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 = ntohs(iph->tot_len);
 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 ((neigh->nud_state & NUD_CONNECTED) && neigh->hh.hh_len) {
 281                         neigh_hh_bridge(&neigh->hh, skb);
 282                         skb->dev = nf_bridge->physindev;
 283                         ret = br_handle_frame_finish(net, sk, skb);
 284                 } else {
 285                         /* the neighbour function below overwrites the complete
 286                          * MAC header, so we save the Ethernet source address and
 287                          * protocol number.
 288                          */
 289                         skb_copy_from_linear_data_offset(skb,
 290                                                          -(ETH_HLEN-ETH_ALEN),
 291                                                          nf_bridge->neigh_header,
 292                                                          ETH_HLEN-ETH_ALEN);
 293                         /* tell br_dev_xmit to continue with forwarding */
 294                         nf_bridge->bridged_dnat = 1;
 295                         /* FIXME Need to refragment */
 296                         ret = neigh->output(neigh, skb);
 297                 }
 298                 neigh_release(neigh);
 299                 return ret;
 300         }
 301 free_skb:
 302         kfree_skb(skb);
 303         return 0;
 304 }
 305
 306 static inline bool
 307 br_nf_ipv4_daddr_was_changed(const struct sk_buff *skb,
 308                              const struct nf_bridge_info *nf_bridge)
 309 {
 310         return ip_hdr(skb)->daddr != nf_bridge->ipv4_daddr;
 311 }
 312
 313 /* This requires some explaining. If DNAT has taken place,
 314  * we will need to fix up the destination Ethernet address.
 315  * This is also true when SNAT takes place (for the reply direction).
 316  *
 317  * There are two cases to consider:
 318  * 1. The packet was DNAT'ed to a device in the same bridge
 319  *    port group as it was received on. We can still bridge
 320  *    the packet.
 321  * 2. The packet was DNAT'ed to a different device, either
 322  *    a non-bridged device or another bridge port group.
 323  *    The packet will need to be routed.
 324  *
 325  * The correct way of distinguishing between these two cases is to
 326  * call ip_route_input() and to look at skb->dst->dev, which is
 327  * changed to the destination device if ip_route_input() succeeds.
 328  *
 329  * Let's first consider the case that ip_route_input() succeeds:
 330  *
 331  * If the output device equals the logical bridge device the packet
 332  * came in on, we can consider this bridging. The corresponding MAC
 333  * address will be obtained in br_nf_pre_routing_finish_bridge.
 334  * Otherwise, the packet is considered to be routed and we just
 335  * change the destination MAC address so that the packet will
 336  * later be passed up to the IP stack to be routed. For a redirected
 337  * packet, ip_route_input() will give back the localhost as output device,
 338  * which differs from the bridge device.
 339  *
 340  * Let's now consider the case that ip_route_input() fails:
 341  *
 342  * This can be because the destination address is martian, in which case
 343  * the packet will be dropped.
 344  * If IP forwarding is disabled, ip_route_input() will fail, while
 345  * ip_route_output_key() can return success. The source
 346  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 347  * thinks we're handling a locally generated packet and won't care
 348  * if IP forwarding is enabled. If the output device equals the logical bridge
 349  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 350  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 351  */
 352 static int br_nf_pre_routing_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
 353 {
 354         struct net_device *dev = skb->dev;
 355         struct iphdr *iph = ip_hdr(skb);
 356         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 357         struct rtable *rt;
 358         int err;
 359
 360         nf_bridge->frag_max_size = IPCB(skb)->frag_max_size;
 361
 362         if (nf_bridge->pkt_otherhost) {
 363                 skb->pkt_type = PACKET_OTHERHOST;
 364                 nf_bridge->pkt_otherhost = false;
 365         }
 366         nf_bridge->in_prerouting = 0;
 367         if (br_nf_ipv4_daddr_was_changed(skb, nf_bridge)) {
 368                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 369                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 370
 371                         /* If err equals -EHOSTUNREACH the error is due to a
 372                          * martian destination or due to the fact that
 373                          * forwarding is disabled. For most martian packets,
 374                          * ip_route_output_key() will fail. It won't fail for 2 types of
 375                          * martian destinations: loopback destinations and destination
 376                          * 0.0.0.0. In both cases the packet will be dropped because the
 377                          * destination is the loopback device and not the bridge. */
 378                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 379                                 goto free_skb;
 380
 381                         rt = ip_route_output(net, iph->daddr, 0,
 382                                              RT_TOS(iph->tos), 0);
 383                         if (!IS_ERR(rt)) {
 384                                 /* - Bridged-and-DNAT'ed traffic doesn't
 385                                  *   require ip_forwarding. */
 386                                 if (rt->dst.dev == dev) {
 387                                         skb_dst_drop(skb);
 388                                         skb_dst_set(skb, &rt->dst);
 389                                         goto bridged_dnat;
 390                                 }
 391                                 ip_rt_put(rt);
 392                         }
 393 free_skb:
 394                         kfree_skb(skb);
 395                         return 0;
 396                 } else {
 397                         if (skb_dst(skb)->dev == dev) {
 398 bridged_dnat:
 399                                 skb->dev = nf_bridge->physindev;
 400                                 nf_bridge_update_protocol(skb);
 401                                 nf_bridge_push_encap_header(skb);
 402                                 br_nf_hook_thresh(NF_BR_PRE_ROUTING,
 403                                                   net, sk, skb, skb->dev,
 404                                                   NULL,
 405                                                   br_nf_pre_routing_finish_bridge);
 406                                 return 0;
 407                         }
 408                         ether_addr_copy(eth_hdr(skb)->h_dest, dev->dev_addr);
 409                         skb->pkt_type = PACKET_HOST;
 410                 }
 411         } else {
 412                 rt = bridge_parent_rtable(nf_bridge->physindev);
 413                 if (!rt) {
 414                         kfree_skb(skb);
 415                         return 0;
 416                 }
 417                 skb_dst_drop(skb);
 418                 skb_dst_set_noref(skb, &rt->dst);
 419         }
 420
 421         skb->dev = nf_bridge->physindev;
 422         nf_bridge_update_protocol(skb);
 423         nf_bridge_push_encap_header(skb);
 424         br_nf_hook_thresh(NF_BR_PRE_ROUTING, net, sk, skb, skb->dev, NULL,
 425                           br_handle_frame_finish);
 426         return 0;
 427 }
 428
 429 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb,
 430                                                const struct net_device *dev,
 431                                                const struct net *net)
 432 {
 433         struct net_device *vlan, *br;
 434         struct brnf_net *brnet = net_generic(net, brnf_net_id);
 435
 436         br = bridge_parent(dev);
 437
 438         if (brnet->pass_vlan_indev == 0 || !skb_vlan_tag_present(skb))
 439                 return br;
 440
 441         vlan = __vlan_find_dev_deep_rcu(br, skb->vlan_proto,
 442                                     skb_vlan_tag_get(skb) & VLAN_VID_MASK);
 443
 444         return vlan ? vlan : br;
 445 }
 446
 447 /* Some common code for IPv4/IPv6 */
 448 struct net_device *setup_pre_routing(struct sk_buff *skb, const struct net *net)
 449 {
 450         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 451
 452         if (skb->pkt_type == PACKET_OTHERHOST) {
 453                 skb->pkt_type = PACKET_HOST;
 454                 nf_bridge->pkt_otherhost = true;
 455         }
 456
 457         nf_bridge->in_prerouting = 1;
 458         nf_bridge->physindev = skb->dev;
 459         skb->dev = brnf_get_logical_dev(skb, skb->dev, net);
 460
 461         if (skb->protocol == htons(ETH_P_8021Q))
 462                 nf_bridge->orig_proto = BRNF_PROTO_8021Q;
 463         else if (skb->protocol == htons(ETH_P_PPP_SES))
 464                 nf_bridge->orig_proto = BRNF_PROTO_PPPOE;
 465
 466         /* Must drop socket now because of tproxy. */
 467         skb_orphan(skb);
 468         return skb->dev;
 469 }
 470
 471 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 472  * Replicate the checks that IPv4 does on packet reception.
 473  * Set skb->dev to the bridge device (i.e. parent of the
 474  * receiving device) to make netfilter happy, the REDIRECT
 475  * target in particular.  Save the original destination IP
 476  * address to be able to detect DNAT afterwards. */
 477 static unsigned int br_nf_pre_routing(void *priv,
 478                                       struct sk_buff *skb,
 479                                       const struct nf_hook_state *state)
 480 {
 481         struct nf_bridge_info *nf_bridge;
 482         struct net_bridge_port *p;
 483         struct net_bridge *br;
 484         __u32 len = nf_bridge_encap_header_len(skb);
 485         struct brnf_net *brnet;
 486
 487         if (unlikely(!pskb_may_pull(skb, len)))
 488                 return NF_DROP;
 489
 490         p = br_port_get_rcu(state->in);
 491         if (p == NULL)
 492                 return NF_DROP;
 493         br = p->br;
 494
 495         brnet = net_generic(state->net, brnf_net_id);
 496         if (IS_IPV6(skb) || is_vlan_ipv6(skb, state->net) ||
 497             is_pppoe_ipv6(skb, state->net)) {
 498                 if (!brnet->call_ip6tables &&
 499                     !br_opt_get(br, BROPT_NF_CALL_IP6TABLES))
 500                         return NF_ACCEPT;
 501                 if (!ipv6_mod_enabled()) {
 502                         pr_warn_once("Module ipv6 is disabled, so call_ip6tables is not supported.");
 503                         return NF_DROP;
 504                 }
 505
 506                 nf_bridge_pull_encap_header_rcsum(skb);
 507                 return br_nf_pre_routing_ipv6(priv, skb, state);
 508         }
 509
 510         if (!brnet->call_iptables && !br_opt_get(br, BROPT_NF_CALL_IPTABLES))
 511                 return NF_ACCEPT;
 512
 513         if (!IS_IP(skb) && !is_vlan_ip(skb, state->net) &&
 514             !is_pppoe_ip(skb, state->net))
 515                 return NF_ACCEPT;
 516
 517         nf_bridge_pull_encap_header_rcsum(skb);
 518
 519         if (br_validate_ipv4(state->net, skb))
 520                 return NF_DROP;
 521
 522         if (!nf_bridge_alloc(skb))
 523                 return NF_DROP;
 524         if (!setup_pre_routing(skb, state->net))
 525                 return NF_DROP;
 526
 527         nf_bridge = nf_bridge_info_get(skb);
 528         nf_bridge->ipv4_daddr = ip_hdr(skb)->daddr;
 529
 530         skb->protocol = htons(ETH_P_IP);
 531         skb->transport_header = skb->network_header + ip_hdr(skb)->ihl * 4;
 532
 533         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, state->net, state->sk, skb,
 534                 skb->dev, NULL,
 535                 br_nf_pre_routing_finish);
 536
 537         return NF_STOLEN;
 538 }
 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, net)) {
 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         br_nf_hook_thresh(NF_BR_FORWARD, net, sk, skb, in, skb->dev,
 567                           br_forward_finish);
 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         nf_bridge = nf_bridge_info_get(skb);
 586         if (!nf_bridge)
 587                 return NF_ACCEPT;
 588
 589         /* Need exclusive nf_bridge_info since we might have multiple
 590          * different physoutdevs. */
 591         if (!nf_bridge_unshare(skb))
 592                 return NF_DROP;
 593
 594         nf_bridge = nf_bridge_info_get(skb);
 595         if (!nf_bridge)
 596                 return NF_DROP;
 597
 598         parent = bridge_parent(state->out);
 599         if (!parent)
 600                 return NF_DROP;
 601
 602         if (IS_IP(skb) || is_vlan_ip(skb, state->net) ||
 603             is_pppoe_ip(skb, state->net))
 604                 pf = NFPROTO_IPV4;
 605         else if (IS_IPV6(skb) || is_vlan_ipv6(skb, state->net) ||
 606                  is_pppoe_ipv6(skb, state->net))
 607                 pf = NFPROTO_IPV6;
 608         else
 609                 return NF_ACCEPT;
 610
 611         nf_bridge_pull_encap_header(skb);
 612
 613         if (skb->pkt_type == PACKET_OTHERHOST) {
 614                 skb->pkt_type = PACKET_HOST;
 615                 nf_bridge->pkt_otherhost = true;
 616         }
 617
 618         if (pf == NFPROTO_IPV4) {
 619                 if (br_validate_ipv4(state->net, skb))
 620                         return NF_DROP;
 621                 IPCB(skb)->frag_max_size = nf_bridge->frag_max_size;
 622         }
 623
 624         if (pf == NFPROTO_IPV6) {
 625                 if (br_validate_ipv6(state->net, skb))
 626                         return NF_DROP;
 627                 IP6CB(skb)->frag_max_size = nf_bridge->frag_max_size;
 628         }
 629
 630         nf_bridge->physoutdev = skb->dev;
 631         if (pf == NFPROTO_IPV4)
 632                 skb->protocol = htons(ETH_P_IP);
 633         else
 634                 skb->protocol = htons(ETH_P_IPV6);
 635
 636         NF_HOOK(pf, NF_INET_FORWARD, state->net, NULL, skb,
 637                 brnf_get_logical_dev(skb, state->in, state->net),
 638                 parent, br_nf_forward_finish);
 639
 640         return NF_STOLEN;
 641 }
 642
 643 static unsigned int br_nf_forward_arp(void *priv,
 644                                       struct sk_buff *skb,
 645                                       const struct nf_hook_state *state)
 646 {
 647         struct net_bridge_port *p;
 648         struct net_bridge *br;
 649         struct net_device **d = (struct net_device **)(skb->cb);
 650         struct brnf_net *brnet;
 651
 652         p = br_port_get_rcu(state->out);
 653         if (p == NULL)
 654                 return NF_ACCEPT;
 655         br = p->br;
 656
 657         brnet = net_generic(state->net, brnf_net_id);
 658         if (!brnet->call_arptables && !br_opt_get(br, BROPT_NF_CALL_ARPTABLES))
 659                 return NF_ACCEPT;
 660
 661         if (!IS_ARP(skb)) {
 662                 if (!is_vlan_arp(skb, state->net))
 663                         return NF_ACCEPT;
 664                 nf_bridge_pull_encap_header(skb);
 665         }
 666
 667         if (unlikely(!pskb_may_pull(skb, sizeof(struct arphdr))))
 668                 return NF_DROP;
 669
 670         if (arp_hdr(skb)->ar_pln != 4) {
 671                 if (is_vlan_arp(skb, state->net))
 672                         nf_bridge_push_encap_header(skb);
 673                 return NF_ACCEPT;
 674         }
 675         *d = state->in;
 676         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, state->net, state->sk, skb,
 677                 state->in, state->out, br_nf_forward_finish);
 678
 679         return NF_STOLEN;
 680 }
 681
 682 static int br_nf_push_frag_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
 683 {
 684         struct brnf_frag_data *data;
 685         int err;
 686
 687         data = this_cpu_ptr(&brnf_frag_data_storage);
 688         err = skb_cow_head(skb, data->size);
 689
 690         if (err) {
 691                 kfree_skb(skb);
 692                 return 0;
 693         }
 694
 695         if (data->vlan_proto)
 696                 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci);
 697
 698         skb_copy_to_linear_data_offset(skb, -data->size, data->mac, data->size);
 699         __skb_push(skb, data->encap_size);
 700
 701         nf_bridge_info_free(skb);
 702         return br_dev_queue_push_xmit(net, sk, skb);
 703 }
 704
 705 static int
 706 br_nf_ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
 707                   int (*output)(struct net *, struct sock *, struct sk_buff *))
 708 {
 709         unsigned int mtu = ip_skb_dst_mtu(sk, skb);
 710         struct iphdr *iph = ip_hdr(skb);
 711
 712         if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->ignore_df) ||
 713                      (IPCB(skb)->frag_max_size &&
 714                       IPCB(skb)->frag_max_size > mtu))) {
 715                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
 716                 kfree_skb(skb);
 717                 return -EMSGSIZE;
 718         }
 719
 720         return ip_do_fragment(net, sk, skb, output);
 721 }
 722
 723 static unsigned int nf_bridge_mtu_reduction(const struct sk_buff *skb)
 724 {
 725         const struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 726
 727         if (nf_bridge->orig_proto == BRNF_PROTO_PPPOE)
 728                 return PPPOE_SES_HLEN;
 729         return 0;
 730 }
 731
 732 static int br_nf_dev_queue_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
 733 {
 734         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 735         unsigned int mtu, mtu_reserved;
 736
 737         mtu_reserved = nf_bridge_mtu_reduction(skb);
 738         mtu = skb->dev->mtu;
 739
 740         if (nf_bridge->pkt_otherhost) {
 741                 skb->pkt_type = PACKET_OTHERHOST;
 742                 nf_bridge->pkt_otherhost = false;
 743         }
 744
 745         if (nf_bridge->frag_max_size && nf_bridge->frag_max_size < mtu)
 746                 mtu = nf_bridge->frag_max_size;
 747
 748         nf_bridge_update_protocol(skb);
 749         nf_bridge_push_encap_header(skb);
 750
 751         if (skb_is_gso(skb) || skb->len + mtu_reserved <= mtu) {
 752                 nf_bridge_info_free(skb);
 753                 return br_dev_queue_push_xmit(net, sk, skb);
 754         }
 755
 756         /* This is wrong! We should preserve the original fragment
 757          * boundaries by preserving frag_list rather than refragmenting.
 758          */
 759         if (IS_ENABLED(CONFIG_NF_DEFRAG_IPV4) &&
 760             skb->protocol == htons(ETH_P_IP)) {
 761                 struct brnf_frag_data *data;
 762
 763                 if (br_validate_ipv4(net, skb))
 764                         goto drop;
 765
 766                 IPCB(skb)->frag_max_size = nf_bridge->frag_max_size;
 767
 768                 data = this_cpu_ptr(&brnf_frag_data_storage);
 769
 770                 if (skb_vlan_tag_present(skb)) {
 771                         data->vlan_tci = skb->vlan_tci;
 772                         data->vlan_proto = skb->vlan_proto;
 773                 } else {
 774                         data->vlan_proto = 0;
 775                 }
 776
 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                 return br_nf_ip_fragment(net, sk, skb, br_nf_push_frag_xmit);
 784         }
 785         if (IS_ENABLED(CONFIG_NF_DEFRAG_IPV6) &&
 786             skb->protocol == htons(ETH_P_IPV6)) {
 787                 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
 788                 struct brnf_frag_data *data;
 789
 790                 if (br_validate_ipv6(net, skb))
 791                         goto drop;
 792
 793                 IP6CB(skb)->frag_max_size = nf_bridge->frag_max_size;
 794
 795                 data = this_cpu_ptr(&brnf_frag_data_storage);
 796                 data->encap_size = nf_bridge_encap_header_len(skb);
 797                 data->size = ETH_HLEN + data->encap_size;
 798
 799                 skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
 800                                                  data->size);
 801
 802                 if (v6ops)
 803                         return v6ops->fragment(net, sk, skb, br_nf_push_frag_xmit);
 804
 805                 kfree_skb(skb);
 806                 return -EMSGSIZE;
 807         }
 808         nf_bridge_info_free(skb);
 809         return br_dev_queue_push_xmit(net, sk, skb);
 810  drop:
 811         kfree_skb(skb);
 812         return 0;
 813 }
 814
 815 /* PF_BRIDGE/POST_ROUTING ********************************************/
 816 static unsigned int br_nf_post_routing(void *priv,
 817                                        struct sk_buff *skb,
 818                                        const struct nf_hook_state *state)
 819 {
 820         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 821         struct net_device *realoutdev = bridge_parent(skb->dev);
 822         u_int8_t pf;
 823
 824         /* if nf_bridge is set, but ->physoutdev is NULL, this packet came in
 825          * on a bridge, but was delivered locally and is now being routed:
 826          *
 827          * POST_ROUTING was already invoked from the ip stack.
 828          */
 829         if (!nf_bridge || !nf_bridge->physoutdev)
 830                 return NF_ACCEPT;
 831
 832         if (!realoutdev)
 833                 return NF_DROP;
 834
 835         if (IS_IP(skb) || is_vlan_ip(skb, state->net) ||
 836             is_pppoe_ip(skb, state->net))
 837                 pf = NFPROTO_IPV4;
 838         else if (IS_IPV6(skb) || is_vlan_ipv6(skb, state->net) ||
 839                  is_pppoe_ipv6(skb, state->net))
 840                 pf = NFPROTO_IPV6;
 841         else
 842                 return NF_ACCEPT;
 843
 844         if (skb->pkt_type == PACKET_OTHERHOST) {
 845                 skb->pkt_type = PACKET_HOST;
 846                 nf_bridge->pkt_otherhost = true;
 847         }
 848
 849         nf_bridge_pull_encap_header(skb);
 850         if (pf == NFPROTO_IPV4)
 851                 skb->protocol = htons(ETH_P_IP);
 852         else
 853                 skb->protocol = htons(ETH_P_IPV6);
 854
 855         NF_HOOK(pf, NF_INET_POST_ROUTING, state->net, state->sk, skb,
 856                 NULL, realoutdev,
 857                 br_nf_dev_queue_xmit);
 858
 859         return NF_STOLEN;
 860 }
 861
 862 /* IP/SABOTAGE *****************************************************/
 863 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 864  * for the second time. */
 865 static unsigned int ip_sabotage_in(void *priv,
 866                                    struct sk_buff *skb,
 867                                    const struct nf_hook_state *state)
 868 {
 869         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 870
 871         if (nf_bridge && !nf_bridge->in_prerouting &&
 872             !netif_is_l3_master(skb->dev) &&
 873             !netif_is_l3_slave(skb->dev)) {
 874                 state->okfn(state->net, state->sk, skb);
 875                 return NF_STOLEN;
 876         }
 877
 878         return NF_ACCEPT;
 879 }
 880
 881 /* This is called when br_netfilter has called into iptables/netfilter,
 882  * and DNAT has taken place on a bridge-forwarded packet.
 883  *
 884  * neigh->output has created a new MAC header, with local br0 MAC
 885  * as saddr.
 886  *
 887  * This restores the original MAC saddr of the bridged packet
 888  * before invoking bridge forward logic to transmit the packet.
 889  */
 890 static void br_nf_pre_routing_finish_bridge_slow(struct sk_buff *skb)
 891 {
 892         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 893
 894         skb_pull(skb, ETH_HLEN);
 895         nf_bridge->bridged_dnat = 0;
 896
 897         BUILD_BUG_ON(sizeof(nf_bridge->neigh_header) != (ETH_HLEN - ETH_ALEN));
 898
 899         skb_copy_to_linear_data_offset(skb, -(ETH_HLEN - ETH_ALEN),
 900                                        nf_bridge->neigh_header,
 901                                        ETH_HLEN - ETH_ALEN);
 902         skb->dev = nf_bridge->physindev;
 903
 904         nf_bridge->physoutdev = NULL;
 905         br_handle_frame_finish(dev_net(skb->dev), NULL, skb);
 906 }
 907
 908 static int br_nf_dev_xmit(struct sk_buff *skb)
 909 {
 910         const struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 911
 912         if (nf_bridge && nf_bridge->bridged_dnat) {
 913                 br_nf_pre_routing_finish_bridge_slow(skb);
 914                 return 1;
 915         }
 916         return 0;
 917 }
 918
 919 static const struct nf_br_ops br_ops = {
 920         .br_dev_xmit_hook =     br_nf_dev_xmit,
 921 };
 922
 923 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 924  * br_dev_queue_push_xmit is called afterwards */
 925 static const struct nf_hook_ops br_nf_ops[] = {
 926         {
 927                 .hook = br_nf_pre_routing,
 928                 .pf = NFPROTO_BRIDGE,
 929                 .hooknum = NF_BR_PRE_ROUTING,
 930                 .priority = NF_BR_PRI_BRNF,
 931         },
 932         {
 933                 .hook = br_nf_forward_ip,
 934                 .pf = NFPROTO_BRIDGE,
 935                 .hooknum = NF_BR_FORWARD,
 936                 .priority = NF_BR_PRI_BRNF - 1,
 937         },
 938         {
 939                 .hook = br_nf_forward_arp,
 940                 .pf = NFPROTO_BRIDGE,
 941                 .hooknum = NF_BR_FORWARD,
 942                 .priority = NF_BR_PRI_BRNF,
 943         },
 944         {
 945                 .hook = br_nf_post_routing,
 946                 .pf = NFPROTO_BRIDGE,
 947                 .hooknum = NF_BR_POST_ROUTING,
 948                 .priority = NF_BR_PRI_LAST,
 949         },
 950         {
 951                 .hook = ip_sabotage_in,
 952                 .pf = NFPROTO_IPV4,
 953                 .hooknum = NF_INET_PRE_ROUTING,
 954                 .priority = NF_IP_PRI_FIRST,
 955         },
 956         {
 957                 .hook = ip_sabotage_in,
 958                 .pf = NFPROTO_IPV6,
 959                 .hooknum = NF_INET_PRE_ROUTING,
 960                 .priority = NF_IP6_PRI_FIRST,
 961         },
 962 };
 963
 964 static int brnf_device_event(struct notifier_block *unused, unsigned long event,
 965                              void *ptr)
 966 {
 967         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 968         struct brnf_net *brnet;
 969         struct net *net;
 970         int ret;
 971
 972         if (event != NETDEV_REGISTER || !netif_is_bridge_master(dev))
 973                 return NOTIFY_DONE;
 974
 975         ASSERT_RTNL();
 976
 977         net = dev_net(dev);
 978         brnet = net_generic(net, brnf_net_id);
 979         if (brnet->enabled)
 980                 return NOTIFY_OK;
 981
 982         ret = nf_register_net_hooks(net, br_nf_ops, ARRAY_SIZE(br_nf_ops));
 983         if (ret)
 984                 return NOTIFY_BAD;
 985
 986         brnet->enabled = true;
 987         return NOTIFY_OK;
 988 }
 989
 990 static struct notifier_block brnf_notifier __read_mostly = {
 991         .notifier_call = brnf_device_event,
 992 };
 993
 994 /* recursively invokes nf_hook_slow (again), skipping already-called
 995  * hooks (< NF_BR_PRI_BRNF).
 996  *
 997  * Called with rcu read lock held.
 998  */
 999 int br_nf_hook_thresh(unsigned int hook, struct net *net,
1000                       struct sock *sk, struct sk_buff *skb,
1001                       struct net_device *indev,
1002                       struct net_device *outdev,
1003                       int (*okfn)(struct net *, struct sock *,
1004                                   struct sk_buff *))
1005 {
1006         const struct nf_hook_entries *e;
1007         struct nf_hook_state state;
1008         struct nf_hook_ops **ops;
1009         unsigned int i;
1010         int ret;
1011
1012         e = rcu_dereference(net->nf.hooks_bridge[hook]);
1013         if (!e)
1014                 return okfn(net, sk, skb);
1015
1016         ops = nf_hook_entries_get_hook_ops(e);
1017         for (i = 0; i < e->num_hook_entries; i++) {
1018                 /* These hooks have already been called */
1019                 if (ops[i]->priority < NF_BR_PRI_BRNF)
1020                         continue;
1021
1022                 /* These hooks have not been called yet, run them. */
1023                 if (ops[i]->priority > NF_BR_PRI_BRNF)
1024                         break;
1025
1026                 /* take a closer look at NF_BR_PRI_BRNF. */
1027                 if (ops[i]->hook == br_nf_pre_routing) {
1028                         /* This hook diverted the skb to this function,
1029                          * hooks after this have not been run yet.
1030                          */
1031                         i++;
1032                         break;
1033                 }
1034         }
1035
1036         nf_hook_state_init(&state, hook, NFPROTO_BRIDGE, indev, outdev,
1037                            sk, net, okfn);
1038
1039         ret = nf_hook_slow(skb, &state, e, i);
1040         if (ret == 1)
1041                 ret = okfn(net, sk, skb);
1042
1043         return ret;
1044 }
1045
1046 #ifdef CONFIG_SYSCTL
1047 static
1048 int brnf_sysctl_call_tables(struct ctl_table *ctl, int write,
1049                             void *buffer, size_t *lenp, loff_t *ppos)
1050 {
1051         int ret;
1052
1053         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1054
1055         if (write && *(int *)(ctl->data))
1056                 *(int *)(ctl->data) = 1;
1057         return ret;
1058 }
1059
1060 static struct ctl_table brnf_table[] = {
1061         {
1062                 .procname       = "bridge-nf-call-arptables",
1063                 .maxlen         = sizeof(int),
1064                 .mode           = 0644,
1065                 .proc_handler   = brnf_sysctl_call_tables,
1066         },
1067         {
1068                 .procname       = "bridge-nf-call-iptables",
1069                 .maxlen         = sizeof(int),
1070                 .mode           = 0644,
1071                 .proc_handler   = brnf_sysctl_call_tables,
1072         },
1073         {
1074                 .procname       = "bridge-nf-call-ip6tables",
1075                 .maxlen         = sizeof(int),
1076                 .mode           = 0644,
1077                 .proc_handler   = brnf_sysctl_call_tables,
1078         },
1079         {
1080                 .procname       = "bridge-nf-filter-vlan-tagged",
1081                 .maxlen         = sizeof(int),
1082                 .mode           = 0644,
1083                 .proc_handler   = brnf_sysctl_call_tables,
1084         },
1085         {
1086                 .procname       = "bridge-nf-filter-pppoe-tagged",
1087                 .maxlen         = sizeof(int),
1088                 .mode           = 0644,
1089                 .proc_handler   = brnf_sysctl_call_tables,
1090         },
1091         {
1092                 .procname       = "bridge-nf-pass-vlan-input-dev",
1093                 .maxlen         = sizeof(int),
1094                 .mode           = 0644,
1095                 .proc_handler   = brnf_sysctl_call_tables,
1096         },
1097         { }
1098 };
1099
1100 static inline void br_netfilter_sysctl_default(struct brnf_net *brnf)
1101 {
1102         brnf->call_iptables = 1;
1103         brnf->call_ip6tables = 1;
1104         brnf->call_arptables = 1;
1105         brnf->filter_vlan_tagged = 0;
1106         brnf->filter_pppoe_tagged = 0;
1107         brnf->pass_vlan_indev = 0;
1108 }
1109
1110 static int br_netfilter_sysctl_init_net(struct net *net)
1111 {
1112         struct ctl_table *table = brnf_table;
1113         struct brnf_net *brnet;
1114
1115         if (!net_eq(net, &init_net)) {
1116                 table = kmemdup(table, sizeof(brnf_table), GFP_KERNEL);
1117                 if (!table)
1118                         return -ENOMEM;
1119         }
1120
1121         brnet = net_generic(net, brnf_net_id);
1122         table[0].data = &brnet->call_arptables;
1123         table[1].data = &brnet->call_iptables;
1124         table[2].data = &brnet->call_ip6tables;
1125         table[3].data = &brnet->filter_vlan_tagged;
1126         table[4].data = &brnet->filter_pppoe_tagged;
1127         table[5].data = &brnet->pass_vlan_indev;
1128
1129         br_netfilter_sysctl_default(brnet);
1130
1131         brnet->ctl_hdr = register_net_sysctl(net, "net/bridge", table);
1132         if (!brnet->ctl_hdr) {
1133                 if (!net_eq(net, &init_net))
1134                         kfree(table);
1135
1136                 return -ENOMEM;
1137         }
1138
1139         return 0;
1140 }
1141
1142 static void br_netfilter_sysctl_exit_net(struct net *net,
1143                                          struct brnf_net *brnet)
1144 {
1145         struct ctl_table *table = brnet->ctl_hdr->ctl_table_arg;
1146
1147         unregister_net_sysctl_table(brnet->ctl_hdr);
1148         if (!net_eq(net, &init_net))
1149                 kfree(table);
1150 }
1151
1152 static int __net_init brnf_init_net(struct net *net)
1153 {
1154         return br_netfilter_sysctl_init_net(net);
1155 }
1156 #endif
1157
1158 static void __net_exit brnf_exit_net(struct net *net)
1159 {
1160         struct brnf_net *brnet;
1161
1162         brnet = net_generic(net, brnf_net_id);
1163         if (brnet->enabled) {
1164                 nf_unregister_net_hooks(net, br_nf_ops, ARRAY_SIZE(br_nf_ops));
1165                 brnet->enabled = false;
1166         }
1167
1168 #ifdef CONFIG_SYSCTL
1169         br_netfilter_sysctl_exit_net(net, brnet);
1170 #endif
1171 }
1172
1173 static struct pernet_operations brnf_net_ops __read_mostly = {
1174 #ifdef CONFIG_SYSCTL
1175         .init = brnf_init_net,
1176 #endif
1177         .exit = brnf_exit_net,
1178         .id   = &brnf_net_id,
1179         .size = sizeof(struct brnf_net),
1180 };
1181
1182 static int __init br_netfilter_init(void)
1183 {
1184         int ret;
1185
1186         ret = register_pernet_subsys(&brnf_net_ops);
1187         if (ret < 0)
1188                 return ret;
1189
1190         ret = register_netdevice_notifier(&brnf_notifier);
1191         if (ret < 0) {
1192                 unregister_pernet_subsys(&brnf_net_ops);
1193                 return ret;
1194         }
1195
1196         RCU_INIT_POINTER(nf_br_ops, &br_ops);
1197         printk(KERN_NOTICE "Bridge firewalling registered\n");
1198         return 0;
1199 }
1200
1201 static void __exit br_netfilter_fini(void)
1202 {
1203         RCU_INIT_POINTER(nf_br_ops, NULL);
1204         unregister_netdevice_notifier(&brnf_notifier);
1205         unregister_pernet_subsys(&brnf_net_ops);
1206 }
1207
1208 module_init(br_netfilter_init);
1209 module_exit(br_netfilter_fini);
1210
1211 MODULE_LICENSE("GPL");
1212 MODULE_AUTHOR("Lennert Buytenhek <buytenh@gnu.org>");
1213 MODULE_AUTHOR("Bart De Schuymer <bdschuym@pandora.be>");
1214 MODULE_DESCRIPTION("Linux ethernet netfilter firewall bridge");