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 void fake_update_pmtu(struct dst_entry *dst, struct sock *sk,
 115                              struct sk_buff *skb, u32 mtu)
 116 {
 117 }
 118
 119 static void fake_redirect(struct dst_entry *dst, struct sock *sk,
 120                           struct sk_buff *skb)
 121 {
 122 }
 123
 124 static u32 *fake_cow_metrics(struct dst_entry *dst, unsigned long old)
 125 {
 126         return NULL;
 127 }
 128
 129 static struct neighbour *fake_neigh_lookup(const struct dst_entry *dst,
 130                                            struct sk_buff *skb,
 131                                            const void *daddr)
 132 {
 133         return NULL;
 134 }
 135
 136 static unsigned int fake_mtu(const struct dst_entry *dst)
 137 {
 138         return dst->dev->mtu;
 139 }
 140
 141 static struct dst_ops fake_dst_ops = {
 142         .family =               AF_INET,
 143         .protocol =             cpu_to_be16(ETH_P_IP),
 144         .update_pmtu =          fake_update_pmtu,
 145         .redirect =             fake_redirect,
 146         .cow_metrics =          fake_cow_metrics,
 147         .neigh_lookup =         fake_neigh_lookup,
 148         .mtu =                  fake_mtu,
 149 };
 150
 151 /*
 152  * Initialize bogus route table used to keep netfilter happy.
 153  * Currently, we fill in the PMTU entry because netfilter
 154  * refragmentation needs it, and the rt_flags entry because
 155  * ipt_REJECT needs it.  Future netfilter modules might
 156  * require us to fill additional fields.
 157  */
 158 static const u32 br_dst_default_metrics[RTAX_MAX] = {
 159         [RTAX_MTU - 1] = 1500,
 160 };
 161
 162 void br_netfilter_rtable_init(struct net_bridge *br)
 163 {
 164         struct rtable *rt = &br->fake_rtable;
 165
 166         atomic_set(&rt->dst.__refcnt, 1);
 167         rt->dst.dev = br->dev;
 168         rt->dst.path = &rt->dst;
 169         dst_init_metrics(&rt->dst, br_dst_default_metrics, true);
 170         rt->dst.flags   = DST_NOXFRM | DST_NOPEER | DST_FAKE_RTABLE;
 171         rt->dst.ops = &fake_dst_ops;
 172 }
 173
 174 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
 175 {
 176         struct net_bridge_port *port;
 177
 178         port = br_port_get_rcu(dev);
 179         return port ? &port->br->fake_rtable : NULL;
 180 }
 181
 182 static inline struct net_device *bridge_parent(const struct net_device *dev)
 183 {
 184         struct net_bridge_port *port;
 185
 186         port = br_port_get_rcu(dev);
 187         return port ? port->br->dev : NULL;
 188 }
 189
 190 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 191 {
 192         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 193         if (likely(skb->nf_bridge))
 194                 atomic_set(&(skb->nf_bridge->use), 1);
 195
 196         return skb->nf_bridge;
 197 }
 198
 199 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 200 {
 201         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 202
 203         if (atomic_read(&nf_bridge->use) > 1) {
 204                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 205
 206                 if (tmp) {
 207                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 208                         atomic_set(&tmp->use, 1);
 209                 }
 210                 nf_bridge_put(nf_bridge);
 211                 nf_bridge = tmp;
 212         }
 213         return nf_bridge;
 214 }
 215
 216 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 217 {
 218         unsigned int len = nf_bridge_encap_header_len(skb);
 219
 220         skb_push(skb, len);
 221         skb->network_header -= len;
 222 }
 223
 224 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 225 {
 226         unsigned int len = nf_bridge_encap_header_len(skb);
 227
 228         skb_pull(skb, len);
 229         skb->network_header += len;
 230 }
 231
 232 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 233 {
 234         unsigned int len = nf_bridge_encap_header_len(skb);
 235
 236         skb_pull_rcsum(skb, len);
 237         skb->network_header += len;
 238 }
 239
 240 static inline void nf_bridge_save_header(struct sk_buff *skb)
 241 {
 242         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 243
 244         skb_copy_from_linear_data_offset(skb, -header_size,
 245                                          skb->nf_bridge->data, header_size);
 246 }
 247
 248 static inline void nf_bridge_update_protocol(struct sk_buff *skb)
 249 {
 250         if (skb->nf_bridge->mask & BRNF_8021Q)
 251                 skb->protocol = htons(ETH_P_8021Q);
 252         else if (skb->nf_bridge->mask & BRNF_PPPoE)
 253                 skb->protocol = htons(ETH_P_PPP_SES);
 254 }
 255
 256 /* When handing a packet over to the IP layer
 257  * check whether we have a skb that is in the
 258  * expected format
 259  */
 260
 261 static int br_parse_ip_options(struct sk_buff *skb)
 262 {
 263         struct ip_options *opt;
 264         const struct iphdr *iph;
 265         struct net_device *dev = skb->dev;
 266         u32 len;
 267
 268         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 269                 goto inhdr_error;
 270
 271         iph = ip_hdr(skb);
 272         opt = &(IPCB(skb)->opt);
 273
 274         /* Basic sanity checks */
 275         if (iph->ihl < 5 || iph->version != 4)
 276                 goto inhdr_error;
 277
 278         if (!pskb_may_pull(skb, iph->ihl*4))
 279                 goto inhdr_error;
 280
 281         iph = ip_hdr(skb);
 282         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 283                 goto inhdr_error;
 284
 285         len = ntohs(iph->tot_len);
 286         if (skb->len < len) {
 287                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
 288                 goto drop;
 289         } else if (len < (iph->ihl*4))
 290                 goto inhdr_error;
 291
 292         if (pskb_trim_rcsum(skb, len)) {
 293                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 294                 goto drop;
 295         }
 296
 297         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 298         if (iph->ihl == 5)
 299                 return 0;
 300
 301         opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
 302         if (ip_options_compile(dev_net(dev), opt, skb))
 303                 goto inhdr_error;
 304
 305         /* Check correct handling of SRR option */
 306         if (unlikely(opt->srr)) {
 307                 struct in_device *in_dev = __in_dev_get_rcu(dev);
 308                 if (in_dev && !IN_DEV_SOURCE_ROUTE(in_dev))
 309                         goto drop;
 310
 311                 if (ip_options_rcv_srr(skb))
 312                         goto drop;
 313         }
 314
 315         return 0;
 316
 317 inhdr_error:
 318         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 319 drop:
 320         return -1;
 321 }
 322
 323 /* Fill in the header for fragmented IP packets handled by
 324  * the IPv4 connection tracking code.
 325  */
 326 int nf_bridge_copy_header(struct sk_buff *skb)
 327 {
 328         int err;
 329         unsigned int header_size;
 330
 331         nf_bridge_update_protocol(skb);
 332         header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 333         err = skb_cow_head(skb, header_size);
 334         if (err)
 335                 return err;
 336
 337         skb_copy_to_linear_data_offset(skb, -header_size,
 338                                        skb->nf_bridge->data, header_size);
 339         __skb_push(skb, nf_bridge_encap_header_len(skb));
 340         return 0;
 341 }
 342
 343 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 344 /* Undo the changes made for ip6tables PREROUTING and continue the
 345  * bridge PRE_ROUTING hook. */
 346 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 347 {
 348         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 349         struct rtable *rt;
 350
 351         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 352                 skb->pkt_type = PACKET_OTHERHOST;
 353                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 354         }
 355         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 356
 357         rt = bridge_parent_rtable(nf_bridge->physindev);
 358         if (!rt) {
 359                 kfree_skb(skb);
 360                 return 0;
 361         }
 362         skb_dst_set_noref(skb, &rt->dst);
 363
 364         skb->dev = nf_bridge->physindev;
 365         nf_bridge_update_protocol(skb);
 366         nf_bridge_push_encap_header(skb);
 367         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 368                        br_handle_frame_finish, 1);
 369
 370         return 0;
 371 }
 372
 373 /* Obtain the correct destination MAC address, while preserving the original
 374  * source MAC address. If we already know this address, we just copy it. If we
 375  * don't, we use the neighbour framework to find out. In both cases, we make
 376  * sure that br_handle_frame_finish() is called afterwards.
 377  */
 378 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 379 {
 380         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 381         struct neighbour *neigh;
 382         struct dst_entry *dst;
 383
 384         skb->dev = bridge_parent(skb->dev);
 385         if (!skb->dev)
 386                 goto free_skb;
 387         dst = skb_dst(skb);
 388         neigh = dst_neigh_lookup_skb(dst, skb);
 389         if (neigh) {
 390                 int ret;
 391
 392                 if (neigh->hh.hh_len) {
 393                         neigh_hh_bridge(&neigh->hh, skb);
 394                         skb->dev = nf_bridge->physindev;
 395                         ret = br_handle_frame_finish(skb);
 396                 } else {
 397                         /* the neighbour function below overwrites the complete
 398                          * MAC header, so we save the Ethernet source address and
 399                          * protocol number.
 400                          */
 401                         skb_copy_from_linear_data_offset(skb,
 402                                                          -(ETH_HLEN-ETH_ALEN),
 403                                                          skb->nf_bridge->data,
 404                                                          ETH_HLEN-ETH_ALEN);
 405                         /* tell br_dev_xmit to continue with forwarding */
 406                         nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 407                         ret = neigh->output(neigh, skb);
 408                 }
 409                 neigh_release(neigh);
 410                 return ret;
 411         }
 412 free_skb:
 413         kfree_skb(skb);
 414         return 0;
 415 }
 416
 417 /* This requires some explaining. If DNAT has taken place,
 418  * we will need to fix up the destination Ethernet address.
 419  *
 420  * There are two cases to consider:
 421  * 1. The packet was DNAT'ed to a device in the same bridge
 422  *    port group as it was received on. We can still bridge
 423  *    the packet.
 424  * 2. The packet was DNAT'ed to a different device, either
 425  *    a non-bridged device or another bridge port group.
 426  *    The packet will need to be routed.
 427  *
 428  * The correct way of distinguishing between these two cases is to
 429  * call ip_route_input() and to look at skb->dst->dev, which is
 430  * changed to the destination device if ip_route_input() succeeds.
 431  *
 432  * Let's first consider the case that ip_route_input() succeeds:
 433  *
 434  * If the output device equals the logical bridge device the packet
 435  * came in on, we can consider this bridging. The corresponding MAC
 436  * address will be obtained in br_nf_pre_routing_finish_bridge.
 437  * Otherwise, the packet is considered to be routed and we just
 438  * change the destination MAC address so that the packet will
 439  * later be passed up to the IP stack to be routed. For a redirected
 440  * packet, ip_route_input() will give back the localhost as output device,
 441  * which differs from the bridge device.
 442  *
 443  * Let's now consider the case that ip_route_input() fails:
 444  *
 445  * This can be because the destination address is martian, in which case
 446  * the packet will be dropped.
 447  * If IP forwarding is disabled, ip_route_input() will fail, while
 448  * ip_route_output_key() can return success. The source
 449  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 450  * thinks we're handling a locally generated packet and won't care
 451  * if IP forwarding is enabled. If the output device equals the logical bridge
 452  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 453  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 454  */
 455 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 456 {
 457         struct net_device *dev = skb->dev;
 458         struct iphdr *iph = ip_hdr(skb);
 459         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 460         struct rtable *rt;
 461         int err;
 462
 463         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 464                 skb->pkt_type = PACKET_OTHERHOST;
 465                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 466         }
 467         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 468         if (dnat_took_place(skb)) {
 469                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 470                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 471
 472                         /* If err equals -EHOSTUNREACH the error is due to a
 473                          * martian destination or due to the fact that
 474                          * forwarding is disabled. For most martian packets,
 475                          * ip_route_output_key() will fail. It won't fail for 2 types of
 476                          * martian destinations: loopback destinations and destination
 477                          * 0.0.0.0. In both cases the packet will be dropped because the
 478                          * destination is the loopback device and not the bridge. */
 479                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 480                                 goto free_skb;
 481
 482                         rt = ip_route_output(dev_net(dev), iph->daddr, 0,
 483                                              RT_TOS(iph->tos), 0);
 484                         if (!IS_ERR(rt)) {
 485                                 /* - Bridged-and-DNAT'ed traffic doesn't
 486                                  *   require ip_forwarding. */
 487                                 if (rt->dst.dev == dev) {
 488                                         skb_dst_set(skb, &rt->dst);
 489                                         goto bridged_dnat;
 490                                 }
 491                                 ip_rt_put(rt);
 492                         }
 493 free_skb:
 494                         kfree_skb(skb);
 495                         return 0;
 496                 } else {
 497                         if (skb_dst(skb)->dev == dev) {
 498 bridged_dnat:
 499                                 skb->dev = nf_bridge->physindev;
 500                                 nf_bridge_update_protocol(skb);
 501                                 nf_bridge_push_encap_header(skb);
 502                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
 503                                                NF_BR_PRE_ROUTING,
 504                                                skb, skb->dev, NULL,
 505                                                br_nf_pre_routing_finish_bridge,
 506                                                1);
 507                                 return 0;
 508                         }
 509                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
 510                         skb->pkt_type = PACKET_HOST;
 511                 }
 512         } else {
 513                 rt = bridge_parent_rtable(nf_bridge->physindev);
 514                 if (!rt) {
 515                         kfree_skb(skb);
 516                         return 0;
 517                 }
 518                 skb_dst_set_noref(skb, &rt->dst);
 519         }
 520
 521         skb->dev = nf_bridge->physindev;
 522         nf_bridge_update_protocol(skb);
 523         nf_bridge_push_encap_header(skb);
 524         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 525                        br_handle_frame_finish, 1);
 526
 527         return 0;
 528 }
 529
 530 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb, const struct net_device *dev)
 531 {
 532         struct net_device *vlan, *br;
 533
 534         br = bridge_parent(dev);
 535         if (brnf_pass_vlan_indev == 0 || !vlan_tx_tag_present(skb))
 536                 return br;
 537
 538         vlan = __vlan_find_dev_deep(br, skb->vlan_proto,
 539                                     vlan_tx_tag_get(skb) & VLAN_VID_MASK);
 540
 541         return vlan ? vlan : br;
 542 }
 543
 544 /* Some common code for IPv4/IPv6 */
 545 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 546 {
 547         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 548
 549         if (skb->pkt_type == PACKET_OTHERHOST) {
 550                 skb->pkt_type = PACKET_HOST;
 551                 nf_bridge->mask |= BRNF_PKT_TYPE;
 552         }
 553
 554         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 555         nf_bridge->physindev = skb->dev;
 556         skb->dev = brnf_get_logical_dev(skb, skb->dev);
 557         if (skb->protocol == htons(ETH_P_8021Q))
 558                 nf_bridge->mask |= BRNF_8021Q;
 559         else if (skb->protocol == htons(ETH_P_PPP_SES))
 560                 nf_bridge->mask |= BRNF_PPPoE;
 561
 562         /* Must drop socket now because of tproxy. */
 563         skb_orphan(skb);
 564         return skb->dev;
 565 }
 566
 567 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 568 static int check_hbh_len(struct sk_buff *skb)
 569 {
 570         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 571         u32 pkt_len;
 572         const unsigned char *nh = skb_network_header(skb);
 573         int off = raw - nh;
 574         int len = (raw[1] + 1) << 3;
 575
 576         if ((raw + len) - skb->data > skb_headlen(skb))
 577                 goto bad;
 578
 579         off += 2;
 580         len -= 2;
 581
 582         while (len > 0) {
 583                 int optlen = nh[off + 1] + 2;
 584
 585                 switch (nh[off]) {
 586                 case IPV6_TLV_PAD1:
 587                         optlen = 1;
 588                         break;
 589
 590                 case IPV6_TLV_PADN:
 591                         break;
 592
 593                 case IPV6_TLV_JUMBO:
 594                         if (nh[off + 1] != 4 || (off & 3) != 2)
 595                                 goto bad;
 596                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 597                         if (pkt_len <= IPV6_MAXPLEN ||
 598                             ipv6_hdr(skb)->payload_len)
 599                                 goto bad;
 600                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 601                                 goto bad;
 602                         if (pskb_trim_rcsum(skb,
 603                                             pkt_len + sizeof(struct ipv6hdr)))
 604                                 goto bad;
 605                         nh = skb_network_header(skb);
 606                         break;
 607                 default:
 608                         if (optlen > len)
 609                                 goto bad;
 610                         break;
 611                 }
 612                 off += optlen;
 613                 len -= optlen;
 614         }
 615         if (len == 0)
 616                 return 0;
 617 bad:
 618         return -1;
 619
 620 }
 621
 622 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 623  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 624 static unsigned int br_nf_pre_routing_ipv6(const struct nf_hook_ops *ops,
 625                                            struct sk_buff *skb,
 626                                            const struct net_device *in,
 627                                            const struct net_device *out,
 628                                            int (*okfn)(struct sk_buff *))
 629 {
 630         const struct ipv6hdr *hdr;
 631         u32 pkt_len;
 632
 633         if (skb->len < sizeof(struct ipv6hdr))
 634                 return NF_DROP;
 635
 636         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 637                 return NF_DROP;
 638
 639         hdr = ipv6_hdr(skb);
 640
 641         if (hdr->version != 6)
 642                 return NF_DROP;
 643
 644         pkt_len = ntohs(hdr->payload_len);
 645
 646         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 647                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 648                         return NF_DROP;
 649                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 650                         return NF_DROP;
 651         }
 652         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 653                 return NF_DROP;
 654
 655         nf_bridge_put(skb->nf_bridge);
 656         if (!nf_bridge_alloc(skb))
 657                 return NF_DROP;
 658         if (!setup_pre_routing(skb))
 659                 return NF_DROP;
 660
 661         skb->protocol = htons(ETH_P_IPV6);
 662         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 663                 br_nf_pre_routing_finish_ipv6);
 664
 665         return NF_STOLEN;
 666 }
 667
 668 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 669  * Replicate the checks that IPv4 does on packet reception.
 670  * Set skb->dev to the bridge device (i.e. parent of the
 671  * receiving device) to make netfilter happy, the REDIRECT
 672  * target in particular.  Save the original destination IP
 673  * address to be able to detect DNAT afterwards. */
 674 static unsigned int br_nf_pre_routing(const struct nf_hook_ops *ops,
 675                                       struct sk_buff *skb,
 676                                       const struct net_device *in,
 677                                       const struct net_device *out,
 678                                       int (*okfn)(struct sk_buff *))
 679 {
 680         struct net_bridge_port *p;
 681         struct net_bridge *br;
 682         __u32 len = nf_bridge_encap_header_len(skb);
 683
 684         if (unlikely(!pskb_may_pull(skb, len)))
 685                 return NF_DROP;
 686
 687         p = br_port_get_rcu(in);
 688         if (p == NULL)
 689                 return NF_DROP;
 690         br = p->br;
 691
 692         if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb)) {
 693                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
 694                         return NF_ACCEPT;
 695
 696                 nf_bridge_pull_encap_header_rcsum(skb);
 697                 return br_nf_pre_routing_ipv6(ops, skb, in, out, okfn);
 698         }
 699
 700         if (!brnf_call_iptables && !br->nf_call_iptables)
 701                 return NF_ACCEPT;
 702
 703         if (!IS_IP(skb) && !IS_VLAN_IP(skb) && !IS_PPPOE_IP(skb))
 704                 return NF_ACCEPT;
 705
 706         nf_bridge_pull_encap_header_rcsum(skb);
 707
 708         if (br_parse_ip_options(skb))
 709                 return NF_DROP;
 710
 711         nf_bridge_put(skb->nf_bridge);
 712         if (!nf_bridge_alloc(skb))
 713                 return NF_DROP;
 714         if (!setup_pre_routing(skb))
 715                 return NF_DROP;
 716         store_orig_dstaddr(skb);
 717         skb->protocol = htons(ETH_P_IP);
 718
 719         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 720                 br_nf_pre_routing_finish);
 721
 722         return NF_STOLEN;
 723 }
 724
 725
 726 /* PF_BRIDGE/LOCAL_IN ************************************************/
 727 /* The packet is locally destined, which requires a real
 728  * dst_entry, so detach the fake one.  On the way up, the
 729  * packet would pass through PRE_ROUTING again (which already
 730  * took place when the packet entered the bridge), but we
 731  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 732  * prevent this from happening. */
 733 static unsigned int br_nf_local_in(const struct nf_hook_ops *ops,
 734                                    struct sk_buff *skb,
 735                                    const struct net_device *in,
 736                                    const struct net_device *out,
 737                                    int (*okfn)(struct sk_buff *))
 738 {
 739         br_drop_fake_rtable(skb);
 740         return NF_ACCEPT;
 741 }
 742
 743 /* PF_BRIDGE/FORWARD *************************************************/
 744 static int br_nf_forward_finish(struct sk_buff *skb)
 745 {
 746         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 747         struct net_device *in;
 748
 749         if (!IS_ARP(skb) && !IS_VLAN_ARP(skb)) {
 750                 in = nf_bridge->physindev;
 751                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 752                         skb->pkt_type = PACKET_OTHERHOST;
 753                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 754                 }
 755                 nf_bridge_update_protocol(skb);
 756         } else {
 757                 in = *((struct net_device **)(skb->cb));
 758         }
 759         nf_bridge_push_encap_header(skb);
 760
 761         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
 762                        skb->dev, br_forward_finish, 1);
 763         return 0;
 764 }
 765
 766
 767 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 768  * netfilter with indev and outdev set to the bridge device,
 769  * but we are still able to filter on the 'real' indev/outdev
 770  * because of the physdev module. For ARP, indev and outdev are the
 771  * bridge ports. */
 772 static unsigned int br_nf_forward_ip(const struct nf_hook_ops *ops,
 773                                      struct sk_buff *skb,
 774                                      const struct net_device *in,
 775                                      const struct net_device *out,
 776                                      int (*okfn)(struct sk_buff *))
 777 {
 778         struct nf_bridge_info *nf_bridge;
 779         struct net_device *parent;
 780         u_int8_t pf;
 781
 782         if (!skb->nf_bridge)
 783                 return NF_ACCEPT;
 784
 785         /* Need exclusive nf_bridge_info since we might have multiple
 786          * different physoutdevs. */
 787         if (!nf_bridge_unshare(skb))
 788                 return NF_DROP;
 789
 790         parent = bridge_parent(out);
 791         if (!parent)
 792                 return NF_DROP;
 793
 794         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 795                 pf = NFPROTO_IPV4;
 796         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 797                 pf = NFPROTO_IPV6;
 798         else
 799                 return NF_ACCEPT;
 800
 801         nf_bridge_pull_encap_header(skb);
 802
 803         nf_bridge = skb->nf_bridge;
 804         if (skb->pkt_type == PACKET_OTHERHOST) {
 805                 skb->pkt_type = PACKET_HOST;
 806                 nf_bridge->mask |= BRNF_PKT_TYPE;
 807         }
 808
 809         if (pf == NFPROTO_IPV4 && br_parse_ip_options(skb))
 810                 return NF_DROP;
 811
 812         /* The physdev module checks on this */
 813         nf_bridge->mask |= BRNF_BRIDGED;
 814         nf_bridge->physoutdev = skb->dev;
 815         if (pf == NFPROTO_IPV4)
 816                 skb->protocol = htons(ETH_P_IP);
 817         else
 818                 skb->protocol = htons(ETH_P_IPV6);
 819
 820         NF_HOOK(pf, NF_INET_FORWARD, skb, brnf_get_logical_dev(skb, in), parent,
 821                 br_nf_forward_finish);
 822
 823         return NF_STOLEN;
 824 }
 825
 826 static unsigned int br_nf_forward_arp(const struct nf_hook_ops *ops,
 827                                       struct sk_buff *skb,
 828                                       const struct net_device *in,
 829                                       const struct net_device *out,
 830                                       int (*okfn)(struct sk_buff *))
 831 {
 832         struct net_bridge_port *p;
 833         struct net_bridge *br;
 834         struct net_device **d = (struct net_device **)(skb->cb);
 835
 836         p = br_port_get_rcu(out);
 837         if (p == NULL)
 838                 return NF_ACCEPT;
 839         br = p->br;
 840
 841         if (!brnf_call_arptables && !br->nf_call_arptables)
 842                 return NF_ACCEPT;
 843
 844         if (!IS_ARP(skb)) {
 845                 if (!IS_VLAN_ARP(skb))
 846                         return NF_ACCEPT;
 847                 nf_bridge_pull_encap_header(skb);
 848         }
 849
 850         if (arp_hdr(skb)->ar_pln != 4) {
 851                 if (IS_VLAN_ARP(skb))
 852                         nf_bridge_push_encap_header(skb);
 853                 return NF_ACCEPT;
 854         }
 855         *d = (struct net_device *)in;
 856         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 857                 (struct net_device *)out, br_nf_forward_finish);
 858
 859         return NF_STOLEN;
 860 }
 861
 862 #if IS_ENABLED(CONFIG_NF_CONNTRACK_IPV4)
 863 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 864 {
 865         int ret;
 866
 867         if (skb->nfct != NULL && skb->protocol == htons(ETH_P_IP) &&
 868             skb->len + nf_bridge_mtu_reduction(skb) > skb->dev->mtu &&
 869             !skb_is_gso(skb)) {
 870                 if (br_parse_ip_options(skb))
 871                         /* Drop invalid packet */
 872                         return NF_DROP;
 873                 ret = ip_fragment(skb, br_dev_queue_push_xmit);
 874         } else
 875                 ret = br_dev_queue_push_xmit(skb);
 876
 877         return ret;
 878 }
 879 #else
 880 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 881 {
 882         return br_dev_queue_push_xmit(skb);
 883 }
 884 #endif
 885
 886 /* PF_BRIDGE/POST_ROUTING ********************************************/
 887 static unsigned int br_nf_post_routing(const struct nf_hook_ops *ops,
 888                                        struct sk_buff *skb,
 889                                        const struct net_device *in,
 890                                        const struct net_device *out,
 891                                        int (*okfn)(struct sk_buff *))
 892 {
 893         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 894         struct net_device *realoutdev = bridge_parent(skb->dev);
 895         u_int8_t pf;
 896
 897         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
 898                 return NF_ACCEPT;
 899
 900         if (!realoutdev)
 901                 return NF_DROP;
 902
 903         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 904                 pf = NFPROTO_IPV4;
 905         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 906                 pf = NFPROTO_IPV6;
 907         else
 908                 return NF_ACCEPT;
 909
 910         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 911          * about the value of skb->pkt_type. */
 912         if (skb->pkt_type == PACKET_OTHERHOST) {
 913                 skb->pkt_type = PACKET_HOST;
 914                 nf_bridge->mask |= BRNF_PKT_TYPE;
 915         }
 916
 917         nf_bridge_pull_encap_header(skb);
 918         nf_bridge_save_header(skb);
 919         if (pf == NFPROTO_IPV4)
 920                 skb->protocol = htons(ETH_P_IP);
 921         else
 922                 skb->protocol = htons(ETH_P_IPV6);
 923
 924         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
 925                 br_nf_dev_queue_xmit);
 926
 927         return NF_STOLEN;
 928 }
 929
 930 /* IP/SABOTAGE *****************************************************/
 931 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 932  * for the second time. */
 933 static unsigned int ip_sabotage_in(const struct nf_hook_ops *ops,
 934                                    struct sk_buff *skb,
 935                                    const struct net_device *in,
 936                                    const struct net_device *out,
 937                                    int (*okfn)(struct sk_buff *))
 938 {
 939         if (skb->nf_bridge &&
 940             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 941                 return NF_STOP;
 942         }
 943
 944         return NF_ACCEPT;
 945 }
 946
 947 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 948  * br_dev_queue_push_xmit is called afterwards */
 949 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
 950         {
 951                 .hook = br_nf_pre_routing,
 952                 .owner = THIS_MODULE,
 953                 .pf = NFPROTO_BRIDGE,
 954                 .hooknum = NF_BR_PRE_ROUTING,
 955                 .priority = NF_BR_PRI_BRNF,
 956         },
 957         {
 958                 .hook = br_nf_local_in,
 959                 .owner = THIS_MODULE,
 960                 .pf = NFPROTO_BRIDGE,
 961                 .hooknum = NF_BR_LOCAL_IN,
 962                 .priority = NF_BR_PRI_BRNF,
 963         },
 964         {
 965                 .hook = br_nf_forward_ip,
 966                 .owner = THIS_MODULE,
 967                 .pf = NFPROTO_BRIDGE,
 968                 .hooknum = NF_BR_FORWARD,
 969                 .priority = NF_BR_PRI_BRNF - 1,
 970         },
 971         {
 972                 .hook = br_nf_forward_arp,
 973                 .owner = THIS_MODULE,
 974                 .pf = NFPROTO_BRIDGE,
 975                 .hooknum = NF_BR_FORWARD,
 976                 .priority = NF_BR_PRI_BRNF,
 977         },
 978         {
 979                 .hook = br_nf_post_routing,
 980                 .owner = THIS_MODULE,
 981                 .pf = NFPROTO_BRIDGE,
 982                 .hooknum = NF_BR_POST_ROUTING,
 983                 .priority = NF_BR_PRI_LAST,
 984         },
 985         {
 986                 .hook = ip_sabotage_in,
 987                 .owner = THIS_MODULE,
 988                 .pf = NFPROTO_IPV4,
 989                 .hooknum = NF_INET_PRE_ROUTING,
 990                 .priority = NF_IP_PRI_FIRST,
 991         },
 992         {
 993                 .hook = ip_sabotage_in,
 994                 .owner = THIS_MODULE,
 995                 .pf = NFPROTO_IPV6,
 996                 .hooknum = NF_INET_PRE_ROUTING,
 997                 .priority = NF_IP6_PRI_FIRST,
 998         },
 999 };
1000
1001 #ifdef CONFIG_SYSCTL
1002 static
1003 int brnf_sysctl_call_tables(struct ctl_table *ctl, int write,
1004                             void __user * buffer, size_t * lenp, loff_t * ppos)
1005 {
1006         int ret;
1007
1008         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1009
1010         if (write && *(int *)(ctl->data))
1011                 *(int *)(ctl->data) = 1;
1012         return ret;
1013 }
1014
1015 static struct ctl_table brnf_table[] = {
1016         {
1017                 .procname       = "bridge-nf-call-arptables",
1018                 .data           = &brnf_call_arptables,
1019                 .maxlen         = sizeof(int),
1020                 .mode           = 0644,
1021                 .proc_handler   = brnf_sysctl_call_tables,
1022         },
1023         {
1024                 .procname       = "bridge-nf-call-iptables",
1025                 .data           = &brnf_call_iptables,
1026                 .maxlen         = sizeof(int),
1027                 .mode           = 0644,
1028                 .proc_handler   = brnf_sysctl_call_tables,
1029         },
1030         {
1031                 .procname       = "bridge-nf-call-ip6tables",
1032                 .data           = &brnf_call_ip6tables,
1033                 .maxlen         = sizeof(int),
1034                 .mode           = 0644,
1035                 .proc_handler   = brnf_sysctl_call_tables,
1036         },
1037         {
1038                 .procname       = "bridge-nf-filter-vlan-tagged",
1039                 .data           = &brnf_filter_vlan_tagged,
1040                 .maxlen         = sizeof(int),
1041                 .mode           = 0644,
1042                 .proc_handler   = brnf_sysctl_call_tables,
1043         },
1044         {
1045                 .procname       = "bridge-nf-filter-pppoe-tagged",
1046                 .data           = &brnf_filter_pppoe_tagged,
1047                 .maxlen         = sizeof(int),
1048                 .mode           = 0644,
1049                 .proc_handler   = brnf_sysctl_call_tables,
1050         },
1051         {
1052                 .procname       = "bridge-nf-pass-vlan-input-dev",
1053                 .data           = &brnf_pass_vlan_indev,
1054                 .maxlen         = sizeof(int),
1055                 .mode           = 0644,
1056                 .proc_handler   = brnf_sysctl_call_tables,
1057         },
1058         { }
1059 };
1060 #endif
1061
1062 int __init br_netfilter_init(void)
1063 {
1064         int ret;
1065
1066         ret = dst_entries_init(&fake_dst_ops);
1067         if (ret < 0)
1068                 return ret;
1069
1070         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1071         if (ret < 0) {
1072                 dst_entries_destroy(&fake_dst_ops);
1073                 return ret;
1074         }
1075 #ifdef CONFIG_SYSCTL
1076         brnf_sysctl_header = register_net_sysctl(&init_net, "net/bridge", brnf_table);
1077         if (brnf_sysctl_header == NULL) {
1078                 printk(KERN_WARNING
1079                        "br_netfilter: can't register to sysctl.\n");
1080                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1081                 dst_entries_destroy(&fake_dst_ops);
1082                 return -ENOMEM;
1083         }
1084 #endif
1085         printk(KERN_NOTICE "Bridge firewalling registered\n");
1086         return 0;
1087 }
1088
1089 void br_netfilter_fini(void)
1090 {
1091         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1092 #ifdef CONFIG_SYSCTL
1093         unregister_net_sysctl_table(brnf_sysctl_header);
1094 #endif
1095         dst_entries_destroy(&fake_dst_ops);
1096 }