net/bridge/br_netfilter_hooks.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 <uapi/linux/netfilter_bridge.h>
  30 #include <linux/netfilter_ipv4.h>
  31 #include <linux/netfilter_ipv6.h>
  32 #include <linux/netfilter_arp.h>
  33 #include <linux/in_route.h>
  34 #include <linux/rculist.h>
  35 #include <linux/inetdevice.h>
  36
  37 #include <net/ip.h>
  38 #include <net/ipv6.h>
  39 #include <net/addrconf.h>
  40 #include <net/route.h>
  41 #include <net/netfilter/br_netfilter.h>
  42 #include <net/netns/generic.h>
  43
  44 #include <linux/uaccess.h>
  45 #include "br_private.h"
  46 #ifdef CONFIG_SYSCTL
  47 #include <linux/sysctl.h>
  48 #endif
  49
  50 static unsigned int brnf_net_id __read_mostly;
  51
  52 struct brnf_net {
  53         bool enabled;
  54 };
  55
  56 #ifdef CONFIG_SYSCTL
  57 static struct ctl_table_header *brnf_sysctl_header;
  58 static int brnf_call_iptables __read_mostly = 1;
  59 static int brnf_call_ip6tables __read_mostly = 1;
  60 static int brnf_call_arptables __read_mostly = 1;
  61 static int brnf_filter_vlan_tagged __read_mostly;
  62 static int brnf_filter_pppoe_tagged __read_mostly;
  63 static int brnf_pass_vlan_indev __read_mostly;
  64 #else
  65 #define brnf_call_iptables 1
  66 #define brnf_call_ip6tables 1
  67 #define brnf_call_arptables 1
  68 #define brnf_filter_vlan_tagged 0
  69 #define brnf_filter_pppoe_tagged 0
  70 #define brnf_pass_vlan_indev 0
  71 #endif
  72
  73 #define IS_IP(skb) \
  74         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IP))
  75
  76 #define IS_IPV6(skb) \
  77         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IPV6))
  78
  79 #define IS_ARP(skb) \
  80         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_ARP))
  81
  82 static inline __be16 vlan_proto(const struct sk_buff *skb)
  83 {
  84         if (skb_vlan_tag_present(skb))
  85                 return skb->protocol;
  86         else if (skb->protocol == htons(ETH_P_8021Q))
  87                 return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  88         else
  89                 return 0;
  90 }
  91
  92 #define IS_VLAN_IP(skb) \
  93         (vlan_proto(skb) == htons(ETH_P_IP) && \
  94          brnf_filter_vlan_tagged)
  95
  96 #define IS_VLAN_IPV6(skb) \
  97         (vlan_proto(skb) == htons(ETH_P_IPV6) && \
  98          brnf_filter_vlan_tagged)
  99
 100 #define IS_VLAN_ARP(skb) \
 101         (vlan_proto(skb) == htons(ETH_P_ARP) && \
 102          brnf_filter_vlan_tagged)
 103
 104 static inline __be16 pppoe_proto(const struct sk_buff *skb)
 105 {
 106         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
 107                             sizeof(struct pppoe_hdr)));
 108 }
 109
 110 #define IS_PPPOE_IP(skb) \
 111         (skb->protocol == htons(ETH_P_PPP_SES) && \
 112          pppoe_proto(skb) == htons(PPP_IP) && \
 113          brnf_filter_pppoe_tagged)
 114
 115 #define IS_PPPOE_IPV6(skb) \
 116         (skb->protocol == htons(ETH_P_PPP_SES) && \
 117          pppoe_proto(skb) == htons(PPP_IPV6) && \
 118          brnf_filter_pppoe_tagged)
 119
 120 /* largest possible L2 header, see br_nf_dev_queue_xmit() */
 121 #define NF_BRIDGE_MAX_MAC_HEADER_LENGTH (PPPOE_SES_HLEN + ETH_HLEN)
 122
 123 struct brnf_frag_data {
 124         char mac[NF_BRIDGE_MAX_MAC_HEADER_LENGTH];
 125         u8 encap_size;
 126         u8 size;
 127         u16 vlan_tci;
 128         __be16 vlan_proto;
 129 };
 130
 131 static DEFINE_PER_CPU(struct brnf_frag_data, brnf_frag_data_storage);
 132
 133 static void nf_bridge_info_free(struct sk_buff *skb)
 134 {
 135         if (skb->nf_bridge) {
 136                 nf_bridge_put(skb->nf_bridge);
 137                 skb->nf_bridge = NULL;
 138         }
 139 }
 140
 141 static inline struct net_device *bridge_parent(const struct net_device *dev)
 142 {
 143         struct net_bridge_port *port;
 144
 145         port = br_port_get_rcu(dev);
 146         return port ? port->br->dev : NULL;
 147 }
 148
 149 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 150 {
 151         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 152
 153         if (refcount_read(&nf_bridge->use) > 1) {
 154                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 155
 156                 if (tmp) {
 157                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 158                         refcount_set(&tmp->use, 1);
 159                 }
 160                 nf_bridge_put(nf_bridge);
 161                 nf_bridge = tmp;
 162         }
 163         return nf_bridge;
 164 }
 165
 166 unsigned int nf_bridge_encap_header_len(const struct sk_buff *skb)
 167 {
 168         switch (skb->protocol) {
 169         case __cpu_to_be16(ETH_P_8021Q):
 170                 return VLAN_HLEN;
 171         case __cpu_to_be16(ETH_P_PPP_SES):
 172                 return PPPOE_SES_HLEN;
 173         default:
 174                 return 0;
 175         }
 176 }
 177
 178 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 179 {
 180         unsigned int len = nf_bridge_encap_header_len(skb);
 181
 182         skb_pull(skb, len);
 183         skb->network_header += len;
 184 }
 185
 186 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 187 {
 188         unsigned int len = nf_bridge_encap_header_len(skb);
 189
 190         skb_pull_rcsum(skb, len);
 191         skb->network_header += len;
 192 }
 193
 194 /* When handing a packet over to the IP layer
 195  * check whether we have a skb that is in the
 196  * expected format
 197  */
 198
 199 static int br_validate_ipv4(struct net *net, struct sk_buff *skb)
 200 {
 201         const struct iphdr *iph;
 202         u32 len;
 203
 204         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 205                 goto inhdr_error;
 206
 207         iph = ip_hdr(skb);
 208
 209         /* Basic sanity checks */
 210         if (iph->ihl < 5 || iph->version != 4)
 211                 goto inhdr_error;
 212
 213         if (!pskb_may_pull(skb, iph->ihl*4))
 214                 goto inhdr_error;
 215
 216         iph = ip_hdr(skb);
 217         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 218                 goto csum_error;
 219
 220         len = ntohs(iph->tot_len);
 221         if (skb->len < len) {
 222                 __IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS);
 223                 goto drop;
 224         } else if (len < (iph->ihl*4))
 225                 goto inhdr_error;
 226
 227         if (pskb_trim_rcsum(skb, len)) {
 228                 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
 229                 goto drop;
 230         }
 231
 232         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 233         /* We should really parse IP options here but until
 234          * somebody who actually uses IP options complains to
 235          * us we'll just silently ignore the options because
 236          * we're lazy!
 237          */
 238         return 0;
 239
 240 csum_error:
 241         __IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS);
 242 inhdr_error:
 243         __IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
 244 drop:
 245         return -1;
 246 }
 247
 248 void nf_bridge_update_protocol(struct sk_buff *skb)
 249 {
 250         switch (skb->nf_bridge->orig_proto) {
 251         case BRNF_PROTO_8021Q:
 252                 skb->protocol = htons(ETH_P_8021Q);
 253                 break;
 254         case BRNF_PROTO_PPPOE:
 255                 skb->protocol = htons(ETH_P_PPP_SES);
 256                 break;
 257         case BRNF_PROTO_UNCHANGED:
 258                 break;
 259         }
 260 }
 261
 262 /* Obtain the correct destination MAC address, while preserving the original
 263  * source MAC address. If we already know this address, we just copy it. If we
 264  * don't, we use the neighbour framework to find out. In both cases, we make
 265  * sure that br_handle_frame_finish() is called afterwards.
 266  */
 267 int br_nf_pre_routing_finish_bridge(struct net *net, struct sock *sk, struct sk_buff *skb)
 268 {
 269         struct neighbour *neigh;
 270         struct dst_entry *dst;
 271
 272         skb->dev = bridge_parent(skb->dev);
 273         if (!skb->dev)
 274                 goto free_skb;
 275         dst = skb_dst(skb);
 276         neigh = dst_neigh_lookup_skb(dst, skb);
 277         if (neigh) {
 278                 struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 279                 int ret;
 280
 281                 if ((neigh->nud_state & NUD_CONNECTED) && neigh->hh.hh_len) {
 282                         neigh_hh_bridge(&neigh->hh, skb);
 283                         skb->dev = nf_bridge->physindev;
 284                         ret = br_handle_frame_finish(net, sk, skb);
 285                 } else {
 286                         /* the neighbour function below overwrites the complete
 287                          * MAC header, so we save the Ethernet source address and
 288                          * protocol number.
 289                          */
 290                         skb_copy_from_linear_data_offset(skb,
 291                                                          -(ETH_HLEN-ETH_ALEN),
 292                                                          nf_bridge->neigh_header,
 293                                                          ETH_HLEN-ETH_ALEN);
 294                         /* tell br_dev_xmit to continue with forwarding */
 295                         nf_bridge->bridged_dnat = 1;
 296                         /* FIXME Need to refragment */
 297                         ret = neigh->output(neigh, skb);
 298                 }
 299                 neigh_release(neigh);
 300                 return ret;
 301         }
 302 free_skb:
 303         kfree_skb(skb);
 304         return 0;
 305 }
 306
 307 static inline bool
 308 br_nf_ipv4_daddr_was_changed(const struct sk_buff *skb,
 309                              const struct nf_bridge_info *nf_bridge)
 310 {
 311         return ip_hdr(skb)->daddr != nf_bridge->ipv4_daddr;
 312 }
 313
 314 /* This requires some explaining. If DNAT has taken place,
 315  * we will need to fix up the destination Ethernet address.
 316  * This is also true when SNAT takes place (for the reply direction).
 317  *
 318  * There are two cases to consider:
 319  * 1. The packet was DNAT'ed to a device in the same bridge
 320  *    port group as it was received on. We can still bridge
 321  *    the packet.
 322  * 2. The packet was DNAT'ed to a different device, either
 323  *    a non-bridged device or another bridge port group.
 324  *    The packet will need to be routed.
 325  *
 326  * The correct way of distinguishing between these two cases is to
 327  * call ip_route_input() and to look at skb->dst->dev, which is
 328  * changed to the destination device if ip_route_input() succeeds.
 329  *
 330  * Let's first consider the case that ip_route_input() succeeds:
 331  *
 332  * If the output device equals the logical bridge device the packet
 333  * came in on, we can consider this bridging. The corresponding MAC
 334  * address will be obtained in br_nf_pre_routing_finish_bridge.
 335  * Otherwise, the packet is considered to be routed and we just
 336  * change the destination MAC address so that the packet will
 337  * later be passed up to the IP stack to be routed. For a redirected
 338  * packet, ip_route_input() will give back the localhost as output device,
 339  * which differs from the bridge device.
 340  *
 341  * Let's now consider the case that ip_route_input() fails:
 342  *
 343  * This can be because the destination address is martian, in which case
 344  * the packet will be dropped.
 345  * If IP forwarding is disabled, ip_route_input() will fail, while
 346  * ip_route_output_key() can return success. The source
 347  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 348  * thinks we're handling a locally generated packet and won't care
 349  * if IP forwarding is enabled. If the output device equals the logical bridge
 350  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 351  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 352  */
 353 static int br_nf_pre_routing_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
 354 {
 355         struct net_device *dev = skb->dev;
 356         struct iphdr *iph = ip_hdr(skb);
 357         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 358         struct rtable *rt;
 359         int err;
 360
 361         nf_bridge->frag_max_size = IPCB(skb)->frag_max_size;
 362
 363         if (nf_bridge->pkt_otherhost) {
 364                 skb->pkt_type = PACKET_OTHERHOST;
 365                 nf_bridge->pkt_otherhost = false;
 366         }
 367         nf_bridge->in_prerouting = 0;
 368         if (br_nf_ipv4_daddr_was_changed(skb, nf_bridge)) {
 369                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 370                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 371
 372                         /* If err equals -EHOSTUNREACH the error is due to a
 373                          * martian destination or due to the fact that
 374                          * forwarding is disabled. For most martian packets,
 375                          * ip_route_output_key() will fail. It won't fail for 2 types of
 376                          * martian destinations: loopback destinations and destination
 377                          * 0.0.0.0. In both cases the packet will be dropped because the
 378                          * destination is the loopback device and not the bridge. */
 379                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 380                                 goto free_skb;
 381
 382                         rt = ip_route_output(net, iph->daddr, 0,
 383                                              RT_TOS(iph->tos), 0);
 384                         if (!IS_ERR(rt)) {
 385                                 /* - Bridged-and-DNAT'ed traffic doesn't
 386                                  *   require ip_forwarding. */
 387                                 if (rt->dst.dev == dev) {
 388                                         skb_dst_set(skb, &rt->dst);
 389                                         goto bridged_dnat;
 390                                 }
 391                                 ip_rt_put(rt);
 392                         }
 393 free_skb:
 394                         kfree_skb(skb);
 395                         return 0;
 396                 } else {
 397                         if (skb_dst(skb)->dev == dev) {
 398 bridged_dnat:
 399                                 skb->dev = nf_bridge->physindev;
 400                                 nf_bridge_update_protocol(skb);
 401                                 nf_bridge_push_encap_header(skb);
 402                                 br_nf_hook_thresh(NF_BR_PRE_ROUTING,
 403                                                   net, sk, skb, skb->dev,
 404                                                   NULL,
 405                                                   br_nf_pre_routing_finish_bridge);
 406                                 return 0;
 407                         }
 408                         ether_addr_copy(eth_hdr(skb)->h_dest, dev->dev_addr);
 409                         skb->pkt_type = PACKET_HOST;
 410                 }
 411         } else {
 412                 rt = bridge_parent_rtable(nf_bridge->physindev);
 413                 if (!rt) {
 414                         kfree_skb(skb);
 415                         return 0;
 416                 }
 417                 skb_dst_set_noref(skb, &rt->dst);
 418         }
 419
 420         skb->dev = nf_bridge->physindev;
 421         nf_bridge_update_protocol(skb);
 422         nf_bridge_push_encap_header(skb);
 423         br_nf_hook_thresh(NF_BR_PRE_ROUTING, net, sk, skb, skb->dev, NULL,
 424                           br_handle_frame_finish);
 425         return 0;
 426 }
 427
 428 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb, const struct net_device *dev)
 429 {
 430         struct net_device *vlan, *br;
 431
 432         br = bridge_parent(dev);
 433         if (brnf_pass_vlan_indev == 0 || !skb_vlan_tag_present(skb))
 434                 return br;
 435
 436         vlan = __vlan_find_dev_deep_rcu(br, skb->vlan_proto,
 437                                     skb_vlan_tag_get(skb) & VLAN_VID_MASK);
 438
 439         return vlan ? vlan : br;
 440 }
 441
 442 /* Some common code for IPv4/IPv6 */
 443 struct net_device *setup_pre_routing(struct sk_buff *skb)
 444 {
 445         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 446
 447         if (skb->pkt_type == PACKET_OTHERHOST) {
 448                 skb->pkt_type = PACKET_HOST;
 449                 nf_bridge->pkt_otherhost = true;
 450         }
 451
 452         nf_bridge->in_prerouting = 1;
 453         nf_bridge->physindev = skb->dev;
 454         skb->dev = brnf_get_logical_dev(skb, skb->dev);
 455
 456         if (skb->protocol == htons(ETH_P_8021Q))
 457                 nf_bridge->orig_proto = BRNF_PROTO_8021Q;
 458         else if (skb->protocol == htons(ETH_P_PPP_SES))
 459                 nf_bridge->orig_proto = BRNF_PROTO_PPPOE;
 460
 461         /* Must drop socket now because of tproxy. */
 462         skb_orphan(skb);
 463         return skb->dev;
 464 }
 465
 466 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 467  * Replicate the checks that IPv4 does on packet reception.
 468  * Set skb->dev to the bridge device (i.e. parent of the
 469  * receiving device) to make netfilter happy, the REDIRECT
 470  * target in particular.  Save the original destination IP
 471  * address to be able to detect DNAT afterwards. */
 472 static unsigned int br_nf_pre_routing(void *priv,
 473                                       struct sk_buff *skb,
 474                                       const struct nf_hook_state *state)
 475 {
 476         struct nf_bridge_info *nf_bridge;
 477         struct net_bridge_port *p;
 478         struct net_bridge *br;
 479         __u32 len = nf_bridge_encap_header_len(skb);
 480
 481         if (unlikely(!pskb_may_pull(skb, len)))
 482                 return NF_DROP;
 483
 484         p = br_port_get_rcu(state->in);
 485         if (p == NULL)
 486                 return NF_DROP;
 487         br = p->br;
 488
 489         if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb)) {
 490                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
 491                         return NF_ACCEPT;
 492
 493                 nf_bridge_pull_encap_header_rcsum(skb);
 494                 return br_nf_pre_routing_ipv6(priv, skb, state);
 495         }
 496
 497         if (!brnf_call_iptables && !br->nf_call_iptables)
 498                 return NF_ACCEPT;
 499
 500         if (!IS_IP(skb) && !IS_VLAN_IP(skb) && !IS_PPPOE_IP(skb))
 501                 return NF_ACCEPT;
 502
 503         nf_bridge_pull_encap_header_rcsum(skb);
 504
 505         if (br_validate_ipv4(state->net, skb))
 506                 return NF_DROP;
 507
 508         nf_bridge_put(skb->nf_bridge);
 509         if (!nf_bridge_alloc(skb))
 510                 return NF_DROP;
 511         if (!setup_pre_routing(skb))
 512                 return NF_DROP;
 513
 514         nf_bridge = nf_bridge_info_get(skb);
 515         nf_bridge->ipv4_daddr = ip_hdr(skb)->daddr;
 516
 517         skb->protocol = htons(ETH_P_IP);
 518         skb->transport_header = skb->network_header + ip_hdr(skb)->ihl * 4;
 519
 520         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, state->net, state->sk, skb,
 521                 skb->dev, NULL,
 522                 br_nf_pre_routing_finish);
 523
 524         return NF_STOLEN;
 525 }
 526
 527
 528 /* PF_BRIDGE/FORWARD *************************************************/
 529 static int br_nf_forward_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
 530 {
 531         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 532         struct net_device *in;
 533
 534         if (!IS_ARP(skb) && !IS_VLAN_ARP(skb)) {
 535
 536                 if (skb->protocol == htons(ETH_P_IP))
 537                         nf_bridge->frag_max_size = IPCB(skb)->frag_max_size;
 538
 539                 if (skb->protocol == htons(ETH_P_IPV6))
 540                         nf_bridge->frag_max_size = IP6CB(skb)->frag_max_size;
 541
 542                 in = nf_bridge->physindev;
 543                 if (nf_bridge->pkt_otherhost) {
 544                         skb->pkt_type = PACKET_OTHERHOST;
 545                         nf_bridge->pkt_otherhost = false;
 546                 }
 547                 nf_bridge_update_protocol(skb);
 548         } else {
 549                 in = *((struct net_device **)(skb->cb));
 550         }
 551         nf_bridge_push_encap_header(skb);
 552
 553         br_nf_hook_thresh(NF_BR_FORWARD, net, sk, skb, in, skb->dev,
 554                           br_forward_finish);
 555         return 0;
 556 }
 557
 558
 559 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 560  * netfilter with indev and outdev set to the bridge device,
 561  * but we are still able to filter on the 'real' indev/outdev
 562  * because of the physdev module. For ARP, indev and outdev are the
 563  * bridge ports. */
 564 static unsigned int br_nf_forward_ip(void *priv,
 565                                      struct sk_buff *skb,
 566                                      const struct nf_hook_state *state)
 567 {
 568         struct nf_bridge_info *nf_bridge;
 569         struct net_device *parent;
 570         u_int8_t pf;
 571
 572         if (!skb->nf_bridge)
 573                 return NF_ACCEPT;
 574
 575         /* Need exclusive nf_bridge_info since we might have multiple
 576          * different physoutdevs. */
 577         if (!nf_bridge_unshare(skb))
 578                 return NF_DROP;
 579
 580         nf_bridge = nf_bridge_info_get(skb);
 581         if (!nf_bridge)
 582                 return NF_DROP;
 583
 584         parent = bridge_parent(state->out);
 585         if (!parent)
 586                 return NF_DROP;
 587
 588         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 589                 pf = NFPROTO_IPV4;
 590         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 591                 pf = NFPROTO_IPV6;
 592         else
 593                 return NF_ACCEPT;
 594
 595         nf_bridge_pull_encap_header(skb);
 596
 597         if (skb->pkt_type == PACKET_OTHERHOST) {
 598                 skb->pkt_type = PACKET_HOST;
 599                 nf_bridge->pkt_otherhost = true;
 600         }
 601
 602         if (pf == NFPROTO_IPV4) {
 603                 if (br_validate_ipv4(state->net, skb))
 604                         return NF_DROP;
 605                 IPCB(skb)->frag_max_size = nf_bridge->frag_max_size;
 606         }
 607
 608         if (pf == NFPROTO_IPV6) {
 609                 if (br_validate_ipv6(state->net, skb))
 610                         return NF_DROP;
 611                 IP6CB(skb)->frag_max_size = nf_bridge->frag_max_size;
 612         }
 613
 614         nf_bridge->physoutdev = skb->dev;
 615         if (pf == NFPROTO_IPV4)
 616                 skb->protocol = htons(ETH_P_IP);
 617         else
 618                 skb->protocol = htons(ETH_P_IPV6);
 619
 620         NF_HOOK(pf, NF_INET_FORWARD, state->net, NULL, skb,
 621                 brnf_get_logical_dev(skb, state->in),
 622                 parent, br_nf_forward_finish);
 623
 624         return NF_STOLEN;
 625 }
 626
 627 static unsigned int br_nf_forward_arp(void *priv,
 628                                       struct sk_buff *skb,
 629                                       const struct nf_hook_state *state)
 630 {
 631         struct net_bridge_port *p;
 632         struct net_bridge *br;
 633         struct net_device **d = (struct net_device **)(skb->cb);
 634
 635         p = br_port_get_rcu(state->out);
 636         if (p == NULL)
 637                 return NF_ACCEPT;
 638         br = p->br;
 639
 640         if (!brnf_call_arptables && !br->nf_call_arptables)
 641                 return NF_ACCEPT;
 642
 643         if (!IS_ARP(skb)) {
 644                 if (!IS_VLAN_ARP(skb))
 645                         return NF_ACCEPT;
 646                 nf_bridge_pull_encap_header(skb);
 647         }
 648
 649         if (unlikely(!pskb_may_pull(skb, sizeof(struct arphdr))))
 650                 return NF_DROP;
 651
 652         if (arp_hdr(skb)->ar_pln != 4) {
 653                 if (IS_VLAN_ARP(skb))
 654                         nf_bridge_push_encap_header(skb);
 655                 return NF_ACCEPT;
 656         }
 657         *d = state->in;
 658         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, state->net, state->sk, skb,
 659                 state->in, state->out, br_nf_forward_finish);
 660
 661         return NF_STOLEN;
 662 }
 663
 664 static int br_nf_push_frag_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
 665 {
 666         struct brnf_frag_data *data;
 667         int err;
 668
 669         data = this_cpu_ptr(&brnf_frag_data_storage);
 670         err = skb_cow_head(skb, data->size);
 671
 672         if (err) {
 673                 kfree_skb(skb);
 674                 return 0;
 675         }
 676
 677         if (data->vlan_tci) {
 678                 skb->vlan_tci = data->vlan_tci;
 679                 skb->vlan_proto = data->vlan_proto;
 680         }
 681
 682         skb_copy_to_linear_data_offset(skb, -data->size, data->mac, data->size);
 683         __skb_push(skb, data->encap_size);
 684
 685         nf_bridge_info_free(skb);
 686         return br_dev_queue_push_xmit(net, sk, skb);
 687 }
 688
 689 static int
 690 br_nf_ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
 691                   int (*output)(struct net *, struct sock *, struct sk_buff *))
 692 {
 693         unsigned int mtu = ip_skb_dst_mtu(sk, skb);
 694         struct iphdr *iph = ip_hdr(skb);
 695
 696         if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->ignore_df) ||
 697                      (IPCB(skb)->frag_max_size &&
 698                       IPCB(skb)->frag_max_size > mtu))) {
 699                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
 700                 kfree_skb(skb);
 701                 return -EMSGSIZE;
 702         }
 703
 704         return ip_do_fragment(net, sk, skb, output);
 705 }
 706
 707 static unsigned int nf_bridge_mtu_reduction(const struct sk_buff *skb)
 708 {
 709         if (skb->nf_bridge->orig_proto == BRNF_PROTO_PPPOE)
 710                 return PPPOE_SES_HLEN;
 711         return 0;
 712 }
 713
 714 static int br_nf_dev_queue_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
 715 {
 716         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 717         unsigned int mtu, mtu_reserved;
 718
 719         mtu_reserved = nf_bridge_mtu_reduction(skb);
 720         mtu = skb->dev->mtu;
 721
 722         if (nf_bridge->pkt_otherhost) {
 723                 skb->pkt_type = PACKET_OTHERHOST;
 724                 nf_bridge->pkt_otherhost = false;
 725         }
 726
 727         if (nf_bridge->frag_max_size && nf_bridge->frag_max_size < mtu)
 728                 mtu = nf_bridge->frag_max_size;
 729
 730         nf_bridge_update_protocol(skb);
 731         nf_bridge_push_encap_header(skb);
 732
 733         if (skb_is_gso(skb) || skb->len + mtu_reserved <= mtu) {
 734                 nf_bridge_info_free(skb);
 735                 return br_dev_queue_push_xmit(net, sk, skb);
 736         }
 737
 738         /* This is wrong! We should preserve the original fragment
 739          * boundaries by preserving frag_list rather than refragmenting.
 740          */
 741         if (IS_ENABLED(CONFIG_NF_DEFRAG_IPV4) &&
 742             skb->protocol == htons(ETH_P_IP)) {
 743                 struct brnf_frag_data *data;
 744
 745                 if (br_validate_ipv4(net, skb))
 746                         goto drop;
 747
 748                 IPCB(skb)->frag_max_size = nf_bridge->frag_max_size;
 749
 750                 data = this_cpu_ptr(&brnf_frag_data_storage);
 751
 752                 data->vlan_tci = skb->vlan_tci;
 753                 data->vlan_proto = skb->vlan_proto;
 754                 data->encap_size = nf_bridge_encap_header_len(skb);
 755                 data->size = ETH_HLEN + data->encap_size;
 756
 757                 skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
 758                                                  data->size);
 759
 760                 return br_nf_ip_fragment(net, sk, skb, br_nf_push_frag_xmit);
 761         }
 762         if (IS_ENABLED(CONFIG_NF_DEFRAG_IPV6) &&
 763             skb->protocol == htons(ETH_P_IPV6)) {
 764                 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
 765                 struct brnf_frag_data *data;
 766
 767                 if (br_validate_ipv6(net, skb))
 768                         goto drop;
 769
 770                 IP6CB(skb)->frag_max_size = nf_bridge->frag_max_size;
 771
 772                 data = this_cpu_ptr(&brnf_frag_data_storage);
 773                 data->encap_size = nf_bridge_encap_header_len(skb);
 774                 data->size = ETH_HLEN + data->encap_size;
 775
 776                 skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
 777                                                  data->size);
 778
 779                 if (v6ops)
 780                         return v6ops->fragment(net, sk, skb, br_nf_push_frag_xmit);
 781
 782                 kfree_skb(skb);
 783                 return -EMSGSIZE;
 784         }
 785         nf_bridge_info_free(skb);
 786         return br_dev_queue_push_xmit(net, sk, skb);
 787  drop:
 788         kfree_skb(skb);
 789         return 0;
 790 }
 791
 792 /* PF_BRIDGE/POST_ROUTING ********************************************/
 793 static unsigned int br_nf_post_routing(void *priv,
 794                                        struct sk_buff *skb,
 795                                        const struct nf_hook_state *state)
 796 {
 797         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 798         struct net_device *realoutdev = bridge_parent(skb->dev);
 799         u_int8_t pf;
 800
 801         /* if nf_bridge is set, but ->physoutdev is NULL, this packet came in
 802          * on a bridge, but was delivered locally and is now being routed:
 803          *
 804          * POST_ROUTING was already invoked from the ip stack.
 805          */
 806         if (!nf_bridge || !nf_bridge->physoutdev)
 807                 return NF_ACCEPT;
 808
 809         if (!realoutdev)
 810                 return NF_DROP;
 811
 812         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 813                 pf = NFPROTO_IPV4;
 814         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 815                 pf = NFPROTO_IPV6;
 816         else
 817                 return NF_ACCEPT;
 818
 819         if (skb->pkt_type == PACKET_OTHERHOST) {
 820                 skb->pkt_type = PACKET_HOST;
 821                 nf_bridge->pkt_otherhost = true;
 822         }
 823
 824         nf_bridge_pull_encap_header(skb);
 825         if (pf == NFPROTO_IPV4)
 826                 skb->protocol = htons(ETH_P_IP);
 827         else
 828                 skb->protocol = htons(ETH_P_IPV6);
 829
 830         NF_HOOK(pf, NF_INET_POST_ROUTING, state->net, state->sk, skb,
 831                 NULL, realoutdev,
 832                 br_nf_dev_queue_xmit);
 833
 834         return NF_STOLEN;
 835 }
 836
 837 /* IP/SABOTAGE *****************************************************/
 838 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 839  * for the second time. */
 840 static unsigned int ip_sabotage_in(void *priv,
 841                                    struct sk_buff *skb,
 842                                    const struct nf_hook_state *state)
 843 {
 844         if (skb->nf_bridge && !skb->nf_bridge->in_prerouting &&
 845             !netif_is_l3_master(skb->dev)) {
 846                 state->okfn(state->net, state->sk, skb);
 847                 return NF_STOLEN;
 848         }
 849
 850         return NF_ACCEPT;
 851 }
 852
 853 /* This is called when br_netfilter has called into iptables/netfilter,
 854  * and DNAT has taken place on a bridge-forwarded packet.
 855  *
 856  * neigh->output has created a new MAC header, with local br0 MAC
 857  * as saddr.
 858  *
 859  * This restores the original MAC saddr of the bridged packet
 860  * before invoking bridge forward logic to transmit the packet.
 861  */
 862 static void br_nf_pre_routing_finish_bridge_slow(struct sk_buff *skb)
 863 {
 864         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 865
 866         skb_pull(skb, ETH_HLEN);
 867         nf_bridge->bridged_dnat = 0;
 868
 869         BUILD_BUG_ON(sizeof(nf_bridge->neigh_header) != (ETH_HLEN - ETH_ALEN));
 870
 871         skb_copy_to_linear_data_offset(skb, -(ETH_HLEN - ETH_ALEN),
 872                                        nf_bridge->neigh_header,
 873                                        ETH_HLEN - ETH_ALEN);
 874         skb->dev = nf_bridge->physindev;
 875
 876         nf_bridge->physoutdev = NULL;
 877         br_handle_frame_finish(dev_net(skb->dev), NULL, skb);
 878 }
 879
 880 static int br_nf_dev_xmit(struct sk_buff *skb)
 881 {
 882         if (skb->nf_bridge && skb->nf_bridge->bridged_dnat) {
 883                 br_nf_pre_routing_finish_bridge_slow(skb);
 884                 return 1;
 885         }
 886         return 0;
 887 }
 888
 889 static const struct nf_br_ops br_ops = {
 890         .br_dev_xmit_hook =     br_nf_dev_xmit,
 891 };
 892
 893 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 894  * br_dev_queue_push_xmit is called afterwards */
 895 static const struct nf_hook_ops br_nf_ops[] = {
 896         {
 897                 .hook = br_nf_pre_routing,
 898                 .pf = NFPROTO_BRIDGE,
 899                 .hooknum = NF_BR_PRE_ROUTING,
 900                 .priority = NF_BR_PRI_BRNF,
 901         },
 902         {
 903                 .hook = br_nf_forward_ip,
 904                 .pf = NFPROTO_BRIDGE,
 905                 .hooknum = NF_BR_FORWARD,
 906                 .priority = NF_BR_PRI_BRNF - 1,
 907         },
 908         {
 909                 .hook = br_nf_forward_arp,
 910                 .pf = NFPROTO_BRIDGE,
 911                 .hooknum = NF_BR_FORWARD,
 912                 .priority = NF_BR_PRI_BRNF,
 913         },
 914         {
 915                 .hook = br_nf_post_routing,
 916                 .pf = NFPROTO_BRIDGE,
 917                 .hooknum = NF_BR_POST_ROUTING,
 918                 .priority = NF_BR_PRI_LAST,
 919         },
 920         {
 921                 .hook = ip_sabotage_in,
 922                 .pf = NFPROTO_IPV4,
 923                 .hooknum = NF_INET_PRE_ROUTING,
 924                 .priority = NF_IP_PRI_FIRST,
 925         },
 926         {
 927                 .hook = ip_sabotage_in,
 928                 .pf = NFPROTO_IPV6,
 929                 .hooknum = NF_INET_PRE_ROUTING,
 930                 .priority = NF_IP6_PRI_FIRST,
 931         },
 932 };
 933
 934 static int brnf_device_event(struct notifier_block *unused, unsigned long event,
 935                              void *ptr)
 936 {
 937         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 938         struct brnf_net *brnet;
 939         struct net *net;
 940         int ret;
 941
 942         if (event != NETDEV_REGISTER || !(dev->priv_flags & IFF_EBRIDGE))
 943                 return NOTIFY_DONE;
 944
 945         ASSERT_RTNL();
 946
 947         net = dev_net(dev);
 948         brnet = net_generic(net, brnf_net_id);
 949         if (brnet->enabled)
 950                 return NOTIFY_OK;
 951
 952         ret = nf_register_net_hooks(net, br_nf_ops, ARRAY_SIZE(br_nf_ops));
 953         if (ret)
 954                 return NOTIFY_BAD;
 955
 956         brnet->enabled = true;
 957         return NOTIFY_OK;
 958 }
 959
 960 static void __net_exit brnf_exit_net(struct net *net)
 961 {
 962         struct brnf_net *brnet = net_generic(net, brnf_net_id);
 963
 964         if (!brnet->enabled)
 965                 return;
 966
 967         nf_unregister_net_hooks(net, br_nf_ops, ARRAY_SIZE(br_nf_ops));
 968         brnet->enabled = false;
 969 }
 970
 971 static struct pernet_operations brnf_net_ops __read_mostly = {
 972         .exit = brnf_exit_net,
 973         .id   = &brnf_net_id,
 974         .size = sizeof(struct brnf_net),
 975 };
 976
 977 static struct notifier_block brnf_notifier __read_mostly = {
 978         .notifier_call = brnf_device_event,
 979 };
 980
 981 /* recursively invokes nf_hook_slow (again), skipping already-called
 982  * hooks (< NF_BR_PRI_BRNF).
 983  *
 984  * Called with rcu read lock held.
 985  */
 986 int br_nf_hook_thresh(unsigned int hook, struct net *net,
 987                       struct sock *sk, struct sk_buff *skb,
 988                       struct net_device *indev,
 989                       struct net_device *outdev,
 990                       int (*okfn)(struct net *, struct sock *,
 991                                   struct sk_buff *))
 992 {
 993         const struct nf_hook_entries *e;
 994         struct nf_hook_state state;
 995         struct nf_hook_ops **ops;
 996         unsigned int i;
 997         int ret;
 998
 999         e = rcu_dereference(net->nf.hooks_bridge[hook]);
1000         if (!e)
1001                 return okfn(net, sk, skb);
1002
1003         ops = nf_hook_entries_get_hook_ops(e);
1004         for (i = 0; i < e->num_hook_entries &&
1005               ops[i]->priority <= NF_BR_PRI_BRNF; i++)
1006                 ;
1007
1008         nf_hook_state_init(&state, hook, NFPROTO_BRIDGE, indev, outdev,
1009                            sk, net, okfn);
1010
1011         ret = nf_hook_slow(skb, &state, e, i);
1012         if (ret == 1)
1013                 ret = okfn(net, sk, skb);
1014
1015         return ret;
1016 }
1017
1018 #ifdef CONFIG_SYSCTL
1019 static
1020 int brnf_sysctl_call_tables(struct ctl_table *ctl, int write,
1021                             void __user *buffer, size_t *lenp, loff_t *ppos)
1022 {
1023         int ret;
1024
1025         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1026
1027         if (write && *(int *)(ctl->data))
1028                 *(int *)(ctl->data) = 1;
1029         return ret;
1030 }
1031
1032 static struct ctl_table brnf_table[] = {
1033         {
1034                 .procname       = "bridge-nf-call-arptables",
1035                 .data           = &brnf_call_arptables,
1036                 .maxlen         = sizeof(int),
1037                 .mode           = 0644,
1038                 .proc_handler   = brnf_sysctl_call_tables,
1039         },
1040         {
1041                 .procname       = "bridge-nf-call-iptables",
1042                 .data           = &brnf_call_iptables,
1043                 .maxlen         = sizeof(int),
1044                 .mode           = 0644,
1045                 .proc_handler   = brnf_sysctl_call_tables,
1046         },
1047         {
1048                 .procname       = "bridge-nf-call-ip6tables",
1049                 .data           = &brnf_call_ip6tables,
1050                 .maxlen         = sizeof(int),
1051                 .mode           = 0644,
1052                 .proc_handler   = brnf_sysctl_call_tables,
1053         },
1054         {
1055                 .procname       = "bridge-nf-filter-vlan-tagged",
1056                 .data           = &brnf_filter_vlan_tagged,
1057                 .maxlen         = sizeof(int),
1058                 .mode           = 0644,
1059                 .proc_handler   = brnf_sysctl_call_tables,
1060         },
1061         {
1062                 .procname       = "bridge-nf-filter-pppoe-tagged",
1063                 .data           = &brnf_filter_pppoe_tagged,
1064                 .maxlen         = sizeof(int),
1065                 .mode           = 0644,
1066                 .proc_handler   = brnf_sysctl_call_tables,
1067         },
1068         {
1069                 .procname       = "bridge-nf-pass-vlan-input-dev",
1070                 .data           = &brnf_pass_vlan_indev,
1071                 .maxlen         = sizeof(int),
1072                 .mode           = 0644,
1073                 .proc_handler   = brnf_sysctl_call_tables,
1074         },
1075         { }
1076 };
1077 #endif
1078
1079 static int __init br_netfilter_init(void)
1080 {
1081         int ret;
1082
1083         ret = register_pernet_subsys(&brnf_net_ops);
1084         if (ret < 0)
1085                 return ret;
1086
1087         ret = register_netdevice_notifier(&brnf_notifier);
1088         if (ret < 0) {
1089                 unregister_pernet_subsys(&brnf_net_ops);
1090                 return ret;
1091         }
1092
1093 #ifdef CONFIG_SYSCTL
1094         brnf_sysctl_header = register_net_sysctl(&init_net, "net/bridge", brnf_table);
1095         if (brnf_sysctl_header == NULL) {
1096                 printk(KERN_WARNING
1097                        "br_netfilter: can't register to sysctl.\n");
1098                 unregister_netdevice_notifier(&brnf_notifier);
1099                 unregister_pernet_subsys(&brnf_net_ops);
1100                 return -ENOMEM;
1101         }
1102 #endif
1103         RCU_INIT_POINTER(nf_br_ops, &br_ops);
1104         printk(KERN_NOTICE "Bridge firewalling registered\n");
1105         return 0;
1106 }
1107
1108 static void __exit br_netfilter_fini(void)
1109 {
1110         RCU_INIT_POINTER(nf_br_ops, NULL);
1111         unregister_netdevice_notifier(&brnf_notifier);
1112         unregister_pernet_subsys(&brnf_net_ops);
1113 #ifdef CONFIG_SYSCTL
1114         unregister_net_sysctl_table(brnf_sysctl_header);
1115 #endif
1116 }
1117
1118 module_init(br_netfilter_init);
1119 module_exit(br_netfilter_fini);
1120
1121 MODULE_LICENSE("GPL");
1122 MODULE_AUTHOR("Lennert Buytenhek <buytenh@gnu.org>");
1123 MODULE_AUTHOR("Bart De Schuymer <bdschuym@pandora.be>");
1124 MODULE_DESCRIPTION("Linux ethernet netfilter firewall bridge");