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