 475  * receiving device) to make netfilter happy, the REDIRECT
 476  * target in particular.  Save the original destination IP
 477  * address to be able to detect DNAT afterwards. */
 478 static unsigned int br_nf_pre_routing(void *priv,
 479                                       struct sk_buff *skb,
 480                                       const struct nf_hook_state *state)
 481 {
 482         struct nf_bridge_info *nf_bridge;
 483         struct net_bridge_port *p;
 484         struct net_bridge *br;
 485         __u32 len = nf_bridge_encap_header_len(skb);
 486         struct brnf_net *brnet;
 487
 488         if (unlikely(!pskb_may_pull(skb, len)))
 489                 return NF_DROP;
 490
 491         p = br_port_get_rcu(state->in);
 492         if (p == NULL)
 493                 return NF_DROP;
 494         br = p->br;
 495
 496         brnet = net_generic(state->net, brnf_net_id);
 497         if (IS_IPV6(skb) || is_vlan_ipv6(skb, state->net) ||
 498             is_pppoe_ipv6(skb, state->net)) {
 499                 if (!brnet->call_ip6tables &&
 500                     !br_opt_get(br, BROPT_NF_CALL_IP6TABLES))
 501                         return NF_ACCEPT;
 502                 if (!ipv6_mod_enabled()) {
 503                         pr_warn_once("Module ipv6 is disabled, so call_ip6tables is not supported.");
 504                         return NF_DROP;
 505                 }
 506
 507                 nf_bridge_pull_encap_header_rcsum(skb);
 508                 return br_nf_pre_routing_ipv6(priv, skb, state);
 509         }
 510
 511         if (!brnet->call_iptables && !br_opt_get(br, BROPT_NF_CALL_IPTABLES))
 512                 return NF_ACCEPT;
 513
 514         if (!IS_IP(skb) && !is_vlan_ip(skb, state->net) &&
 515             !is_pppoe_ip(skb, state->net))
 516                 return NF_ACCEPT;
 517
 518         nf_bridge_pull_encap_header_rcsum(skb);
 519
 520         if (br_validate_ipv4(state->net, skb))
 521                 return NF_DROP;
 522
 523         if (!nf_bridge_alloc(skb))
 524                 return NF_DROP;
 525         if (!setup_pre_routing(skb, state->net))
 526                 return NF_DROP;
 527
 528         nf_bridge = nf_bridge_info_get(skb);
 529         nf_bridge->ipv4_daddr = ip_hdr(skb)->daddr;
 530
 531         skb->protocol = htons(ETH_P_IP);
 532         skb->transport_header = skb->network_header + ip_hdr(skb)->ihl * 4;
 533
 534         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, state->net, state->sk, skb,
 535                 skb->dev, NULL,
 536                 br_nf_pre_routing_finish);
 537
 538         return NF_STOLEN;
 539 }
 540
 541
 542 /* PF_BRIDGE/FORWARD *************************************************/
 543 static int br_nf_forward_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
 544 {
 545         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 546         struct net_device *in;
 547
 548         if (!IS_ARP(skb) && !is_vlan_arp(skb, net)) {
 549
 550                 if (skb->protocol == htons(ETH_P_IP))
 551                         nf_bridge->frag_max_size = IPCB(skb)->frag_max_size;
 552
 553                 if (skb->protocol == htons(ETH_P_IPV6))
 554                         nf_bridge->frag_max_size = IP6CB(skb)->frag_max_size;
 555
 556                 in = nf_bridge->physindev;
 557                 if (nf_bridge->pkt_otherhost) {
 558                         skb->pkt_type = PACKET_OTHERHOST;
 559                         nf_bridge->pkt_otherhost = false;
 560                 }
 561                 nf_bridge_update_protocol(skb);
 562         } else {
 563                 in = *((struct net_device **)(skb->cb));
 564         }
 565         nf_bridge_push_encap_header(skb);
 566
 567         br_nf_hook_thresh(NF_BR_FORWARD, net, sk, skb, in, skb->dev,
 568                           br_forward_finish);
 569         return 0;
 570 }
 571
 572
 573 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 574  * netfilter with indev and outdev set to the bridge device,
 575  * but we are still able to filter on the 'real' indev/outdev
 576  * because of the physdev module. For ARP, indev and outdev are the
 577  * bridge ports. */
 578 static unsigned int br_nf_forward_ip(void *priv,
 579                                      struct sk_buff *skb,
 580                                      const struct nf_hook_state *state)
 581 {
 582         struct nf_bridge_info *nf_bridge;
 583         struct net_device *parent;
 584         u_int8_t pf;
 585
 586         nf_bridge = nf_bridge_info_get(skb);
 587         if (!nf_bridge)
 588                 return NF_ACCEPT;
 589
 590         /* Need exclusive nf_bridge_info since we might have multiple
 591          * different physoutdevs. */
 592         if (!nf_bridge_unshare(skb))
 593                 return NF_DROP;
 594
 595         nf_bridge = nf_bridge_info_get(skb);
 596         if (!nf_bridge)
 597                 return NF_DROP;
 598
 599         parent = bridge_parent(state->out);
 600         if (!parent)
 601                 return NF_DROP;
 602
 603         if (IS_IP(skb) || is_vlan_ip(skb, state->net) ||
 604             is_pppoe_ip(skb, state->net))
 605                 pf = NFPROTO_IPV4;
 606         else if (IS_IPV6(skb) || is_vlan_ipv6(skb, state->net) ||
 607                  is_pppoe_ipv6(skb, state->net))
 608                 pf = NFPROTO_IPV6;
 609         else
 610                 return NF_ACCEPT;
 611
 612         nf_bridge_pull_encap_header(skb);
 613
 614         if (skb->pkt_type == PACKET_OTHERHOST) {
 615                 skb->pkt_type = PACKET_HOST;
 616                 nf_bridge->pkt_otherhost = true;
 617         }
 618
 619         if (pf == NFPROTO_IPV4) {
 620                 if (br_validate_ipv4(state->net, skb))
 621                         return NF_DROP;
 622                 IPCB(skb)->frag_max_size = nf_bridge->frag_max_size;
 623         }
 624
 625         if (pf == NFPROTO_IPV6) {
 626                 if (br_validate_ipv6(state->net, skb))
 627                         return NF_DROP;
 628                 IP6CB(skb)->frag_max_size = nf_bridge->frag_max_size;
 629         }
 630
 631         nf_bridge->physoutdev = skb->dev;
 632         if (pf == NFPROTO_IPV4)
 633                 skb->protocol = htons(ETH_P_IP);
 634         else
 635                 skb->protocol = htons(ETH_P_IPV6);
 636
 637         NF_HOOK(pf, NF_INET_FORWARD, state->net, NULL, skb,
 638                 brnf_get_logical_dev(skb, state->in, state->net),
 639                 parent, br_nf_forward_finish);
 640
 641         return NF_STOLEN;
 642 }
 643
 644 static unsigned int br_nf_forward_arp(void *priv,
 645                                       struct sk_buff *skb,
 646                                       const struct nf_hook_state *state)
 647 {
 648         struct net_bridge_port *p;
 649         struct net_bridge *br;
 650         struct net_device **d = (struct net_device **)(skb->cb);
 651         struct brnf_net *brnet;
 652
 653         p = br_port_get_rcu(state->out);
 654         if (p == NULL)
 655                 return NF_ACCEPT;
 656         br = p->br;
 657
 658         brnet = net_generic(state->net, brnf_net_id);
 659         if (!brnet->call_arptables && !br_opt_get(br, BROPT_NF_CALL_ARPTABLES))
 660                 return NF_ACCEPT;
 661
 662         if (!IS_ARP(skb)) {
 663                 if (!is_vlan_arp(skb, state->net))
 664                         return NF_ACCEPT;
 665                 nf_bridge_pull_encap_header(skb);
 666         }
 667
 668         if (unlikely(!pskb_may_pull(skb, sizeof(struct arphdr))))
 669                 return NF_DROP;
 670
 671         if (arp_hdr(skb)->ar_pln != 4) {
 672                 if (is_vlan_arp(skb, state->net))
 673                         nf_bridge_push_encap_header(skb);
 674                 return NF_ACCEPT;
 675         }
 676         *d = state->in;
 677         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, state->net, state->sk, skb,
 678                 state->in, state->out, br_nf_forward_finish);
 679
 680         return NF_STOLEN;
 681 }
 682
 683 static int br_nf_push_frag_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
 684 {
 685         struct brnf_frag_data *data;
 686         int err;
 687
 688         data = this_cpu_ptr(&brnf_frag_data_storage);
 689         err = skb_cow_head(skb, data->size);
 690
 691         if (err) {
 692                 kfree_skb(skb);
 693                 return 0;
 694         }
 695
 696         if (data->vlan_proto)
 697                 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci);
 698
 699         skb_copy_to_linear_data_offset(skb, -data->size, data->mac, data->size);
 700         __skb_push(skb, data->encap_size);
 701
 702         nf_bridge_info_free(skb);
 703         return br_dev_queue_push_xmit(net, sk, skb);
 704 }
 705
 706 static int
 707 br_nf_ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
 708                   int (*output)(struct net *, struct sock *, struct sk_buff *))
 709 {
 710         unsigned int mtu = ip_skb_dst_mtu(sk, skb);
 711         struct iphdr *iph = ip_hdr(skb);
 712
 713         if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->ignore_df) ||
 714                      (IPCB(skb)->frag_max_size &&
 715                       IPCB(skb)->frag_max_size > mtu))) {
 716                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
 717                 kfree_skb(skb);
 718                 return -EMSGSIZE;
 719         }
 720
 721         return ip_do_fragment(net, sk, skb, output);
 722 }
 723
 724 static unsigned int nf_bridge_mtu_reduction(const struct sk_buff *skb)
 725 {
 726         const struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 727
 728         if (nf_bridge->orig_proto == BRNF_PROTO_PPPOE)
 729                 return PPPOE_SES_HLEN;
 730         return 0;
 731 }
 732
 733 static int br_nf_dev_queue_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
 734 {
 735         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 736         unsigned int mtu, mtu_reserved;
 737
 738         mtu_reserved = nf_bridge_mtu_reduction(skb);
 739         mtu = skb->dev->mtu;
 740
 741         if (nf_bridge->pkt_otherhost) {
 742                 skb->pkt_type = PACKET_OTHERHOST;
 743                 nf_bridge->pkt_otherhost = false;
 744         }
 745
 746         if (nf_bridge->frag_max_size && nf_bridge->frag_max_size < mtu)
 747                 mtu = nf_bridge->frag_max_size;
 748
 749         nf_bridge_update_protocol(skb);
 750         nf_bridge_push_encap_header(skb);
 751
 752         if (skb_is_gso(skb) || skb->len + mtu_reserved <= mtu) {
 753                 nf_bridge_info_free(skb);
 754                 return br_dev_queue_push_xmit(net, sk, skb);
 755         }
 756
 757         /* This is wrong! We should preserve the original fragment
 758          * boundaries by preserving frag_list rather than refragmenting.
 759          */
 760         if (IS_ENABLED(CONFIG_NF_DEFRAG_IPV4) &&
 761             skb->protocol == htons(ETH_P_IP)) {
 762                 struct brnf_frag_data *data;
 763
 764                 if (br_validate_ipv4(net, skb))
 765                         goto drop;
 766
 767                 IPCB(skb)->frag_max_size = nf_bridge->frag_max_size;
 768
 769                 data = this_cpu_ptr(&brnf_frag_data_storage);
 770
 771                 if (skb_vlan_tag_present(skb)) {
 772                         data->vlan_tci = skb->vlan_tci;
 773                         data->vlan_proto = skb->vlan_proto;
 774                 } else {
 775                         data->vlan_proto = 0;
 776                 }
 777
 778                 data->encap_size = nf_bridge_encap_header_len(skb);
 779                 data->size = ETH_HLEN + data->encap_size;
 780
 781                 skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
 782                                                  data->size);
 783
 784                 return br_nf_ip_fragment(net, sk, skb, br_nf_push_frag_xmit);
 785         }
 786         if (IS_ENABLED(CONFIG_NF_DEFRAG_IPV6) &&
 787             skb->protocol == htons(ETH_P_IPV6)) {
 788                 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
 789                 struct brnf_frag_data *data;
 790
 791                 if (br_validate_ipv6(net, skb))
 792                         goto drop;
 793
 794                 IP6CB(skb)->frag_max_size = nf_bridge->frag_max_size;
 795
 796                 data = this_cpu_ptr(&brnf_frag_data_storage);
 797                 data->encap_size = nf_bridge_encap_header_len(skb);
 798                 data->size = ETH_HLEN + data->encap_size;
 799
 800                 skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
 801                                                  data->size);
 802
 803                 if (v6ops)
 804                         return v6ops->fragment(net, sk, skb, br_nf_push_frag_xmit);
 805
 806                 kfree_skb(skb);
 807                 return -EMSGSIZE;
 808         }
 809         nf_bridge_info_free(skb);
 810         return br_dev_queue_push_xmit(net, sk, skb);
 811  drop:
 812         kfree_skb(skb);
 813         return 0;
 814 }
 815
 816 /* PF_BRIDGE/POST_ROUTING ********************************************/
 817 static unsigned int br_nf_post_routing(void *priv,
 818                                        struct sk_buff *skb,
 819                                        const struct nf_hook_state *state)
 820 {
 821         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 822         struct net_device *realoutdev = bridge_parent(skb->dev);
 823         u_int8_t pf;
 824
 825         /* if nf_bridge is set, but ->physoutdev is NULL, this packet came in
 826          * on a bridge, but was delivered locally and is now being routed:
 827          *
 828          * POST_ROUTING was already invoked from the ip stack.
 829          */
 830         if (!nf_bridge || !nf_bridge->physoutdev)
 831                 return NF_ACCEPT;
 832
 833         if (!realoutdev)
 834                 return NF_DROP;
 835
 836         if (IS_IP(skb) || is_vlan_ip(skb, state->net) ||
 837             is_pppoe_ip(skb, state->net))
 838                 pf = NFPROTO_IPV4;
 839         else if (IS_IPV6(skb) || is_vlan_ipv6(skb, state->net) ||
 840                  is_pppoe_ipv6(skb, state->net))
 841                 pf = NFPROTO_IPV6;
 842         else
 843                 return NF_ACCEPT;
 844
 845         if (skb->pkt_type == PACKET_OTHERHOST) {
 846                 skb->pkt_type = PACKET_HOST;
 847                 nf_bridge->pkt_otherhost = true;
 848         }
 849
 850         nf_bridge_pull_encap_header(skb);
 851         if (pf == NFPROTO_IPV4)
 852                 skb->protocol = htons(ETH_P_IP);
 853         else
 854                 skb->protocol = htons(ETH_P_IPV6);
 855
 856         NF_HOOK(pf, NF_INET_POST_ROUTING, state->net, state->sk, skb,
 857                 NULL, realoutdev,
 858                 br_nf_dev_queue_xmit);
 859
 860         return NF_STOLEN;
 861 }
 862
 863 /* IP/SABOTAGE *****************************************************/
 864 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 865  * for the second time. */
 866 static unsigned int ip_sabotage_in(void *priv,
 867                                    struct sk_buff *skb,
 868                                    const struct nf_hook_state *state)
 869 {
 870         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 871
 872         if (nf_bridge) {
 873                 if (nf_bridge->sabotage_in_done)
 874                         return NF_ACCEPT;
 875
 876                 if (!nf_bridge->in_prerouting &&
 877                     !netif_is_l3_master(skb->dev) &&
 878                     !netif_is_l3_slave(skb->dev)) {
 879                         nf_bridge->sabotage_in_done = 1;
 880                         state->okfn(state->net, state->sk, skb);
 881                         return NF_STOLEN;
 882                 }
 883         }
 884
 885         return NF_ACCEPT;
 886 }
 887
 888 /* This is called when br_netfilter has called into iptables/netfilter,
 889  * and DNAT has taken place on a bridge-forwarded packet.
 890  *
 891  * neigh->output has created a new MAC header, with local br0 MAC
 892  * as saddr.
 893  *
 894  * This restores the original MAC saddr of the bridged packet
 895  * before invoking bridge forward logic to transmit the packet.
 896  */
 897 static void br_nf_pre_routing_finish_bridge_slow(struct sk_buff *skb)
 898 {
 899         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 900
 901         skb_pull(skb, ETH_HLEN);
 902         nf_bridge->bridged_dnat = 0;
 903
 904         BUILD_BUG_ON(sizeof(nf_bridge->neigh_header) != (ETH_HLEN - ETH_ALEN));
 905
 906         skb_copy_to_linear_data_offset(skb, -(ETH_HLEN - ETH_ALEN),
 907                                        nf_bridge->neigh_header,
 908                                        ETH_HLEN - ETH_ALEN);
 909         skb->dev = nf_bridge->physindev;
 910
 911         nf_bridge->physoutdev = NULL;
 912         br_handle_frame_finish(dev_net(skb->dev), NULL, skb);
 913 }
 914
 915 static int br_nf_dev_xmit(struct sk_buff *skb)
 916 {
 917         const struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 918
 919         if (nf_bridge && nf_bridge->bridged_dnat) {
 920                 br_nf_pre_routing_finish_bridge_slow(skb);
 921                 return 1;
 922         }
 923         return 0;
 924 }
 925
 926 static const struct nf_br_ops br_ops = {
 927         .br_dev_xmit_hook =     br_nf_dev_xmit,
 928 };
 929
 930 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 931  * br_dev_queue_push_xmit is called afterwards */
 932 static const struct nf_hook_ops br_nf_ops[] = {
 933         {
 934                 .hook = br_nf_pre_routing,
 935                 .pf = NFPROTO_BRIDGE,
 936                 .hooknum = NF_BR_PRE_ROUTING,
 937                 .priority = NF_BR_PRI_BRNF,
 938         },
 939         {
 940                 .hook = br_nf_forward_ip,
 941                 .pf = NFPROTO_BRIDGE,
 942                 .hooknum = NF_BR_FORWARD,
 943                 .priority = NF_BR_PRI_BRNF - 1,
 944         },
 945         {
 946                 .hook = br_nf_forward_arp,
 947                 .pf = NFPROTO_BRIDGE,
 948                 .hooknum = NF_BR_FORWARD,
 949                 .priority = NF_BR_PRI_BRNF,
 950         },
 951         {
 952                 .hook = br_nf_post_routing,
 953                 .pf = NFPROTO_BRIDGE,
 954                 .hooknum = NF_BR_POST_ROUTING,
 955                 .priority = NF_BR_PRI_LAST,
 956         },
 957         {
 958                 .hook = ip_sabotage_in,
 959                 .pf = NFPROTO_IPV4,
 960                 .hooknum = NF_INET_PRE_ROUTING,
 961                 .priority = NF_IP_PRI_FIRST,
 962         },
 963         {
 964                 .hook = ip_sabotage_in,
 965                 .pf = NFPROTO_IPV6,
 966                 .hooknum = NF_INET_PRE_ROUTING,
 967                 .priority = NF_IP6_PRI_FIRST,
 968         },
 969 };
 970
 971 static int brnf_device_event(struct notifier_block *unused, unsigned long event,
 972                              void *ptr)
 973 {
 974         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 975         struct brnf_net *brnet;
 976         struct net *net;
 977         int ret;
 978
 979         if (event != NETDEV_REGISTER || !netif_is_bridge_master(dev))
 980                 return NOTIFY_DONE;
 981
 982         ASSERT_RTNL();
 983
 984         net = dev_net(dev);
 985         brnet = net_generic(net, brnf_net_id);
 986         if (brnet->enabled)
 987                 return NOTIFY_OK;
 988
 989         ret = nf_register_net_hooks(net, br_nf_ops, ARRAY_SIZE(br_nf_ops));
 990         if (ret)
 991                 return NOTIFY_BAD;
 992
 993         brnet->enabled = true;
 994         return NOTIFY_OK;
 995 }
 996
 997 static struct notifier_block brnf_notifier __read_mostly = {
 998         .notifier_call = brnf_device_event,
 999 };
1000
1001 /* recursively invokes nf_hook_slow (again), skipping already-called
1002  * hooks (< NF_BR_PRI_BRNF).
1003  *
1004  * Called with rcu read lock held.
1005  */
1006 int br_nf_hook_thresh(unsigned int hook, struct net *net,
1007                       struct sock *sk, struct sk_buff *skb,
1008                       struct net_device *indev,
1009                       struct net_device *outdev,
1010                       int (*okfn)(struct net *, struct sock *,
1011                                   struct sk_buff *))
1012 {
1013         const struct nf_hook_entries *e;
1014         struct nf_hook_state state;
1015         struct nf_hook_ops **ops;
1016         unsigned int i;
1017         int ret;
1018
1019         e = rcu_dereference(net->nf.hooks_bridge[hook]);
1020         if (!e)
1021                 return okfn(net, sk, skb);
1022
1023         ops = nf_hook_entries_get_hook_ops(e);
1024         for (i = 0; i < e->num_hook_entries; i++) {
1025                 /* These hooks have already been called */
1026                 if (ops[i]->priority < NF_BR_PRI_BRNF)
1027                         continue;
1028
1029                 /* These hooks have not been called yet, run them. */
1030                 if (ops[i]->priority > NF_BR_PRI_BRNF)
1031                         break;
1032
1033                 /* take a closer look at NF_BR_PRI_BRNF. */
1034                 if (ops[i]->hook == br_nf_pre_routing) {
1035                         /* This hook diverted the skb to this function,
1036                          * hooks after this have not been run yet.
1037                          */
1038                         i++;
1039                         break;
1040                 }
1041         }
1042
1043         nf_hook_state_init(&state, hook, NFPROTO_BRIDGE, indev, outdev,
1044                            sk, net, okfn);
1045
1046         ret = nf_hook_slow(skb, &state, e, i);
1047         if (ret == 1)
1048                 ret = okfn(net, sk, skb);
1049
1050         return ret;
1051 }
1052
1053 #ifdef CONFIG_SYSCTL
1054 static
1055 int brnf_sysctl_call_tables(struct ctl_table *ctl, int write,
1056                             void *buffer, size_t *lenp, loff_t *ppos)
1057 {
1058         int ret;
1059
1060         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1061
1062         if (write && *(int *)(ctl->data))
1063                 *(int *)(ctl->data) = 1;
1064         return ret;
1065 }
1066
1067 static struct ctl_table brnf_table[] = {
1068         {
1069                 .procname       = "bridge-nf-call-arptables",
1070                 .maxlen         = sizeof(int),
1071                 .mode           = 0644,
1072                 .proc_handler   = brnf_sysctl_call_tables,
1073         },
1074         {
1075                 .procname       = "bridge-nf-call-iptables",
1076                 .maxlen         = sizeof(int),
1077                 .mode           = 0644,
1078                 .proc_handler   = brnf_sysctl_call_tables,
1079         },
1080         {
1081                 .procname       = "bridge-nf-call-ip6tables",
1082                 .maxlen         = sizeof(int),
1083                 .mode           = 0644,
1084                 .proc_handler   = brnf_sysctl_call_tables,
1085         },
1086         {
1087                 .procname       = "bridge-nf-filter-vlan-tagged",
1088                 .maxlen         = sizeof(int),
1089                 .mode           = 0644,
1090                 .proc_handler   = brnf_sysctl_call_tables,
1091         },
1092         {
1093                 .procname       = "bridge-nf-filter-pppoe-tagged",
1094                 .maxlen         = sizeof(int),
1095                 .mode           = 0644,
1096                 .proc_handler   = brnf_sysctl_call_tables,
1097         },
1098         {
1099                 .procname       = "bridge-nf-pass-vlan-input-dev",
1100                 .maxlen         = sizeof(int),
1101                 .mode           = 0644,
1102                 .proc_handler   = brnf_sysctl_call_tables,
1103         },
1104         { }
1105 };
1106
1107 static inline void br_netfilter_sysctl_default(struct brnf_net *brnf)
1108 {
1109         brnf->call_iptables = 1;
1110         brnf->call_ip6tables = 1;
1111         brnf->call_arptables = 1;
1112         brnf->filter_vlan_tagged = 0;
1113         brnf->filter_pppoe_tagged = 0;
1114         brnf->pass_vlan_indev = 0;
1115 }
1116
1117 static int br_netfilter_sysctl_init_net(struct net *net)
1118 {
1119         struct ctl_table *table = brnf_table;
1120         struct brnf_net *brnet;
1121
1122         if (!net_eq(net, &init_net)) {
1123                 table = kmemdup(table, sizeof(brnf_table), GFP_KERNEL);
1124                 if (!table)
1125                         return -ENOMEM;
1126         }
1127
1128         brnet = net_generic(net, brnf_net_id);
1129         table[0].data = &brnet->call_arptables;
1130         table[1].data = &brnet->call_iptables;
1131         table[2].data = &brnet->call_ip6tables;
1132         table[3].data = &brnet->filter_vlan_tagged;
1133         table[4].data = &brnet->filter_pppoe_tagged;
1134         table[5].data = &brnet->pass_vlan_indev;
1135
1136         br_netfilter_sysctl_default(brnet);
1137
1138         brnet->ctl_hdr = register_net_sysctl(net, "net/bridge", table);
1139         if (!brnet->ctl_hdr) {
1140                 if (!net_eq(net, &init_net))
1141                         kfree(table);
1142
1143                 return -ENOMEM;
1144         }
1145
1146         return 0;
1147 }
1148
1149 static void br_netfilter_sysctl_exit_net(struct net *net,
1150                                          struct brnf_net *brnet)
1151 {
1152         struct ctl_table *table = brnet->ctl_hdr->ctl_table_arg;
1153
1154         unregister_net_sysctl_table(brnet->ctl_hdr);
1155         if (!net_eq(net, &init_net))
1156                 kfree(table);
1157 }
1158
1159 static int __net_init brnf_init_net(struct net *net)
1160 {
1161         return br_netfilter_sysctl_init_net(net);
1162 }
1163 #endif
1164
1165 static void __net_exit brnf_exit_net(struct net *net)
1166 {
1167         struct brnf_net *brnet;
1168
1169         brnet = net_generic(net, brnf_net_id);
1170         if (brnet->enabled) {
1171                 nf_unregister_net_hooks(net, br_nf_ops, ARRAY_SIZE(br_nf_ops));
1172                 brnet->enabled = false;
1173         }
1174
1175 #ifdef CONFIG_SYSCTL
1176         br_netfilter_sysctl_exit_net(net, brnet);
1177 #endif
1178 }
1179
1180 static struct pernet_operations brnf_net_ops __read_mostly = {
1181 #ifdef CONFIG_SYSCTL
1182         .init = brnf_init_net,
1183 #endif
1184         .exit = brnf_exit_net,
1185         .id   = &brnf_net_id,
1186         .size = sizeof(struct brnf_net),
1187 };
1188
1189 static int __init br_netfilter_init(void)
1190 {
1191         int ret;
1192
1193         ret = register_pernet_subsys(&brnf_net_ops);
1194         if (ret < 0)
1195                 return ret;
1196
1197         ret = register_netdevice_notifier(&brnf_notifier);
1198         if (ret < 0) {
1199                 unregister_pernet_subsys(&brnf_net_ops);
1200                 return ret;
1201         }
1202
1203         RCU_INIT_POINTER(nf_br_ops, &br_ops);
1204         printk(KERN_NOTICE "Bridge firewalling registered\n");
1205         return 0;
1206 }
1207
1208 static void __exit br_netfilter_fini(void)
1209 {
1210         RCU_INIT_POINTER(nf_br_ops, NULL);
1211         unregister_netdevice_notifier(&brnf_notifier);
1212         unregister_pernet_subsys(&brnf_net_ops);
1213 }
1214
1215 module_init(br_netfilter_init);
1216 module_exit(br_netfilter_fini);
1217
1218 MODULE_LICENSE("GPL");
1219 MODULE_AUTHOR("Lennert Buytenhek <buytenh@gnu.org>");
1220 MODULE_AUTHOR("Bart De Schuymer <bdschuym@pandora.be>");
1221 MODULE_DESCRIPTION("Linux ethernet netfilter firewall bridge");