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
  39 #include <asm/uaccess.h>
  40 #include "br_private.h"
  41 #ifdef CONFIG_SYSCTL
  42 #include <linux/sysctl.h>
  43 #endif
  44
  45 #define skb_origaddr(skb)        (((struct bridge_skb_cb *) \
  46                                  (skb->nf_bridge->data))->daddr.ipv4)
  47 #define store_orig_dstaddr(skb)  (skb_origaddr(skb) = ip_hdr(skb)->daddr)
  48 #define dnat_took_place(skb)     (skb_origaddr(skb) != ip_hdr(skb)->daddr)
  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         (!vlan_tx_tag_present(skb) && skb->protocol == htons(ETH_P_IP))
  69
  70 #define IS_IPV6(skb) \
  71         (!vlan_tx_tag_present(skb) && skb->protocol == htons(ETH_P_IPV6))
  72
  73 #define IS_ARP(skb) \
  74         (!vlan_tx_tag_present(skb) && skb->protocol == htons(ETH_P_ARP))
  75
  76 static inline __be16 vlan_proto(const struct sk_buff *skb)
  77 {
  78         if (vlan_tx_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         struct ip_options *opt;
 196         const struct iphdr *iph;
 197         struct net_device *dev = skb->dev;
 198         u32 len;
 199
 200         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 201                 goto inhdr_error;
 202
 203         iph = ip_hdr(skb);
 204         opt = &(IPCB(skb)->opt);
 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 inhdr_error;
 216
 217         len = ntohs(iph->tot_len);
 218         if (skb->len < len) {
 219                 IP_INC_STATS_BH(dev_net(dev), 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_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 226                 goto drop;
 227         }
 228
 229         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 230         if (iph->ihl == 5)
 231                 return 0;
 232
 233         opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
 234         if (ip_options_compile(dev_net(dev), opt, skb))
 235                 goto inhdr_error;
 236
 237         /* Check correct handling of SRR option */
 238         if (unlikely(opt->srr)) {
 239                 struct in_device *in_dev = __in_dev_get_rcu(dev);
 240                 if (in_dev && !IN_DEV_SOURCE_ROUTE(in_dev))
 241                         goto drop;
 242
 243                 if (ip_options_rcv_srr(skb))
 244                         goto drop;
 245         }
 246
 247         return 0;
 248
 249 inhdr_error:
 250         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 251 drop:
 252         return -1;
 253 }
 254
 255 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 256 /* Undo the changes made for ip6tables PREROUTING and continue the
 257  * bridge PRE_ROUTING hook. */
 258 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 259 {
 260         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 261         struct rtable *rt;
 262
 263         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 264                 skb->pkt_type = PACKET_OTHERHOST;
 265                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 266         }
 267         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 268
 269         rt = bridge_parent_rtable(nf_bridge->physindev);
 270         if (!rt) {
 271                 kfree_skb(skb);
 272                 return 0;
 273         }
 274         skb_dst_set_noref(skb, &rt->dst);
 275
 276         skb->dev = nf_bridge->physindev;
 277         nf_bridge_update_protocol(skb);
 278         nf_bridge_push_encap_header(skb);
 279         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 280                        br_handle_frame_finish, 1);
 281
 282         return 0;
 283 }
 284
 285 /* Obtain the correct destination MAC address, while preserving the original
 286  * source MAC address. If we already know this address, we just copy it. If we
 287  * don't, we use the neighbour framework to find out. In both cases, we make
 288  * sure that br_handle_frame_finish() is called afterwards.
 289  */
 290 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 291 {
 292         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 293         struct neighbour *neigh;
 294         struct dst_entry *dst;
 295
 296         skb->dev = bridge_parent(skb->dev);
 297         if (!skb->dev)
 298                 goto free_skb;
 299         dst = skb_dst(skb);
 300         neigh = dst_neigh_lookup_skb(dst, skb);
 301         if (neigh) {
 302                 int ret;
 303
 304                 if (neigh->hh.hh_len) {
 305                         neigh_hh_bridge(&neigh->hh, skb);
 306                         skb->dev = nf_bridge->physindev;
 307                         ret = br_handle_frame_finish(skb);
 308                 } else {
 309                         /* the neighbour function below overwrites the complete
 310                          * MAC header, so we save the Ethernet source address and
 311                          * protocol number.
 312                          */
 313                         skb_copy_from_linear_data_offset(skb,
 314                                                          -(ETH_HLEN-ETH_ALEN),
 315                                                          skb->nf_bridge->data,
 316                                                          ETH_HLEN-ETH_ALEN);
 317                         /* tell br_dev_xmit to continue with forwarding */
 318                         nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 319                         /* FIXME Need to refragment */
 320                         ret = neigh->output(neigh, skb);
 321                 }
 322                 neigh_release(neigh);
 323                 return ret;
 324         }
 325 free_skb:
 326         kfree_skb(skb);
 327         return 0;
 328 }
 329
 330 /* This requires some explaining. If DNAT has taken place,
 331  * we will need to fix up the destination Ethernet address.
 332  *
 333  * There are two cases to consider:
 334  * 1. The packet was DNAT'ed to a device in the same bridge
 335  *    port group as it was received on. We can still bridge
 336  *    the packet.
 337  * 2. The packet was DNAT'ed to a different device, either
 338  *    a non-bridged device or another bridge port group.
 339  *    The packet will need to be routed.
 340  *
 341  * The correct way of distinguishing between these two cases is to
 342  * call ip_route_input() and to look at skb->dst->dev, which is
 343  * changed to the destination device if ip_route_input() succeeds.
 344  *
 345  * Let's first consider the case that ip_route_input() succeeds:
 346  *
 347  * If the output device equals the logical bridge device the packet
 348  * came in on, we can consider this bridging. The corresponding MAC
 349  * address will be obtained in br_nf_pre_routing_finish_bridge.
 350  * Otherwise, the packet is considered to be routed and we just
 351  * change the destination MAC address so that the packet will
 352  * later be passed up to the IP stack to be routed. For a redirected
 353  * packet, ip_route_input() will give back the localhost as output device,
 354  * which differs from the bridge device.
 355  *
 356  * Let's now consider the case that ip_route_input() fails:
 357  *
 358  * This can be because the destination address is martian, in which case
 359  * the packet will be dropped.
 360  * If IP forwarding is disabled, ip_route_input() will fail, while
 361  * ip_route_output_key() can return success. The source
 362  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 363  * thinks we're handling a locally generated packet and won't care
 364  * if IP forwarding is enabled. If the output device equals the logical bridge
 365  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 366  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 367  */
 368 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 369 {
 370         struct net_device *dev = skb->dev;
 371         struct iphdr *iph = ip_hdr(skb);
 372         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 373         struct rtable *rt;
 374         int err;
 375         int frag_max_size;
 376
 377         frag_max_size = IPCB(skb)->frag_max_size;
 378         BR_INPUT_SKB_CB(skb)->frag_max_size = frag_max_size;
 379
 380         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 381                 skb->pkt_type = PACKET_OTHERHOST;
 382                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 383         }
 384         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 385         if (dnat_took_place(skb)) {
 386                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 387                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 388
 389                         /* If err equals -EHOSTUNREACH the error is due to a
 390                          * martian destination or due to the fact that
 391                          * forwarding is disabled. For most martian packets,
 392                          * ip_route_output_key() will fail. It won't fail for 2 types of
 393                          * martian destinations: loopback destinations and destination
 394                          * 0.0.0.0. In both cases the packet will be dropped because the
 395                          * destination is the loopback device and not the bridge. */
 396                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 397                                 goto free_skb;
 398
 399                         rt = ip_route_output(dev_net(dev), iph->daddr, 0,
 400                                              RT_TOS(iph->tos), 0);
 401                         if (!IS_ERR(rt)) {
 402                                 /* - Bridged-and-DNAT'ed traffic doesn't
 403                                  *   require ip_forwarding. */
 404                                 if (rt->dst.dev == dev) {
 405                                         skb_dst_set(skb, &rt->dst);
 406                                         goto bridged_dnat;
 407                                 }
 408                                 ip_rt_put(rt);
 409                         }
 410 free_skb:
 411                         kfree_skb(skb);
 412                         return 0;
 413                 } else {
 414                         if (skb_dst(skb)->dev == dev) {
 415 bridged_dnat:
 416                                 skb->dev = nf_bridge->physindev;
 417                                 nf_bridge_update_protocol(skb);
 418                                 nf_bridge_push_encap_header(skb);
 419                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
 420                                                NF_BR_PRE_ROUTING,
 421                                                skb, skb->dev, NULL,
 422                                                br_nf_pre_routing_finish_bridge,
 423                                                1);
 424                                 return 0;
 425                         }
 426                         ether_addr_copy(eth_hdr(skb)->h_dest, dev->dev_addr);
 427                         skb->pkt_type = PACKET_HOST;
 428                 }
 429         } else {
 430                 rt = bridge_parent_rtable(nf_bridge->physindev);
 431                 if (!rt) {
 432                         kfree_skb(skb);
 433                         return 0;
 434                 }
 435                 skb_dst_set_noref(skb, &rt->dst);
 436         }
 437
 438         skb->dev = nf_bridge->physindev;
 439         nf_bridge_update_protocol(skb);
 440         nf_bridge_push_encap_header(skb);
 441         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 442                        br_handle_frame_finish, 1);
 443
 444         return 0;
 445 }
 446
 447 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb, const struct net_device *dev)
 448 {
 449         struct net_device *vlan, *br;
 450
 451         br = bridge_parent(dev);
 452         if (brnf_pass_vlan_indev == 0 || !vlan_tx_tag_present(skb))
 453                 return br;
 454
 455         vlan = __vlan_find_dev_deep_rcu(br, skb->vlan_proto,
 456                                     vlan_tx_tag_get(skb) & VLAN_VID_MASK);
 457
 458         return vlan ? vlan : br;
 459 }
 460
 461 /* Some common code for IPv4/IPv6 */
 462 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 463 {
 464         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 465
 466         if (skb->pkt_type == PACKET_OTHERHOST) {
 467                 skb->pkt_type = PACKET_HOST;
 468                 nf_bridge->mask |= BRNF_PKT_TYPE;
 469         }
 470
 471         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 472         nf_bridge->physindev = skb->dev;
 473         skb->dev = brnf_get_logical_dev(skb, skb->dev);
 474         if (skb->protocol == htons(ETH_P_8021Q))
 475                 nf_bridge->mask |= BRNF_8021Q;
 476         else if (skb->protocol == htons(ETH_P_PPP_SES))
 477                 nf_bridge->mask |= BRNF_PPPoE;
 478
 479         /* Must drop socket now because of tproxy. */
 480         skb_orphan(skb);
 481         return skb->dev;
 482 }
 483
 484 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 485 static int check_hbh_len(struct sk_buff *skb)
 486 {
 487         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 488         u32 pkt_len;
 489         const unsigned char *nh = skb_network_header(skb);
 490         int off = raw - nh;
 491         int len = (raw[1] + 1) << 3;
 492
 493         if ((raw + len) - skb->data > skb_headlen(skb))
 494                 goto bad;
 495
 496         off += 2;
 497         len -= 2;
 498
 499         while (len > 0) {
 500                 int optlen = nh[off + 1] + 2;
 501
 502                 switch (nh[off]) {
 503                 case IPV6_TLV_PAD1:
 504                         optlen = 1;
 505                         break;
 506
 507                 case IPV6_TLV_PADN:
 508                         break;
 509
 510                 case IPV6_TLV_JUMBO:
 511                         if (nh[off + 1] != 4 || (off & 3) != 2)
 512                                 goto bad;
 513                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 514                         if (pkt_len <= IPV6_MAXPLEN ||
 515                             ipv6_hdr(skb)->payload_len)
 516                                 goto bad;
 517                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 518                                 goto bad;
 519                         if (pskb_trim_rcsum(skb,
 520                                             pkt_len + sizeof(struct ipv6hdr)))
 521                                 goto bad;
 522                         nh = skb_network_header(skb);
 523                         break;
 524                 default:
 525                         if (optlen > len)
 526                                 goto bad;
 527                         break;
 528                 }
 529                 off += optlen;
 530                 len -= optlen;
 531         }
 532         if (len == 0)
 533                 return 0;
 534 bad:
 535         return -1;
 536
 537 }
 538
 539 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 540  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 541 static unsigned int br_nf_pre_routing_ipv6(const struct nf_hook_ops *ops,
 542                                            struct sk_buff *skb,
 543                                            const struct net_device *in,
 544                                            const struct net_device *out,
 545                                            int (*okfn)(struct sk_buff *))
 546 {
 547         const struct ipv6hdr *hdr;
 548         u32 pkt_len;
 549
 550         if (skb->len < sizeof(struct ipv6hdr))
 551                 return NF_DROP;
 552
 553         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 554                 return NF_DROP;
 555
 556         hdr = ipv6_hdr(skb);
 557
 558         if (hdr->version != 6)
 559                 return NF_DROP;
 560
 561         pkt_len = ntohs(hdr->payload_len);
 562
 563         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 564                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 565                         return NF_DROP;
 566                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 567                         return NF_DROP;
 568         }
 569         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 570                 return NF_DROP;
 571
 572         nf_bridge_put(skb->nf_bridge);
 573         if (!nf_bridge_alloc(skb))
 574                 return NF_DROP;
 575         if (!setup_pre_routing(skb))
 576                 return NF_DROP;
 577
 578         skb->protocol = htons(ETH_P_IPV6);
 579         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 580                 br_nf_pre_routing_finish_ipv6);
 581
 582         return NF_STOLEN;
 583 }
 584
 585 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 586  * Replicate the checks that IPv4 does on packet reception.
 587  * Set skb->dev to the bridge device (i.e. parent of the
 588  * receiving device) to make netfilter happy, the REDIRECT
 589  * target in particular.  Save the original destination IP
 590  * address to be able to detect DNAT afterwards. */
 591 static unsigned int br_nf_pre_routing(const struct nf_hook_ops *ops,
 592                                       struct sk_buff *skb,
 593                                       const struct net_device *in,
 594                                       const struct net_device *out,
 595                                       int (*okfn)(struct sk_buff *))
 596 {
 597         struct net_bridge_port *p;
 598         struct net_bridge *br;
 599         __u32 len = nf_bridge_encap_header_len(skb);
 600
 601         if (unlikely(!pskb_may_pull(skb, len)))
 602                 return NF_DROP;
 603
 604         p = br_port_get_rcu(in);
 605         if (p == NULL)
 606                 return NF_DROP;
 607         br = p->br;
 608
 609         if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb)) {
 610                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
 611                         return NF_ACCEPT;
 612
 613                 nf_bridge_pull_encap_header_rcsum(skb);
 614                 return br_nf_pre_routing_ipv6(ops, skb, in, out, okfn);
 615         }
 616
 617         if (!brnf_call_iptables && !br->nf_call_iptables)
 618                 return NF_ACCEPT;
 619
 620         if (!IS_IP(skb) && !IS_VLAN_IP(skb) && !IS_PPPOE_IP(skb))
 621                 return NF_ACCEPT;
 622
 623         nf_bridge_pull_encap_header_rcsum(skb);
 624
 625         if (br_parse_ip_options(skb))
 626                 return NF_DROP;
 627
 628         nf_bridge_put(skb->nf_bridge);
 629         if (!nf_bridge_alloc(skb))
 630                 return NF_DROP;
 631         if (!setup_pre_routing(skb))
 632                 return NF_DROP;
 633         store_orig_dstaddr(skb);
 634         skb->protocol = htons(ETH_P_IP);
 635
 636         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 637                 br_nf_pre_routing_finish);
 638
 639         return NF_STOLEN;
 640 }
 641
 642
 643 /* PF_BRIDGE/LOCAL_IN ************************************************/
 644 /* The packet is locally destined, which requires a real
 645  * dst_entry, so detach the fake one.  On the way up, the
 646  * packet would pass through PRE_ROUTING again (which already
 647  * took place when the packet entered the bridge), but we
 648  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 649  * prevent this from happening. */
 650 static unsigned int br_nf_local_in(const struct nf_hook_ops *ops,
 651                                    struct sk_buff *skb,
 652                                    const struct net_device *in,
 653                                    const struct net_device *out,
 654                                    int (*okfn)(struct sk_buff *))
 655 {
 656         br_drop_fake_rtable(skb);
 657         return NF_ACCEPT;
 658 }
 659
 660 /* PF_BRIDGE/FORWARD *************************************************/
 661 static int br_nf_forward_finish(struct sk_buff *skb)
 662 {
 663         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 664         struct net_device *in;
 665
 666         if (!IS_ARP(skb) && !IS_VLAN_ARP(skb)) {
 667                 in = nf_bridge->physindev;
 668                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 669                         skb->pkt_type = PACKET_OTHERHOST;
 670                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 671                 }
 672                 nf_bridge_update_protocol(skb);
 673         } else {
 674                 in = *((struct net_device **)(skb->cb));
 675         }
 676         nf_bridge_push_encap_header(skb);
 677
 678         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
 679                        skb->dev, br_forward_finish, 1);
 680         return 0;
 681 }
 682
 683
 684 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 685  * netfilter with indev and outdev set to the bridge device,
 686  * but we are still able to filter on the 'real' indev/outdev
 687  * because of the physdev module. For ARP, indev and outdev are the
 688  * bridge ports. */
 689 static unsigned int br_nf_forward_ip(const struct nf_hook_ops *ops,
 690                                      struct sk_buff *skb,
 691                                      const struct net_device *in,
 692                                      const struct net_device *out,
 693                                      int (*okfn)(struct sk_buff *))
 694 {
 695         struct nf_bridge_info *nf_bridge;
 696         struct net_device *parent;
 697         u_int8_t pf;
 698
 699         if (!skb->nf_bridge)
 700                 return NF_ACCEPT;
 701
 702         /* Need exclusive nf_bridge_info since we might have multiple
 703          * different physoutdevs. */
 704         if (!nf_bridge_unshare(skb))
 705                 return NF_DROP;
 706
 707         parent = bridge_parent(out);
 708         if (!parent)
 709                 return NF_DROP;
 710
 711         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 712                 pf = NFPROTO_IPV4;
 713         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 714                 pf = NFPROTO_IPV6;
 715         else
 716                 return NF_ACCEPT;
 717
 718         nf_bridge_pull_encap_header(skb);
 719
 720         nf_bridge = skb->nf_bridge;
 721         if (skb->pkt_type == PACKET_OTHERHOST) {
 722                 skb->pkt_type = PACKET_HOST;
 723                 nf_bridge->mask |= BRNF_PKT_TYPE;
 724         }
 725
 726         if (pf == NFPROTO_IPV4 && br_parse_ip_options(skb))
 727                 return NF_DROP;
 728
 729         /* The physdev module checks on this */
 730         nf_bridge->mask |= BRNF_BRIDGED;
 731         nf_bridge->physoutdev = skb->dev;
 732         if (pf == NFPROTO_IPV4)
 733                 skb->protocol = htons(ETH_P_IP);
 734         else
 735                 skb->protocol = htons(ETH_P_IPV6);
 736
 737         NF_HOOK(pf, NF_INET_FORWARD, skb, brnf_get_logical_dev(skb, in), parent,
 738                 br_nf_forward_finish);
 739
 740         return NF_STOLEN;
 741 }
 742
 743 static unsigned int br_nf_forward_arp(const struct nf_hook_ops *ops,
 744                                       struct sk_buff *skb,
 745                                       const struct net_device *in,
 746                                       const struct net_device *out,
 747                                       int (*okfn)(struct sk_buff *))
 748 {
 749         struct net_bridge_port *p;
 750         struct net_bridge *br;
 751         struct net_device **d = (struct net_device **)(skb->cb);
 752
 753         p = br_port_get_rcu(out);
 754         if (p == NULL)
 755                 return NF_ACCEPT;
 756         br = p->br;
 757
 758         if (!brnf_call_arptables && !br->nf_call_arptables)
 759                 return NF_ACCEPT;
 760
 761         if (!IS_ARP(skb)) {
 762                 if (!IS_VLAN_ARP(skb))
 763                         return NF_ACCEPT;
 764                 nf_bridge_pull_encap_header(skb);
 765         }
 766
 767         if (arp_hdr(skb)->ar_pln != 4) {
 768                 if (IS_VLAN_ARP(skb))
 769                         nf_bridge_push_encap_header(skb);
 770                 return NF_ACCEPT;
 771         }
 772         *d = (struct net_device *)in;
 773         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 774                 (struct net_device *)out, br_nf_forward_finish);
 775
 776         return NF_STOLEN;
 777 }
 778
 779 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4)
 780 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 781 {
 782         int ret;
 783         int frag_max_size;
 784
 785         /* This is wrong! We should preserve the original fragment
 786          * boundaries by preserving frag_list rather than refragmenting.
 787          */
 788         if (skb->protocol == htons(ETH_P_IP) &&
 789             skb->len + nf_bridge_mtu_reduction(skb) > skb->dev->mtu &&
 790             !skb_is_gso(skb)) {
 791                 frag_max_size = BR_INPUT_SKB_CB(skb)->frag_max_size;
 792                 if (br_parse_ip_options(skb))
 793                         /* Drop invalid packet */
 794                         return NF_DROP;
 795                 IPCB(skb)->frag_max_size = frag_max_size;
 796                 ret = ip_fragment(skb, br_dev_queue_push_xmit);
 797         } else
 798                 ret = br_dev_queue_push_xmit(skb);
 799
 800         return ret;
 801 }
 802 #else
 803 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 804 {
 805         return br_dev_queue_push_xmit(skb);
 806 }
 807 #endif
 808
 809 /* PF_BRIDGE/POST_ROUTING ********************************************/
 810 static unsigned int br_nf_post_routing(const struct nf_hook_ops *ops,
 811                                        struct sk_buff *skb,
 812                                        const struct net_device *in,
 813                                        const struct net_device *out,
 814                                        int (*okfn)(struct sk_buff *))
 815 {
 816         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 817         struct net_device *realoutdev = bridge_parent(skb->dev);
 818         u_int8_t pf;
 819
 820         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
 821                 return NF_ACCEPT;
 822
 823         if (!realoutdev)
 824                 return NF_DROP;
 825
 826         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 827                 pf = NFPROTO_IPV4;
 828         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 829                 pf = NFPROTO_IPV6;
 830         else
 831                 return NF_ACCEPT;
 832
 833         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 834          * about the value of skb->pkt_type. */
 835         if (skb->pkt_type == PACKET_OTHERHOST) {
 836                 skb->pkt_type = PACKET_HOST;
 837                 nf_bridge->mask |= BRNF_PKT_TYPE;
 838         }
 839
 840         nf_bridge_pull_encap_header(skb);
 841         nf_bridge_save_header(skb);
 842         if (pf == NFPROTO_IPV4)
 843                 skb->protocol = htons(ETH_P_IP);
 844         else
 845                 skb->protocol = htons(ETH_P_IPV6);
 846
 847         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
 848                 br_nf_dev_queue_xmit);
 849
 850         return NF_STOLEN;
 851 }
 852
 853 /* IP/SABOTAGE *****************************************************/
 854 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 855  * for the second time. */
 856 static unsigned int ip_sabotage_in(const struct nf_hook_ops *ops,
 857                                    struct sk_buff *skb,
 858                                    const struct net_device *in,
 859                                    const struct net_device *out,
 860                                    int (*okfn)(struct sk_buff *))
 861 {
 862         if (skb->nf_bridge &&
 863             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 864                 return NF_STOP;
 865         }
 866
 867         return NF_ACCEPT;
 868 }
 869
 870 void br_netfilter_enable(void)
 871 {
 872 }
 873 EXPORT_SYMBOL_GPL(br_netfilter_enable);
 874
 875 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 876  * br_dev_queue_push_xmit is called afterwards */
 877 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
 878         {
 879                 .hook = br_nf_pre_routing,
 880                 .owner = THIS_MODULE,
 881                 .pf = NFPROTO_BRIDGE,
 882                 .hooknum = NF_BR_PRE_ROUTING,
 883                 .priority = NF_BR_PRI_BRNF,
 884         },
 885         {
 886                 .hook = br_nf_local_in,
 887                 .owner = THIS_MODULE,
 888                 .pf = NFPROTO_BRIDGE,
 889                 .hooknum = NF_BR_LOCAL_IN,
 890                 .priority = NF_BR_PRI_BRNF,
 891         },
 892         {
 893                 .hook = br_nf_forward_ip,
 894                 .owner = THIS_MODULE,
 895                 .pf = NFPROTO_BRIDGE,
 896                 .hooknum = NF_BR_FORWARD,
 897                 .priority = NF_BR_PRI_BRNF - 1,
 898         },
 899         {
 900                 .hook = br_nf_forward_arp,
 901                 .owner = THIS_MODULE,
 902                 .pf = NFPROTO_BRIDGE,
 903                 .hooknum = NF_BR_FORWARD,
 904                 .priority = NF_BR_PRI_BRNF,
 905         },
 906         {
 907                 .hook = br_nf_post_routing,
 908                 .owner = THIS_MODULE,
 909                 .pf = NFPROTO_BRIDGE,
 910                 .hooknum = NF_BR_POST_ROUTING,
 911                 .priority = NF_BR_PRI_LAST,
 912         },
 913         {
 914                 .hook = ip_sabotage_in,
 915                 .owner = THIS_MODULE,
 916                 .pf = NFPROTO_IPV4,
 917                 .hooknum = NF_INET_PRE_ROUTING,
 918                 .priority = NF_IP_PRI_FIRST,
 919         },
 920         {
 921                 .hook = ip_sabotage_in,
 922                 .owner = THIS_MODULE,
 923                 .pf = NFPROTO_IPV6,
 924                 .hooknum = NF_INET_PRE_ROUTING,
 925                 .priority = NF_IP6_PRI_FIRST,
 926         },
 927 };
 928
 929 #ifdef CONFIG_SYSCTL
 930 static
 931 int brnf_sysctl_call_tables(struct ctl_table *ctl, int write,
 932                             void __user *buffer, size_t *lenp, loff_t *ppos)
 933 {
 934         int ret;
 935
 936         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
 937
 938         if (write && *(int *)(ctl->data))
 939                 *(int *)(ctl->data) = 1;
 940         return ret;
 941 }
 942
 943 static struct ctl_table brnf_table[] = {
 944         {
 945                 .procname       = "bridge-nf-call-arptables",
 946                 .data           = &brnf_call_arptables,
 947                 .maxlen         = sizeof(int),
 948                 .mode           = 0644,
 949                 .proc_handler   = brnf_sysctl_call_tables,
 950         },
 951         {
 952                 .procname       = "bridge-nf-call-iptables",
 953                 .data           = &brnf_call_iptables,
 954                 .maxlen         = sizeof(int),
 955                 .mode           = 0644,
 956                 .proc_handler   = brnf_sysctl_call_tables,
 957         },
 958         {
 959                 .procname       = "bridge-nf-call-ip6tables",
 960                 .data           = &brnf_call_ip6tables,
 961                 .maxlen         = sizeof(int),
 962                 .mode           = 0644,
 963                 .proc_handler   = brnf_sysctl_call_tables,
 964         },
 965         {
 966                 .procname       = "bridge-nf-filter-vlan-tagged",
 967                 .data           = &brnf_filter_vlan_tagged,
 968                 .maxlen         = sizeof(int),
 969                 .mode           = 0644,
 970                 .proc_handler   = brnf_sysctl_call_tables,
 971         },
 972         {
 973                 .procname       = "bridge-nf-filter-pppoe-tagged",
 974                 .data           = &brnf_filter_pppoe_tagged,
 975                 .maxlen         = sizeof(int),
 976                 .mode           = 0644,
 977                 .proc_handler   = brnf_sysctl_call_tables,
 978         },
 979         {
 980                 .procname       = "bridge-nf-pass-vlan-input-dev",
 981                 .data           = &brnf_pass_vlan_indev,
 982                 .maxlen         = sizeof(int),
 983                 .mode           = 0644,
 984                 .proc_handler   = brnf_sysctl_call_tables,
 985         },
 986         { }
 987 };
 988 #endif
 989
 990 static int __init br_netfilter_init(void)
 991 {
 992         int ret;
 993
 994         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 995         if (ret < 0)
 996                 return ret;
 997
 998 #ifdef CONFIG_SYSCTL
 999         brnf_sysctl_header = register_net_sysctl(&init_net, "net/bridge", brnf_table);
1000         if (brnf_sysctl_header == NULL) {
1001                 printk(KERN_WARNING
1002                        "br_netfilter: can't register to sysctl.\n");
1003                 ret = -ENOMEM;
1004                 goto err1;
1005         }
1006 #endif
1007         printk(KERN_NOTICE "Bridge firewalling registered\n");
1008         return 0;
1009 err1:
1010         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1011         return ret;
1012 }
1013
1014 static void __exit br_netfilter_fini(void)
1015 {
1016         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1017 #ifdef CONFIG_SYSCTL
1018         unregister_net_sysctl_table(brnf_sysctl_header);
1019 #endif
1020 }
1021
1022 module_init(br_netfilter_init);
1023 module_exit(br_netfilter_fini);
1024
1025 MODULE_LICENSE("GPL");
1026 MODULE_AUTHOR("Lennert Buytenhek <buytenh@gnu.org>");
1027 MODULE_AUTHOR("Bart De Schuymer <bdschuym@pandora.be>");
1028 MODULE_DESCRIPTION("Linux ethernet netfilter firewall bridge");