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