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