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 #else
  58 #define brnf_call_iptables 1
  59 #define brnf_call_ip6tables 1
  60 #define brnf_call_arptables 1
  61 #define brnf_filter_vlan_tagged 0
  62 #define brnf_filter_pppoe_tagged 0
  63 #endif
  64
  65 static inline __be16 vlan_proto(const struct sk_buff *skb)
  66 {
  67         if (vlan_tx_tag_present(skb))
  68                 return skb->protocol;
  69         else if (skb->protocol == htons(ETH_P_8021Q))
  70                 return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  71         else
  72                 return 0;
  73 }
  74
  75 #define IS_VLAN_IP(skb) \
  76         (vlan_proto(skb) == htons(ETH_P_IP) && \
  77          brnf_filter_vlan_tagged)
  78
  79 #define IS_VLAN_IPV6(skb) \
  80         (vlan_proto(skb) == htons(ETH_P_IPV6) && \
  81          brnf_filter_vlan_tagged)
  82
  83 #define IS_VLAN_ARP(skb) \
  84         (vlan_proto(skb) == htons(ETH_P_ARP) && \
  85          brnf_filter_vlan_tagged)
  86
  87 static inline __be16 pppoe_proto(const struct sk_buff *skb)
  88 {
  89         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
  90                             sizeof(struct pppoe_hdr)));
  91 }
  92
  93 #define IS_PPPOE_IP(skb) \
  94         (skb->protocol == htons(ETH_P_PPP_SES) && \
  95          pppoe_proto(skb) == htons(PPP_IP) && \
  96          brnf_filter_pppoe_tagged)
  97
  98 #define IS_PPPOE_IPV6(skb) \
  99         (skb->protocol == htons(ETH_P_PPP_SES) && \
 100          pppoe_proto(skb) == htons(PPP_IPV6) && \
 101          brnf_filter_pppoe_tagged)
 102
 103 static void fake_update_pmtu(struct dst_entry *dst, u32 mtu)
 104 {
 105 }
 106
 107 static u32 *fake_cow_metrics(struct dst_entry *dst, unsigned long old)
 108 {
 109         return NULL;
 110 }
 111
 112 static struct neighbour *fake_neigh_lookup(const struct dst_entry *dst, const void *daddr)
 113 {
 114         return NULL;
 115 }
 116
 117 static unsigned int fake_mtu(const struct dst_entry *dst)
 118 {
 119         return dst->dev->mtu;
 120 }
 121
 122 static struct dst_ops fake_dst_ops = {
 123         .family =               AF_INET,
 124         .protocol =             cpu_to_be16(ETH_P_IP),
 125         .update_pmtu =          fake_update_pmtu,
 126         .cow_metrics =          fake_cow_metrics,
 127         .neigh_lookup =         fake_neigh_lookup,
 128         .mtu =                  fake_mtu,
 129 };
 130
 131 /*
 132  * Initialize bogus route table used to keep netfilter happy.
 133  * Currently, we fill in the PMTU entry because netfilter
 134  * refragmentation needs it, and the rt_flags entry because
 135  * ipt_REJECT needs it.  Future netfilter modules might
 136  * require us to fill additional fields.
 137  */
 138 static const u32 br_dst_default_metrics[RTAX_MAX] = {
 139         [RTAX_MTU - 1] = 1500,
 140 };
 141
 142 void br_netfilter_rtable_init(struct net_bridge *br)
 143 {
 144         struct rtable *rt = &br->fake_rtable;
 145
 146         atomic_set(&rt->dst.__refcnt, 1);
 147         rt->dst.dev = br->dev;
 148         rt->dst.path = &rt->dst;
 149         dst_init_metrics(&rt->dst, br_dst_default_metrics, true);
 150         rt->dst.flags   = DST_NOXFRM | DST_NOPEER | DST_FAKE_RTABLE;
 151         rt->dst.ops = &fake_dst_ops;
 152 }
 153
 154 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
 155 {
 156         struct net_bridge_port *port;
 157
 158         port = br_port_get_rcu(dev);
 159         return port ? &port->br->fake_rtable : NULL;
 160 }
 161
 162 static inline struct net_device *bridge_parent(const struct net_device *dev)
 163 {
 164         struct net_bridge_port *port;
 165
 166         port = br_port_get_rcu(dev);
 167         return port ? port->br->dev : NULL;
 168 }
 169
 170 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 171 {
 172         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 173         if (likely(skb->nf_bridge))
 174                 atomic_set(&(skb->nf_bridge->use), 1);
 175
 176         return skb->nf_bridge;
 177 }
 178
 179 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 180 {
 181         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 182
 183         if (atomic_read(&nf_bridge->use) > 1) {
 184                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 185
 186                 if (tmp) {
 187                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 188                         atomic_set(&tmp->use, 1);
 189                 }
 190                 nf_bridge_put(nf_bridge);
 191                 nf_bridge = tmp;
 192         }
 193         return nf_bridge;
 194 }
 195
 196 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 197 {
 198         unsigned int len = nf_bridge_encap_header_len(skb);
 199
 200         skb_push(skb, len);
 201         skb->network_header -= len;
 202 }
 203
 204 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 205 {
 206         unsigned int len = nf_bridge_encap_header_len(skb);
 207
 208         skb_pull(skb, len);
 209         skb->network_header += len;
 210 }
 211
 212 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 213 {
 214         unsigned int len = nf_bridge_encap_header_len(skb);
 215
 216         skb_pull_rcsum(skb, len);
 217         skb->network_header += len;
 218 }
 219
 220 static inline void nf_bridge_save_header(struct sk_buff *skb)
 221 {
 222         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 223
 224         skb_copy_from_linear_data_offset(skb, -header_size,
 225                                          skb->nf_bridge->data, header_size);
 226 }
 227
 228 static inline void nf_bridge_update_protocol(struct sk_buff *skb)
 229 {
 230         if (skb->nf_bridge->mask & BRNF_8021Q)
 231                 skb->protocol = htons(ETH_P_8021Q);
 232         else if (skb->nf_bridge->mask & BRNF_PPPoE)
 233                 skb->protocol = htons(ETH_P_PPP_SES);
 234 }
 235
 236 /* When handing a packet over to the IP layer
 237  * check whether we have a skb that is in the
 238  * expected format
 239  */
 240
 241 static int br_parse_ip_options(struct sk_buff *skb)
 242 {
 243         struct ip_options *opt;
 244         const struct iphdr *iph;
 245         struct net_device *dev = skb->dev;
 246         u32 len;
 247
 248         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 249                 goto inhdr_error;
 250
 251         iph = ip_hdr(skb);
 252         opt = &(IPCB(skb)->opt);
 253
 254         /* Basic sanity checks */
 255         if (iph->ihl < 5 || iph->version != 4)
 256                 goto inhdr_error;
 257
 258         if (!pskb_may_pull(skb, iph->ihl*4))
 259                 goto inhdr_error;
 260
 261         iph = ip_hdr(skb);
 262         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 263                 goto inhdr_error;
 264
 265         len = ntohs(iph->tot_len);
 266         if (skb->len < len) {
 267                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
 268                 goto drop;
 269         } else if (len < (iph->ihl*4))
 270                 goto inhdr_error;
 271
 272         if (pskb_trim_rcsum(skb, len)) {
 273                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 274                 goto drop;
 275         }
 276
 277         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 278         if (iph->ihl == 5)
 279                 return 0;
 280
 281         opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
 282         if (ip_options_compile(dev_net(dev), opt, skb))
 283                 goto inhdr_error;
 284
 285         /* Check correct handling of SRR option */
 286         if (unlikely(opt->srr)) {
 287                 struct in_device *in_dev = __in_dev_get_rcu(dev);
 288                 if (in_dev && !IN_DEV_SOURCE_ROUTE(in_dev))
 289                         goto drop;
 290
 291                 if (ip_options_rcv_srr(skb))
 292                         goto drop;
 293         }
 294
 295         return 0;
 296
 297 inhdr_error:
 298         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 299 drop:
 300         return -1;
 301 }
 302
 303 /* Fill in the header for fragmented IP packets handled by
 304  * the IPv4 connection tracking code.
 305  */
 306 int nf_bridge_copy_header(struct sk_buff *skb)
 307 {
 308         int err;
 309         unsigned int header_size;
 310
 311         nf_bridge_update_protocol(skb);
 312         header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 313         err = skb_cow_head(skb, header_size);
 314         if (err)
 315                 return err;
 316
 317         skb_copy_to_linear_data_offset(skb, -header_size,
 318                                        skb->nf_bridge->data, header_size);
 319         __skb_push(skb, nf_bridge_encap_header_len(skb));
 320         return 0;
 321 }
 322
 323 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 324 /* Undo the changes made for ip6tables PREROUTING and continue the
 325  * bridge PRE_ROUTING hook. */
 326 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 327 {
 328         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 329         struct rtable *rt;
 330
 331         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 332                 skb->pkt_type = PACKET_OTHERHOST;
 333                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 334         }
 335         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 336
 337         rt = bridge_parent_rtable(nf_bridge->physindev);
 338         if (!rt) {
 339                 kfree_skb(skb);
 340                 return 0;
 341         }
 342         skb_dst_set_noref(skb, &rt->dst);
 343
 344         skb->dev = nf_bridge->physindev;
 345         nf_bridge_update_protocol(skb);
 346         nf_bridge_push_encap_header(skb);
 347         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 348                        br_handle_frame_finish, 1);
 349
 350         return 0;
 351 }
 352
 353 /* Obtain the correct destination MAC address, while preserving the original
 354  * source MAC address. If we already know this address, we just copy it. If we
 355  * don't, we use the neighbour framework to find out. In both cases, we make
 356  * sure that br_handle_frame_finish() is called afterwards.
 357  */
 358 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 359 {
 360         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 361         struct neighbour *neigh;
 362         struct dst_entry *dst;
 363
 364         skb->dev = bridge_parent(skb->dev);
 365         if (!skb->dev)
 366                 goto free_skb;
 367         dst = skb_dst(skb);
 368         neigh = dst_get_neighbour(dst);
 369         if (neigh->hh.hh_len) {
 370                 neigh_hh_bridge(&neigh->hh, skb);
 371                 skb->dev = nf_bridge->physindev;
 372                 return br_handle_frame_finish(skb);
 373         } else {
 374                 /* the neighbour function below overwrites the complete
 375                  * MAC header, so we save the Ethernet source address and
 376                  * protocol number. */
 377                 skb_copy_from_linear_data_offset(skb, -(ETH_HLEN-ETH_ALEN), skb->nf_bridge->data, ETH_HLEN-ETH_ALEN);
 378                 /* tell br_dev_xmit to continue with forwarding */
 379                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 380                 return neigh->output(neigh, skb);
 381         }
 382 free_skb:
 383         kfree_skb(skb);
 384         return 0;
 385 }
 386
 387 /* This requires some explaining. If DNAT has taken place,
 388  * we will need to fix up the destination Ethernet address.
 389  *
 390  * There are two cases to consider:
 391  * 1. The packet was DNAT'ed to a device in the same bridge
 392  *    port group as it was received on. We can still bridge
 393  *    the packet.
 394  * 2. The packet was DNAT'ed to a different device, either
 395  *    a non-bridged device or another bridge port group.
 396  *    The packet will need to be routed.
 397  *
 398  * The correct way of distinguishing between these two cases is to
 399  * call ip_route_input() and to look at skb->dst->dev, which is
 400  * changed to the destination device if ip_route_input() succeeds.
 401  *
 402  * Let's first consider the case that ip_route_input() succeeds:
 403  *
 404  * If the output device equals the logical bridge device the packet
 405  * came in on, we can consider this bridging. The corresponding MAC
 406  * address will be obtained in br_nf_pre_routing_finish_bridge.
 407  * Otherwise, the packet is considered to be routed and we just
 408  * change the destination MAC address so that the packet will
 409  * later be passed up to the IP stack to be routed. For a redirected
 410  * packet, ip_route_input() will give back the localhost as output device,
 411  * which differs from the bridge device.
 412  *
 413  * Let's now consider the case that ip_route_input() fails:
 414  *
 415  * This can be because the destination address is martian, in which case
 416  * the packet will be dropped.
 417  * If IP forwarding is disabled, ip_route_input() will fail, while
 418  * ip_route_output_key() can return success. The source
 419  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 420  * thinks we're handling a locally generated packet and won't care
 421  * if IP forwarding is enabled. If the output device equals the logical bridge
 422  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 423  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 424  */
 425 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 426 {
 427         struct net_device *dev = skb->dev;
 428         struct iphdr *iph = ip_hdr(skb);
 429         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 430         struct rtable *rt;
 431         int err;
 432
 433         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 434                 skb->pkt_type = PACKET_OTHERHOST;
 435                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 436         }
 437         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 438         if (dnat_took_place(skb)) {
 439                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 440                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 441
 442                         /* If err equals -EHOSTUNREACH the error is due to a
 443                          * martian destination or due to the fact that
 444                          * forwarding is disabled. For most martian packets,
 445                          * ip_route_output_key() will fail. It won't fail for 2 types of
 446                          * martian destinations: loopback destinations and destination
 447                          * 0.0.0.0. In both cases the packet will be dropped because the
 448                          * destination is the loopback device and not the bridge. */
 449                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 450                                 goto free_skb;
 451
 452                         rt = ip_route_output(dev_net(dev), iph->daddr, 0,
 453                                              RT_TOS(iph->tos), 0);
 454                         if (!IS_ERR(rt)) {
 455                                 /* - Bridged-and-DNAT'ed traffic doesn't
 456                                  *   require ip_forwarding. */
 457                                 if (rt->dst.dev == dev) {
 458                                         skb_dst_set(skb, &rt->dst);
 459                                         goto bridged_dnat;
 460                                 }
 461                                 ip_rt_put(rt);
 462                         }
 463 free_skb:
 464                         kfree_skb(skb);
 465                         return 0;
 466                 } else {
 467                         if (skb_dst(skb)->dev == dev) {
 468 bridged_dnat:
 469                                 skb->dev = nf_bridge->physindev;
 470                                 nf_bridge_update_protocol(skb);
 471                                 nf_bridge_push_encap_header(skb);
 472                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
 473                                                NF_BR_PRE_ROUTING,
 474                                                skb, skb->dev, NULL,
 475                                                br_nf_pre_routing_finish_bridge,
 476                                                1);
 477                                 return 0;
 478                         }
 479                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
 480                         skb->pkt_type = PACKET_HOST;
 481                 }
 482         } else {
 483                 rt = bridge_parent_rtable(nf_bridge->physindev);
 484                 if (!rt) {
 485                         kfree_skb(skb);
 486                         return 0;
 487                 }
 488                 skb_dst_set_noref(skb, &rt->dst);
 489         }
 490
 491         skb->dev = nf_bridge->physindev;
 492         nf_bridge_update_protocol(skb);
 493         nf_bridge_push_encap_header(skb);
 494         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 495                        br_handle_frame_finish, 1);
 496
 497         return 0;
 498 }
 499
 500 /* Some common code for IPv4/IPv6 */
 501 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 502 {
 503         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 504
 505         if (skb->pkt_type == PACKET_OTHERHOST) {
 506                 skb->pkt_type = PACKET_HOST;
 507                 nf_bridge->mask |= BRNF_PKT_TYPE;
 508         }
 509
 510         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 511         nf_bridge->physindev = skb->dev;
 512         skb->dev = bridge_parent(skb->dev);
 513         if (skb->protocol == htons(ETH_P_8021Q))
 514                 nf_bridge->mask |= BRNF_8021Q;
 515         else if (skb->protocol == htons(ETH_P_PPP_SES))
 516                 nf_bridge->mask |= BRNF_PPPoE;
 517
 518         return skb->dev;
 519 }
 520
 521 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 522 static int check_hbh_len(struct sk_buff *skb)
 523 {
 524         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 525         u32 pkt_len;
 526         const unsigned char *nh = skb_network_header(skb);
 527         int off = raw - nh;
 528         int len = (raw[1] + 1) << 3;
 529
 530         if ((raw + len) - skb->data > skb_headlen(skb))
 531                 goto bad;
 532
 533         off += 2;
 534         len -= 2;
 535
 536         while (len > 0) {
 537                 int optlen = nh[off + 1] + 2;
 538
 539                 switch (nh[off]) {
 540                 case IPV6_TLV_PAD0:
 541                         optlen = 1;
 542                         break;
 543
 544                 case IPV6_TLV_PADN:
 545                         break;
 546
 547                 case IPV6_TLV_JUMBO:
 548                         if (nh[off + 1] != 4 || (off & 3) != 2)
 549                                 goto bad;
 550                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 551                         if (pkt_len <= IPV6_MAXPLEN ||
 552                             ipv6_hdr(skb)->payload_len)
 553                                 goto bad;
 554                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 555                                 goto bad;
 556                         if (pskb_trim_rcsum(skb,
 557                                             pkt_len + sizeof(struct ipv6hdr)))
 558                                 goto bad;
 559                         nh = skb_network_header(skb);
 560                         break;
 561                 default:
 562                         if (optlen > len)
 563                                 goto bad;
 564                         break;
 565                 }
 566                 off += optlen;
 567                 len -= optlen;
 568         }
 569         if (len == 0)
 570                 return 0;
 571 bad:
 572         return -1;
 573
 574 }
 575
 576 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 577  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 578 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 579                                            struct sk_buff *skb,
 580                                            const struct net_device *in,
 581                                            const struct net_device *out,
 582                                            int (*okfn)(struct sk_buff *))
 583 {
 584         const struct ipv6hdr *hdr;
 585         u32 pkt_len;
 586
 587         if (skb->len < sizeof(struct ipv6hdr))
 588                 return NF_DROP;
 589
 590         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 591                 return NF_DROP;
 592
 593         hdr = ipv6_hdr(skb);
 594
 595         if (hdr->version != 6)
 596                 return NF_DROP;
 597
 598         pkt_len = ntohs(hdr->payload_len);
 599
 600         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 601                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 602                         return NF_DROP;
 603                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 604                         return NF_DROP;
 605         }
 606         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 607                 return NF_DROP;
 608
 609         nf_bridge_put(skb->nf_bridge);
 610         if (!nf_bridge_alloc(skb))
 611                 return NF_DROP;
 612         if (!setup_pre_routing(skb))
 613                 return NF_DROP;
 614
 615         skb->protocol = htons(ETH_P_IPV6);
 616         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 617                 br_nf_pre_routing_finish_ipv6);
 618
 619         return NF_STOLEN;
 620 }
 621
 622 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 623  * Replicate the checks that IPv4 does on packet reception.
 624  * Set skb->dev to the bridge device (i.e. parent of the
 625  * receiving device) to make netfilter happy, the REDIRECT
 626  * target in particular.  Save the original destination IP
 627  * address to be able to detect DNAT afterwards. */
 628 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
 629                                       const struct net_device *in,
 630                                       const struct net_device *out,
 631                                       int (*okfn)(struct sk_buff *))
 632 {
 633         struct net_bridge_port *p;
 634         struct net_bridge *br;
 635         __u32 len = nf_bridge_encap_header_len(skb);
 636
 637         if (unlikely(!pskb_may_pull(skb, len)))
 638                 return NF_DROP;
 639
 640         p = br_port_get_rcu(in);
 641         if (p == NULL)
 642                 return NF_DROP;
 643         br = p->br;
 644
 645         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 646             IS_PPPOE_IPV6(skb)) {
 647                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
 648                         return NF_ACCEPT;
 649
 650                 nf_bridge_pull_encap_header_rcsum(skb);
 651                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 652         }
 653
 654         if (!brnf_call_iptables && !br->nf_call_iptables)
 655                 return NF_ACCEPT;
 656
 657         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
 658             !IS_PPPOE_IP(skb))
 659                 return NF_ACCEPT;
 660
 661         nf_bridge_pull_encap_header_rcsum(skb);
 662
 663         if (br_parse_ip_options(skb))
 664                 return NF_DROP;
 665
 666         nf_bridge_put(skb->nf_bridge);
 667         if (!nf_bridge_alloc(skb))
 668                 return NF_DROP;
 669         if (!setup_pre_routing(skb))
 670                 return NF_DROP;
 671         store_orig_dstaddr(skb);
 672         skb->protocol = htons(ETH_P_IP);
 673
 674         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 675                 br_nf_pre_routing_finish);
 676
 677         return NF_STOLEN;
 678 }
 679
 680
 681 /* PF_BRIDGE/LOCAL_IN ************************************************/
 682 /* The packet is locally destined, which requires a real
 683  * dst_entry, so detach the fake one.  On the way up, the
 684  * packet would pass through PRE_ROUTING again (which already
 685  * took place when the packet entered the bridge), but we
 686  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 687  * prevent this from happening. */
 688 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
 689                                    const struct net_device *in,
 690                                    const struct net_device *out,
 691                                    int (*okfn)(struct sk_buff *))
 692 {
 693         br_drop_fake_rtable(skb);
 694         return NF_ACCEPT;
 695 }
 696
 697 /* PF_BRIDGE/FORWARD *************************************************/
 698 static int br_nf_forward_finish(struct sk_buff *skb)
 699 {
 700         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 701         struct net_device *in;
 702
 703         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
 704                 in = nf_bridge->physindev;
 705                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 706                         skb->pkt_type = PACKET_OTHERHOST;
 707                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 708                 }
 709                 nf_bridge_update_protocol(skb);
 710         } else {
 711                 in = *((struct net_device **)(skb->cb));
 712         }
 713         nf_bridge_push_encap_header(skb);
 714
 715         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
 716                        skb->dev, br_forward_finish, 1);
 717         return 0;
 718 }
 719
 720 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 721  * netfilter with indev and outdev set to the bridge device,
 722  * but we are still able to filter on the 'real' indev/outdev
 723  * because of the physdev module. For ARP, indev and outdev are the
 724  * bridge ports. */
 725 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
 726                                      const struct net_device *in,
 727                                      const struct net_device *out,
 728                                      int (*okfn)(struct sk_buff *))
 729 {
 730         struct nf_bridge_info *nf_bridge;
 731         struct net_device *parent;
 732         u_int8_t pf;
 733
 734         if (!skb->nf_bridge)
 735                 return NF_ACCEPT;
 736
 737         /* Need exclusive nf_bridge_info since we might have multiple
 738          * different physoutdevs. */
 739         if (!nf_bridge_unshare(skb))
 740                 return NF_DROP;
 741
 742         parent = bridge_parent(out);
 743         if (!parent)
 744                 return NF_DROP;
 745
 746         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 747             IS_PPPOE_IP(skb))
 748                 pf = PF_INET;
 749         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 750                  IS_PPPOE_IPV6(skb))
 751                 pf = PF_INET6;
 752         else
 753                 return NF_ACCEPT;
 754
 755         nf_bridge_pull_encap_header(skb);
 756
 757         nf_bridge = skb->nf_bridge;
 758         if (skb->pkt_type == PACKET_OTHERHOST) {
 759                 skb->pkt_type = PACKET_HOST;
 760                 nf_bridge->mask |= BRNF_PKT_TYPE;
 761         }
 762
 763         if (pf == PF_INET && br_parse_ip_options(skb))
 764                 return NF_DROP;
 765
 766         /* The physdev module checks on this */
 767         nf_bridge->mask |= BRNF_BRIDGED;
 768         nf_bridge->physoutdev = skb->dev;
 769         if (pf == PF_INET)
 770                 skb->protocol = htons(ETH_P_IP);
 771         else
 772                 skb->protocol = htons(ETH_P_IPV6);
 773
 774         NF_HOOK(pf, NF_INET_FORWARD, skb, bridge_parent(in), parent,
 775                 br_nf_forward_finish);
 776
 777         return NF_STOLEN;
 778 }
 779
 780 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
 781                                       const struct net_device *in,
 782                                       const struct net_device *out,
 783                                       int (*okfn)(struct sk_buff *))
 784 {
 785         struct net_bridge_port *p;
 786         struct net_bridge *br;
 787         struct net_device **d = (struct net_device **)(skb->cb);
 788
 789         p = br_port_get_rcu(out);
 790         if (p == NULL)
 791                 return NF_ACCEPT;
 792         br = p->br;
 793
 794         if (!brnf_call_arptables && !br->nf_call_arptables)
 795                 return NF_ACCEPT;
 796
 797         if (skb->protocol != htons(ETH_P_ARP)) {
 798                 if (!IS_VLAN_ARP(skb))
 799                         return NF_ACCEPT;
 800                 nf_bridge_pull_encap_header(skb);
 801         }
 802
 803         if (arp_hdr(skb)->ar_pln != 4) {
 804                 if (IS_VLAN_ARP(skb))
 805                         nf_bridge_push_encap_header(skb);
 806                 return NF_ACCEPT;
 807         }
 808         *d = (struct net_device *)in;
 809         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 810                 (struct net_device *)out, br_nf_forward_finish);
 811
 812         return NF_STOLEN;
 813 }
 814
 815 #if defined(CONFIG_NF_CONNTRACK_IPV4) || defined(CONFIG_NF_CONNTRACK_IPV4_MODULE)
 816 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 817 {
 818         int ret;
 819
 820         if (skb->nfct != NULL && skb->protocol == htons(ETH_P_IP) &&
 821             skb->len + nf_bridge_mtu_reduction(skb) > skb->dev->mtu &&
 822             !skb_is_gso(skb)) {
 823                 if (br_parse_ip_options(skb))
 824                         /* Drop invalid packet */
 825                         goto drop;
 826                 ret = ip_fragment(skb, br_dev_queue_push_xmit);
 827         } else
 828                 ret = br_dev_queue_push_xmit(skb);
 829
 830         return ret;
 831  drop:
 832         kfree_skb(skb);
 833         return 0;
 834 }
 835 #else
 836 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 837 {
 838         return br_dev_queue_push_xmit(skb);
 839 }
 840 #endif
 841
 842 /* PF_BRIDGE/POST_ROUTING ********************************************/
 843 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
 844                                        const struct net_device *in,
 845                                        const struct net_device *out,
 846                                        int (*okfn)(struct sk_buff *))
 847 {
 848         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 849         struct net_device *realoutdev = bridge_parent(skb->dev);
 850         u_int8_t pf;
 851
 852         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
 853                 return NF_ACCEPT;
 854
 855         if (!realoutdev)
 856                 return NF_DROP;
 857
 858         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 859             IS_PPPOE_IP(skb))
 860                 pf = PF_INET;
 861         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 862                  IS_PPPOE_IPV6(skb))
 863                 pf = PF_INET6;
 864         else
 865                 return NF_ACCEPT;
 866
 867         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 868          * about the value of skb->pkt_type. */
 869         if (skb->pkt_type == PACKET_OTHERHOST) {
 870                 skb->pkt_type = PACKET_HOST;
 871                 nf_bridge->mask |= BRNF_PKT_TYPE;
 872         }
 873
 874         nf_bridge_pull_encap_header(skb);
 875         nf_bridge_save_header(skb);
 876         if (pf == PF_INET)
 877                 skb->protocol = htons(ETH_P_IP);
 878         else
 879                 skb->protocol = htons(ETH_P_IPV6);
 880
 881         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
 882                 br_nf_dev_queue_xmit);
 883
 884         return NF_STOLEN;
 885 }
 886
 887 /* IP/SABOTAGE *****************************************************/
 888 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 889  * for the second time. */
 890 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
 891                                    const struct net_device *in,
 892                                    const struct net_device *out,
 893                                    int (*okfn)(struct sk_buff *))
 894 {
 895         if (skb->nf_bridge &&
 896             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 897                 return NF_STOP;
 898         }
 899
 900         return NF_ACCEPT;
 901 }
 902
 903 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 904  * br_dev_queue_push_xmit is called afterwards */
 905 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
 906         {
 907                 .hook = br_nf_pre_routing,
 908                 .owner = THIS_MODULE,
 909                 .pf = PF_BRIDGE,
 910                 .hooknum = NF_BR_PRE_ROUTING,
 911                 .priority = NF_BR_PRI_BRNF,
 912         },
 913         {
 914                 .hook = br_nf_local_in,
 915                 .owner = THIS_MODULE,
 916                 .pf = PF_BRIDGE,
 917                 .hooknum = NF_BR_LOCAL_IN,
 918                 .priority = NF_BR_PRI_BRNF,
 919         },
 920         {
 921                 .hook = br_nf_forward_ip,
 922                 .owner = THIS_MODULE,
 923                 .pf = PF_BRIDGE,
 924                 .hooknum = NF_BR_FORWARD,
 925                 .priority = NF_BR_PRI_BRNF - 1,
 926         },
 927         {
 928                 .hook = br_nf_forward_arp,
 929                 .owner = THIS_MODULE,
 930                 .pf = PF_BRIDGE,
 931                 .hooknum = NF_BR_FORWARD,
 932                 .priority = NF_BR_PRI_BRNF,
 933         },
 934         {
 935                 .hook = br_nf_post_routing,
 936                 .owner = THIS_MODULE,
 937                 .pf = PF_BRIDGE,
 938                 .hooknum = NF_BR_POST_ROUTING,
 939                 .priority = NF_BR_PRI_LAST,
 940         },
 941         {
 942                 .hook = ip_sabotage_in,
 943                 .owner = THIS_MODULE,
 944                 .pf = PF_INET,
 945                 .hooknum = NF_INET_PRE_ROUTING,
 946                 .priority = NF_IP_PRI_FIRST,
 947         },
 948         {
 949                 .hook = ip_sabotage_in,
 950                 .owner = THIS_MODULE,
 951                 .pf = PF_INET6,
 952                 .hooknum = NF_INET_PRE_ROUTING,
 953                 .priority = NF_IP6_PRI_FIRST,
 954         },
 955 };
 956
 957 #ifdef CONFIG_SYSCTL
 958 static
 959 int brnf_sysctl_call_tables(ctl_table * ctl, int write,
 960                             void __user * buffer, size_t * lenp, loff_t * ppos)
 961 {
 962         int ret;
 963
 964         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
 965
 966         if (write && *(int *)(ctl->data))
 967                 *(int *)(ctl->data) = 1;
 968         return ret;
 969 }
 970
 971 static ctl_table brnf_table[] = {
 972         {
 973                 .procname       = "bridge-nf-call-arptables",
 974                 .data           = &brnf_call_arptables,
 975                 .maxlen         = sizeof(int),
 976                 .mode           = 0644,
 977                 .proc_handler   = brnf_sysctl_call_tables,
 978         },
 979         {
 980                 .procname       = "bridge-nf-call-iptables",
 981                 .data           = &brnf_call_iptables,
 982                 .maxlen         = sizeof(int),
 983                 .mode           = 0644,
 984                 .proc_handler   = brnf_sysctl_call_tables,
 985         },
 986         {
 987                 .procname       = "bridge-nf-call-ip6tables",
 988                 .data           = &brnf_call_ip6tables,
 989                 .maxlen         = sizeof(int),
 990                 .mode           = 0644,
 991                 .proc_handler   = brnf_sysctl_call_tables,
 992         },
 993         {
 994                 .procname       = "bridge-nf-filter-vlan-tagged",
 995                 .data           = &brnf_filter_vlan_tagged,
 996                 .maxlen         = sizeof(int),
 997                 .mode           = 0644,
 998                 .proc_handler   = brnf_sysctl_call_tables,
 999         },
1000         {
1001                 .procname       = "bridge-nf-filter-pppoe-tagged",
1002                 .data           = &brnf_filter_pppoe_tagged,
1003                 .maxlen         = sizeof(int),
1004                 .mode           = 0644,
1005                 .proc_handler   = brnf_sysctl_call_tables,
1006         },
1007         { }
1008 };
1009
1010 static struct ctl_path brnf_path[] = {
1011         { .procname = "net", },
1012         { .procname = "bridge", },
1013         { }
1014 };
1015 #endif
1016
1017 int __init br_netfilter_init(void)
1018 {
1019         int ret;
1020
1021         ret = dst_entries_init(&fake_dst_ops);
1022         if (ret < 0)
1023                 return ret;
1024
1025         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1026         if (ret < 0) {
1027                 dst_entries_destroy(&fake_dst_ops);
1028                 return ret;
1029         }
1030 #ifdef CONFIG_SYSCTL
1031         brnf_sysctl_header = register_sysctl_paths(brnf_path, brnf_table);
1032         if (brnf_sysctl_header == NULL) {
1033                 printk(KERN_WARNING
1034                        "br_netfilter: can't register to sysctl.\n");
1035                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1036                 dst_entries_destroy(&fake_dst_ops);
1037                 return -ENOMEM;
1038         }
1039 #endif
1040         printk(KERN_NOTICE "Bridge firewalling registered\n");
1041         return 0;
1042 }
1043
1044 void br_netfilter_fini(void)
1045 {
1046         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1047 #ifdef CONFIG_SYSCTL
1048         unregister_sysctl_table(brnf_sysctl_header);
1049 #endif
1050         dst_entries_destroy(&fake_dst_ops);
1051 }