net/bridge/br_netfilter_hooks.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Handle firewalling
   4  *      Linux ethernet bridge
   5  *
   6  *      Authors:
   7  *      Lennert Buytenhek               <buytenh@gnu.org>
   8  *      Bart De Schuymer                <bdschuym@pandora.be>
   9  *
  10  *      Lennert dedicates this file to Kerstin Wurdinger.
  11  */
  12
  13 #include <linux/module.h>
  14 #include <linux/kernel.h>
  15 #include <linux/slab.h>
  16 #include <linux/ip.h>
  17 #include <linux/netdevice.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/if_arp.h>
  20 #include <linux/if_ether.h>
  21 #include <linux/if_vlan.h>
  22 #include <linux/if_pppox.h>
  23 #include <linux/ppp_defs.h>
  24 #include <linux/netfilter_bridge.h>
  25 #include <uapi/linux/netfilter_bridge.h>
  26 #include <linux/netfilter_ipv4.h>
  27 #include <linux/netfilter_ipv6.h>
  28 #include <linux/netfilter_arp.h>
  29 #include <linux/in_route.h>
  30 #include <linux/rculist.h>
  31 #include <linux/inetdevice.h>
  32
  33 #include <net/ip.h>
  34 #include <net/ipv6.h>
  35 #include <net/addrconf.h>
  36 #include <net/route.h>
  37 #include <net/netfilter/br_netfilter.h>
  38 #include <net/netns/generic.h>
  39
  40 #include <linux/uaccess.h>
  41 #include "br_private.h"
  42 #ifdef CONFIG_SYSCTL
  43 #include <linux/sysctl.h>
  44 #endif
  45
  46 static unsigned int brnf_net_id __read_mostly;
  47
  48 struct brnf_net {
  49         bool enabled;
  50 };
  51
  52 #ifdef CONFIG_SYSCTL
  53 static struct ctl_table_header *brnf_sysctl_header;
  54 static int brnf_call_iptables __read_mostly = 1;
  55 static int brnf_call_ip6tables __read_mostly = 1;
  56 static int brnf_call_arptables __read_mostly = 1;
  57 static int brnf_filter_vlan_tagged __read_mostly;
  58 static int brnf_filter_pppoe_tagged __read_mostly;
  59 static int brnf_pass_vlan_indev __read_mostly;
  60 #else
  61 #define brnf_call_iptables 1
  62 #define brnf_call_ip6tables 1
  63 #define brnf_call_arptables 1
  64 #define brnf_filter_vlan_tagged 0
  65 #define brnf_filter_pppoe_tagged 0
  66 #define brnf_pass_vlan_indev 0
  67 #endif
  68
  69 #define IS_IP(skb) \
  70         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IP))
  71
  72 #define IS_IPV6(skb) \
  73         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IPV6))
  74
  75 #define IS_ARP(skb) \
  76         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_ARP))
  77
  78 static inline __be16 vlan_proto(const struct sk_buff *skb)
  79 {
  80         if (skb_vlan_tag_present(skb))
  81                 return skb->protocol;
  82         else if (skb->protocol == htons(ETH_P_8021Q))
  83                 return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  84         else
  85                 return 0;
  86 }
  87
  88 #define IS_VLAN_IP(skb) \
  89         (vlan_proto(skb) == htons(ETH_P_IP) && \
  90          brnf_filter_vlan_tagged)
  91
  92 #define IS_VLAN_IPV6(skb) \
  93         (vlan_proto(skb) == htons(ETH_P_IPV6) && \
  94          brnf_filter_vlan_tagged)
  95
  96 #define IS_VLAN_ARP(skb) \
  97         (vlan_proto(skb) == htons(ETH_P_ARP) && \
  98          brnf_filter_vlan_tagged)
  99
 100 static inline __be16 pppoe_proto(const struct sk_buff *skb)
 101 {
 102         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
 103                             sizeof(struct pppoe_hdr)));
 104 }
 105
 106 #define IS_PPPOE_IP(skb) \
 107         (skb->protocol == htons(ETH_P_PPP_SES) && \
 108          pppoe_proto(skb) == htons(PPP_IP) && \
 109          brnf_filter_pppoe_tagged)
 110
 111 #define IS_PPPOE_IPV6(skb) \
 112         (skb->protocol == htons(ETH_P_PPP_SES) && \
 113          pppoe_proto(skb) == htons(PPP_IPV6) && \
 114          brnf_filter_pppoe_tagged)
 115
 116 /* largest possible L2 header, see br_nf_dev_queue_xmit() */
 117 #define NF_BRIDGE_MAX_MAC_HEADER_LENGTH (PPPOE_SES_HLEN + ETH_HLEN)
 118
 119 struct brnf_frag_data {
 120         char mac[NF_BRIDGE_MAX_MAC_HEADER_LENGTH];
 121         u8 encap_size;
 122         u8 size;
 123         u16 vlan_tci;
 124         __be16 vlan_proto;
 125 };
 126
 127 static DEFINE_PER_CPU(struct brnf_frag_data, brnf_frag_data_storage);
 128
 129 static void nf_bridge_info_free(struct sk_buff *skb)
 130 {
 131         skb_ext_del(skb, SKB_EXT_BRIDGE_NF);
 132 }
 133
 134 static inline struct net_device *bridge_parent(const struct net_device *dev)
 135 {
 136         struct net_bridge_port *port;
 137
 138         port = br_port_get_rcu(dev);
 139         return port ? port->br->dev : NULL;
 140 }
 141
 142 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 143 {
 144         return skb_ext_add(skb, SKB_EXT_BRIDGE_NF);
 145 }
 146
 147 unsigned int nf_bridge_encap_header_len(const struct sk_buff *skb)
 148 {
 149         switch (skb->protocol) {
 150         case __cpu_to_be16(ETH_P_8021Q):
 151                 return VLAN_HLEN;
 152         case __cpu_to_be16(ETH_P_PPP_SES):
 153                 return PPPOE_SES_HLEN;
 154         default:
 155                 return 0;
 156         }
 157 }
 158
 159 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 160 {
 161         unsigned int len = nf_bridge_encap_header_len(skb);
 162
 163         skb_pull(skb, len);
 164         skb->network_header += len;
 165 }
 166
 167 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 168 {
 169         unsigned int len = nf_bridge_encap_header_len(skb);
 170
 171         skb_pull_rcsum(skb, len);
 172         skb->network_header += len;
 173 }
 174
 175 /* When handing a packet over to the IP layer
 176  * check whether we have a skb that is in the
 177  * expected format
 178  */
 179
 180 static int br_validate_ipv4(struct net *net, struct sk_buff *skb)
 181 {
 182         const struct iphdr *iph;
 183         u32 len;
 184
 185         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 186                 goto inhdr_error;
 187
 188         iph = ip_hdr(skb);
 189
 190         /* Basic sanity checks */
 191         if (iph->ihl < 5 || iph->version != 4)
 192                 goto inhdr_error;
 193
 194         if (!pskb_may_pull(skb, iph->ihl*4))
 195                 goto inhdr_error;
 196
 197         iph = ip_hdr(skb);
 198         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 199                 goto csum_error;
 200
 201         len = ntohs(iph->tot_len);
 202         if (skb->len < len) {
 203                 __IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS);
 204                 goto drop;
 205         } else if (len < (iph->ihl*4))
 206                 goto inhdr_error;
 207
 208         if (pskb_trim_rcsum(skb, len)) {
 209                 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
 210                 goto drop;
 211         }
 212
 213         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 214         /* We should really parse IP options here but until
 215          * somebody who actually uses IP options complains to
 216          * us we'll just silently ignore the options because
 217          * we're lazy!
 218          */
 219         return 0;
 220
 221 csum_error:
 222         __IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS);
 223 inhdr_error:
 224         __IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
 225 drop:
 226         return -1;
 227 }
 228
 229 void nf_bridge_update_protocol(struct sk_buff *skb)
 230 {
 231         const struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 232
 233         switch (nf_bridge->orig_proto) {
 234         case BRNF_PROTO_8021Q:
 235                 skb->protocol = htons(ETH_P_8021Q);
 236                 break;
 237         case BRNF_PROTO_PPPOE:
 238                 skb->protocol = htons(ETH_P_PPP_SES);
 239                 break;
 240         case BRNF_PROTO_UNCHANGED:
 241                 break;
 242         }
 243 }
 244
 245 /* Obtain the correct destination MAC address, while preserving the original
 246  * source MAC address. If we already know this address, we just copy it. If we
 247  * don't, we use the neighbour framework to find out. In both cases, we make
 248  * sure that br_handle_frame_finish() is called afterwards.
 249  */
 250 int br_nf_pre_routing_finish_bridge(struct net *net, struct sock *sk, struct sk_buff *skb)
 251 {
 252         struct neighbour *neigh;
 253         struct dst_entry *dst;
 254
 255         skb->dev = bridge_parent(skb->dev);
 256         if (!skb->dev)
 257                 goto free_skb;
 258         dst = skb_dst(skb);
 259         neigh = dst_neigh_lookup_skb(dst, skb);
 260         if (neigh) {
 261                 struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 262                 int ret;
 263
 264                 if ((neigh->nud_state & NUD_CONNECTED) && neigh->hh.hh_len) {
 265                         neigh_hh_bridge(&neigh->hh, skb);
 266                         skb->dev = nf_bridge->physindev;
 267                         ret = br_handle_frame_finish(net, sk, skb);
 268                 } else {
 269                         /* the neighbour function below overwrites the complete
 270                          * MAC header, so we save the Ethernet source address and
 271                          * protocol number.
 272                          */
 273                         skb_copy_from_linear_data_offset(skb,
 274                                                          -(ETH_HLEN-ETH_ALEN),
 275                                                          nf_bridge->neigh_header,
 276                                                          ETH_HLEN-ETH_ALEN);
 277                         /* tell br_dev_xmit to continue with forwarding */
 278                         nf_bridge->bridged_dnat = 1;
 279                         /* FIXME Need to refragment */
 280                         ret = neigh->output(neigh, skb);
 281                 }
 282                 neigh_release(neigh);
 283                 return ret;
 284         }
 285 free_skb:
 286         kfree_skb(skb);
 287         return 0;
 288 }
 289
 290 static inline bool
 291 br_nf_ipv4_daddr_was_changed(const struct sk_buff *skb,
 292                              const struct nf_bridge_info *nf_bridge)
 293 {
 294         return ip_hdr(skb)->daddr != nf_bridge->ipv4_daddr;
 295 }
 296
 297 /* This requires some explaining. If DNAT has taken place,
 298  * we will need to fix up the destination Ethernet address.
 299  * This is also true when SNAT takes place (for the reply direction).
 300  *
 301  * There are two cases to consider:
 302  * 1. The packet was DNAT'ed to a device in the same bridge
 303  *    port group as it was received on. We can still bridge
 304  *    the packet.
 305  * 2. The packet was DNAT'ed to a different device, either
 306  *    a non-bridged device or another bridge port group.
 307  *    The packet will need to be routed.
 308  *
 309  * The correct way of distinguishing between these two cases is to
 310  * call ip_route_input() and to look at skb->dst->dev, which is
 311  * changed to the destination device if ip_route_input() succeeds.
 312  *
 313  * Let's first consider the case that ip_route_input() succeeds:
 314  *
 315  * If the output device equals the logical bridge device the packet
 316  * came in on, we can consider this bridging. The corresponding MAC
 317  * address will be obtained in br_nf_pre_routing_finish_bridge.
 318  * Otherwise, the packet is considered to be routed and we just
 319  * change the destination MAC address so that the packet will
 320  * later be passed up to the IP stack to be routed. For a redirected
 321  * packet, ip_route_input() will give back the localhost as output device,
 322  * which differs from the bridge device.
 323  *
 324  * Let's now consider the case that ip_route_input() fails:
 325  *
 326  * This can be because the destination address is martian, in which case
 327  * the packet will be dropped.
 328  * If IP forwarding is disabled, ip_route_input() will fail, while
 329  * ip_route_output_key() can return success. The source
 330  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 331  * thinks we're handling a locally generated packet and won't care
 332  * if IP forwarding is enabled. If the output device equals the logical bridge
 333  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 334  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 335  */
 336 static int br_nf_pre_routing_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
 337 {
 338         struct net_device *dev = skb->dev;
 339         struct iphdr *iph = ip_hdr(skb);
 340         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 341         struct rtable *rt;
 342         int err;
 343
 344         nf_bridge->frag_max_size = IPCB(skb)->frag_max_size;
 345
 346         if (nf_bridge->pkt_otherhost) {
 347                 skb->pkt_type = PACKET_OTHERHOST;
 348                 nf_bridge->pkt_otherhost = false;
 349         }
 350         nf_bridge->in_prerouting = 0;
 351         if (br_nf_ipv4_daddr_was_changed(skb, nf_bridge)) {
 352                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 353                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 354
 355                         /* If err equals -EHOSTUNREACH the error is due to a
 356                          * martian destination or due to the fact that
 357                          * forwarding is disabled. For most martian packets,
 358                          * ip_route_output_key() will fail. It won't fail for 2 types of
 359                          * martian destinations: loopback destinations and destination
 360                          * 0.0.0.0. In both cases the packet will be dropped because the
 361                          * destination is the loopback device and not the bridge. */
 362                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 363                                 goto free_skb;
 364
 365                         rt = ip_route_output(net, iph->daddr, 0,
 366                                              RT_TOS(iph->tos), 0);
 367                         if (!IS_ERR(rt)) {
 368                                 /* - Bridged-and-DNAT'ed traffic doesn't
 369                                  *   require ip_forwarding. */
 370                                 if (rt->dst.dev == dev) {
 371                                         skb_dst_set(skb, &rt->dst);
 372                                         goto bridged_dnat;
 373                                 }
 374                                 ip_rt_put(rt);
 375                         }
 376 free_skb:
 377                         kfree_skb(skb);
 378                         return 0;
 379                 } else {
 380                         if (skb_dst(skb)->dev == dev) {
 381 bridged_dnat:
 382                                 skb->dev = nf_bridge->physindev;
 383                                 nf_bridge_update_protocol(skb);
 384                                 nf_bridge_push_encap_header(skb);
 385                                 br_nf_hook_thresh(NF_BR_PRE_ROUTING,
 386                                                   net, sk, skb, skb->dev,
 387                                                   NULL,
 388                                                   br_nf_pre_routing_finish_bridge);
 389                                 return 0;
 390                         }
 391                         ether_addr_copy(eth_hdr(skb)->h_dest, dev->dev_addr);
 392                         skb->pkt_type = PACKET_HOST;
 393                 }
 394         } else {
 395                 rt = bridge_parent_rtable(nf_bridge->physindev);
 396                 if (!rt) {
 397                         kfree_skb(skb);
 398                         return 0;
 399                 }
 400                 skb_dst_set_noref(skb, &rt->dst);
 401         }
 402
 403         skb->dev = nf_bridge->physindev;
 404         nf_bridge_update_protocol(skb);
 405         nf_bridge_push_encap_header(skb);
 406         br_nf_hook_thresh(NF_BR_PRE_ROUTING, net, sk, skb, skb->dev, NULL,
 407                           br_handle_frame_finish);
 408         return 0;
 409 }
 410
 411 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb, const struct net_device *dev)
 412 {
 413         struct net_device *vlan, *br;
 414
 415         br = bridge_parent(dev);
 416         if (brnf_pass_vlan_indev == 0 || !skb_vlan_tag_present(skb))
 417                 return br;
 418
 419         vlan = __vlan_find_dev_deep_rcu(br, skb->vlan_proto,
 420                                     skb_vlan_tag_get(skb) & VLAN_VID_MASK);
 421
 422         return vlan ? vlan : br;
 423 }
 424
 425 /* Some common code for IPv4/IPv6 */
 426 struct net_device *setup_pre_routing(struct sk_buff *skb)
 427 {
 428         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 429
 430         if (skb->pkt_type == PACKET_OTHERHOST) {
 431                 skb->pkt_type = PACKET_HOST;
 432                 nf_bridge->pkt_otherhost = true;
 433         }
 434
 435         nf_bridge->in_prerouting = 1;
 436         nf_bridge->physindev = skb->dev;
 437         skb->dev = brnf_get_logical_dev(skb, skb->dev);
 438
 439         if (skb->protocol == htons(ETH_P_8021Q))
 440                 nf_bridge->orig_proto = BRNF_PROTO_8021Q;
 441         else if (skb->protocol == htons(ETH_P_PPP_SES))
 442                 nf_bridge->orig_proto = BRNF_PROTO_PPPOE;
 443
 444         /* Must drop socket now because of tproxy. */
 445         skb_orphan(skb);
 446         return skb->dev;
 447 }
 448
 449 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 450  * Replicate the checks that IPv4 does on packet reception.
 451  * Set skb->dev to the bridge device (i.e. parent of the
 452  * receiving device) to make netfilter happy, the REDIRECT
 453  * target in particular.  Save the original destination IP
 454  * address to be able to detect DNAT afterwards. */
 455 static unsigned int br_nf_pre_routing(void *priv,
 456                                       struct sk_buff *skb,
 457                                       const struct nf_hook_state *state)
 458 {
 459         struct nf_bridge_info *nf_bridge;
 460         struct net_bridge_port *p;
 461         struct net_bridge *br;
 462         __u32 len = nf_bridge_encap_header_len(skb);
 463
 464         if (unlikely(!pskb_may_pull(skb, len)))
 465                 return NF_DROP;
 466
 467         p = br_port_get_rcu(state->in);
 468         if (p == NULL)
 469                 return NF_DROP;
 470         br = p->br;
 471
 472         if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb)) {
 473                 if (!brnf_call_ip6tables &&
 474                     !br_opt_get(br, BROPT_NF_CALL_IP6TABLES))
 475                         return NF_ACCEPT;
 476
 477                 nf_bridge_pull_encap_header_rcsum(skb);
 478                 return br_nf_pre_routing_ipv6(priv, skb, state);
 479         }
 480
 481         if (!brnf_call_iptables && !br_opt_get(br, BROPT_NF_CALL_IPTABLES))
 482                 return NF_ACCEPT;
 483
 484         if (!IS_IP(skb) && !IS_VLAN_IP(skb) && !IS_PPPOE_IP(skb))
 485                 return NF_ACCEPT;
 486
 487         nf_bridge_pull_encap_header_rcsum(skb);
 488
 489         if (br_validate_ipv4(state->net, skb))
 490                 return NF_DROP;
 491
 492         if (!nf_bridge_alloc(skb))
 493                 return NF_DROP;
 494         if (!setup_pre_routing(skb))
 495                 return NF_DROP;
 496
 497         nf_bridge = nf_bridge_info_get(skb);
 498         nf_bridge->ipv4_daddr = ip_hdr(skb)->daddr;
 499
 500         skb->protocol = htons(ETH_P_IP);
 501         skb->transport_header = skb->network_header + ip_hdr(skb)->ihl * 4;
 502
 503         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, state->net, state->sk, skb,
 504                 skb->dev, NULL,
 505                 br_nf_pre_routing_finish);
 506
 507         return NF_STOLEN;
 508 }
 509
 510
 511 /* PF_BRIDGE/FORWARD *************************************************/
 512 static int br_nf_forward_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
 513 {
 514         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 515         struct net_device *in;
 516
 517         if (!IS_ARP(skb) && !IS_VLAN_ARP(skb)) {
 518
 519                 if (skb->protocol == htons(ETH_P_IP))
 520                         nf_bridge->frag_max_size = IPCB(skb)->frag_max_size;
 521
 522                 if (skb->protocol == htons(ETH_P_IPV6))
 523                         nf_bridge->frag_max_size = IP6CB(skb)->frag_max_size;
 524
 525                 in = nf_bridge->physindev;
 526                 if (nf_bridge->pkt_otherhost) {
 527                         skb->pkt_type = PACKET_OTHERHOST;
 528                         nf_bridge->pkt_otherhost = false;
 529                 }
 530                 nf_bridge_update_protocol(skb);
 531         } else {
 532                 in = *((struct net_device **)(skb->cb));
 533         }
 534         nf_bridge_push_encap_header(skb);
 535
 536         br_nf_hook_thresh(NF_BR_FORWARD, net, sk, skb, in, skb->dev,
 537                           br_forward_finish);
 538         return 0;
 539 }
 540
 541
 542 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 543  * netfilter with indev and outdev set to the bridge device,
 544  * but we are still able to filter on the 'real' indev/outdev
 545  * because of the physdev module. For ARP, indev and outdev are the
 546  * bridge ports. */
 547 static unsigned int br_nf_forward_ip(void *priv,
 548                                      struct sk_buff *skb,
 549                                      const struct nf_hook_state *state)
 550 {
 551         struct nf_bridge_info *nf_bridge;
 552         struct net_device *parent;
 553         u_int8_t pf;
 554
 555         nf_bridge = nf_bridge_info_get(skb);
 556         if (!nf_bridge)
 557                 return NF_ACCEPT;
 558
 559         /* Need exclusive nf_bridge_info since we might have multiple
 560          * different physoutdevs. */
 561         if (!nf_bridge_unshare(skb))
 562                 return NF_DROP;
 563
 564         nf_bridge = nf_bridge_info_get(skb);
 565         if (!nf_bridge)
 566                 return NF_DROP;
 567
 568         parent = bridge_parent(state->out);
 569         if (!parent)
 570                 return NF_DROP;
 571
 572         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 573                 pf = NFPROTO_IPV4;
 574         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 575                 pf = NFPROTO_IPV6;
 576         else
 577                 return NF_ACCEPT;
 578
 579         nf_bridge_pull_encap_header(skb);
 580
 581         if (skb->pkt_type == PACKET_OTHERHOST) {
 582                 skb->pkt_type = PACKET_HOST;
 583                 nf_bridge->pkt_otherhost = true;
 584         }
 585
 586         if (pf == NFPROTO_IPV4) {
 587                 if (br_validate_ipv4(state->net, skb))
 588                         return NF_DROP;
 589                 IPCB(skb)->frag_max_size = nf_bridge->frag_max_size;
 590         }
 591
 592         if (pf == NFPROTO_IPV6) {
 593                 if (br_validate_ipv6(state->net, skb))
 594                         return NF_DROP;
 595                 IP6CB(skb)->frag_max_size = nf_bridge->frag_max_size;
 596         }
 597
 598         nf_bridge->physoutdev = skb->dev;
 599         if (pf == NFPROTO_IPV4)
 600                 skb->protocol = htons(ETH_P_IP);
 601         else
 602                 skb->protocol = htons(ETH_P_IPV6);
 603
 604         NF_HOOK(pf, NF_INET_FORWARD, state->net, NULL, skb,
 605                 brnf_get_logical_dev(skb, state->in),
 606                 parent, br_nf_forward_finish);
 607
 608         return NF_STOLEN;
 609 }
 610
 611 static unsigned int br_nf_forward_arp(void *priv,
 612                                       struct sk_buff *skb,
 613                                       const struct nf_hook_state *state)
 614 {
 615         struct net_bridge_port *p;
 616         struct net_bridge *br;
 617         struct net_device **d = (struct net_device **)(skb->cb);
 618
 619         p = br_port_get_rcu(state->out);
 620         if (p == NULL)
 621                 return NF_ACCEPT;
 622         br = p->br;
 623
 624         if (!brnf_call_arptables && !br_opt_get(br, BROPT_NF_CALL_ARPTABLES))
 625                 return NF_ACCEPT;
 626
 627         if (!IS_ARP(skb)) {
 628                 if (!IS_VLAN_ARP(skb))
 629                         return NF_ACCEPT;
 630                 nf_bridge_pull_encap_header(skb);
 631         }
 632
 633         if (arp_hdr(skb)->ar_pln != 4) {
 634                 if (IS_VLAN_ARP(skb))
 635                         nf_bridge_push_encap_header(skb);
 636                 return NF_ACCEPT;
 637         }
 638         *d = state->in;
 639         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, state->net, state->sk, skb,
 640                 state->in, state->out, br_nf_forward_finish);
 641
 642         return NF_STOLEN;
 643 }
 644
 645 static int br_nf_push_frag_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
 646 {
 647         struct brnf_frag_data *data;
 648         int err;
 649
 650         data = this_cpu_ptr(&brnf_frag_data_storage);
 651         err = skb_cow_head(skb, data->size);
 652
 653         if (err) {
 654                 kfree_skb(skb);
 655                 return 0;
 656         }
 657
 658         if (data->vlan_proto)
 659                 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci);
 660
 661         skb_copy_to_linear_data_offset(skb, -data->size, data->mac, data->size);
 662         __skb_push(skb, data->encap_size);
 663
 664         nf_bridge_info_free(skb);
 665         return br_dev_queue_push_xmit(net, sk, skb);
 666 }
 667
 668 static int
 669 br_nf_ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
 670                   int (*output)(struct net *, struct sock *, struct sk_buff *))
 671 {
 672         unsigned int mtu = ip_skb_dst_mtu(sk, skb);
 673         struct iphdr *iph = ip_hdr(skb);
 674
 675         if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->ignore_df) ||
 676                      (IPCB(skb)->frag_max_size &&
 677                       IPCB(skb)->frag_max_size > mtu))) {
 678                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
 679                 kfree_skb(skb);
 680                 return -EMSGSIZE;
 681         }
 682
 683         return ip_do_fragment(net, sk, skb, output);
 684 }
 685
 686 static unsigned int nf_bridge_mtu_reduction(const struct sk_buff *skb)
 687 {
 688         const struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 689
 690         if (nf_bridge->orig_proto == BRNF_PROTO_PPPOE)
 691                 return PPPOE_SES_HLEN;
 692         return 0;
 693 }
 694
 695 static int br_nf_dev_queue_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
 696 {
 697         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 698         unsigned int mtu, mtu_reserved;
 699
 700         mtu_reserved = nf_bridge_mtu_reduction(skb);
 701         mtu = skb->dev->mtu;
 702
 703         if (nf_bridge->frag_max_size && nf_bridge->frag_max_size < mtu)
 704                 mtu = nf_bridge->frag_max_size;
 705
 706         if (skb_is_gso(skb) || skb->len + mtu_reserved <= mtu) {
 707                 nf_bridge_info_free(skb);
 708                 return br_dev_queue_push_xmit(net, sk, skb);
 709         }
 710
 711         /* This is wrong! We should preserve the original fragment
 712          * boundaries by preserving frag_list rather than refragmenting.
 713          */
 714         if (IS_ENABLED(CONFIG_NF_DEFRAG_IPV4) &&
 715             skb->protocol == htons(ETH_P_IP)) {
 716                 struct brnf_frag_data *data;
 717
 718                 if (br_validate_ipv4(net, skb))
 719                         goto drop;
 720
 721                 IPCB(skb)->frag_max_size = nf_bridge->frag_max_size;
 722
 723                 nf_bridge_update_protocol(skb);
 724
 725                 data = this_cpu_ptr(&brnf_frag_data_storage);
 726
 727                 if (skb_vlan_tag_present(skb)) {
 728                         data->vlan_tci = skb->vlan_tci;
 729                         data->vlan_proto = skb->vlan_proto;
 730                 } else {
 731                         data->vlan_proto = 0;
 732                 }
 733
 734                 data->encap_size = nf_bridge_encap_header_len(skb);
 735                 data->size = ETH_HLEN + data->encap_size;
 736
 737                 skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
 738                                                  data->size);
 739
 740                 return br_nf_ip_fragment(net, sk, skb, br_nf_push_frag_xmit);
 741         }
 742         if (IS_ENABLED(CONFIG_NF_DEFRAG_IPV6) &&
 743             skb->protocol == htons(ETH_P_IPV6)) {
 744                 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
 745                 struct brnf_frag_data *data;
 746
 747                 if (br_validate_ipv6(net, skb))
 748                         goto drop;
 749
 750                 IP6CB(skb)->frag_max_size = nf_bridge->frag_max_size;
 751
 752                 nf_bridge_update_protocol(skb);
 753
 754                 data = this_cpu_ptr(&brnf_frag_data_storage);
 755                 data->encap_size = nf_bridge_encap_header_len(skb);
 756                 data->size = ETH_HLEN + data->encap_size;
 757
 758                 skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
 759                                                  data->size);
 760
 761                 if (v6ops)
 762                         return v6ops->fragment(net, sk, skb, br_nf_push_frag_xmit);
 763
 764                 kfree_skb(skb);
 765                 return -EMSGSIZE;
 766         }
 767         nf_bridge_info_free(skb);
 768         return br_dev_queue_push_xmit(net, sk, skb);
 769  drop:
 770         kfree_skb(skb);
 771         return 0;
 772 }
 773
 774 /* PF_BRIDGE/POST_ROUTING ********************************************/
 775 static unsigned int br_nf_post_routing(void *priv,
 776                                        struct sk_buff *skb,
 777                                        const struct nf_hook_state *state)
 778 {
 779         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 780         struct net_device *realoutdev = bridge_parent(skb->dev);
 781         u_int8_t pf;
 782
 783         /* if nf_bridge is set, but ->physoutdev is NULL, this packet came in
 784          * on a bridge, but was delivered locally and is now being routed:
 785          *
 786          * POST_ROUTING was already invoked from the ip stack.
 787          */
 788         if (!nf_bridge || !nf_bridge->physoutdev)
 789                 return NF_ACCEPT;
 790
 791         if (!realoutdev)
 792                 return NF_DROP;
 793
 794         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 795                 pf = NFPROTO_IPV4;
 796         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 797                 pf = NFPROTO_IPV6;
 798         else
 799                 return NF_ACCEPT;
 800
 801         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 802          * about the value of skb->pkt_type. */
 803         if (skb->pkt_type == PACKET_OTHERHOST) {
 804                 skb->pkt_type = PACKET_HOST;
 805                 nf_bridge->pkt_otherhost = true;
 806         }
 807
 808         nf_bridge_pull_encap_header(skb);
 809         if (pf == NFPROTO_IPV4)
 810                 skb->protocol = htons(ETH_P_IP);
 811         else
 812                 skb->protocol = htons(ETH_P_IPV6);
 813
 814         NF_HOOK(pf, NF_INET_POST_ROUTING, state->net, state->sk, skb,
 815                 NULL, realoutdev,
 816                 br_nf_dev_queue_xmit);
 817
 818         return NF_STOLEN;
 819 }
 820
 821 /* IP/SABOTAGE *****************************************************/
 822 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 823  * for the second time. */
 824 static unsigned int ip_sabotage_in(void *priv,
 825                                    struct sk_buff *skb,
 826                                    const struct nf_hook_state *state)
 827 {
 828         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 829
 830         if (nf_bridge && !nf_bridge->in_prerouting &&
 831             !netif_is_l3_master(skb->dev) &&
 832             !netif_is_l3_slave(skb->dev)) {
 833                 state->okfn(state->net, state->sk, skb);
 834                 return NF_STOLEN;
 835         }
 836
 837         return NF_ACCEPT;
 838 }
 839
 840 /* This is called when br_netfilter has called into iptables/netfilter,
 841  * and DNAT has taken place on a bridge-forwarded packet.
 842  *
 843  * neigh->output has created a new MAC header, with local br0 MAC
 844  * as saddr.
 845  *
 846  * This restores the original MAC saddr of the bridged packet
 847  * before invoking bridge forward logic to transmit the packet.
 848  */
 849 static void br_nf_pre_routing_finish_bridge_slow(struct sk_buff *skb)
 850 {
 851         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 852
 853         skb_pull(skb, ETH_HLEN);
 854         nf_bridge->bridged_dnat = 0;
 855
 856         BUILD_BUG_ON(sizeof(nf_bridge->neigh_header) != (ETH_HLEN - ETH_ALEN));
 857
 858         skb_copy_to_linear_data_offset(skb, -(ETH_HLEN - ETH_ALEN),
 859                                        nf_bridge->neigh_header,
 860                                        ETH_HLEN - ETH_ALEN);
 861         skb->dev = nf_bridge->physindev;
 862
 863         nf_bridge->physoutdev = NULL;
 864         br_handle_frame_finish(dev_net(skb->dev), NULL, skb);
 865 }
 866
 867 static int br_nf_dev_xmit(struct sk_buff *skb)
 868 {
 869         const struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 870
 871         if (nf_bridge && nf_bridge->bridged_dnat) {
 872                 br_nf_pre_routing_finish_bridge_slow(skb);
 873                 return 1;
 874         }
 875         return 0;
 876 }
 877
 878 static const struct nf_br_ops br_ops = {
 879         .br_dev_xmit_hook =     br_nf_dev_xmit,
 880 };
 881
 882 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 883  * br_dev_queue_push_xmit is called afterwards */
 884 static const struct nf_hook_ops br_nf_ops[] = {
 885         {
 886                 .hook = br_nf_pre_routing,
 887                 .pf = NFPROTO_BRIDGE,
 888                 .hooknum = NF_BR_PRE_ROUTING,
 889                 .priority = NF_BR_PRI_BRNF,
 890         },
 891         {
 892                 .hook = br_nf_forward_ip,
 893                 .pf = NFPROTO_BRIDGE,
 894                 .hooknum = NF_BR_FORWARD,
 895                 .priority = NF_BR_PRI_BRNF - 1,
 896         },
 897         {
 898                 .hook = br_nf_forward_arp,
 899                 .pf = NFPROTO_BRIDGE,
 900                 .hooknum = NF_BR_FORWARD,
 901                 .priority = NF_BR_PRI_BRNF,
 902         },
 903         {
 904                 .hook = br_nf_post_routing,
 905                 .pf = NFPROTO_BRIDGE,
 906                 .hooknum = NF_BR_POST_ROUTING,
 907                 .priority = NF_BR_PRI_LAST,
 908         },
 909         {
 910                 .hook = ip_sabotage_in,
 911                 .pf = NFPROTO_IPV4,
 912                 .hooknum = NF_INET_PRE_ROUTING,
 913                 .priority = NF_IP_PRI_FIRST,
 914         },
 915         {
 916                 .hook = ip_sabotage_in,
 917                 .pf = NFPROTO_IPV6,
 918                 .hooknum = NF_INET_PRE_ROUTING,
 919                 .priority = NF_IP6_PRI_FIRST,
 920         },
 921 };
 922
 923 static int brnf_device_event(struct notifier_block *unused, unsigned long event,
 924                              void *ptr)
 925 {
 926         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 927         struct brnf_net *brnet;
 928         struct net *net;
 929         int ret;
 930
 931         if (event != NETDEV_REGISTER || !(dev->priv_flags & IFF_EBRIDGE))
 932                 return NOTIFY_DONE;
 933
 934         ASSERT_RTNL();
 935
 936         net = dev_net(dev);
 937         brnet = net_generic(net, brnf_net_id);
 938         if (brnet->enabled)
 939                 return NOTIFY_OK;
 940
 941         ret = nf_register_net_hooks(net, br_nf_ops, ARRAY_SIZE(br_nf_ops));
 942         if (ret)
 943                 return NOTIFY_BAD;
 944
 945         brnet->enabled = true;
 946         return NOTIFY_OK;
 947 }
 948
 949 static void __net_exit brnf_exit_net(struct net *net)
 950 {
 951         struct brnf_net *brnet = net_generic(net, brnf_net_id);
 952
 953         if (!brnet->enabled)
 954                 return;
 955
 956         nf_unregister_net_hooks(net, br_nf_ops, ARRAY_SIZE(br_nf_ops));
 957         brnet->enabled = false;
 958 }
 959
 960 static struct pernet_operations brnf_net_ops __read_mostly = {
 961         .exit = brnf_exit_net,
 962         .id   = &brnf_net_id,
 963         .size = sizeof(struct brnf_net),
 964 };
 965
 966 static struct notifier_block brnf_notifier __read_mostly = {
 967         .notifier_call = brnf_device_event,
 968 };
 969
 970 /* recursively invokes nf_hook_slow (again), skipping already-called
 971  * hooks (< NF_BR_PRI_BRNF).
 972  *
 973  * Called with rcu read lock held.
 974  */
 975 int br_nf_hook_thresh(unsigned int hook, struct net *net,
 976                       struct sock *sk, struct sk_buff *skb,
 977                       struct net_device *indev,
 978                       struct net_device *outdev,
 979                       int (*okfn)(struct net *, struct sock *,
 980                                   struct sk_buff *))
 981 {
 982         const struct nf_hook_entries *e;
 983         struct nf_hook_state state;
 984         struct nf_hook_ops **ops;
 985         unsigned int i;
 986         int ret;
 987
 988         e = rcu_dereference(net->nf.hooks_bridge[hook]);
 989         if (!e)
 990                 return okfn(net, sk, skb);
 991
 992         ops = nf_hook_entries_get_hook_ops(e);
 993         for (i = 0; i < e->num_hook_entries &&
 994               ops[i]->priority <= NF_BR_PRI_BRNF; i++)
 995                 ;
 996
 997         nf_hook_state_init(&state, hook, NFPROTO_BRIDGE, indev, outdev,
 998                            sk, net, okfn);
 999
1000         ret = nf_hook_slow(skb, &state, e, i);
1001         if (ret == 1)
1002                 ret = okfn(net, sk, skb);
1003
1004         return ret;
1005 }
1006
1007 #ifdef CONFIG_SYSCTL
1008 static
1009 int brnf_sysctl_call_tables(struct ctl_table *ctl, int write,
1010                             void __user *buffer, size_t *lenp, loff_t *ppos)
1011 {
1012         int ret;
1013
1014         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1015
1016         if (write && *(int *)(ctl->data))
1017                 *(int *)(ctl->data) = 1;
1018         return ret;
1019 }
1020
1021 static struct ctl_table brnf_table[] = {
1022         {
1023                 .procname       = "bridge-nf-call-arptables",
1024                 .data           = &brnf_call_arptables,
1025                 .maxlen         = sizeof(int),
1026                 .mode           = 0644,
1027                 .proc_handler   = brnf_sysctl_call_tables,
1028         },
1029         {
1030                 .procname       = "bridge-nf-call-iptables",
1031                 .data           = &brnf_call_iptables,
1032                 .maxlen         = sizeof(int),
1033                 .mode           = 0644,
1034                 .proc_handler   = brnf_sysctl_call_tables,
1035         },
1036         {
1037                 .procname       = "bridge-nf-call-ip6tables",
1038                 .data           = &brnf_call_ip6tables,
1039                 .maxlen         = sizeof(int),
1040                 .mode           = 0644,
1041                 .proc_handler   = brnf_sysctl_call_tables,
1042         },
1043         {
1044                 .procname       = "bridge-nf-filter-vlan-tagged",
1045                 .data           = &brnf_filter_vlan_tagged,
1046                 .maxlen         = sizeof(int),
1047                 .mode           = 0644,
1048                 .proc_handler   = brnf_sysctl_call_tables,
1049         },
1050         {
1051                 .procname       = "bridge-nf-filter-pppoe-tagged",
1052                 .data           = &brnf_filter_pppoe_tagged,
1053                 .maxlen         = sizeof(int),
1054                 .mode           = 0644,
1055                 .proc_handler   = brnf_sysctl_call_tables,
1056         },
1057         {
1058                 .procname       = "bridge-nf-pass-vlan-input-dev",
1059                 .data           = &brnf_pass_vlan_indev,
1060                 .maxlen         = sizeof(int),
1061                 .mode           = 0644,
1062                 .proc_handler   = brnf_sysctl_call_tables,
1063         },
1064         { }
1065 };
1066 #endif
1067
1068 static int __init br_netfilter_init(void)
1069 {
1070         int ret;
1071
1072         ret = register_pernet_subsys(&brnf_net_ops);
1073         if (ret < 0)
1074                 return ret;
1075
1076         ret = register_netdevice_notifier(&brnf_notifier);
1077         if (ret < 0) {
1078                 unregister_pernet_subsys(&brnf_net_ops);
1079                 return ret;
1080         }
1081
1082 #ifdef CONFIG_SYSCTL
1083         brnf_sysctl_header = register_net_sysctl(&init_net, "net/bridge", brnf_table);
1084         if (brnf_sysctl_header == NULL) {
1085                 printk(KERN_WARNING
1086                        "br_netfilter: can't register to sysctl.\n");
1087                 unregister_netdevice_notifier(&brnf_notifier);
1088                 unregister_pernet_subsys(&brnf_net_ops);
1089                 return -ENOMEM;
1090         }
1091 #endif
1092         RCU_INIT_POINTER(nf_br_ops, &br_ops);
1093         printk(KERN_NOTICE "Bridge firewalling registered\n");
1094         return 0;
1095 }
1096
1097 static void __exit br_netfilter_fini(void)
1098 {
1099         RCU_INIT_POINTER(nf_br_ops, NULL);
1100         unregister_netdevice_notifier(&brnf_notifier);
1101         unregister_pernet_subsys(&brnf_net_ops);
1102 #ifdef CONFIG_SYSCTL
1103         unregister_net_sysctl_table(brnf_sysctl_header);
1104 #endif
1105 }
1106
1107 module_init(br_netfilter_init);
1108 module_exit(br_netfilter_fini);
1109
1110 MODULE_LICENSE("GPL");
1111 MODULE_AUTHOR("Lennert Buytenhek <buytenh@gnu.org>");
1112 MODULE_AUTHOR("Bart De Schuymer <bdschuym@pandora.be>");
1113 MODULE_DESCRIPTION("Linux ethernet netfilter firewall bridge");