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 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
  47 #include <net/netfilter/nf_conntrack_core.h>
  48 #endif
  49
  50 static unsigned int brnf_net_id __read_mostly;
  51
  52 struct brnf_net {
  53         bool enabled;
  54
  55 #ifdef CONFIG_SYSCTL
  56         struct ctl_table_header *ctl_hdr;
  57 #endif
  58
  59         /* default value is 1 */
  60         int call_iptables;
  61         int call_ip6tables;
  62         int call_arptables;
  63
  64         /* default value is 0 */
  65         int filter_vlan_tagged;
  66         int filter_pppoe_tagged;
  67         int pass_vlan_indev;
  68 };
  69
  70 #define IS_IP(skb) \
  71         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IP))
  72
  73 #define IS_IPV6(skb) \
  74         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IPV6))
  75
  76 #define IS_ARP(skb) \
  77         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_ARP))
  78
  79 static inline __be16 vlan_proto(const struct sk_buff *skb)
  80 {
  81         if (skb_vlan_tag_present(skb))
  82                 return skb->protocol;
  83         else if (skb->protocol == htons(ETH_P_8021Q))
  84                 return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  85         else
  86                 return 0;
  87 }
  88
  89 static inline bool is_vlan_ip(const struct sk_buff *skb, const struct net *net)
  90 {
  91         struct brnf_net *brnet = net_generic(net, brnf_net_id);
  92
  93         return vlan_proto(skb) == htons(ETH_P_IP) && brnet->filter_vlan_tagged;
  94 }
  95
  96 static inline bool is_vlan_ipv6(const struct sk_buff *skb,
  97                                 const struct net *net)
  98 {
  99         struct brnf_net *brnet = net_generic(net, brnf_net_id);
 100
 101         return vlan_proto(skb) == htons(ETH_P_IPV6) &&
 102                brnet->filter_vlan_tagged;
 103 }
 104
 105 static inline bool is_vlan_arp(const struct sk_buff *skb, const struct net *net)
 106 {
 107         struct brnf_net *brnet = net_generic(net, brnf_net_id);
 108
 109         return vlan_proto(skb) == htons(ETH_P_ARP) && brnet->filter_vlan_tagged;
 110 }
 111
 112 static inline __be16 pppoe_proto(const struct sk_buff *skb)
 113 {
 114         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
 115                             sizeof(struct pppoe_hdr)));
 116 }
 117
 118 static inline bool is_pppoe_ip(const struct sk_buff *skb, const struct net *net)
 119 {
 120         struct brnf_net *brnet = net_generic(net, brnf_net_id);
 121
 122         return skb->protocol == htons(ETH_P_PPP_SES) &&
 123                pppoe_proto(skb) == htons(PPP_IP) && brnet->filter_pppoe_tagged;
 124 }
 125
 126 static inline bool is_pppoe_ipv6(const struct sk_buff *skb,
 127                                  const struct net *net)
 128 {
 129         struct brnf_net *brnet = net_generic(net, brnf_net_id);
 130
 131         return skb->protocol == htons(ETH_P_PPP_SES) &&
 132                pppoe_proto(skb) == htons(PPP_IPV6) &&
 133                brnet->filter_pppoe_tagged;
 134 }
 135
 136 /* largest possible L2 header, see br_nf_dev_queue_xmit() */
 137 #define NF_BRIDGE_MAX_MAC_HEADER_LENGTH (PPPOE_SES_HLEN + ETH_HLEN)
 138
 139 struct brnf_frag_data {
 140         char mac[NF_BRIDGE_MAX_MAC_HEADER_LENGTH];
 141         u8 encap_size;
 142         u8 size;
 143         u16 vlan_tci;
 144         __be16 vlan_proto;
 145 };
 146
 147 static DEFINE_PER_CPU(struct brnf_frag_data, brnf_frag_data_storage);
 148
 149 static void nf_bridge_info_free(struct sk_buff *skb)
 150 {
 151         skb_ext_del(skb, SKB_EXT_BRIDGE_NF);
 152 }
 153
 154 static inline struct net_device *bridge_parent(const struct net_device *dev)
 155 {
 156         struct net_bridge_port *port;
 157
 158         port = br_port_get_rcu(dev);
 159         return port ? port->br->dev : NULL;
 160 }
 161
 162 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 163 {
 164         return skb_ext_add(skb, SKB_EXT_BRIDGE_NF);
 165 }
 166
 167 unsigned int nf_bridge_encap_header_len(const struct sk_buff *skb)
 168 {
 169         switch (skb->protocol) {
 170         case __cpu_to_be16(ETH_P_8021Q):
 171                 return VLAN_HLEN;
 172         case __cpu_to_be16(ETH_P_PPP_SES):
 173                 return PPPOE_SES_HLEN;
 174         default:
 175                 return 0;
 176         }
 177 }
 178
 179 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 180 {
 181         unsigned int len = nf_bridge_encap_header_len(skb);
 182
 183         skb_pull(skb, len);
 184         skb->network_header += len;
 185 }
 186
 187 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 188 {
 189         unsigned int len = nf_bridge_encap_header_len(skb);
 190
 191         skb_pull_rcsum(skb, len);
 192         skb->network_header += len;
 193 }
 194
 195 /* When handing a packet over to the IP layer
 196  * check whether we have a skb that is in the
 197  * expected format
 198  */
 199
 200 static int br_validate_ipv4(struct net *net, struct sk_buff *skb)
 201 {
 202         const struct iphdr *iph;
 203         u32 len;
 204
 205         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 206                 goto inhdr_error;
 207
 208         iph = ip_hdr(skb);
 209
 210         /* Basic sanity checks */
 211         if (iph->ihl < 5 || iph->version != 4)
 212                 goto inhdr_error;
 213
 214         if (!pskb_may_pull(skb, iph->ihl*4))
 215                 goto inhdr_error;
 216
 217         iph = ip_hdr(skb);
 218         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 219                 goto csum_error;
 220
 221         len = skb_ip_totlen(skb);
 222         if (skb->len < len) {
 223                 __IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS);
 224                 goto drop;
 225         } else if (len < (iph->ihl*4))
 226                 goto inhdr_error;
 227
 228         if (pskb_trim_rcsum(skb, len)) {
 229                 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
 230                 goto drop;
 231         }
 232
 233         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 234         /* We should really parse IP options here but until
 235          * somebody who actually uses IP options complains to
 236          * us we'll just silently ignore the options because
 237          * we're lazy!
 238          */
 239         return 0;
 240
 241 csum_error:
 242         __IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS);
 243 inhdr_error:
 244         __IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
 245 drop:
 246         return -1;
 247 }
 248
 249 void nf_bridge_update_protocol(struct sk_buff *skb)
 250 {
 251         const struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 252
 253         switch (nf_bridge->orig_proto) {
 254         case BRNF_PROTO_8021Q:
 255                 skb->protocol = htons(ETH_P_8021Q);
 256                 break;
 257         case BRNF_PROTO_PPPOE:
 258                 skb->protocol = htons(ETH_P_PPP_SES);
 259                 break;
 260         case BRNF_PROTO_UNCHANGED:
 261                 break;
 262         }
 263 }
 264
 265 /* Obtain the correct destination MAC address, while preserving the original
 266  * source MAC address. If we already know this address, we just copy it. If we
 267  * don't, we use the neighbour framework to find out. In both cases, we make
 268  * sure that br_handle_frame_finish() is called afterwards.
 269  */
 270 int br_nf_pre_routing_finish_bridge(struct net *net, struct sock *sk, struct sk_buff *skb)
 271 {
 272         struct neighbour *neigh;
 273         struct dst_entry *dst;
 274
 275         skb->dev = bridge_parent(skb->dev);
 276         if (!skb->dev)
 277                 goto free_skb;
 278         dst = skb_dst(skb);
 279         neigh = dst_neigh_lookup_skb(dst, skb);
 280         if (neigh) {
 281                 struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 282                 int ret;
 283
 284                 if ((READ_ONCE(neigh->nud_state) & NUD_CONNECTED) &&
 285                     READ_ONCE(neigh->hh.hh_len)) {
 286                         struct net_device *br_indev;
 287
 288                         br_indev = nf_bridge_get_physindev(skb, net);
 289                         if (!br_indev) {
 290                                 neigh_release(neigh);
 291                                 goto free_skb;
 292                         }
 293
 294                         neigh_hh_bridge(&neigh->hh, skb);
 295                         skb->dev = br_indev;
 296
 297                         ret = br_handle_frame_finish(net, sk, skb);
 298                 } else {
 299                         /* the neighbour function below overwrites the complete
 300                          * MAC header, so we save the Ethernet source address and
 301                          * protocol number.
 302                          */
 303                         skb_copy_from_linear_data_offset(skb,
 304                                                          -(ETH_HLEN-ETH_ALEN),
 305                                                          nf_bridge->neigh_header,
 306                                                          ETH_HLEN-ETH_ALEN);
 307                         /* tell br_dev_xmit to continue with forwarding */
 308                         nf_bridge->bridged_dnat = 1;
 309                         /* FIXME Need to refragment */
 310                         ret = READ_ONCE(neigh->output)(neigh, skb);
 311                 }
 312                 neigh_release(neigh);
 313                 return ret;
 314         }
 315 free_skb:
 316         kfree_skb(skb);
 317         return 0;
 318 }
 319
 320 static inline bool
 321 br_nf_ipv4_daddr_was_changed(const struct sk_buff *skb,
 322                              const struct nf_bridge_info *nf_bridge)
 323 {
 324         return ip_hdr(skb)->daddr != nf_bridge->ipv4_daddr;
 325 }
 326
 327 /* This requires some explaining. If DNAT has taken place,
 328  * we will need to fix up the destination Ethernet address.
 329  * This is also true when SNAT takes place (for the reply direction).
 330  *
 331  * There are two cases to consider:
 332  * 1. The packet was DNAT'ed to a device in the same bridge
 333  *    port group as it was received on. We can still bridge
 334  *    the packet.
 335  * 2. The packet was DNAT'ed to a different device, either
 336  *    a non-bridged device or another bridge port group.
 337  *    The packet will need to be routed.
 338  *
 339  * The correct way of distinguishing between these two cases is to
 340  * call ip_route_input() and to look at skb->dst->dev, which is
 341  * changed to the destination device if ip_route_input() succeeds.
 342  *
 343  * Let's first consider the case that ip_route_input() succeeds:
 344  *
 345  * If the output device equals the logical bridge device the packet
 346  * came in on, we can consider this bridging. The corresponding MAC
 347  * address will be obtained in br_nf_pre_routing_finish_bridge.
 348  * Otherwise, the packet is considered to be routed and we just
 349  * change the destination MAC address so that the packet will
 350  * later be passed up to the IP stack to be routed. For a redirected
 351  * packet, ip_route_input() will give back the localhost as output device,
 352  * which differs from the bridge device.
 353  *
 354  * Let's now consider the case that ip_route_input() fails:
 355  *
 356  * This can be because the destination address is martian, in which case
 357  * the packet will be dropped.
 358  * If IP forwarding is disabled, ip_route_input() will fail, while
 359  * ip_route_output_key() can return success. The source
 360  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 361  * thinks we're handling a locally generated packet and won't care
 362  * if IP forwarding is enabled. If the output device equals the logical bridge
 363  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 364  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 365  */
 366 static int br_nf_pre_routing_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
 367 {
 368         struct net_device *dev = skb->dev, *br_indev;
 369         struct iphdr *iph = ip_hdr(skb);
 370         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 371         struct rtable *rt;
 372         int err;
 373
 374         br_indev = nf_bridge_get_physindev(skb, net);
 375         if (!br_indev) {
 376                 kfree_skb(skb);
 377                 return 0;
 378         }
 379
 380         nf_bridge->frag_max_size = IPCB(skb)->frag_max_size;
 381
 382         if (nf_bridge->pkt_otherhost) {
 383                 skb->pkt_type = PACKET_OTHERHOST;
 384                 nf_bridge->pkt_otherhost = false;
 385         }
 386         nf_bridge->in_prerouting = 0;
 387         if (br_nf_ipv4_daddr_was_changed(skb, nf_bridge)) {
 388                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 389                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 390
 391                         /* If err equals -EHOSTUNREACH the error is due to a
 392                          * martian destination or due to the fact that
 393                          * forwarding is disabled. For most martian packets,
 394                          * ip_route_output_key() will fail. It won't fail for 2 types of
 395                          * martian destinations: loopback destinations and destination
 396                          * 0.0.0.0. In both cases the packet will be dropped because the
 397                          * destination is the loopback device and not the bridge. */
 398                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 399                                 goto free_skb;
 400
 401                         rt = ip_route_output(net, iph->daddr, 0,
 402                                              RT_TOS(iph->tos), 0);
 403                         if (!IS_ERR(rt)) {
 404                                 /* - Bridged-and-DNAT'ed traffic doesn't
 405                                  *   require ip_forwarding. */
 406                                 if (rt->dst.dev == dev) {
 407                                         skb_dst_drop(skb);
 408                                         skb_dst_set(skb, &rt->dst);
 409                                         goto bridged_dnat;
 410                                 }
 411                                 ip_rt_put(rt);
 412                         }
 413 free_skb:
 414                         kfree_skb(skb);
 415                         return 0;
 416                 } else {
 417                         if (skb_dst(skb)->dev == dev) {
 418 bridged_dnat:
 419                                 skb->dev = br_indev;
 420                                 nf_bridge_update_protocol(skb);
 421                                 nf_bridge_push_encap_header(skb);
 422                                 br_nf_hook_thresh(NF_BR_PRE_ROUTING,
 423                                                   net, sk, skb, skb->dev,
 424                                                   NULL,
 425                                                   br_nf_pre_routing_finish_bridge);
 426                                 return 0;
 427                         }
 428                         ether_addr_copy(eth_hdr(skb)->h_dest, dev->dev_addr);
 429                         skb->pkt_type = PACKET_HOST;
 430                 }
 431         } else {
 432                 rt = bridge_parent_rtable(br_indev);
 433                 if (!rt) {
 434                         kfree_skb(skb);
 435                         return 0;
 436                 }
 437                 skb_dst_drop(skb);
 438                 skb_dst_set_noref(skb, &rt->dst);
 439         }
 440
 441         skb->dev = br_indev;
 442         nf_bridge_update_protocol(skb);
 443         nf_bridge_push_encap_header(skb);
 444         br_nf_hook_thresh(NF_BR_PRE_ROUTING, net, sk, skb, skb->dev, NULL,
 445                           br_handle_frame_finish);
 446         return 0;
 447 }
 448
 449 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb,
 450                                                const struct net_device *dev,
 451                                                const struct net *net)
 452 {
 453         struct net_device *vlan, *br;
 454         struct brnf_net *brnet = net_generic(net, brnf_net_id);
 455
 456         br = bridge_parent(dev);
 457
 458         if (brnet->pass_vlan_indev == 0 || !skb_vlan_tag_present(skb))
 459                 return br;
 460
 461         vlan = __vlan_find_dev_deep_rcu(br, skb->vlan_proto,
 462                                     skb_vlan_tag_get(skb) & VLAN_VID_MASK);
 463
 464         return vlan ? vlan : br;
 465 }
 466
 467 /* Some common code for IPv4/IPv6 */
 468 struct net_device *setup_pre_routing(struct sk_buff *skb, const struct net *net)
 469 {
 470         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 471
 472         if (skb->pkt_type == PACKET_OTHERHOST) {
 473                 skb->pkt_type = PACKET_HOST;
 474                 nf_bridge->pkt_otherhost = true;
 475         }
 476
 477         nf_bridge->in_prerouting = 1;
 478         nf_bridge->physinif = skb->dev->ifindex;
 479         skb->dev = brnf_get_logical_dev(skb, skb->dev, net);
 480
 481         if (skb->protocol == htons(ETH_P_8021Q))
 482                 nf_bridge->orig_proto = BRNF_PROTO_8021Q;
 483         else if (skb->protocol == htons(ETH_P_PPP_SES))
 484                 nf_bridge->orig_proto = BRNF_PROTO_PPPOE;
 485
 486         /* Must drop socket now because of tproxy. */
 487         skb_orphan(skb);
 488         return skb->dev;
 489 }
 490
 491 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 492  * Replicate the checks that IPv4 does on packet reception.
 493  * Set skb->dev to the bridge device (i.e. parent of the
 494  * receiving device) to make netfilter happy, the REDIRECT
 495  * target in particular.  Save the original destination IP
 496  * address to be able to detect DNAT afterwards. */
 497 static unsigned int br_nf_pre_routing(void *priv,
 498                                       struct sk_buff *skb,
 499                                       const struct nf_hook_state *state)
 500 {
 501         struct nf_bridge_info *nf_bridge;
 502         struct net_bridge_port *p;
 503         struct net_bridge *br;
 504         __u32 len = nf_bridge_encap_header_len(skb);
 505         struct brnf_net *brnet;
 506
 507         if (unlikely(!pskb_may_pull(skb, len)))
 508                 return NF_DROP_REASON(skb, SKB_DROP_REASON_PKT_TOO_SMALL, 0);
 509
 510         p = br_port_get_rcu(state->in);
 511         if (p == NULL)
 512                 return NF_DROP_REASON(skb, SKB_DROP_REASON_DEV_READY, 0);
 513         br = p->br;
 514
 515         brnet = net_generic(state->net, brnf_net_id);
 516         if (IS_IPV6(skb) || is_vlan_ipv6(skb, state->net) ||
 517             is_pppoe_ipv6(skb, state->net)) {
 518                 if (!brnet->call_ip6tables &&
 519                     !br_opt_get(br, BROPT_NF_CALL_IP6TABLES))
 520                         return NF_ACCEPT;
 521                 if (!ipv6_mod_enabled()) {
 522                         pr_warn_once("Module ipv6 is disabled, so call_ip6tables is not supported.");
 523                         return NF_DROP_REASON(skb, SKB_DROP_REASON_IPV6DISABLED, 0);
 524                 }
 525
 526                 nf_bridge_pull_encap_header_rcsum(skb);
 527                 return br_nf_pre_routing_ipv6(priv, skb, state);
 528         }
 529
 530         if (!brnet->call_iptables && !br_opt_get(br, BROPT_NF_CALL_IPTABLES))
 531                 return NF_ACCEPT;
 532
 533         if (!IS_IP(skb) && !is_vlan_ip(skb, state->net) &&
 534             !is_pppoe_ip(skb, state->net))
 535                 return NF_ACCEPT;
 536
 537         nf_bridge_pull_encap_header_rcsum(skb);
 538
 539         if (br_validate_ipv4(state->net, skb))
 540                 return NF_DROP_REASON(skb, SKB_DROP_REASON_IP_INHDR, 0);
 541
 542         if (!nf_bridge_alloc(skb))
 543                 return NF_DROP_REASON(skb, SKB_DROP_REASON_NOMEM, 0);
 544         if (!setup_pre_routing(skb, state->net))
 545                 return NF_DROP_REASON(skb, SKB_DROP_REASON_DEV_READY, 0);
 546
 547         nf_bridge = nf_bridge_info_get(skb);
 548         nf_bridge->ipv4_daddr = ip_hdr(skb)->daddr;
 549
 550         skb->protocol = htons(ETH_P_IP);
 551         skb->transport_header = skb->network_header + ip_hdr(skb)->ihl * 4;
 552
 553         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, state->net, state->sk, skb,
 554                 skb->dev, NULL,
 555                 br_nf_pre_routing_finish);
 556
 557         return NF_STOLEN;
 558 }
 559
 560 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 561 /* conntracks' nf_confirm logic cannot handle cloned skbs referencing
 562  * the same nf_conn entry, which will happen for multicast (broadcast)
 563  * Frames on bridges.
 564  *
 565  * Example:
 566  *      macvlan0
 567  *      br0
 568  *  ethX  ethY
 569  *
 570  * ethX (or Y) receives multicast or broadcast packet containing
 571  * an IP packet, not yet in conntrack table.
 572  *
 573  * 1. skb passes through bridge and fake-ip (br_netfilter)Prerouting.
 574  *    -> skb->_nfct now references a unconfirmed entry
 575  * 2. skb is broad/mcast packet. bridge now passes clones out on each bridge
 576  *    interface.
 577  * 3. skb gets passed up the stack.
 578  * 4. In macvlan case, macvlan driver retains clone(s) of the mcast skb
 579  *    and schedules a work queue to send them out on the lower devices.
 580  *
 581  *    The clone skb->_nfct is not a copy, it is the same entry as the
 582  *    original skb.  The macvlan rx handler then returns RX_HANDLER_PASS.
 583  * 5. Normal conntrack hooks (in NF_INET_LOCAL_IN) confirm the orig skb.
 584  *
 585  * The Macvlan broadcast worker and normal confirm path will race.
 586  *
 587  * This race will not happen if step 2 already confirmed a clone. In that
 588  * case later steps perform skb_clone() with skb->_nfct already confirmed (in
 589  * hash table).  This works fine.
 590  *
 591  * But such confirmation won't happen when eb/ip/nftables rules dropped the
 592  * packets before they reached the nf_confirm step in postrouting.
 593  *
 594  * Work around this problem by explicit confirmation of the entry at
 595  * LOCAL_IN time, before upper layer has a chance to clone the unconfirmed
 596  * entry.
 597  *
 598  */
 599 static unsigned int br_nf_local_in(void *priv,
 600                                    struct sk_buff *skb,
 601                                    const struct nf_hook_state *state)
 602 {
 603         struct nf_conntrack *nfct = skb_nfct(skb);
 604         const struct nf_ct_hook *ct_hook;
 605         struct nf_conn *ct;
 606         int ret;
 607
 608         if (!nfct || skb->pkt_type == PACKET_HOST)
 609                 return NF_ACCEPT;
 610
 611         ct = container_of(nfct, struct nf_conn, ct_general);
 612         if (likely(nf_ct_is_confirmed(ct)))
 613                 return NF_ACCEPT;
 614
 615         WARN_ON_ONCE(skb_shared(skb));
 616         WARN_ON_ONCE(refcount_read(&nfct->use) != 1);
 617
 618         /* We can't call nf_confirm here, it would create a dependency
 619          * on nf_conntrack module.
 620          */
 621         ct_hook = rcu_dereference(nf_ct_hook);
 622         if (!ct_hook) {
 623                 skb->_nfct = 0ul;
 624                 nf_conntrack_put(nfct);
 625                 return NF_ACCEPT;
 626         }
 627
 628         nf_bridge_pull_encap_header(skb);
 629         ret = ct_hook->confirm(skb);
 630         switch (ret & NF_VERDICT_MASK) {
 631         case NF_STOLEN:
 632                 return NF_STOLEN;
 633         default:
 634                 nf_bridge_push_encap_header(skb);
 635                 break;
 636         }
 637
 638         ct = container_of(nfct, struct nf_conn, ct_general);
 639         WARN_ON_ONCE(!nf_ct_is_confirmed(ct));
 640
 641         return ret;
 642 }
 643 #endif
 644
 645 /* PF_BRIDGE/FORWARD *************************************************/
 646 static int br_nf_forward_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
 647 {
 648         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 649         struct net_device *in;
 650
 651         if (!IS_ARP(skb) && !is_vlan_arp(skb, net)) {
 652
 653                 if (skb->protocol == htons(ETH_P_IP))
 654                         nf_bridge->frag_max_size = IPCB(skb)->frag_max_size;
 655
 656                 if (skb->protocol == htons(ETH_P_IPV6))
 657                         nf_bridge->frag_max_size = IP6CB(skb)->frag_max_size;
 658
 659                 in = nf_bridge_get_physindev(skb, net);
 660                 if (!in) {
 661                         kfree_skb(skb);
 662                         return 0;
 663                 }
 664                 if (nf_bridge->pkt_otherhost) {
 665                         skb->pkt_type = PACKET_OTHERHOST;
 666                         nf_bridge->pkt_otherhost = false;
 667                 }
 668                 nf_bridge_update_protocol(skb);
 669         } else {
 670                 in = *((struct net_device **)(skb->cb));
 671         }
 672         nf_bridge_push_encap_header(skb);
 673
 674         br_nf_hook_thresh(NF_BR_FORWARD, net, sk, skb, in, skb->dev,
 675                           br_forward_finish);
 676         return 0;
 677 }
 678
 679
 680 static unsigned int br_nf_forward_ip(struct sk_buff *skb,
 681                                      const struct nf_hook_state *state,
 682                                      u8 pf)
 683 {
 684         struct nf_bridge_info *nf_bridge;
 685         struct net_device *parent;
 686
 687         nf_bridge = nf_bridge_info_get(skb);
 688         if (!nf_bridge)
 689                 return NF_ACCEPT;
 690
 691         /* Need exclusive nf_bridge_info since we might have multiple
 692          * different physoutdevs. */
 693         if (!nf_bridge_unshare(skb))
 694                 return NF_DROP_REASON(skb, SKB_DROP_REASON_NOMEM, 0);
 695
 696         nf_bridge = nf_bridge_info_get(skb);
 697         if (!nf_bridge)
 698                 return NF_DROP_REASON(skb, SKB_DROP_REASON_NOMEM, 0);
 699
 700         parent = bridge_parent(state->out);
 701         if (!parent)
 702                 return NF_DROP_REASON(skb, SKB_DROP_REASON_DEV_READY, 0);
 703
 704         nf_bridge_pull_encap_header(skb);
 705
 706         if (skb->pkt_type == PACKET_OTHERHOST) {
 707                 skb->pkt_type = PACKET_HOST;
 708                 nf_bridge->pkt_otherhost = true;
 709         }
 710
 711         if (pf == NFPROTO_IPV4) {
 712                 if (br_validate_ipv4(state->net, skb))
 713                         return NF_DROP_REASON(skb, SKB_DROP_REASON_IP_INHDR, 0);
 714                 IPCB(skb)->frag_max_size = nf_bridge->frag_max_size;
 715                 skb->protocol = htons(ETH_P_IP);
 716         } else if (pf == NFPROTO_IPV6) {
 717                 if (br_validate_ipv6(state->net, skb))
 718                         return NF_DROP_REASON(skb, SKB_DROP_REASON_IP_INHDR, 0);
 719                 IP6CB(skb)->frag_max_size = nf_bridge->frag_max_size;
 720                 skb->protocol = htons(ETH_P_IPV6);
 721         } else {
 722                 WARN_ON_ONCE(1);
 723                 return NF_DROP;
 724         }
 725
 726         nf_bridge->physoutdev = skb->dev;
 727
 728         NF_HOOK(pf, NF_INET_FORWARD, state->net, NULL, skb,
 729                 brnf_get_logical_dev(skb, state->in, state->net),
 730                 parent, br_nf_forward_finish);
 731
 732         return NF_STOLEN;
 733 }
 734
 735 static unsigned int br_nf_forward_arp(struct sk_buff *skb,
 736                                       const struct nf_hook_state *state)
 737 {
 738         struct net_bridge_port *p;
 739         struct net_bridge *br;
 740         struct net_device **d = (struct net_device **)(skb->cb);
 741         struct brnf_net *brnet;
 742
 743         p = br_port_get_rcu(state->out);
 744         if (p == NULL)
 745                 return NF_ACCEPT;
 746         br = p->br;
 747
 748         brnet = net_generic(state->net, brnf_net_id);
 749         if (!brnet->call_arptables && !br_opt_get(br, BROPT_NF_CALL_ARPTABLES))
 750                 return NF_ACCEPT;
 751
 752         if (is_vlan_arp(skb, state->net))
 753                 nf_bridge_pull_encap_header(skb);
 754
 755         if (unlikely(!pskb_may_pull(skb, sizeof(struct arphdr))))
 756                 return NF_DROP_REASON(skb, SKB_DROP_REASON_PKT_TOO_SMALL, 0);
 757
 758         if (arp_hdr(skb)->ar_pln != 4) {
 759                 if (is_vlan_arp(skb, state->net))
 760                         nf_bridge_push_encap_header(skb);
 761                 return NF_ACCEPT;
 762         }
 763         *d = state->in;
 764         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, state->net, state->sk, skb,
 765                 state->in, state->out, br_nf_forward_finish);
 766
 767         return NF_STOLEN;
 768 }
 769
 770 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 771  * netfilter with indev and outdev set to the bridge device,
 772  * but we are still able to filter on the 'real' indev/outdev
 773  * because of the physdev module. For ARP, indev and outdev are the
 774  * bridge ports.
 775  */
 776 static unsigned int br_nf_forward(void *priv,
 777                                   struct sk_buff *skb,
 778                                   const struct nf_hook_state *state)
 779 {
 780         if (IS_IP(skb) || is_vlan_ip(skb, state->net) ||
 781             is_pppoe_ip(skb, state->net))
 782                 return br_nf_forward_ip(skb, state, NFPROTO_IPV4);
 783         if (IS_IPV6(skb) || is_vlan_ipv6(skb, state->net) ||
 784             is_pppoe_ipv6(skb, state->net))
 785                 return br_nf_forward_ip(skb, state, NFPROTO_IPV6);
 786         if (IS_ARP(skb) || is_vlan_arp(skb, state->net))
 787                 return br_nf_forward_arp(skb, state);
 788
 789         return NF_ACCEPT;
 790 }
 791
 792 static int br_nf_push_frag_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
 793 {
 794         struct brnf_frag_data *data;
 795         int err;
 796
 797         data = this_cpu_ptr(&brnf_frag_data_storage);
 798         err = skb_cow_head(skb, data->size);
 799
 800         if (err) {
 801                 kfree_skb(skb);
 802                 return 0;
 803         }
 804
 805         if (data->vlan_proto)
 806                 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci);
 807
 808         skb_copy_to_linear_data_offset(skb, -data->size, data->mac, data->size);
 809         __skb_push(skb, data->encap_size);
 810
 811         nf_bridge_info_free(skb);
 812         return br_dev_queue_push_xmit(net, sk, skb);
 813 }
 814
 815 static int
 816 br_nf_ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
 817                   int (*output)(struct net *, struct sock *, struct sk_buff *))
 818 {
 819         unsigned int mtu = ip_skb_dst_mtu(sk, skb);
 820         struct iphdr *iph = ip_hdr(skb);
 821
 822         if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->ignore_df) ||
 823                      (IPCB(skb)->frag_max_size &&
 824                       IPCB(skb)->frag_max_size > mtu))) {
 825                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
 826                 kfree_skb(skb);
 827                 return -EMSGSIZE;
 828         }
 829
 830         return ip_do_fragment(net, sk, skb, output);
 831 }
 832
 833 static unsigned int nf_bridge_mtu_reduction(const struct sk_buff *skb)
 834 {
 835         const struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 836
 837         if (nf_bridge->orig_proto == BRNF_PROTO_PPPOE)
 838                 return PPPOE_SES_HLEN;
 839         return 0;
 840 }
 841
 842 static int br_nf_dev_queue_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
 843 {
 844         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 845         unsigned int mtu, mtu_reserved;
 846
 847         mtu_reserved = nf_bridge_mtu_reduction(skb);
 848         mtu = skb->dev->mtu;
 849
 850         if (nf_bridge->pkt_otherhost) {
 851                 skb->pkt_type = PACKET_OTHERHOST;
 852                 nf_bridge->pkt_otherhost = false;
 853         }
 854
 855         if (nf_bridge->frag_max_size && nf_bridge->frag_max_size < mtu)
 856                 mtu = nf_bridge->frag_max_size;
 857
 858         nf_bridge_update_protocol(skb);
 859         nf_bridge_push_encap_header(skb);
 860
 861         if (skb_is_gso(skb) || skb->len + mtu_reserved <= mtu) {
 862                 nf_bridge_info_free(skb);
 863                 return br_dev_queue_push_xmit(net, sk, skb);
 864         }
 865
 866         /* This is wrong! We should preserve the original fragment
 867          * boundaries by preserving frag_list rather than refragmenting.
 868          */
 869         if (IS_ENABLED(CONFIG_NF_DEFRAG_IPV4) &&
 870             skb->protocol == htons(ETH_P_IP)) {
 871                 struct brnf_frag_data *data;
 872
 873                 if (br_validate_ipv4(net, skb))
 874                         goto drop;
 875
 876                 IPCB(skb)->frag_max_size = nf_bridge->frag_max_size;
 877
 878                 data = this_cpu_ptr(&brnf_frag_data_storage);
 879
 880                 if (skb_vlan_tag_present(skb)) {
 881                         data->vlan_tci = skb->vlan_tci;
 882                         data->vlan_proto = skb->vlan_proto;
 883                 } else {
 884                         data->vlan_proto = 0;
 885                 }
 886
 887                 data->encap_size = nf_bridge_encap_header_len(skb);
 888                 data->size = ETH_HLEN + data->encap_size;
 889
 890                 skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
 891                                                  data->size);
 892
 893                 return br_nf_ip_fragment(net, sk, skb, br_nf_push_frag_xmit);
 894         }
 895         if (IS_ENABLED(CONFIG_NF_DEFRAG_IPV6) &&
 896             skb->protocol == htons(ETH_P_IPV6)) {
 897                 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
 898                 struct brnf_frag_data *data;
 899
 900                 if (br_validate_ipv6(net, skb))
 901                         goto drop;
 902
 903                 IP6CB(skb)->frag_max_size = nf_bridge->frag_max_size;
 904
 905                 data = this_cpu_ptr(&brnf_frag_data_storage);
 906                 data->encap_size = nf_bridge_encap_header_len(skb);
 907                 data->size = ETH_HLEN + data->encap_size;
 908
 909                 skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
 910                                                  data->size);
 911
 912                 if (v6ops)
 913                         return v6ops->fragment(net, sk, skb, br_nf_push_frag_xmit);
 914
 915                 kfree_skb(skb);
 916                 return -EMSGSIZE;
 917         }
 918         nf_bridge_info_free(skb);
 919         return br_dev_queue_push_xmit(net, sk, skb);
 920  drop:
 921         kfree_skb(skb);
 922         return 0;
 923 }
 924
 925 /* PF_BRIDGE/POST_ROUTING ********************************************/
 926 static unsigned int br_nf_post_routing(void *priv,
 927                                        struct sk_buff *skb,
 928                                        const struct nf_hook_state *state)
 929 {
 930         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 931         struct net_device *realoutdev = bridge_parent(skb->dev);
 932         u_int8_t pf;
 933
 934         /* if nf_bridge is set, but ->physoutdev is NULL, this packet came in
 935          * on a bridge, but was delivered locally and is now being routed:
 936          *
 937          * POST_ROUTING was already invoked from the ip stack.
 938          */
 939         if (!nf_bridge || !nf_bridge->physoutdev)
 940                 return NF_ACCEPT;
 941
 942         if (!realoutdev)
 943                 return NF_DROP_REASON(skb, SKB_DROP_REASON_DEV_READY, 0);
 944
 945         if (IS_IP(skb) || is_vlan_ip(skb, state->net) ||
 946             is_pppoe_ip(skb, state->net))
 947                 pf = NFPROTO_IPV4;
 948         else if (IS_IPV6(skb) || is_vlan_ipv6(skb, state->net) ||
 949                  is_pppoe_ipv6(skb, state->net))
 950                 pf = NFPROTO_IPV6;
 951         else
 952                 return NF_ACCEPT;
 953
 954         if (skb->pkt_type == PACKET_OTHERHOST) {
 955                 skb->pkt_type = PACKET_HOST;
 956                 nf_bridge->pkt_otherhost = true;
 957         }
 958
 959         nf_bridge_pull_encap_header(skb);
 960         if (pf == NFPROTO_IPV4)
 961                 skb->protocol = htons(ETH_P_IP);
 962         else
 963                 skb->protocol = htons(ETH_P_IPV6);
 964
 965         NF_HOOK(pf, NF_INET_POST_ROUTING, state->net, state->sk, skb,
 966                 NULL, realoutdev,
 967                 br_nf_dev_queue_xmit);
 968
 969         return NF_STOLEN;
 970 }
 971
 972 /* IP/SABOTAGE *****************************************************/
 973 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 974  * for the second time. */
 975 static unsigned int ip_sabotage_in(void *priv,
 976                                    struct sk_buff *skb,
 977                                    const struct nf_hook_state *state)
 978 {
 979         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 980
 981         if (nf_bridge) {
 982                 if (nf_bridge->sabotage_in_done)
 983                         return NF_ACCEPT;
 984
 985                 if (!nf_bridge->in_prerouting &&
 986                     !netif_is_l3_master(skb->dev) &&
 987                     !netif_is_l3_slave(skb->dev)) {
 988                         nf_bridge->sabotage_in_done = 1;
 989                         state->okfn(state->net, state->sk, skb);
 990                         return NF_STOLEN;
 991                 }
 992         }
 993
 994         return NF_ACCEPT;
 995 }
 996
 997 /* This is called when br_netfilter has called into iptables/netfilter,
 998  * and DNAT has taken place on a bridge-forwarded packet.
 999  *
1000  * neigh->output has created a new MAC header, with local br0 MAC
1001  * as saddr.
1002  *
1003  * This restores the original MAC saddr of the bridged packet
1004  * before invoking bridge forward logic to transmit the packet.
1005  */
1006 static void br_nf_pre_routing_finish_bridge_slow(struct sk_buff *skb)
1007 {
1008         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
1009         struct net_device *br_indev;
1010
1011         br_indev = nf_bridge_get_physindev(skb, dev_net(skb->dev));
1012         if (!br_indev) {
1013                 kfree_skb(skb);
1014                 return;
1015         }
1016
1017         skb_pull(skb, ETH_HLEN);
1018         nf_bridge->bridged_dnat = 0;
1019
1020         BUILD_BUG_ON(sizeof(nf_bridge->neigh_header) != (ETH_HLEN - ETH_ALEN));
1021
1022         skb_copy_to_linear_data_offset(skb, -(ETH_HLEN - ETH_ALEN),
1023                                        nf_bridge->neigh_header,
1024                                        ETH_HLEN - ETH_ALEN);
1025         skb->dev = br_indev;
1026
1027         nf_bridge->physoutdev = NULL;
1028         br_handle_frame_finish(dev_net(skb->dev), NULL, skb);
1029 }
1030
1031 static int br_nf_dev_xmit(struct sk_buff *skb)
1032 {
1033         const struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
1034
1035         if (nf_bridge && nf_bridge->bridged_dnat) {
1036                 br_nf_pre_routing_finish_bridge_slow(skb);
1037                 return 1;
1038         }
1039         return 0;
1040 }
1041
1042 static const struct nf_br_ops br_ops = {
1043         .br_dev_xmit_hook =     br_nf_dev_xmit,
1044 };
1045
1046 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
1047  * br_dev_queue_push_xmit is called afterwards */
1048 static const struct nf_hook_ops br_nf_ops[] = {
1049         {
1050                 .hook = br_nf_pre_routing,
1051                 .pf = NFPROTO_BRIDGE,
1052                 .hooknum = NF_BR_PRE_ROUTING,
1053                 .priority = NF_BR_PRI_BRNF,
1054         },
1055 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1056         {
1057                 .hook = br_nf_local_in,
1058                 .pf = NFPROTO_BRIDGE,
1059                 .hooknum = NF_BR_LOCAL_IN,
1060                 .priority = NF_BR_PRI_LAST,
1061         },
1062 #endif
1063         {
1064                 .hook = br_nf_forward,
1065                 .pf = NFPROTO_BRIDGE,
1066                 .hooknum = NF_BR_FORWARD,
1067                 .priority = NF_BR_PRI_BRNF,
1068         },
1069         {
1070                 .hook = br_nf_post_routing,
1071                 .pf = NFPROTO_BRIDGE,
1072                 .hooknum = NF_BR_POST_ROUTING,
1073                 .priority = NF_BR_PRI_LAST,
1074         },
1075         {
1076                 .hook = ip_sabotage_in,
1077                 .pf = NFPROTO_IPV4,
1078                 .hooknum = NF_INET_PRE_ROUTING,
1079                 .priority = NF_IP_PRI_FIRST,
1080         },
1081         {
1082                 .hook = ip_sabotage_in,
1083                 .pf = NFPROTO_IPV6,
1084                 .hooknum = NF_INET_PRE_ROUTING,
1085                 .priority = NF_IP6_PRI_FIRST,
1086         },
1087 };
1088
1089 static int brnf_device_event(struct notifier_block *unused, unsigned long event,
1090                              void *ptr)
1091 {
1092         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1093         struct brnf_net *brnet;
1094         struct net *net;
1095         int ret;
1096
1097         if (event != NETDEV_REGISTER || !netif_is_bridge_master(dev))
1098                 return NOTIFY_DONE;
1099
1100         ASSERT_RTNL();
1101
1102         net = dev_net(dev);
1103         brnet = net_generic(net, brnf_net_id);
1104         if (brnet->enabled)
1105                 return NOTIFY_OK;
1106
1107         ret = nf_register_net_hooks(net, br_nf_ops, ARRAY_SIZE(br_nf_ops));
1108         if (ret)
1109                 return NOTIFY_BAD;
1110
1111         brnet->enabled = true;
1112         return NOTIFY_OK;
1113 }
1114
1115 static struct notifier_block brnf_notifier __read_mostly = {
1116         .notifier_call = brnf_device_event,
1117 };
1118
1119 /* recursively invokes nf_hook_slow (again), skipping already-called
1120  * hooks (< NF_BR_PRI_BRNF).
1121  *
1122  * Called with rcu read lock held.
1123  */
1124 int br_nf_hook_thresh(unsigned int hook, struct net *net,
1125                       struct sock *sk, struct sk_buff *skb,
1126                       struct net_device *indev,
1127                       struct net_device *outdev,
1128                       int (*okfn)(struct net *, struct sock *,
1129                                   struct sk_buff *))
1130 {
1131         const struct nf_hook_entries *e;
1132         struct nf_hook_state state;
1133         struct nf_hook_ops **ops;
1134         unsigned int i;
1135         int ret;
1136
1137         e = rcu_dereference(net->nf.hooks_bridge[hook]);
1138         if (!e)
1139                 return okfn(net, sk, skb);
1140
1141         ops = nf_hook_entries_get_hook_ops(e);
1142         for (i = 0; i < e->num_hook_entries; i++) {
1143                 /* These hooks have already been called */
1144                 if (ops[i]->priority < NF_BR_PRI_BRNF)
1145                         continue;
1146
1147                 /* These hooks have not been called yet, run them. */
1148                 if (ops[i]->priority > NF_BR_PRI_BRNF)
1149                         break;
1150
1151                 /* take a closer look at NF_BR_PRI_BRNF. */
1152                 if (ops[i]->hook == br_nf_pre_routing) {
1153                         /* This hook diverted the skb to this function,
1154                          * hooks after this have not been run yet.
1155                          */
1156                         i++;
1157                         break;
1158                 }
1159         }
1160
1161         nf_hook_state_init(&state, hook, NFPROTO_BRIDGE, indev, outdev,
1162                            sk, net, okfn);
1163
1164         ret = nf_hook_slow(skb, &state, e, i);
1165         if (ret == 1)
1166                 ret = okfn(net, sk, skb);
1167
1168         return ret;
1169 }
1170
1171 #ifdef CONFIG_SYSCTL
1172 static
1173 int brnf_sysctl_call_tables(struct ctl_table *ctl, int write,
1174                             void *buffer, size_t *lenp, loff_t *ppos)
1175 {
1176         int ret;
1177
1178         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1179
1180         if (write && *(int *)(ctl->data))
1181                 *(int *)(ctl->data) = 1;
1182         return ret;
1183 }
1184
1185 static struct ctl_table brnf_table[] = {
1186         {
1187                 .procname       = "bridge-nf-call-arptables",
1188                 .maxlen         = sizeof(int),
1189                 .mode           = 0644,
1190                 .proc_handler   = brnf_sysctl_call_tables,
1191         },
1192         {
1193                 .procname       = "bridge-nf-call-iptables",
1194                 .maxlen         = sizeof(int),
1195                 .mode           = 0644,
1196                 .proc_handler   = brnf_sysctl_call_tables,
1197         },
1198         {
1199                 .procname       = "bridge-nf-call-ip6tables",
1200                 .maxlen         = sizeof(int),
1201                 .mode           = 0644,
1202                 .proc_handler   = brnf_sysctl_call_tables,
1203         },
1204         {
1205                 .procname       = "bridge-nf-filter-vlan-tagged",
1206                 .maxlen         = sizeof(int),
1207                 .mode           = 0644,
1208                 .proc_handler   = brnf_sysctl_call_tables,
1209         },
1210         {
1211                 .procname       = "bridge-nf-filter-pppoe-tagged",
1212                 .maxlen         = sizeof(int),
1213                 .mode           = 0644,
1214                 .proc_handler   = brnf_sysctl_call_tables,
1215         },
1216         {
1217                 .procname       = "bridge-nf-pass-vlan-input-dev",
1218                 .maxlen         = sizeof(int),
1219                 .mode           = 0644,
1220                 .proc_handler   = brnf_sysctl_call_tables,
1221         },
1222         { }
1223 };
1224
1225 static inline void br_netfilter_sysctl_default(struct brnf_net *brnf)
1226 {
1227         brnf->call_iptables = 1;
1228         brnf->call_ip6tables = 1;
1229         brnf->call_arptables = 1;
1230         brnf->filter_vlan_tagged = 0;
1231         brnf->filter_pppoe_tagged = 0;
1232         brnf->pass_vlan_indev = 0;
1233 }
1234
1235 static int br_netfilter_sysctl_init_net(struct net *net)
1236 {
1237         struct ctl_table *table = brnf_table;
1238         struct brnf_net *brnet;
1239
1240         if (!net_eq(net, &init_net)) {
1241                 table = kmemdup(table, sizeof(brnf_table), GFP_KERNEL);
1242                 if (!table)
1243                         return -ENOMEM;
1244         }
1245
1246         brnet = net_generic(net, brnf_net_id);
1247         table[0].data = &brnet->call_arptables;
1248         table[1].data = &brnet->call_iptables;
1249         table[2].data = &brnet->call_ip6tables;
1250         table[3].data = &brnet->filter_vlan_tagged;
1251         table[4].data = &brnet->filter_pppoe_tagged;
1252         table[5].data = &brnet->pass_vlan_indev;
1253
1254         br_netfilter_sysctl_default(brnet);
1255
1256         brnet->ctl_hdr = register_net_sysctl_sz(net, "net/bridge", table,
1257                                                 ARRAY_SIZE(brnf_table));
1258         if (!brnet->ctl_hdr) {
1259                 if (!net_eq(net, &init_net))
1260                         kfree(table);
1261
1262                 return -ENOMEM;
1263         }
1264
1265         return 0;
1266 }
1267
1268 static void br_netfilter_sysctl_exit_net(struct net *net,
1269                                          struct brnf_net *brnet)
1270 {
1271         struct ctl_table *table = brnet->ctl_hdr->ctl_table_arg;
1272
1273         unregister_net_sysctl_table(brnet->ctl_hdr);
1274         if (!net_eq(net, &init_net))
1275                 kfree(table);
1276 }
1277
1278 static int __net_init brnf_init_net(struct net *net)
1279 {
1280         return br_netfilter_sysctl_init_net(net);
1281 }
1282 #endif
1283
1284 static void __net_exit brnf_exit_net(struct net *net)
1285 {
1286         struct brnf_net *brnet;
1287
1288         brnet = net_generic(net, brnf_net_id);
1289         if (brnet->enabled) {
1290                 nf_unregister_net_hooks(net, br_nf_ops, ARRAY_SIZE(br_nf_ops));
1291                 brnet->enabled = false;
1292         }
1293
1294 #ifdef CONFIG_SYSCTL
1295         br_netfilter_sysctl_exit_net(net, brnet);
1296 #endif
1297 }
1298
1299 static struct pernet_operations brnf_net_ops __read_mostly = {
1300 #ifdef CONFIG_SYSCTL
1301         .init = brnf_init_net,
1302 #endif
1303         .exit = brnf_exit_net,
1304         .id   = &brnf_net_id,
1305         .size = sizeof(struct brnf_net),
1306 };
1307
1308 static int __init br_netfilter_init(void)
1309 {
1310         int ret;
1311
1312         ret = register_pernet_subsys(&brnf_net_ops);
1313         if (ret < 0)
1314                 return ret;
1315
1316         ret = register_netdevice_notifier(&brnf_notifier);
1317         if (ret < 0) {
1318                 unregister_pernet_subsys(&brnf_net_ops);
1319                 return ret;
1320         }
1321
1322         RCU_INIT_POINTER(nf_br_ops, &br_ops);
1323         printk(KERN_NOTICE "Bridge firewalling registered\n");
1324         return 0;
1325 }
1326
1327 static void __exit br_netfilter_fini(void)
1328 {
1329         RCU_INIT_POINTER(nf_br_ops, NULL);
1330         unregister_netdevice_notifier(&brnf_notifier);
1331         unregister_pernet_subsys(&brnf_net_ops);
1332 }
1333
1334 module_init(br_netfilter_init);
1335 module_exit(br_netfilter_fini);
1336
1337 MODULE_LICENSE("GPL");
1338 MODULE_AUTHOR("Lennert Buytenhek <buytenh@gnu.org>");
1339 MODULE_AUTHOR("Bart De Schuymer <bdschuym@pandora.be>");
1340 MODULE_DESCRIPTION("Linux ethernet netfilter firewall bridge");