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 #define IS_IP(skb) \
  66         (!vlan_tx_tag_present(skb) && skb->protocol == htons(ETH_P_IP))
  67
  68 #define IS_IPV6(skb) \
  69         (!vlan_tx_tag_present(skb) && skb->protocol == htons(ETH_P_IPV6))
  70
  71 #define IS_ARP(skb) \
  72         (!vlan_tx_tag_present(skb) && skb->protocol == htons(ETH_P_ARP))
  73
  74 static inline __be16 vlan_proto(const struct sk_buff *skb)
  75 {
  76         if (vlan_tx_tag_present(skb))
  77                 return skb->protocol;
  78         else if (skb->protocol == htons(ETH_P_8021Q))
  79                 return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  80         else
  81                 return 0;
  82 }
  83
  84 #define IS_VLAN_IP(skb) \
  85         (vlan_proto(skb) == htons(ETH_P_IP) && \
  86          brnf_filter_vlan_tagged)
  87
  88 #define IS_VLAN_IPV6(skb) \
  89         (vlan_proto(skb) == htons(ETH_P_IPV6) && \
  90          brnf_filter_vlan_tagged)
  91
  92 #define IS_VLAN_ARP(skb) \
  93         (vlan_proto(skb) == htons(ETH_P_ARP) && \
  94          brnf_filter_vlan_tagged)
  95
  96 static inline __be16 pppoe_proto(const struct sk_buff *skb)
  97 {
  98         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
  99                             sizeof(struct pppoe_hdr)));
 100 }
 101
 102 #define IS_PPPOE_IP(skb) \
 103         (skb->protocol == htons(ETH_P_PPP_SES) && \
 104          pppoe_proto(skb) == htons(PPP_IP) && \
 105          brnf_filter_pppoe_tagged)
 106
 107 #define IS_PPPOE_IPV6(skb) \
 108         (skb->protocol == htons(ETH_P_PPP_SES) && \
 109          pppoe_proto(skb) == htons(PPP_IPV6) && \
 110          brnf_filter_pppoe_tagged)
 111
 112 static void fake_update_pmtu(struct dst_entry *dst, u32 mtu)
 113 {
 114 }
 115
 116 static u32 *fake_cow_metrics(struct dst_entry *dst, unsigned long old)
 117 {
 118         return NULL;
 119 }
 120
 121 static struct neighbour *fake_neigh_lookup(const struct dst_entry *dst, const void *daddr)
 122 {
 123         return NULL;
 124 }
 125
 126 static unsigned int fake_mtu(const struct dst_entry *dst)
 127 {
 128         return dst->dev->mtu;
 129 }
 130
 131 static struct dst_ops fake_dst_ops = {
 132         .family =               AF_INET,
 133         .protocol =             cpu_to_be16(ETH_P_IP),
 134         .update_pmtu =          fake_update_pmtu,
 135         .cow_metrics =          fake_cow_metrics,
 136         .neigh_lookup =         fake_neigh_lookup,
 137         .mtu =                  fake_mtu,
 138 };
 139
 140 /*
 141  * Initialize bogus route table used to keep netfilter happy.
 142  * Currently, we fill in the PMTU entry because netfilter
 143  * refragmentation needs it, and the rt_flags entry because
 144  * ipt_REJECT needs it.  Future netfilter modules might
 145  * require us to fill additional fields.
 146  */
 147 static const u32 br_dst_default_metrics[RTAX_MAX] = {
 148         [RTAX_MTU - 1] = 1500,
 149 };
 150
 151 void br_netfilter_rtable_init(struct net_bridge *br)
 152 {
 153         struct rtable *rt = &br->fake_rtable;
 154
 155         atomic_set(&rt->dst.__refcnt, 1);
 156         rt->dst.dev = br->dev;
 157         rt->dst.path = &rt->dst;
 158         dst_init_metrics(&rt->dst, br_dst_default_metrics, true);
 159         rt->dst.flags   = DST_NOXFRM | DST_NOPEER;
 160         rt->dst.ops = &fake_dst_ops;
 161 }
 162
 163 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
 164 {
 165         struct net_bridge_port *port;
 166
 167         port = br_port_get_rcu(dev);
 168         return port ? &port->br->fake_rtable : NULL;
 169 }
 170
 171 static inline struct net_device *bridge_parent(const struct net_device *dev)
 172 {
 173         struct net_bridge_port *port;
 174
 175         port = br_port_get_rcu(dev);
 176         return port ? port->br->dev : NULL;
 177 }
 178
 179 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 180 {
 181         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 182         if (likely(skb->nf_bridge))
 183                 atomic_set(&(skb->nf_bridge->use), 1);
 184
 185         return skb->nf_bridge;
 186 }
 187
 188 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 189 {
 190         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 191
 192         if (atomic_read(&nf_bridge->use) > 1) {
 193                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 194
 195                 if (tmp) {
 196                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 197                         atomic_set(&tmp->use, 1);
 198                 }
 199                 nf_bridge_put(nf_bridge);
 200                 nf_bridge = tmp;
 201         }
 202         return nf_bridge;
 203 }
 204
 205 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 206 {
 207         unsigned int len = nf_bridge_encap_header_len(skb);
 208
 209         skb_push(skb, len);
 210         skb->network_header -= len;
 211 }
 212
 213 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 214 {
 215         unsigned int len = nf_bridge_encap_header_len(skb);
 216
 217         skb_pull(skb, len);
 218         skb->network_header += len;
 219 }
 220
 221 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 222 {
 223         unsigned int len = nf_bridge_encap_header_len(skb);
 224
 225         skb_pull_rcsum(skb, len);
 226         skb->network_header += len;
 227 }
 228
 229 static inline void nf_bridge_save_header(struct sk_buff *skb)
 230 {
 231         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 232
 233         skb_copy_from_linear_data_offset(skb, -header_size,
 234                                          skb->nf_bridge->data, header_size);
 235 }
 236
 237 static inline void nf_bridge_update_protocol(struct sk_buff *skb)
 238 {
 239         if (skb->nf_bridge->mask & BRNF_8021Q)
 240                 skb->protocol = htons(ETH_P_8021Q);
 241         else if (skb->nf_bridge->mask & BRNF_PPPoE)
 242                 skb->protocol = htons(ETH_P_PPP_SES);
 243 }
 244
 245 /* When handing a packet over to the IP layer
 246  * check whether we have a skb that is in the
 247  * expected format
 248  */
 249
 250 static int br_parse_ip_options(struct sk_buff *skb)
 251 {
 252         struct ip_options *opt;
 253         const struct iphdr *iph;
 254         struct net_device *dev = skb->dev;
 255         u32 len;
 256
 257         iph = ip_hdr(skb);
 258         opt = &(IPCB(skb)->opt);
 259
 260         /* Basic sanity checks */
 261         if (iph->ihl < 5 || iph->version != 4)
 262                 goto inhdr_error;
 263
 264         if (!pskb_may_pull(skb, iph->ihl*4))
 265                 goto inhdr_error;
 266
 267         iph = ip_hdr(skb);
 268         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 269                 goto inhdr_error;
 270
 271         len = ntohs(iph->tot_len);
 272         if (skb->len < len) {
 273                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
 274                 goto drop;
 275         } else if (len < (iph->ihl*4))
 276                 goto inhdr_error;
 277
 278         if (pskb_trim_rcsum(skb, len)) {
 279                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 280                 goto drop;
 281         }
 282
 283         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 284         if (iph->ihl == 5)
 285                 return 0;
 286
 287         opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
 288         if (ip_options_compile(dev_net(dev), opt, skb))
 289                 goto inhdr_error;
 290
 291         /* Check correct handling of SRR option */
 292         if (unlikely(opt->srr)) {
 293                 struct in_device *in_dev = __in_dev_get_rcu(dev);
 294                 if (in_dev && !IN_DEV_SOURCE_ROUTE(in_dev))
 295                         goto drop;
 296
 297                 if (ip_options_rcv_srr(skb))
 298                         goto drop;
 299         }
 300
 301         return 0;
 302
 303 inhdr_error:
 304         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 305 drop:
 306         return -1;
 307 }
 308
 309 /* Fill in the header for fragmented IP packets handled by
 310  * the IPv4 connection tracking code.
 311  */
 312 int nf_bridge_copy_header(struct sk_buff *skb)
 313 {
 314         int err;
 315         unsigned int header_size;
 316
 317         nf_bridge_update_protocol(skb);
 318         header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 319         err = skb_cow_head(skb, header_size);
 320         if (err)
 321                 return err;
 322
 323         skb_copy_to_linear_data_offset(skb, -header_size,
 324                                        skb->nf_bridge->data, header_size);
 325         __skb_push(skb, nf_bridge_encap_header_len(skb));
 326         return 0;
 327 }
 328
 329 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 330 /* Undo the changes made for ip6tables PREROUTING and continue the
 331  * bridge PRE_ROUTING hook. */
 332 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 333 {
 334         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 335         struct rtable *rt;
 336
 337         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 338                 skb->pkt_type = PACKET_OTHERHOST;
 339                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 340         }
 341         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 342
 343         rt = bridge_parent_rtable(nf_bridge->physindev);
 344         if (!rt) {
 345                 kfree_skb(skb);
 346                 return 0;
 347         }
 348         skb_dst_set_noref(skb, &rt->dst);
 349
 350         skb->dev = nf_bridge->physindev;
 351         nf_bridge_update_protocol(skb);
 352         nf_bridge_push_encap_header(skb);
 353         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 354                        br_handle_frame_finish, 1);
 355
 356         return 0;
 357 }
 358
 359 /* Obtain the correct destination MAC address, while preserving the original
 360  * source MAC address. If we already know this address, we just copy it. If we
 361  * don't, we use the neighbour framework to find out. In both cases, we make
 362  * sure that br_handle_frame_finish() is called afterwards.
 363  */
 364 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 365 {
 366         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 367         struct neighbour *neigh;
 368         struct dst_entry *dst;
 369
 370         skb->dev = bridge_parent(skb->dev);
 371         if (!skb->dev)
 372                 goto free_skb;
 373         dst = skb_dst(skb);
 374         neigh = dst_get_neighbour_noref(dst);
 375         if (neigh->hh.hh_len) {
 376                 neigh_hh_bridge(&neigh->hh, skb);
 377                 skb->dev = nf_bridge->physindev;
 378                 return br_handle_frame_finish(skb);
 379         } else {
 380                 /* the neighbour function below overwrites the complete
 381                  * MAC header, so we save the Ethernet source address and
 382                  * protocol number. */
 383                 skb_copy_from_linear_data_offset(skb, -(ETH_HLEN-ETH_ALEN), skb->nf_bridge->data, ETH_HLEN-ETH_ALEN);
 384                 /* tell br_dev_xmit to continue with forwarding */
 385                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 386                 return neigh->output(neigh, skb);
 387         }
 388 free_skb:
 389         kfree_skb(skb);
 390         return 0;
 391 }
 392
 393 /* This requires some explaining. If DNAT has taken place,
 394  * we will need to fix up the destination Ethernet address.
 395  *
 396  * There are two cases to consider:
 397  * 1. The packet was DNAT'ed to a device in the same bridge
 398  *    port group as it was received on. We can still bridge
 399  *    the packet.
 400  * 2. The packet was DNAT'ed to a different device, either
 401  *    a non-bridged device or another bridge port group.
 402  *    The packet will need to be routed.
 403  *
 404  * The correct way of distinguishing between these two cases is to
 405  * call ip_route_input() and to look at skb->dst->dev, which is
 406  * changed to the destination device if ip_route_input() succeeds.
 407  *
 408  * Let's first consider the case that ip_route_input() succeeds:
 409  *
 410  * If the output device equals the logical bridge device the packet
 411  * came in on, we can consider this bridging. The corresponding MAC
 412  * address will be obtained in br_nf_pre_routing_finish_bridge.
 413  * Otherwise, the packet is considered to be routed and we just
 414  * change the destination MAC address so that the packet will
 415  * later be passed up to the IP stack to be routed. For a redirected
 416  * packet, ip_route_input() will give back the localhost as output device,
 417  * which differs from the bridge device.
 418  *
 419  * Let's now consider the case that ip_route_input() fails:
 420  *
 421  * This can be because the destination address is martian, in which case
 422  * the packet will be dropped.
 423  * If IP forwarding is disabled, ip_route_input() will fail, while
 424  * ip_route_output_key() can return success. The source
 425  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 426  * thinks we're handling a locally generated packet and won't care
 427  * if IP forwarding is enabled. If the output device equals the logical bridge
 428  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 429  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 430  */
 431 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 432 {
 433         struct net_device *dev = skb->dev;
 434         struct iphdr *iph = ip_hdr(skb);
 435         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 436         struct rtable *rt;
 437         int err;
 438
 439         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 440                 skb->pkt_type = PACKET_OTHERHOST;
 441                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 442         }
 443         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 444         if (dnat_took_place(skb)) {
 445                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 446                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 447
 448                         /* If err equals -EHOSTUNREACH the error is due to a
 449                          * martian destination or due to the fact that
 450                          * forwarding is disabled. For most martian packets,
 451                          * ip_route_output_key() will fail. It won't fail for 2 types of
 452                          * martian destinations: loopback destinations and destination
 453                          * 0.0.0.0. In both cases the packet will be dropped because the
 454                          * destination is the loopback device and not the bridge. */
 455                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 456                                 goto free_skb;
 457
 458                         rt = ip_route_output(dev_net(dev), iph->daddr, 0,
 459                                              RT_TOS(iph->tos), 0);
 460                         if (!IS_ERR(rt)) {
 461                                 /* - Bridged-and-DNAT'ed traffic doesn't
 462                                  *   require ip_forwarding. */
 463                                 if (rt->dst.dev == dev) {
 464                                         skb_dst_set(skb, &rt->dst);
 465                                         goto bridged_dnat;
 466                                 }
 467                                 ip_rt_put(rt);
 468                         }
 469 free_skb:
 470                         kfree_skb(skb);
 471                         return 0;
 472                 } else {
 473                         if (skb_dst(skb)->dev == dev) {
 474 bridged_dnat:
 475                                 skb->dev = nf_bridge->physindev;
 476                                 nf_bridge_update_protocol(skb);
 477                                 nf_bridge_push_encap_header(skb);
 478                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
 479                                                NF_BR_PRE_ROUTING,
 480                                                skb, skb->dev, NULL,
 481                                                br_nf_pre_routing_finish_bridge,
 482                                                1);
 483                                 return 0;
 484                         }
 485                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
 486                         skb->pkt_type = PACKET_HOST;
 487                 }
 488         } else {
 489                 rt = bridge_parent_rtable(nf_bridge->physindev);
 490                 if (!rt) {
 491                         kfree_skb(skb);
 492                         return 0;
 493                 }
 494                 skb_dst_set_noref(skb, &rt->dst);
 495         }
 496
 497         skb->dev = nf_bridge->physindev;
 498         nf_bridge_update_protocol(skb);
 499         nf_bridge_push_encap_header(skb);
 500         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 501                        br_handle_frame_finish, 1);
 502
 503         return 0;
 504 }
 505
 506 /* Some common code for IPv4/IPv6 */
 507 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 508 {
 509         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 510
 511         if (skb->pkt_type == PACKET_OTHERHOST) {
 512                 skb->pkt_type = PACKET_HOST;
 513                 nf_bridge->mask |= BRNF_PKT_TYPE;
 514         }
 515
 516         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 517         nf_bridge->physindev = skb->dev;
 518         skb->dev = bridge_parent(skb->dev);
 519         if (skb->protocol == htons(ETH_P_8021Q))
 520                 nf_bridge->mask |= BRNF_8021Q;
 521         else if (skb->protocol == htons(ETH_P_PPP_SES))
 522                 nf_bridge->mask |= BRNF_PPPoE;
 523
 524         return skb->dev;
 525 }
 526
 527 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 528 static int check_hbh_len(struct sk_buff *skb)
 529 {
 530         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 531         u32 pkt_len;
 532         const unsigned char *nh = skb_network_header(skb);
 533         int off = raw - nh;
 534         int len = (raw[1] + 1) << 3;
 535
 536         if ((raw + len) - skb->data > skb_headlen(skb))
 537                 goto bad;
 538
 539         off += 2;
 540         len -= 2;
 541
 542         while (len > 0) {
 543                 int optlen = nh[off + 1] + 2;
 544
 545                 switch (nh[off]) {
 546                 case IPV6_TLV_PAD0:
 547                         optlen = 1;
 548                         break;
 549
 550                 case IPV6_TLV_PADN:
 551                         break;
 552
 553                 case IPV6_TLV_JUMBO:
 554                         if (nh[off + 1] != 4 || (off & 3) != 2)
 555                                 goto bad;
 556                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 557                         if (pkt_len <= IPV6_MAXPLEN ||
 558                             ipv6_hdr(skb)->payload_len)
 559                                 goto bad;
 560                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 561                                 goto bad;
 562                         if (pskb_trim_rcsum(skb,
 563                                             pkt_len + sizeof(struct ipv6hdr)))
 564                                 goto bad;
 565                         nh = skb_network_header(skb);
 566                         break;
 567                 default:
 568                         if (optlen > len)
 569                                 goto bad;
 570                         break;
 571                 }
 572                 off += optlen;
 573                 len -= optlen;
 574         }
 575         if (len == 0)
 576                 return 0;
 577 bad:
 578         return -1;
 579
 580 }
 581
 582 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 583  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 584 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 585                                            struct sk_buff *skb,
 586                                            const struct net_device *in,
 587                                            const struct net_device *out,
 588                                            int (*okfn)(struct sk_buff *))
 589 {
 590         const struct ipv6hdr *hdr;
 591         u32 pkt_len;
 592
 593         if (skb->len < sizeof(struct ipv6hdr))
 594                 return NF_DROP;
 595
 596         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 597                 return NF_DROP;
 598
 599         hdr = ipv6_hdr(skb);
 600
 601         if (hdr->version != 6)
 602                 return NF_DROP;
 603
 604         pkt_len = ntohs(hdr->payload_len);
 605
 606         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 607                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 608                         return NF_DROP;
 609                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 610                         return NF_DROP;
 611         }
 612         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 613                 return NF_DROP;
 614
 615         nf_bridge_put(skb->nf_bridge);
 616         if (!nf_bridge_alloc(skb))
 617                 return NF_DROP;
 618         if (!setup_pre_routing(skb))
 619                 return NF_DROP;
 620
 621         skb->protocol = htons(ETH_P_IPV6);
 622         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 623                 br_nf_pre_routing_finish_ipv6);
 624
 625         return NF_STOLEN;
 626 }
 627
 628 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 629  * Replicate the checks that IPv4 does on packet reception.
 630  * Set skb->dev to the bridge device (i.e. parent of the
 631  * receiving device) to make netfilter happy, the REDIRECT
 632  * target in particular.  Save the original destination IP
 633  * address to be able to detect DNAT afterwards. */
 634 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
 635                                       const struct net_device *in,
 636                                       const struct net_device *out,
 637                                       int (*okfn)(struct sk_buff *))
 638 {
 639         struct net_bridge_port *p;
 640         struct net_bridge *br;
 641         __u32 len = nf_bridge_encap_header_len(skb);
 642
 643         if (unlikely(!pskb_may_pull(skb, len)))
 644                 return NF_DROP;
 645
 646         p = br_port_get_rcu(in);
 647         if (p == NULL)
 648                 return NF_DROP;
 649         br = p->br;
 650
 651         if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb)) {
 652                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
 653                         return NF_ACCEPT;
 654
 655                 nf_bridge_pull_encap_header_rcsum(skb);
 656                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 657         }
 658
 659         if (!brnf_call_iptables && !br->nf_call_iptables)
 660                 return NF_ACCEPT;
 661
 662         if (!IS_IP(skb) && !IS_VLAN_IP(skb) && !IS_PPPOE_IP(skb))
 663                 return NF_ACCEPT;
 664
 665         nf_bridge_pull_encap_header_rcsum(skb);
 666
 667         if (br_parse_ip_options(skb))
 668                 return NF_DROP;
 669
 670         nf_bridge_put(skb->nf_bridge);
 671         if (!nf_bridge_alloc(skb))
 672                 return NF_DROP;
 673         if (!setup_pre_routing(skb))
 674                 return NF_DROP;
 675         store_orig_dstaddr(skb);
 676         skb->protocol = htons(ETH_P_IP);
 677
 678         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 679                 br_nf_pre_routing_finish);
 680
 681         return NF_STOLEN;
 682 }
 683
 684
 685 /* PF_BRIDGE/LOCAL_IN ************************************************/
 686 /* The packet is locally destined, which requires a real
 687  * dst_entry, so detach the fake one.  On the way up, the
 688  * packet would pass through PRE_ROUTING again (which already
 689  * took place when the packet entered the bridge), but we
 690  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 691  * prevent this from happening. */
 692 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
 693                                    const struct net_device *in,
 694                                    const struct net_device *out,
 695                                    int (*okfn)(struct sk_buff *))
 696 {
 697         struct rtable *rt = skb_rtable(skb);
 698
 699         if (rt && rt == bridge_parent_rtable(in))
 700                 skb_dst_drop(skb);
 701
 702         return NF_ACCEPT;
 703 }
 704
 705 /* PF_BRIDGE/FORWARD *************************************************/
 706 static int br_nf_forward_finish(struct sk_buff *skb)
 707 {
 708         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 709         struct net_device *in;
 710
 711         if (!IS_ARP(skb) && !IS_VLAN_ARP(skb)) {
 712                 in = nf_bridge->physindev;
 713                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 714                         skb->pkt_type = PACKET_OTHERHOST;
 715                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 716                 }
 717                 nf_bridge_update_protocol(skb);
 718         } else {
 719                 in = *((struct net_device **)(skb->cb));
 720         }
 721         nf_bridge_push_encap_header(skb);
 722
 723         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
 724                        skb->dev, br_forward_finish, 1);
 725         return 0;
 726 }
 727
 728
 729 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 730  * netfilter with indev and outdev set to the bridge device,
 731  * but we are still able to filter on the 'real' indev/outdev
 732  * because of the physdev module. For ARP, indev and outdev are the
 733  * bridge ports. */
 734 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
 735                                      const struct net_device *in,
 736                                      const struct net_device *out,
 737                                      int (*okfn)(struct sk_buff *))
 738 {
 739         struct nf_bridge_info *nf_bridge;
 740         struct net_device *parent;
 741         u_int8_t pf;
 742
 743         if (!skb->nf_bridge)
 744                 return NF_ACCEPT;
 745
 746         /* Need exclusive nf_bridge_info since we might have multiple
 747          * different physoutdevs. */
 748         if (!nf_bridge_unshare(skb))
 749                 return NF_DROP;
 750
 751         parent = bridge_parent(out);
 752         if (!parent)
 753                 return NF_DROP;
 754
 755         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 756                 pf = PF_INET;
 757         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 758                 pf = PF_INET6;
 759         else
 760                 return NF_ACCEPT;
 761
 762         nf_bridge_pull_encap_header(skb);
 763
 764         nf_bridge = skb->nf_bridge;
 765         if (skb->pkt_type == PACKET_OTHERHOST) {
 766                 skb->pkt_type = PACKET_HOST;
 767                 nf_bridge->mask |= BRNF_PKT_TYPE;
 768         }
 769
 770         if (pf == PF_INET && br_parse_ip_options(skb))
 771                 return NF_DROP;
 772
 773         /* The physdev module checks on this */
 774         nf_bridge->mask |= BRNF_BRIDGED;
 775         nf_bridge->physoutdev = skb->dev;
 776         if (pf == PF_INET)
 777                 skb->protocol = htons(ETH_P_IP);
 778         else
 779                 skb->protocol = htons(ETH_P_IPV6);
 780
 781         NF_HOOK(pf, NF_INET_FORWARD, skb, bridge_parent(in), parent,
 782                 br_nf_forward_finish);
 783
 784         return NF_STOLEN;
 785 }
 786
 787 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
 788                                       const struct net_device *in,
 789                                       const struct net_device *out,
 790                                       int (*okfn)(struct sk_buff *))
 791 {
 792         struct net_bridge_port *p;
 793         struct net_bridge *br;
 794         struct net_device **d = (struct net_device **)(skb->cb);
 795
 796         p = br_port_get_rcu(out);
 797         if (p == NULL)
 798                 return NF_ACCEPT;
 799         br = p->br;
 800
 801         if (!brnf_call_arptables && !br->nf_call_arptables)
 802                 return NF_ACCEPT;
 803
 804         if (!IS_ARP(skb)) {
 805                 if (!IS_VLAN_ARP(skb))
 806                         return NF_ACCEPT;
 807                 nf_bridge_pull_encap_header(skb);
 808         }
 809
 810         if (arp_hdr(skb)->ar_pln != 4) {
 811                 if (IS_VLAN_ARP(skb))
 812                         nf_bridge_push_encap_header(skb);
 813                 return NF_ACCEPT;
 814         }
 815         *d = (struct net_device *)in;
 816         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 817                 (struct net_device *)out, br_nf_forward_finish);
 818
 819         return NF_STOLEN;
 820 }
 821
 822 #if IS_ENABLED(CONFIG_NF_CONNTRACK_IPV4)
 823 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 824 {
 825         int ret;
 826
 827         if (skb->nfct != NULL && skb->protocol == htons(ETH_P_IP) &&
 828             skb->len + nf_bridge_mtu_reduction(skb) > skb->dev->mtu &&
 829             !skb_is_gso(skb)) {
 830                 if (br_parse_ip_options(skb))
 831                         /* Drop invalid packet */
 832                         return NF_DROP;
 833                 ret = ip_fragment(skb, br_dev_queue_push_xmit);
 834         } else
 835                 ret = br_dev_queue_push_xmit(skb);
 836
 837         return ret;
 838 }
 839 #else
 840 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 841 {
 842         return br_dev_queue_push_xmit(skb);
 843 }
 844 #endif
 845
 846 /* PF_BRIDGE/POST_ROUTING ********************************************/
 847 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
 848                                        const struct net_device *in,
 849                                        const struct net_device *out,
 850                                        int (*okfn)(struct sk_buff *))
 851 {
 852         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 853         struct net_device *realoutdev = bridge_parent(skb->dev);
 854         u_int8_t pf;
 855
 856         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
 857                 return NF_ACCEPT;
 858
 859         if (!realoutdev)
 860                 return NF_DROP;
 861
 862         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 863                 pf = PF_INET;
 864         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 865                 pf = PF_INET6;
 866         else
 867                 return NF_ACCEPT;
 868
 869         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 870          * about the value of skb->pkt_type. */
 871         if (skb->pkt_type == PACKET_OTHERHOST) {
 872                 skb->pkt_type = PACKET_HOST;
 873                 nf_bridge->mask |= BRNF_PKT_TYPE;
 874         }
 875
 876         nf_bridge_pull_encap_header(skb);
 877         nf_bridge_save_header(skb);
 878         if (pf == PF_INET)
 879                 skb->protocol = htons(ETH_P_IP);
 880         else
 881                 skb->protocol = htons(ETH_P_IPV6);
 882
 883         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
 884                 br_nf_dev_queue_xmit);
 885
 886         return NF_STOLEN;
 887 }
 888
 889 /* IP/SABOTAGE *****************************************************/
 890 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 891  * for the second time. */
 892 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
 893                                    const struct net_device *in,
 894                                    const struct net_device *out,
 895                                    int (*okfn)(struct sk_buff *))
 896 {
 897         if (skb->nf_bridge &&
 898             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 899                 return NF_STOP;
 900         }
 901
 902         return NF_ACCEPT;
 903 }
 904
 905 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 906  * br_dev_queue_push_xmit is called afterwards */
 907 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
 908         {
 909                 .hook = br_nf_pre_routing,
 910                 .owner = THIS_MODULE,
 911                 .pf = PF_BRIDGE,
 912                 .hooknum = NF_BR_PRE_ROUTING,
 913                 .priority = NF_BR_PRI_BRNF,
 914         },
 915         {
 916                 .hook = br_nf_local_in,
 917                 .owner = THIS_MODULE,
 918                 .pf = PF_BRIDGE,
 919                 .hooknum = NF_BR_LOCAL_IN,
 920                 .priority = NF_BR_PRI_BRNF,
 921         },
 922         {
 923                 .hook = br_nf_forward_ip,
 924                 .owner = THIS_MODULE,
 925                 .pf = PF_BRIDGE,
 926                 .hooknum = NF_BR_FORWARD,
 927                 .priority = NF_BR_PRI_BRNF - 1,
 928         },
 929         {
 930                 .hook = br_nf_forward_arp,
 931                 .owner = THIS_MODULE,
 932                 .pf = PF_BRIDGE,
 933                 .hooknum = NF_BR_FORWARD,
 934                 .priority = NF_BR_PRI_BRNF,
 935         },
 936         {
 937                 .hook = br_nf_post_routing,
 938                 .owner = THIS_MODULE,
 939                 .pf = PF_BRIDGE,
 940                 .hooknum = NF_BR_POST_ROUTING,
 941                 .priority = NF_BR_PRI_LAST,
 942         },
 943         {
 944                 .hook = ip_sabotage_in,
 945                 .owner = THIS_MODULE,
 946                 .pf = PF_INET,
 947                 .hooknum = NF_INET_PRE_ROUTING,
 948                 .priority = NF_IP_PRI_FIRST,
 949         },
 950         {
 951                 .hook = ip_sabotage_in,
 952                 .owner = THIS_MODULE,
 953                 .pf = PF_INET6,
 954                 .hooknum = NF_INET_PRE_ROUTING,
 955                 .priority = NF_IP6_PRI_FIRST,
 956         },
 957 };
 958
 959 #ifdef CONFIG_SYSCTL
 960 static
 961 int brnf_sysctl_call_tables(ctl_table * ctl, int write,
 962                             void __user * buffer, size_t * lenp, loff_t * ppos)
 963 {
 964         int ret;
 965
 966         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
 967
 968         if (write && *(int *)(ctl->data))
 969                 *(int *)(ctl->data) = 1;
 970         return ret;
 971 }
 972
 973 static ctl_table brnf_table[] = {
 974         {
 975                 .procname       = "bridge-nf-call-arptables",
 976                 .data           = &brnf_call_arptables,
 977                 .maxlen         = sizeof(int),
 978                 .mode           = 0644,
 979                 .proc_handler   = brnf_sysctl_call_tables,
 980         },
 981         {
 982                 .procname       = "bridge-nf-call-iptables",
 983                 .data           = &brnf_call_iptables,
 984                 .maxlen         = sizeof(int),
 985                 .mode           = 0644,
 986                 .proc_handler   = brnf_sysctl_call_tables,
 987         },
 988         {
 989                 .procname       = "bridge-nf-call-ip6tables",
 990                 .data           = &brnf_call_ip6tables,
 991                 .maxlen         = sizeof(int),
 992                 .mode           = 0644,
 993                 .proc_handler   = brnf_sysctl_call_tables,
 994         },
 995         {
 996                 .procname       = "bridge-nf-filter-vlan-tagged",
 997                 .data           = &brnf_filter_vlan_tagged,
 998                 .maxlen         = sizeof(int),
 999                 .mode           = 0644,
1000                 .proc_handler   = brnf_sysctl_call_tables,
1001         },
1002         {
1003                 .procname       = "bridge-nf-filter-pppoe-tagged",
1004                 .data           = &brnf_filter_pppoe_tagged,
1005                 .maxlen         = sizeof(int),
1006                 .mode           = 0644,
1007                 .proc_handler   = brnf_sysctl_call_tables,
1008         },
1009         { }
1010 };
1011
1012 static struct ctl_path brnf_path[] = {
1013         { .procname = "net", },
1014         { .procname = "bridge", },
1015         { }
1016 };
1017 #endif
1018
1019 int __init br_netfilter_init(void)
1020 {
1021         int ret;
1022
1023         ret = dst_entries_init(&fake_dst_ops);
1024         if (ret < 0)
1025                 return ret;
1026
1027         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1028         if (ret < 0) {
1029                 dst_entries_destroy(&fake_dst_ops);
1030                 return ret;
1031         }
1032 #ifdef CONFIG_SYSCTL
1033         brnf_sysctl_header = register_sysctl_paths(brnf_path, brnf_table);
1034         if (brnf_sysctl_header == NULL) {
1035                 printk(KERN_WARNING
1036                        "br_netfilter: can't register to sysctl.\n");
1037                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1038                 dst_entries_destroy(&fake_dst_ops);
1039                 return -ENOMEM;
1040         }
1041 #endif
1042         printk(KERN_NOTICE "Bridge firewalling registered\n");
1043         return 0;
1044 }
1045
1046 void br_netfilter_fini(void)
1047 {
1048         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1049 #ifdef CONFIG_SYSCTL
1050         unregister_sysctl_table(brnf_sysctl_header);
1051 #endif
1052         dst_entries_destroy(&fake_dst_ops);
1053 }