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