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, const void *daddr)
 124 {
 125         return NULL;
 126 }
 127
 128 static unsigned int fake_mtu(const struct dst_entry *dst)
 129 {
 130         return dst->dev->mtu;
 131 }
 132
 133 static struct dst_ops fake_dst_ops = {
 134         .family =               AF_INET,
 135         .protocol =             cpu_to_be16(ETH_P_IP),
 136         .update_pmtu =          fake_update_pmtu,
 137         .cow_metrics =          fake_cow_metrics,
 138         .neigh_lookup =         fake_neigh_lookup,
 139         .mtu =                  fake_mtu,
 140 };
 141
 142 /*
 143  * Initialize bogus route table used to keep netfilter happy.
 144  * Currently, we fill in the PMTU entry because netfilter
 145  * refragmentation needs it, and the rt_flags entry because
 146  * ipt_REJECT needs it.  Future netfilter modules might
 147  * require us to fill additional fields.
 148  */
 149 static const u32 br_dst_default_metrics[RTAX_MAX] = {
 150         [RTAX_MTU - 1] = 1500,
 151 };
 152
 153 void br_netfilter_rtable_init(struct net_bridge *br)
 154 {
 155         struct rtable *rt = &br->fake_rtable;
 156
 157         atomic_set(&rt->dst.__refcnt, 1);
 158         rt->dst.dev = br->dev;
 159         rt->dst.path = &rt->dst;
 160         dst_init_metrics(&rt->dst, br_dst_default_metrics, true);
 161         rt->dst.flags   = DST_NOXFRM | DST_NOPEER | DST_FAKE_RTABLE;
 162         rt->dst.ops = &fake_dst_ops;
 163 }
 164
 165 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
 166 {
 167         struct net_bridge_port *port;
 168
 169         port = br_port_get_rcu(dev);
 170         return port ? &port->br->fake_rtable : NULL;
 171 }
 172
 173 static inline struct net_device *bridge_parent(const struct net_device *dev)
 174 {
 175         struct net_bridge_port *port;
 176
 177         port = br_port_get_rcu(dev);
 178         return port ? port->br->dev : NULL;
 179 }
 180
 181 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 182 {
 183         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 184         if (likely(skb->nf_bridge))
 185                 atomic_set(&(skb->nf_bridge->use), 1);
 186
 187         return skb->nf_bridge;
 188 }
 189
 190 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 191 {
 192         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 193
 194         if (atomic_read(&nf_bridge->use) > 1) {
 195                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 196
 197                 if (tmp) {
 198                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 199                         atomic_set(&tmp->use, 1);
 200                 }
 201                 nf_bridge_put(nf_bridge);
 202                 nf_bridge = tmp;
 203         }
 204         return nf_bridge;
 205 }
 206
 207 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 208 {
 209         unsigned int len = nf_bridge_encap_header_len(skb);
 210
 211         skb_push(skb, len);
 212         skb->network_header -= len;
 213 }
 214
 215 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 216 {
 217         unsigned int len = nf_bridge_encap_header_len(skb);
 218
 219         skb_pull(skb, len);
 220         skb->network_header += len;
 221 }
 222
 223 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 224 {
 225         unsigned int len = nf_bridge_encap_header_len(skb);
 226
 227         skb_pull_rcsum(skb, len);
 228         skb->network_header += len;
 229 }
 230
 231 static inline void nf_bridge_save_header(struct sk_buff *skb)
 232 {
 233         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 234
 235         skb_copy_from_linear_data_offset(skb, -header_size,
 236                                          skb->nf_bridge->data, header_size);
 237 }
 238
 239 static inline void nf_bridge_update_protocol(struct sk_buff *skb)
 240 {
 241         if (skb->nf_bridge->mask & BRNF_8021Q)
 242                 skb->protocol = htons(ETH_P_8021Q);
 243         else if (skb->nf_bridge->mask & BRNF_PPPoE)
 244                 skb->protocol = htons(ETH_P_PPP_SES);
 245 }
 246
 247 /* When handing a packet over to the IP layer
 248  * check whether we have a skb that is in the
 249  * expected format
 250  */
 251
 252 static int br_parse_ip_options(struct sk_buff *skb)
 253 {
 254         struct ip_options *opt;
 255         const struct iphdr *iph;
 256         struct net_device *dev = skb->dev;
 257         u32 len;
 258
 259         iph = ip_hdr(skb);
 260         opt = &(IPCB(skb)->opt);
 261
 262         /* Basic sanity checks */
 263         if (iph->ihl < 5 || iph->version != 4)
 264                 goto inhdr_error;
 265
 266         if (!pskb_may_pull(skb, iph->ihl*4))
 267                 goto inhdr_error;
 268
 269         iph = ip_hdr(skb);
 270         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 271                 goto inhdr_error;
 272
 273         len = ntohs(iph->tot_len);
 274         if (skb->len < len) {
 275                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
 276                 goto drop;
 277         } else if (len < (iph->ihl*4))
 278                 goto inhdr_error;
 279
 280         if (pskb_trim_rcsum(skb, len)) {
 281                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 282                 goto drop;
 283         }
 284
 285         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 286         if (iph->ihl == 5)
 287                 return 0;
 288
 289         opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
 290         if (ip_options_compile(dev_net(dev), opt, skb))
 291                 goto inhdr_error;
 292
 293         /* Check correct handling of SRR option */
 294         if (unlikely(opt->srr)) {
 295                 struct in_device *in_dev = __in_dev_get_rcu(dev);
 296                 if (in_dev && !IN_DEV_SOURCE_ROUTE(in_dev))
 297                         goto drop;
 298
 299                 if (ip_options_rcv_srr(skb))
 300                         goto drop;
 301         }
 302
 303         return 0;
 304
 305 inhdr_error:
 306         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 307 drop:
 308         return -1;
 309 }
 310
 311 /* Fill in the header for fragmented IP packets handled by
 312  * the IPv4 connection tracking code.
 313  */
 314 int nf_bridge_copy_header(struct sk_buff *skb)
 315 {
 316         int err;
 317         unsigned int header_size;
 318
 319         nf_bridge_update_protocol(skb);
 320         header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 321         err = skb_cow_head(skb, header_size);
 322         if (err)
 323                 return err;
 324
 325         skb_copy_to_linear_data_offset(skb, -header_size,
 326                                        skb->nf_bridge->data, header_size);
 327         __skb_push(skb, nf_bridge_encap_header_len(skb));
 328         return 0;
 329 }
 330
 331 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 332 /* Undo the changes made for ip6tables PREROUTING and continue the
 333  * bridge PRE_ROUTING hook. */
 334 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 335 {
 336         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 337         struct rtable *rt;
 338
 339         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 340                 skb->pkt_type = PACKET_OTHERHOST;
 341                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 342         }
 343         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 344
 345         rt = bridge_parent_rtable(nf_bridge->physindev);
 346         if (!rt) {
 347                 kfree_skb(skb);
 348                 return 0;
 349         }
 350         skb_dst_set_noref(skb, &rt->dst);
 351
 352         skb->dev = nf_bridge->physindev;
 353         nf_bridge_update_protocol(skb);
 354         nf_bridge_push_encap_header(skb);
 355         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 356                        br_handle_frame_finish, 1);
 357
 358         return 0;
 359 }
 360
 361 /* Obtain the correct destination MAC address, while preserving the original
 362  * source MAC address. If we already know this address, we just copy it. If we
 363  * don't, we use the neighbour framework to find out. In both cases, we make
 364  * sure that br_handle_frame_finish() is called afterwards.
 365  */
 366 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 367 {
 368         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 369         struct neighbour *neigh;
 370         struct dst_entry *dst;
 371
 372         skb->dev = bridge_parent(skb->dev);
 373         if (!skb->dev)
 374                 goto free_skb;
 375         dst = skb_dst(skb);
 376         neigh = dst_get_neighbour_noref(dst);
 377         if (neigh->hh.hh_len) {
 378                 neigh_hh_bridge(&neigh->hh, skb);
 379                 skb->dev = nf_bridge->physindev;
 380                 return br_handle_frame_finish(skb);
 381         } else {
 382                 /* the neighbour function below overwrites the complete
 383                  * MAC header, so we save the Ethernet source address and
 384                  * protocol number. */
 385                 skb_copy_from_linear_data_offset(skb, -(ETH_HLEN-ETH_ALEN), skb->nf_bridge->data, ETH_HLEN-ETH_ALEN);
 386                 /* tell br_dev_xmit to continue with forwarding */
 387                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 388                 return neigh->output(neigh, skb);
 389         }
 390 free_skb:
 391         kfree_skb(skb);
 392         return 0;
 393 }
 394
 395 /* This requires some explaining. If DNAT has taken place,
 396  * we will need to fix up the destination Ethernet address.
 397  *
 398  * There are two cases to consider:
 399  * 1. The packet was DNAT'ed to a device in the same bridge
 400  *    port group as it was received on. We can still bridge
 401  *    the packet.
 402  * 2. The packet was DNAT'ed to a different device, either
 403  *    a non-bridged device or another bridge port group.
 404  *    The packet will need to be routed.
 405  *
 406  * The correct way of distinguishing between these two cases is to
 407  * call ip_route_input() and to look at skb->dst->dev, which is
 408  * changed to the destination device if ip_route_input() succeeds.
 409  *
 410  * Let's first consider the case that ip_route_input() succeeds:
 411  *
 412  * If the output device equals the logical bridge device the packet
 413  * came in on, we can consider this bridging. The corresponding MAC
 414  * address will be obtained in br_nf_pre_routing_finish_bridge.
 415  * Otherwise, the packet is considered to be routed and we just
 416  * change the destination MAC address so that the packet will
 417  * later be passed up to the IP stack to be routed. For a redirected
 418  * packet, ip_route_input() will give back the localhost as output device,
 419  * which differs from the bridge device.
 420  *
 421  * Let's now consider the case that ip_route_input() fails:
 422  *
 423  * This can be because the destination address is martian, in which case
 424  * the packet will be dropped.
 425  * If IP forwarding is disabled, ip_route_input() will fail, while
 426  * ip_route_output_key() can return success. The source
 427  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 428  * thinks we're handling a locally generated packet and won't care
 429  * if IP forwarding is enabled. If the output device equals the logical bridge
 430  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 431  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 432  */
 433 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 434 {
 435         struct net_device *dev = skb->dev;
 436         struct iphdr *iph = ip_hdr(skb);
 437         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 438         struct rtable *rt;
 439         int err;
 440
 441         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 442                 skb->pkt_type = PACKET_OTHERHOST;
 443                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 444         }
 445         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 446         if (dnat_took_place(skb)) {
 447                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 448                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 449
 450                         /* If err equals -EHOSTUNREACH the error is due to a
 451                          * martian destination or due to the fact that
 452                          * forwarding is disabled. For most martian packets,
 453                          * ip_route_output_key() will fail. It won't fail for 2 types of
 454                          * martian destinations: loopback destinations and destination
 455                          * 0.0.0.0. In both cases the packet will be dropped because the
 456                          * destination is the loopback device and not the bridge. */
 457                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 458                                 goto free_skb;
 459
 460                         rt = ip_route_output(dev_net(dev), iph->daddr, 0,
 461                                              RT_TOS(iph->tos), 0);
 462                         if (!IS_ERR(rt)) {
 463                                 /* - Bridged-and-DNAT'ed traffic doesn't
 464                                  *   require ip_forwarding. */
 465                                 if (rt->dst.dev == dev) {
 466                                         skb_dst_set(skb, &rt->dst);
 467                                         goto bridged_dnat;
 468                                 }
 469                                 ip_rt_put(rt);
 470                         }
 471 free_skb:
 472                         kfree_skb(skb);
 473                         return 0;
 474                 } else {
 475                         if (skb_dst(skb)->dev == dev) {
 476 bridged_dnat:
 477                                 skb->dev = nf_bridge->physindev;
 478                                 nf_bridge_update_protocol(skb);
 479                                 nf_bridge_push_encap_header(skb);
 480                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
 481                                                NF_BR_PRE_ROUTING,
 482                                                skb, skb->dev, NULL,
 483                                                br_nf_pre_routing_finish_bridge,
 484                                                1);
 485                                 return 0;
 486                         }
 487                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
 488                         skb->pkt_type = PACKET_HOST;
 489                 }
 490         } else {
 491                 rt = bridge_parent_rtable(nf_bridge->physindev);
 492                 if (!rt) {
 493                         kfree_skb(skb);
 494                         return 0;
 495                 }
 496                 skb_dst_set_noref(skb, &rt->dst);
 497         }
 498
 499         skb->dev = nf_bridge->physindev;
 500         nf_bridge_update_protocol(skb);
 501         nf_bridge_push_encap_header(skb);
 502         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 503                        br_handle_frame_finish, 1);
 504
 505         return 0;
 506 }
 507
 508 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb, const struct net_device *dev)
 509 {
 510         struct net_device *vlan, *br;
 511
 512         br = bridge_parent(dev);
 513         if (brnf_pass_vlan_indev == 0 || !vlan_tx_tag_present(skb))
 514                 return br;
 515
 516         vlan = __vlan_find_dev_deep(br, vlan_tx_tag_get(skb) & VLAN_VID_MASK);
 517
 518         return vlan ? vlan : br;
 519 }
 520
 521 /* Some common code for IPv4/IPv6 */
 522 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 523 {
 524         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 525
 526         if (skb->pkt_type == PACKET_OTHERHOST) {
 527                 skb->pkt_type = PACKET_HOST;
 528                 nf_bridge->mask |= BRNF_PKT_TYPE;
 529         }
 530
 531         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 532         nf_bridge->physindev = skb->dev;
 533         skb->dev = brnf_get_logical_dev(skb, skb->dev);
 534         if (skb->protocol == htons(ETH_P_8021Q))
 535                 nf_bridge->mask |= BRNF_8021Q;
 536         else if (skb->protocol == htons(ETH_P_PPP_SES))
 537                 nf_bridge->mask |= BRNF_PPPoE;
 538
 539         return skb->dev;
 540 }
 541
 542 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 543 static int check_hbh_len(struct sk_buff *skb)
 544 {
 545         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 546         u32 pkt_len;
 547         const unsigned char *nh = skb_network_header(skb);
 548         int off = raw - nh;
 549         int len = (raw[1] + 1) << 3;
 550
 551         if ((raw + len) - skb->data > skb_headlen(skb))
 552                 goto bad;
 553
 554         off += 2;
 555         len -= 2;
 556
 557         while (len > 0) {
 558                 int optlen = nh[off + 1] + 2;
 559
 560                 switch (nh[off]) {
 561                 case IPV6_TLV_PAD1:
 562                         optlen = 1;
 563                         break;
 564
 565                 case IPV6_TLV_PADN:
 566                         break;
 567
 568                 case IPV6_TLV_JUMBO:
 569                         if (nh[off + 1] != 4 || (off & 3) != 2)
 570                                 goto bad;
 571                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 572                         if (pkt_len <= IPV6_MAXPLEN ||
 573                             ipv6_hdr(skb)->payload_len)
 574                                 goto bad;
 575                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 576                                 goto bad;
 577                         if (pskb_trim_rcsum(skb,
 578                                             pkt_len + sizeof(struct ipv6hdr)))
 579                                 goto bad;
 580                         nh = skb_network_header(skb);
 581                         break;
 582                 default:
 583                         if (optlen > len)
 584                                 goto bad;
 585                         break;
 586                 }
 587                 off += optlen;
 588                 len -= optlen;
 589         }
 590         if (len == 0)
 591                 return 0;
 592 bad:
 593         return -1;
 594
 595 }
 596
 597 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 598  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 599 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 600                                            struct sk_buff *skb,
 601                                            const struct net_device *in,
 602                                            const struct net_device *out,
 603                                            int (*okfn)(struct sk_buff *))
 604 {
 605         const struct ipv6hdr *hdr;
 606         u32 pkt_len;
 607
 608         if (skb->len < sizeof(struct ipv6hdr))
 609                 return NF_DROP;
 610
 611         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 612                 return NF_DROP;
 613
 614         hdr = ipv6_hdr(skb);
 615
 616         if (hdr->version != 6)
 617                 return NF_DROP;
 618
 619         pkt_len = ntohs(hdr->payload_len);
 620
 621         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 622                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 623                         return NF_DROP;
 624                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 625                         return NF_DROP;
 626         }
 627         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 628                 return NF_DROP;
 629
 630         nf_bridge_put(skb->nf_bridge);
 631         if (!nf_bridge_alloc(skb))
 632                 return NF_DROP;
 633         if (!setup_pre_routing(skb))
 634                 return NF_DROP;
 635
 636         skb->protocol = htons(ETH_P_IPV6);
 637         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 638                 br_nf_pre_routing_finish_ipv6);
 639
 640         return NF_STOLEN;
 641 }
 642
 643 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 644  * Replicate the checks that IPv4 does on packet reception.
 645  * Set skb->dev to the bridge device (i.e. parent of the
 646  * receiving device) to make netfilter happy, the REDIRECT
 647  * target in particular.  Save the original destination IP
 648  * address to be able to detect DNAT afterwards. */
 649 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
 650                                       const struct net_device *in,
 651                                       const struct net_device *out,
 652                                       int (*okfn)(struct sk_buff *))
 653 {
 654         struct net_bridge_port *p;
 655         struct net_bridge *br;
 656         __u32 len = nf_bridge_encap_header_len(skb);
 657
 658         if (unlikely(!pskb_may_pull(skb, len)))
 659                 return NF_DROP;
 660
 661         p = br_port_get_rcu(in);
 662         if (p == NULL)
 663                 return NF_DROP;
 664         br = p->br;
 665
 666         if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb)) {
 667                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
 668                         return NF_ACCEPT;
 669
 670                 nf_bridge_pull_encap_header_rcsum(skb);
 671                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 672         }
 673
 674         if (!brnf_call_iptables && !br->nf_call_iptables)
 675                 return NF_ACCEPT;
 676
 677         if (!IS_IP(skb) && !IS_VLAN_IP(skb) && !IS_PPPOE_IP(skb))
 678                 return NF_ACCEPT;
 679
 680         nf_bridge_pull_encap_header_rcsum(skb);
 681
 682         if (br_parse_ip_options(skb))
 683                 return NF_DROP;
 684
 685         nf_bridge_put(skb->nf_bridge);
 686         if (!nf_bridge_alloc(skb))
 687                 return NF_DROP;
 688         if (!setup_pre_routing(skb))
 689                 return NF_DROP;
 690         store_orig_dstaddr(skb);
 691         skb->protocol = htons(ETH_P_IP);
 692
 693         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 694                 br_nf_pre_routing_finish);
 695
 696         return NF_STOLEN;
 697 }
 698
 699
 700 /* PF_BRIDGE/LOCAL_IN ************************************************/
 701 /* The packet is locally destined, which requires a real
 702  * dst_entry, so detach the fake one.  On the way up, the
 703  * packet would pass through PRE_ROUTING again (which already
 704  * took place when the packet entered the bridge), but we
 705  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 706  * prevent this from happening. */
 707 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
 708                                    const struct net_device *in,
 709                                    const struct net_device *out,
 710                                    int (*okfn)(struct sk_buff *))
 711 {
 712         br_drop_fake_rtable(skb);
 713         return NF_ACCEPT;
 714 }
 715
 716 /* PF_BRIDGE/FORWARD *************************************************/
 717 static int br_nf_forward_finish(struct sk_buff *skb)
 718 {
 719         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 720         struct net_device *in;
 721
 722         if (!IS_ARP(skb) && !IS_VLAN_ARP(skb)) {
 723                 in = nf_bridge->physindev;
 724                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 725                         skb->pkt_type = PACKET_OTHERHOST;
 726                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 727                 }
 728                 nf_bridge_update_protocol(skb);
 729         } else {
 730                 in = *((struct net_device **)(skb->cb));
 731         }
 732         nf_bridge_push_encap_header(skb);
 733
 734         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
 735                        skb->dev, br_forward_finish, 1);
 736         return 0;
 737 }
 738
 739
 740 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 741  * netfilter with indev and outdev set to the bridge device,
 742  * but we are still able to filter on the 'real' indev/outdev
 743  * because of the physdev module. For ARP, indev and outdev are the
 744  * bridge ports. */
 745 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
 746                                      const struct net_device *in,
 747                                      const struct net_device *out,
 748                                      int (*okfn)(struct sk_buff *))
 749 {
 750         struct nf_bridge_info *nf_bridge;
 751         struct net_device *parent;
 752         u_int8_t pf;
 753
 754         if (!skb->nf_bridge)
 755                 return NF_ACCEPT;
 756
 757         /* Need exclusive nf_bridge_info since we might have multiple
 758          * different physoutdevs. */
 759         if (!nf_bridge_unshare(skb))
 760                 return NF_DROP;
 761
 762         parent = bridge_parent(out);
 763         if (!parent)
 764                 return NF_DROP;
 765
 766         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 767                 pf = PF_INET;
 768         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 769                 pf = PF_INET6;
 770         else
 771                 return NF_ACCEPT;
 772
 773         nf_bridge_pull_encap_header(skb);
 774
 775         nf_bridge = skb->nf_bridge;
 776         if (skb->pkt_type == PACKET_OTHERHOST) {
 777                 skb->pkt_type = PACKET_HOST;
 778                 nf_bridge->mask |= BRNF_PKT_TYPE;
 779         }
 780
 781         if (pf == PF_INET && br_parse_ip_options(skb))
 782                 return NF_DROP;
 783
 784         /* The physdev module checks on this */
 785         nf_bridge->mask |= BRNF_BRIDGED;
 786         nf_bridge->physoutdev = skb->dev;
 787         if (pf == PF_INET)
 788                 skb->protocol = htons(ETH_P_IP);
 789         else
 790                 skb->protocol = htons(ETH_P_IPV6);
 791
 792         NF_HOOK(pf, NF_INET_FORWARD, skb, brnf_get_logical_dev(skb, in), parent,
 793                 br_nf_forward_finish);
 794
 795         return NF_STOLEN;
 796 }
 797
 798 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
 799                                       const struct net_device *in,
 800                                       const struct net_device *out,
 801                                       int (*okfn)(struct sk_buff *))
 802 {
 803         struct net_bridge_port *p;
 804         struct net_bridge *br;
 805         struct net_device **d = (struct net_device **)(skb->cb);
 806
 807         p = br_port_get_rcu(out);
 808         if (p == NULL)
 809                 return NF_ACCEPT;
 810         br = p->br;
 811
 812         if (!brnf_call_arptables && !br->nf_call_arptables)
 813                 return NF_ACCEPT;
 814
 815         if (!IS_ARP(skb)) {
 816                 if (!IS_VLAN_ARP(skb))
 817                         return NF_ACCEPT;
 818                 nf_bridge_pull_encap_header(skb);
 819         }
 820
 821         if (arp_hdr(skb)->ar_pln != 4) {
 822                 if (IS_VLAN_ARP(skb))
 823                         nf_bridge_push_encap_header(skb);
 824                 return NF_ACCEPT;
 825         }
 826         *d = (struct net_device *)in;
 827         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 828                 (struct net_device *)out, br_nf_forward_finish);
 829
 830         return NF_STOLEN;
 831 }
 832
 833 #if IS_ENABLED(CONFIG_NF_CONNTRACK_IPV4)
 834 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 835 {
 836         int ret;
 837
 838         if (skb->nfct != NULL && skb->protocol == htons(ETH_P_IP) &&
 839             skb->len + nf_bridge_mtu_reduction(skb) > skb->dev->mtu &&
 840             !skb_is_gso(skb)) {
 841                 if (br_parse_ip_options(skb))
 842                         /* Drop invalid packet */
 843                         return NF_DROP;
 844                 ret = ip_fragment(skb, br_dev_queue_push_xmit);
 845         } else
 846                 ret = br_dev_queue_push_xmit(skb);
 847
 848         return ret;
 849 }
 850 #else
 851 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 852 {
 853         return br_dev_queue_push_xmit(skb);
 854 }
 855 #endif
 856
 857 /* PF_BRIDGE/POST_ROUTING ********************************************/
 858 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
 859                                        const struct net_device *in,
 860                                        const struct net_device *out,
 861                                        int (*okfn)(struct sk_buff *))
 862 {
 863         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 864         struct net_device *realoutdev = bridge_parent(skb->dev);
 865         u_int8_t pf;
 866
 867         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
 868                 return NF_ACCEPT;
 869
 870         if (!realoutdev)
 871                 return NF_DROP;
 872
 873         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 874                 pf = PF_INET;
 875         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 876                 pf = PF_INET6;
 877         else
 878                 return NF_ACCEPT;
 879
 880         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 881          * about the value of skb->pkt_type. */
 882         if (skb->pkt_type == PACKET_OTHERHOST) {
 883                 skb->pkt_type = PACKET_HOST;
 884                 nf_bridge->mask |= BRNF_PKT_TYPE;
 885         }
 886
 887         nf_bridge_pull_encap_header(skb);
 888         nf_bridge_save_header(skb);
 889         if (pf == PF_INET)
 890                 skb->protocol = htons(ETH_P_IP);
 891         else
 892                 skb->protocol = htons(ETH_P_IPV6);
 893
 894         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
 895                 br_nf_dev_queue_xmit);
 896
 897         return NF_STOLEN;
 898 }
 899
 900 /* IP/SABOTAGE *****************************************************/
 901 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 902  * for the second time. */
 903 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
 904                                    const struct net_device *in,
 905                                    const struct net_device *out,
 906                                    int (*okfn)(struct sk_buff *))
 907 {
 908         if (skb->nf_bridge &&
 909             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 910                 return NF_STOP;
 911         }
 912
 913         return NF_ACCEPT;
 914 }
 915
 916 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 917  * br_dev_queue_push_xmit is called afterwards */
 918 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
 919         {
 920                 .hook = br_nf_pre_routing,
 921                 .owner = THIS_MODULE,
 922                 .pf = PF_BRIDGE,
 923                 .hooknum = NF_BR_PRE_ROUTING,
 924                 .priority = NF_BR_PRI_BRNF,
 925         },
 926         {
 927                 .hook = br_nf_local_in,
 928                 .owner = THIS_MODULE,
 929                 .pf = PF_BRIDGE,
 930                 .hooknum = NF_BR_LOCAL_IN,
 931                 .priority = NF_BR_PRI_BRNF,
 932         },
 933         {
 934                 .hook = br_nf_forward_ip,
 935                 .owner = THIS_MODULE,
 936                 .pf = PF_BRIDGE,
 937                 .hooknum = NF_BR_FORWARD,
 938                 .priority = NF_BR_PRI_BRNF - 1,
 939         },
 940         {
 941                 .hook = br_nf_forward_arp,
 942                 .owner = THIS_MODULE,
 943                 .pf = PF_BRIDGE,
 944                 .hooknum = NF_BR_FORWARD,
 945                 .priority = NF_BR_PRI_BRNF,
 946         },
 947         {
 948                 .hook = br_nf_post_routing,
 949                 .owner = THIS_MODULE,
 950                 .pf = PF_BRIDGE,
 951                 .hooknum = NF_BR_POST_ROUTING,
 952                 .priority = NF_BR_PRI_LAST,
 953         },
 954         {
 955                 .hook = ip_sabotage_in,
 956                 .owner = THIS_MODULE,
 957                 .pf = PF_INET,
 958                 .hooknum = NF_INET_PRE_ROUTING,
 959                 .priority = NF_IP_PRI_FIRST,
 960         },
 961         {
 962                 .hook = ip_sabotage_in,
 963                 .owner = THIS_MODULE,
 964                 .pf = PF_INET6,
 965                 .hooknum = NF_INET_PRE_ROUTING,
 966                 .priority = NF_IP6_PRI_FIRST,
 967         },
 968 };
 969
 970 #ifdef CONFIG_SYSCTL
 971 static
 972 int brnf_sysctl_call_tables(ctl_table * ctl, int write,
 973                             void __user * buffer, size_t * lenp, loff_t * ppos)
 974 {
 975         int ret;
 976
 977         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
 978
 979         if (write && *(int *)(ctl->data))
 980                 *(int *)(ctl->data) = 1;
 981         return ret;
 982 }
 983
 984 static ctl_table brnf_table[] = {
 985         {
 986                 .procname       = "bridge-nf-call-arptables",
 987                 .data           = &brnf_call_arptables,
 988                 .maxlen         = sizeof(int),
 989                 .mode           = 0644,
 990                 .proc_handler   = brnf_sysctl_call_tables,
 991         },
 992         {
 993                 .procname       = "bridge-nf-call-iptables",
 994                 .data           = &brnf_call_iptables,
 995                 .maxlen         = sizeof(int),
 996                 .mode           = 0644,
 997                 .proc_handler   = brnf_sysctl_call_tables,
 998         },
 999         {
1000                 .procname       = "bridge-nf-call-ip6tables",
1001                 .data           = &brnf_call_ip6tables,
1002                 .maxlen         = sizeof(int),
1003                 .mode           = 0644,
1004                 .proc_handler   = brnf_sysctl_call_tables,
1005         },
1006         {
1007                 .procname       = "bridge-nf-filter-vlan-tagged",
1008                 .data           = &brnf_filter_vlan_tagged,
1009                 .maxlen         = sizeof(int),
1010                 .mode           = 0644,
1011                 .proc_handler   = brnf_sysctl_call_tables,
1012         },
1013         {
1014                 .procname       = "bridge-nf-filter-pppoe-tagged",
1015                 .data           = &brnf_filter_pppoe_tagged,
1016                 .maxlen         = sizeof(int),
1017                 .mode           = 0644,
1018                 .proc_handler   = brnf_sysctl_call_tables,
1019         },
1020         {
1021                 .procname       = "bridge-nf-pass-vlan-input-dev",
1022                 .data           = &brnf_pass_vlan_indev,
1023                 .maxlen         = sizeof(int),
1024                 .mode           = 0644,
1025                 .proc_handler   = brnf_sysctl_call_tables,
1026         },
1027         { }
1028 };
1029 #endif
1030
1031 int __init br_netfilter_init(void)
1032 {
1033         int ret;
1034
1035         ret = dst_entries_init(&fake_dst_ops);
1036         if (ret < 0)
1037                 return ret;
1038
1039         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1040         if (ret < 0) {
1041                 dst_entries_destroy(&fake_dst_ops);
1042                 return ret;
1043         }
1044 #ifdef CONFIG_SYSCTL
1045         brnf_sysctl_header = register_net_sysctl(&init_net, "net/bridge", brnf_table);
1046         if (brnf_sysctl_header == NULL) {
1047                 printk(KERN_WARNING
1048                        "br_netfilter: can't register to sysctl.\n");
1049                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1050                 dst_entries_destroy(&fake_dst_ops);
1051                 return -ENOMEM;
1052         }
1053 #endif
1054         printk(KERN_NOTICE "Bridge firewalling registered\n");
1055         return 0;
1056 }
1057
1058 void br_netfilter_fini(void)
1059 {
1060         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1061 #ifdef CONFIG_SYSCTL
1062         unregister_net_sysctl_table(brnf_sysctl_header);
1063 #endif
1064         dst_entries_destroy(&fake_dst_ops);
1065 }