net/bridge/br_netfilter.c

   1 /*
   2  *      Handle firewalling
   3  *      Linux ethernet bridge
   4  *
   5  *      Authors:
   6  *      Lennert Buytenhek               <buytenh@gnu.org>
   7  *      Bart De Schuymer                <bdschuym@pandora.be>
   8  *
   9  *      This program is free software; you can redistribute it and/or
  10  *      modify it under the terms of the GNU General Public License
  11  *      as published by the Free Software Foundation; either version
  12  *      2 of the License, or (at your option) any later version.
  13  *
  14  *      Lennert dedicates this file to Kerstin Wurdinger.
  15  */
  16
  17 #include <linux/module.h>
  18 #include <linux/kernel.h>
  19 #include <linux/slab.h>
  20 #include <linux/ip.h>
  21 #include <linux/netdevice.h>
  22 #include <linux/skbuff.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_ether.h>
  25 #include <linux/if_vlan.h>
  26 #include <linux/if_pppox.h>
  27 #include <linux/ppp_defs.h>
  28 #include <linux/netfilter_bridge.h>
  29 #include <linux/netfilter_ipv4.h>
  30 #include <linux/netfilter_ipv6.h>
  31 #include <linux/netfilter_arp.h>
  32 #include <linux/in_route.h>
  33 #include <linux/inetdevice.h>
  34
  35 #include <net/ip.h>
  36 #include <net/ipv6.h>
  37 #include <net/route.h>
  38 #include <net/netfilter/br_netfilter.h>
  39
  40 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
  41 #include <net/netfilter/nf_conntrack.h>
  42 #endif
  43
  44 #include <asm/uaccess.h>
  45 #include "br_private.h"
  46 #ifdef CONFIG_SYSCTL
  47 #include <linux/sysctl.h>
  48 #endif
  49
  50 #ifdef CONFIG_SYSCTL
  51 static struct ctl_table_header *brnf_sysctl_header;
  52 static int brnf_call_iptables __read_mostly = 1;
  53 static int brnf_call_ip6tables __read_mostly = 1;
  54 static int brnf_call_arptables __read_mostly = 1;
  55 static int brnf_filter_vlan_tagged __read_mostly = 0;
  56 static int brnf_filter_pppoe_tagged __read_mostly = 0;
  57 static int brnf_pass_vlan_indev __read_mostly = 0;
  58 #else
  59 #define brnf_call_iptables 1
  60 #define brnf_call_ip6tables 1
  61 #define brnf_call_arptables 1
  62 #define brnf_filter_vlan_tagged 0
  63 #define brnf_filter_pppoe_tagged 0
  64 #define brnf_pass_vlan_indev 0
  65 #endif
  66
  67 #define IS_IP(skb) \
  68         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IP))
  69
  70 #define IS_IPV6(skb) \
  71         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IPV6))
  72
  73 #define IS_ARP(skb) \
  74         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_ARP))
  75
  76 static inline __be16 vlan_proto(const struct sk_buff *skb)
  77 {
  78         if (skb_vlan_tag_present(skb))
  79                 return skb->protocol;
  80         else if (skb->protocol == htons(ETH_P_8021Q))
  81                 return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  82         else
  83                 return 0;
  84 }
  85
  86 #define IS_VLAN_IP(skb) \
  87         (vlan_proto(skb) == htons(ETH_P_IP) && \
  88          brnf_filter_vlan_tagged)
  89
  90 #define IS_VLAN_IPV6(skb) \
  91         (vlan_proto(skb) == htons(ETH_P_IPV6) && \
  92          brnf_filter_vlan_tagged)
  93
  94 #define IS_VLAN_ARP(skb) \
  95         (vlan_proto(skb) == htons(ETH_P_ARP) && \
  96          brnf_filter_vlan_tagged)
  97
  98 static inline __be16 pppoe_proto(const struct sk_buff *skb)
  99 {
 100         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
 101                             sizeof(struct pppoe_hdr)));
 102 }
 103
 104 #define IS_PPPOE_IP(skb) \
 105         (skb->protocol == htons(ETH_P_PPP_SES) && \
 106          pppoe_proto(skb) == htons(PPP_IP) && \
 107          brnf_filter_pppoe_tagged)
 108
 109 #define IS_PPPOE_IPV6(skb) \
 110         (skb->protocol == htons(ETH_P_PPP_SES) && \
 111          pppoe_proto(skb) == htons(PPP_IPV6) && \
 112          brnf_filter_pppoe_tagged)
 113
 114 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
 115 {
 116         struct net_bridge_port *port;
 117
 118         port = br_port_get_rcu(dev);
 119         return port ? &port->br->fake_rtable : NULL;
 120 }
 121
 122 static inline struct net_device *bridge_parent(const struct net_device *dev)
 123 {
 124         struct net_bridge_port *port;
 125
 126         port = br_port_get_rcu(dev);
 127         return port ? port->br->dev : NULL;
 128 }
 129
 130 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 131 {
 132         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 133         if (likely(skb->nf_bridge))
 134                 atomic_set(&(skb->nf_bridge->use), 1);
 135
 136         return skb->nf_bridge;
 137 }
 138
 139 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 140 {
 141         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 142
 143         if (atomic_read(&nf_bridge->use) > 1) {
 144                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 145
 146                 if (tmp) {
 147                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 148                         atomic_set(&tmp->use, 1);
 149                 }
 150                 nf_bridge_put(nf_bridge);
 151                 nf_bridge = tmp;
 152         }
 153         return nf_bridge;
 154 }
 155
 156 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 157 {
 158         unsigned int len = nf_bridge_encap_header_len(skb);
 159
 160         skb_push(skb, len);
 161         skb->network_header -= len;
 162 }
 163
 164 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 165 {
 166         unsigned int len = nf_bridge_encap_header_len(skb);
 167
 168         skb_pull(skb, len);
 169         skb->network_header += len;
 170 }
 171
 172 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 173 {
 174         unsigned int len = nf_bridge_encap_header_len(skb);
 175
 176         skb_pull_rcsum(skb, len);
 177         skb->network_header += len;
 178 }
 179
 180 static inline void nf_bridge_save_header(struct sk_buff *skb)
 181 {
 182         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 183
 184         skb_copy_from_linear_data_offset(skb, -header_size,
 185                                          skb->nf_bridge->data, header_size);
 186 }
 187
 188 /* When handing a packet over to the IP layer
 189  * check whether we have a skb that is in the
 190  * expected format
 191  */
 192
 193 static int br_parse_ip_options(struct sk_buff *skb)
 194 {
 195         const struct iphdr *iph;
 196         struct net_device *dev = skb->dev;
 197         u32 len;
 198
 199         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 200                 goto inhdr_error;
 201
 202         iph = ip_hdr(skb);
 203
 204         /* Basic sanity checks */
 205         if (iph->ihl < 5 || iph->version != 4)
 206                 goto inhdr_error;
 207
 208         if (!pskb_may_pull(skb, iph->ihl*4))
 209                 goto inhdr_error;
 210
 211         iph = ip_hdr(skb);
 212         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 213                 goto inhdr_error;
 214
 215         len = ntohs(iph->tot_len);
 216         if (skb->len < len) {
 217                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
 218                 goto drop;
 219         } else if (len < (iph->ihl*4))
 220                 goto inhdr_error;
 221
 222         if (pskb_trim_rcsum(skb, len)) {
 223                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 224                 goto drop;
 225         }
 226
 227         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 228         /* We should really parse IP options here but until
 229          * somebody who actually uses IP options complains to
 230          * us we'll just silently ignore the options because
 231          * we're lazy!
 232          */
 233         return 0;
 234
 235 inhdr_error:
 236         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 237 drop:
 238         return -1;
 239 }
 240
 241 static void nf_bridge_update_protocol(struct sk_buff *skb)
 242 {
 243         if (skb->nf_bridge->mask & BRNF_8021Q)
 244                 skb->protocol = htons(ETH_P_8021Q);
 245         else if (skb->nf_bridge->mask & BRNF_PPPoE)
 246                 skb->protocol = htons(ETH_P_PPP_SES);
 247 }
 248
 249 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 250 /* Undo the changes made for ip6tables PREROUTING and continue the
 251  * bridge PRE_ROUTING hook. */
 252 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 253 {
 254         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 255         struct rtable *rt;
 256
 257         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 258                 skb->pkt_type = PACKET_OTHERHOST;
 259                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 260         }
 261         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 262
 263         rt = bridge_parent_rtable(nf_bridge->physindev);
 264         if (!rt) {
 265                 kfree_skb(skb);
 266                 return 0;
 267         }
 268         skb_dst_set_noref(skb, &rt->dst);
 269
 270         skb->dev = nf_bridge->physindev;
 271         nf_bridge_update_protocol(skb);
 272         nf_bridge_push_encap_header(skb);
 273         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 274                        br_handle_frame_finish, 1);
 275
 276         return 0;
 277 }
 278
 279 /* Obtain the correct destination MAC address, while preserving the original
 280  * source MAC address. If we already know this address, we just copy it. If we
 281  * don't, we use the neighbour framework to find out. In both cases, we make
 282  * sure that br_handle_frame_finish() is called afterwards.
 283  */
 284 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 285 {
 286         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 287         struct neighbour *neigh;
 288         struct dst_entry *dst;
 289
 290         skb->dev = bridge_parent(skb->dev);
 291         if (!skb->dev)
 292                 goto free_skb;
 293         dst = skb_dst(skb);
 294         neigh = dst_neigh_lookup_skb(dst, skb);
 295         if (neigh) {
 296                 int ret;
 297
 298                 if (neigh->hh.hh_len) {
 299                         neigh_hh_bridge(&neigh->hh, skb);
 300                         skb->dev = nf_bridge->physindev;
 301                         ret = br_handle_frame_finish(skb);
 302                 } else {
 303                         /* the neighbour function below overwrites the complete
 304                          * MAC header, so we save the Ethernet source address and
 305                          * protocol number.
 306                          */
 307                         skb_copy_from_linear_data_offset(skb,
 308                                                          -(ETH_HLEN-ETH_ALEN),
 309                                                          skb->nf_bridge->data,
 310                                                          ETH_HLEN-ETH_ALEN);
 311                         /* tell br_dev_xmit to continue with forwarding */
 312                         nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 313                         /* FIXME Need to refragment */
 314                         ret = neigh->output(neigh, skb);
 315                 }
 316                 neigh_release(neigh);
 317                 return ret;
 318         }
 319 free_skb:
 320         kfree_skb(skb);
 321         return 0;
 322 }
 323
 324 static bool dnat_took_place(const struct sk_buff *skb)
 325 {
 326 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 327         enum ip_conntrack_info ctinfo;
 328         struct nf_conn *ct;
 329
 330         ct = nf_ct_get(skb, &ctinfo);
 331         if (!ct || nf_ct_is_untracked(ct))
 332                 return false;
 333
 334         return test_bit(IPS_DST_NAT_BIT, &ct->status);
 335 #else
 336         return false;
 337 #endif
 338 }
 339
 340 /* This requires some explaining. If DNAT has taken place,
 341  * we will need to fix up the destination Ethernet address.
 342  *
 343  * There are two cases to consider:
 344  * 1. The packet was DNAT'ed to a device in the same bridge
 345  *    port group as it was received on. We can still bridge
 346  *    the packet.
 347  * 2. The packet was DNAT'ed to a different device, either
 348  *    a non-bridged device or another bridge port group.
 349  *    The packet will need to be routed.
 350  *
 351  * The correct way of distinguishing between these two cases is to
 352  * call ip_route_input() and to look at skb->dst->dev, which is
 353  * changed to the destination device if ip_route_input() succeeds.
 354  *
 355  * Let's first consider the case that ip_route_input() succeeds:
 356  *
 357  * If the output device equals the logical bridge device the packet
 358  * came in on, we can consider this bridging. The corresponding MAC
 359  * address will be obtained in br_nf_pre_routing_finish_bridge.
 360  * Otherwise, the packet is considered to be routed and we just
 361  * change the destination MAC address so that the packet will
 362  * later be passed up to the IP stack to be routed. For a redirected
 363  * packet, ip_route_input() will give back the localhost as output device,
 364  * which differs from the bridge device.
 365  *
 366  * Let's now consider the case that ip_route_input() fails:
 367  *
 368  * This can be because the destination address is martian, in which case
 369  * the packet will be dropped.
 370  * If IP forwarding is disabled, ip_route_input() will fail, while
 371  * ip_route_output_key() can return success. The source
 372  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 373  * thinks we're handling a locally generated packet and won't care
 374  * if IP forwarding is enabled. If the output device equals the logical bridge
 375  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 376  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 377  */
 378 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 379 {
 380         struct net_device *dev = skb->dev;
 381         struct iphdr *iph = ip_hdr(skb);
 382         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 383         struct rtable *rt;
 384         int err;
 385         int frag_max_size;
 386
 387         frag_max_size = IPCB(skb)->frag_max_size;
 388         BR_INPUT_SKB_CB(skb)->frag_max_size = frag_max_size;
 389
 390         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 391                 skb->pkt_type = PACKET_OTHERHOST;
 392                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 393         }
 394         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 395         if (dnat_took_place(skb)) {
 396                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 397                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 398
 399                         /* If err equals -EHOSTUNREACH the error is due to a
 400                          * martian destination or due to the fact that
 401                          * forwarding is disabled. For most martian packets,
 402                          * ip_route_output_key() will fail. It won't fail for 2 types of
 403                          * martian destinations: loopback destinations and destination
 404                          * 0.0.0.0. In both cases the packet will be dropped because the
 405                          * destination is the loopback device and not the bridge. */
 406                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 407                                 goto free_skb;
 408
 409                         rt = ip_route_output(dev_net(dev), iph->daddr, 0,
 410                                              RT_TOS(iph->tos), 0);
 411                         if (!IS_ERR(rt)) {
 412                                 /* - Bridged-and-DNAT'ed traffic doesn't
 413                                  *   require ip_forwarding. */
 414                                 if (rt->dst.dev == dev) {
 415                                         skb_dst_set(skb, &rt->dst);
 416                                         goto bridged_dnat;
 417                                 }
 418                                 ip_rt_put(rt);
 419                         }
 420 free_skb:
 421                         kfree_skb(skb);
 422                         return 0;
 423                 } else {
 424                         if (skb_dst(skb)->dev == dev) {
 425 bridged_dnat:
 426                                 skb->dev = nf_bridge->physindev;
 427                                 nf_bridge_update_protocol(skb);
 428                                 nf_bridge_push_encap_header(skb);
 429                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
 430                                                NF_BR_PRE_ROUTING,
 431                                                skb, skb->dev, NULL,
 432                                                br_nf_pre_routing_finish_bridge,
 433                                                1);
 434                                 return 0;
 435                         }
 436                         ether_addr_copy(eth_hdr(skb)->h_dest, dev->dev_addr);
 437                         skb->pkt_type = PACKET_HOST;
 438                 }
 439         } else {
 440                 rt = bridge_parent_rtable(nf_bridge->physindev);
 441                 if (!rt) {
 442                         kfree_skb(skb);
 443                         return 0;
 444                 }
 445                 skb_dst_set_noref(skb, &rt->dst);
 446         }
 447
 448         skb->dev = nf_bridge->physindev;
 449         nf_bridge_update_protocol(skb);
 450         nf_bridge_push_encap_header(skb);
 451         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 452                        br_handle_frame_finish, 1);
 453
 454         return 0;
 455 }
 456
 457 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb, const struct net_device *dev)
 458 {
 459         struct net_device *vlan, *br;
 460
 461         br = bridge_parent(dev);
 462         if (brnf_pass_vlan_indev == 0 || !skb_vlan_tag_present(skb))
 463                 return br;
 464
 465         vlan = __vlan_find_dev_deep_rcu(br, skb->vlan_proto,
 466                                     skb_vlan_tag_get(skb) & VLAN_VID_MASK);
 467
 468         return vlan ? vlan : br;
 469 }
 470
 471 /* Some common code for IPv4/IPv6 */
 472 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 473 {
 474         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 475
 476         if (skb->pkt_type == PACKET_OTHERHOST) {
 477                 skb->pkt_type = PACKET_HOST;
 478                 nf_bridge->mask |= BRNF_PKT_TYPE;
 479         }
 480
 481         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 482         nf_bridge->physindev = skb->dev;
 483         skb->dev = brnf_get_logical_dev(skb, skb->dev);
 484         if (skb->protocol == htons(ETH_P_8021Q))
 485                 nf_bridge->mask |= BRNF_8021Q;
 486         else if (skb->protocol == htons(ETH_P_PPP_SES))
 487                 nf_bridge->mask |= BRNF_PPPoE;
 488
 489         /* Must drop socket now because of tproxy. */
 490         skb_orphan(skb);
 491         return skb->dev;
 492 }
 493
 494 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 495 static int check_hbh_len(struct sk_buff *skb)
 496 {
 497         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 498         u32 pkt_len;
 499         const unsigned char *nh = skb_network_header(skb);
 500         int off = raw - nh;
 501         int len = (raw[1] + 1) << 3;
 502
 503         if ((raw + len) - skb->data > skb_headlen(skb))
 504                 goto bad;
 505
 506         off += 2;
 507         len -= 2;
 508
 509         while (len > 0) {
 510                 int optlen = nh[off + 1] + 2;
 511
 512                 switch (nh[off]) {
 513                 case IPV6_TLV_PAD1:
 514                         optlen = 1;
 515                         break;
 516
 517                 case IPV6_TLV_PADN:
 518                         break;
 519
 520                 case IPV6_TLV_JUMBO:
 521                         if (nh[off + 1] != 4 || (off & 3) != 2)
 522                                 goto bad;
 523                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 524                         if (pkt_len <= IPV6_MAXPLEN ||
 525                             ipv6_hdr(skb)->payload_len)
 526                                 goto bad;
 527                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 528                                 goto bad;
 529                         if (pskb_trim_rcsum(skb,
 530                                             pkt_len + sizeof(struct ipv6hdr)))
 531                                 goto bad;
 532                         nh = skb_network_header(skb);
 533                         break;
 534                 default:
 535                         if (optlen > len)
 536                                 goto bad;
 537                         break;
 538                 }
 539                 off += optlen;
 540                 len -= optlen;
 541         }
 542         if (len == 0)
 543                 return 0;
 544 bad:
 545         return -1;
 546
 547 }
 548
 549 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 550  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 551 static unsigned int br_nf_pre_routing_ipv6(const struct nf_hook_ops *ops,
 552                                            struct sk_buff *skb,
 553                                            const struct net_device *in,
 554                                            const struct net_device *out,
 555                                            int (*okfn)(struct sk_buff *))
 556 {
 557         const struct ipv6hdr *hdr;
 558         u32 pkt_len;
 559
 560         if (skb->len < sizeof(struct ipv6hdr))
 561                 return NF_DROP;
 562
 563         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 564                 return NF_DROP;
 565
 566         hdr = ipv6_hdr(skb);
 567
 568         if (hdr->version != 6)
 569                 return NF_DROP;
 570
 571         pkt_len = ntohs(hdr->payload_len);
 572
 573         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 574                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 575                         return NF_DROP;
 576                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 577                         return NF_DROP;
 578         }
 579         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 580                 return NF_DROP;
 581
 582         nf_bridge_put(skb->nf_bridge);
 583         if (!nf_bridge_alloc(skb))
 584                 return NF_DROP;
 585         if (!setup_pre_routing(skb))
 586                 return NF_DROP;
 587
 588         skb->protocol = htons(ETH_P_IPV6);
 589         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 590                 br_nf_pre_routing_finish_ipv6);
 591
 592         return NF_STOLEN;
 593 }
 594
 595 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 596  * Replicate the checks that IPv4 does on packet reception.
 597  * Set skb->dev to the bridge device (i.e. parent of the
 598  * receiving device) to make netfilter happy, the REDIRECT
 599  * target in particular.  Save the original destination IP
 600  * address to be able to detect DNAT afterwards. */
 601 static unsigned int br_nf_pre_routing(const struct nf_hook_ops *ops,
 602                                       struct sk_buff *skb,
 603                                       const struct net_device *in,
 604                                       const struct net_device *out,
 605                                       int (*okfn)(struct sk_buff *))
 606 {
 607         struct net_bridge_port *p;
 608         struct net_bridge *br;
 609         __u32 len = nf_bridge_encap_header_len(skb);
 610
 611         if (unlikely(!pskb_may_pull(skb, len)))
 612                 return NF_DROP;
 613
 614         p = br_port_get_rcu(in);
 615         if (p == NULL)
 616                 return NF_DROP;
 617         br = p->br;
 618
 619         if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb)) {
 620                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
 621                         return NF_ACCEPT;
 622
 623                 nf_bridge_pull_encap_header_rcsum(skb);
 624                 return br_nf_pre_routing_ipv6(ops, skb, in, out, okfn);
 625         }
 626
 627         if (!brnf_call_iptables && !br->nf_call_iptables)
 628                 return NF_ACCEPT;
 629
 630         if (!IS_IP(skb) && !IS_VLAN_IP(skb) && !IS_PPPOE_IP(skb))
 631                 return NF_ACCEPT;
 632
 633         nf_bridge_pull_encap_header_rcsum(skb);
 634
 635         if (br_parse_ip_options(skb))
 636                 return NF_DROP;
 637
 638         nf_bridge_put(skb->nf_bridge);
 639         if (!nf_bridge_alloc(skb))
 640                 return NF_DROP;
 641         if (!setup_pre_routing(skb))
 642                 return NF_DROP;
 643
 644         skb->protocol = htons(ETH_P_IP);
 645
 646         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 647                 br_nf_pre_routing_finish);
 648
 649         return NF_STOLEN;
 650 }
 651
 652
 653 /* PF_BRIDGE/LOCAL_IN ************************************************/
 654 /* The packet is locally destined, which requires a real
 655  * dst_entry, so detach the fake one.  On the way up, the
 656  * packet would pass through PRE_ROUTING again (which already
 657  * took place when the packet entered the bridge), but we
 658  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 659  * prevent this from happening. */
 660 static unsigned int br_nf_local_in(const struct nf_hook_ops *ops,
 661                                    struct sk_buff *skb,
 662                                    const struct net_device *in,
 663                                    const struct net_device *out,
 664                                    int (*okfn)(struct sk_buff *))
 665 {
 666         br_drop_fake_rtable(skb);
 667         return NF_ACCEPT;
 668 }
 669
 670 /* PF_BRIDGE/FORWARD *************************************************/
 671 static int br_nf_forward_finish(struct sk_buff *skb)
 672 {
 673         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 674         struct net_device *in;
 675
 676         if (!IS_ARP(skb) && !IS_VLAN_ARP(skb)) {
 677                 in = nf_bridge->physindev;
 678                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 679                         skb->pkt_type = PACKET_OTHERHOST;
 680                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 681                 }
 682                 nf_bridge_update_protocol(skb);
 683         } else {
 684                 in = *((struct net_device **)(skb->cb));
 685         }
 686         nf_bridge_push_encap_header(skb);
 687
 688         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
 689                        skb->dev, br_forward_finish, 1);
 690         return 0;
 691 }
 692
 693
 694 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 695  * netfilter with indev and outdev set to the bridge device,
 696  * but we are still able to filter on the 'real' indev/outdev
 697  * because of the physdev module. For ARP, indev and outdev are the
 698  * bridge ports. */
 699 static unsigned int br_nf_forward_ip(const struct nf_hook_ops *ops,
 700                                      struct sk_buff *skb,
 701                                      const struct net_device *in,
 702                                      const struct net_device *out,
 703                                      int (*okfn)(struct sk_buff *))
 704 {
 705         struct nf_bridge_info *nf_bridge;
 706         struct net_device *parent;
 707         u_int8_t pf;
 708
 709         if (!skb->nf_bridge)
 710                 return NF_ACCEPT;
 711
 712         /* Need exclusive nf_bridge_info since we might have multiple
 713          * different physoutdevs. */
 714         if (!nf_bridge_unshare(skb))
 715                 return NF_DROP;
 716
 717         parent = bridge_parent(out);
 718         if (!parent)
 719                 return NF_DROP;
 720
 721         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 722                 pf = NFPROTO_IPV4;
 723         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 724                 pf = NFPROTO_IPV6;
 725         else
 726                 return NF_ACCEPT;
 727
 728         nf_bridge_pull_encap_header(skb);
 729
 730         nf_bridge = skb->nf_bridge;
 731         if (skb->pkt_type == PACKET_OTHERHOST) {
 732                 skb->pkt_type = PACKET_HOST;
 733                 nf_bridge->mask |= BRNF_PKT_TYPE;
 734         }
 735
 736         if (pf == NFPROTO_IPV4 && br_parse_ip_options(skb))
 737                 return NF_DROP;
 738
 739         nf_bridge->physoutdev = skb->dev;
 740         if (pf == NFPROTO_IPV4)
 741                 skb->protocol = htons(ETH_P_IP);
 742         else
 743                 skb->protocol = htons(ETH_P_IPV6);
 744
 745         NF_HOOK(pf, NF_INET_FORWARD, skb, brnf_get_logical_dev(skb, in), parent,
 746                 br_nf_forward_finish);
 747
 748         return NF_STOLEN;
 749 }
 750
 751 static unsigned int br_nf_forward_arp(const struct nf_hook_ops *ops,
 752                                       struct sk_buff *skb,
 753                                       const struct net_device *in,
 754                                       const struct net_device *out,
 755                                       int (*okfn)(struct sk_buff *))
 756 {
 757         struct net_bridge_port *p;
 758         struct net_bridge *br;
 759         struct net_device **d = (struct net_device **)(skb->cb);
 760
 761         p = br_port_get_rcu(out);
 762         if (p == NULL)
 763                 return NF_ACCEPT;
 764         br = p->br;
 765
 766         if (!brnf_call_arptables && !br->nf_call_arptables)
 767                 return NF_ACCEPT;
 768
 769         if (!IS_ARP(skb)) {
 770                 if (!IS_VLAN_ARP(skb))
 771                         return NF_ACCEPT;
 772                 nf_bridge_pull_encap_header(skb);
 773         }
 774
 775         if (arp_hdr(skb)->ar_pln != 4) {
 776                 if (IS_VLAN_ARP(skb))
 777                         nf_bridge_push_encap_header(skb);
 778                 return NF_ACCEPT;
 779         }
 780         *d = (struct net_device *)in;
 781         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 782                 (struct net_device *)out, br_nf_forward_finish);
 783
 784         return NF_STOLEN;
 785 }
 786
 787 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4)
 788 static bool nf_bridge_copy_header(struct sk_buff *skb)
 789 {
 790         int err;
 791         unsigned int header_size;
 792
 793         nf_bridge_update_protocol(skb);
 794         header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 795         err = skb_cow_head(skb, header_size);
 796         if (err)
 797                 return false;
 798
 799         skb_copy_to_linear_data_offset(skb, -header_size,
 800                                        skb->nf_bridge->data, header_size);
 801         __skb_push(skb, nf_bridge_encap_header_len(skb));
 802         return true;
 803 }
 804
 805 static int br_nf_push_frag_xmit(struct sk_buff *skb)
 806 {
 807         if (!nf_bridge_copy_header(skb)) {
 808                 kfree_skb(skb);
 809                 return 0;
 810         }
 811
 812         return br_dev_queue_push_xmit(skb);
 813 }
 814
 815 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 816 {
 817         int ret;
 818         int frag_max_size;
 819         unsigned int mtu_reserved;
 820
 821         if (skb_is_gso(skb) || skb->protocol != htons(ETH_P_IP))
 822                 return br_dev_queue_push_xmit(skb);
 823
 824         mtu_reserved = nf_bridge_mtu_reduction(skb);
 825         /* This is wrong! We should preserve the original fragment
 826          * boundaries by preserving frag_list rather than refragmenting.
 827          */
 828         if (skb->len + mtu_reserved > skb->dev->mtu) {
 829                 frag_max_size = BR_INPUT_SKB_CB(skb)->frag_max_size;
 830                 if (br_parse_ip_options(skb))
 831                         /* Drop invalid packet */
 832                         return NF_DROP;
 833                 IPCB(skb)->frag_max_size = frag_max_size;
 834                 ret = ip_fragment(skb, br_nf_push_frag_xmit);
 835         } else
 836                 ret = br_dev_queue_push_xmit(skb);
 837
 838         return ret;
 839 }
 840 #else
 841 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 842 {
 843         return br_dev_queue_push_xmit(skb);
 844 }
 845 #endif
 846
 847 /* PF_BRIDGE/POST_ROUTING ********************************************/
 848 static unsigned int br_nf_post_routing(const struct nf_hook_ops *ops,
 849                                        struct sk_buff *skb,
 850                                        const struct net_device *in,
 851                                        const struct net_device *out,
 852                                        int (*okfn)(struct sk_buff *))
 853 {
 854         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 855         struct net_device *realoutdev = bridge_parent(skb->dev);
 856         u_int8_t pf;
 857
 858         /* if nf_bridge is set, but ->physoutdev is NULL, this packet came in
 859          * on a bridge, but was delivered locally and is now being routed:
 860          *
 861          * POST_ROUTING was already invoked from the ip stack.
 862          */
 863         if (!nf_bridge || !nf_bridge->physoutdev)
 864                 return NF_ACCEPT;
 865
 866         if (!realoutdev)
 867                 return NF_DROP;
 868
 869         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 870                 pf = NFPROTO_IPV4;
 871         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 872                 pf = NFPROTO_IPV6;
 873         else
 874                 return NF_ACCEPT;
 875
 876         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 877          * about the value of skb->pkt_type. */
 878         if (skb->pkt_type == PACKET_OTHERHOST) {
 879                 skb->pkt_type = PACKET_HOST;
 880                 nf_bridge->mask |= BRNF_PKT_TYPE;
 881         }
 882
 883         nf_bridge_pull_encap_header(skb);
 884         nf_bridge_save_header(skb);
 885         if (pf == NFPROTO_IPV4)
 886                 skb->protocol = htons(ETH_P_IP);
 887         else
 888                 skb->protocol = htons(ETH_P_IPV6);
 889
 890         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
 891                 br_nf_dev_queue_xmit);
 892
 893         return NF_STOLEN;
 894 }
 895
 896 /* IP/SABOTAGE *****************************************************/
 897 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 898  * for the second time. */
 899 static unsigned int ip_sabotage_in(const struct nf_hook_ops *ops,
 900                                    struct sk_buff *skb,
 901                                    const struct net_device *in,
 902                                    const struct net_device *out,
 903                                    int (*okfn)(struct sk_buff *))
 904 {
 905         if (skb->nf_bridge &&
 906             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 907                 return NF_STOP;
 908         }
 909
 910         return NF_ACCEPT;
 911 }
 912
 913 /* This is called when br_netfilter has called into iptables/netfilter,
 914  * and DNAT has taken place on a bridge-forwarded packet.
 915  *
 916  * neigh->output has created a new MAC header, with local br0 MAC
 917  * as saddr.
 918  *
 919  * This restores the original MAC saddr of the bridged packet
 920  * before invoking bridge forward logic to transmit the packet.
 921  */
 922 static void br_nf_pre_routing_finish_bridge_slow(struct sk_buff *skb)
 923 {
 924         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 925
 926         skb_pull(skb, ETH_HLEN);
 927         nf_bridge->mask &= ~BRNF_BRIDGED_DNAT;
 928
 929         skb_copy_to_linear_data_offset(skb, -(ETH_HLEN-ETH_ALEN),
 930                                        skb->nf_bridge->data, ETH_HLEN-ETH_ALEN);
 931         skb->dev = nf_bridge->physindev;
 932         br_handle_frame_finish(skb);
 933 }
 934
 935 static int br_nf_dev_xmit(struct sk_buff *skb)
 936 {
 937         if (skb->nf_bridge && (skb->nf_bridge->mask & BRNF_BRIDGED_DNAT)) {
 938                 br_nf_pre_routing_finish_bridge_slow(skb);
 939                 return 1;
 940         }
 941         return 0;
 942 }
 943
 944 static const struct nf_br_ops br_ops = {
 945         .br_dev_xmit_hook =     br_nf_dev_xmit,
 946 };
 947
 948 void br_netfilter_enable(void)
 949 {
 950 }
 951 EXPORT_SYMBOL_GPL(br_netfilter_enable);
 952
 953 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 954  * br_dev_queue_push_xmit is called afterwards */
 955 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
 956         {
 957                 .hook = br_nf_pre_routing,
 958                 .owner = THIS_MODULE,
 959                 .pf = NFPROTO_BRIDGE,
 960                 .hooknum = NF_BR_PRE_ROUTING,
 961                 .priority = NF_BR_PRI_BRNF,
 962         },
 963         {
 964                 .hook = br_nf_local_in,
 965                 .owner = THIS_MODULE,
 966                 .pf = NFPROTO_BRIDGE,
 967                 .hooknum = NF_BR_LOCAL_IN,
 968                 .priority = NF_BR_PRI_BRNF,
 969         },
 970         {
 971                 .hook = br_nf_forward_ip,
 972                 .owner = THIS_MODULE,
 973                 .pf = NFPROTO_BRIDGE,
 974                 .hooknum = NF_BR_FORWARD,
 975                 .priority = NF_BR_PRI_BRNF - 1,
 976         },
 977         {
 978                 .hook = br_nf_forward_arp,
 979                 .owner = THIS_MODULE,
 980                 .pf = NFPROTO_BRIDGE,
 981                 .hooknum = NF_BR_FORWARD,
 982                 .priority = NF_BR_PRI_BRNF,
 983         },
 984         {
 985                 .hook = br_nf_post_routing,
 986                 .owner = THIS_MODULE,
 987                 .pf = NFPROTO_BRIDGE,
 988                 .hooknum = NF_BR_POST_ROUTING,
 989                 .priority = NF_BR_PRI_LAST,
 990         },
 991         {
 992                 .hook = ip_sabotage_in,
 993                 .owner = THIS_MODULE,
 994                 .pf = NFPROTO_IPV4,
 995                 .hooknum = NF_INET_PRE_ROUTING,
 996                 .priority = NF_IP_PRI_FIRST,
 997         },
 998         {
 999                 .hook = ip_sabotage_in,
1000                 .owner = THIS_MODULE,
1001                 .pf = NFPROTO_IPV6,
1002                 .hooknum = NF_INET_PRE_ROUTING,
1003                 .priority = NF_IP6_PRI_FIRST,
1004         },
1005 };
1006
1007 #ifdef CONFIG_SYSCTL
1008 static
1009 int brnf_sysctl_call_tables(struct ctl_table *ctl, int write,
1010                             void __user *buffer, size_t *lenp, loff_t *ppos)
1011 {
1012         int ret;
1013
1014         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1015
1016         if (write && *(int *)(ctl->data))
1017                 *(int *)(ctl->data) = 1;
1018         return ret;
1019 }
1020
1021 static struct ctl_table brnf_table[] = {
1022         {
1023                 .procname       = "bridge-nf-call-arptables",
1024                 .data           = &brnf_call_arptables,
1025                 .maxlen         = sizeof(int),
1026                 .mode           = 0644,
1027                 .proc_handler   = brnf_sysctl_call_tables,
1028         },
1029         {
1030                 .procname       = "bridge-nf-call-iptables",
1031                 .data           = &brnf_call_iptables,
1032                 .maxlen         = sizeof(int),
1033                 .mode           = 0644,
1034                 .proc_handler   = brnf_sysctl_call_tables,
1035         },
1036         {
1037                 .procname       = "bridge-nf-call-ip6tables",
1038                 .data           = &brnf_call_ip6tables,
1039                 .maxlen         = sizeof(int),
1040                 .mode           = 0644,
1041                 .proc_handler   = brnf_sysctl_call_tables,
1042         },
1043         {
1044                 .procname       = "bridge-nf-filter-vlan-tagged",
1045                 .data           = &brnf_filter_vlan_tagged,
1046                 .maxlen         = sizeof(int),
1047                 .mode           = 0644,
1048                 .proc_handler   = brnf_sysctl_call_tables,
1049         },
1050         {
1051                 .procname       = "bridge-nf-filter-pppoe-tagged",
1052                 .data           = &brnf_filter_pppoe_tagged,
1053                 .maxlen         = sizeof(int),
1054                 .mode           = 0644,
1055                 .proc_handler   = brnf_sysctl_call_tables,
1056         },
1057         {
1058                 .procname       = "bridge-nf-pass-vlan-input-dev",
1059                 .data           = &brnf_pass_vlan_indev,
1060                 .maxlen         = sizeof(int),
1061                 .mode           = 0644,
1062                 .proc_handler   = brnf_sysctl_call_tables,
1063         },
1064         { }
1065 };
1066 #endif
1067
1068 static int __init br_netfilter_init(void)
1069 {
1070         int ret;
1071
1072         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1073         if (ret < 0)
1074                 return ret;
1075
1076 #ifdef CONFIG_SYSCTL
1077         brnf_sysctl_header = register_net_sysctl(&init_net, "net/bridge", brnf_table);
1078         if (brnf_sysctl_header == NULL) {
1079                 printk(KERN_WARNING
1080                        "br_netfilter: can't register to sysctl.\n");
1081                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1082                 return -ENOMEM;
1083         }
1084 #endif
1085         RCU_INIT_POINTER(nf_br_ops, &br_ops);
1086         printk(KERN_NOTICE "Bridge firewalling registered\n");
1087         return 0;
1088 }
1089
1090 static void __exit br_netfilter_fini(void)
1091 {
1092         RCU_INIT_POINTER(nf_br_ops, NULL);
1093         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1094 #ifdef CONFIG_SYSCTL
1095         unregister_net_sysctl_table(brnf_sysctl_header);
1096 #endif
1097 }
1098
1099 module_init(br_netfilter_init);
1100 module_exit(br_netfilter_fini);
1101
1102 MODULE_LICENSE("GPL");
1103 MODULE_AUTHOR("Lennert Buytenhek <buytenh@gnu.org>");
1104 MODULE_AUTHOR("Bart De Schuymer <bdschuym@pandora.be>");
1105 MODULE_DESCRIPTION("Linux ethernet netfilter firewall bridge");