net/bridge/br_input.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Handle incoming frames
   4  *      Linux ethernet bridge
   5  *
   6  *      Authors:
   7  *      Lennert Buytenhek               <buytenh@gnu.org>
   8  */
   9
  10 #include <linux/slab.h>
  11 #include <linux/kernel.h>
  12 #include <linux/netdevice.h>
  13 #include <linux/etherdevice.h>
  14 #include <linux/netfilter_bridge.h>
  15 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE
  16 #include <net/netfilter/nf_queue.h>
  17 #endif
  18 #include <linux/neighbour.h>
  19 #include <net/arp.h>
  20 #include <net/dsa.h>
  21 #include <linux/export.h>
  22 #include <linux/rculist.h>
  23 #include "br_private.h"
  24 #include "br_private_tunnel.h"
  25
  26 static int
  27 br_netif_receive_skb(struct net *net, struct sock *sk, struct sk_buff *skb)
  28 {
  29         br_drop_fake_rtable(skb);
  30         return netif_receive_skb(skb);
  31 }
  32
  33 static int br_pass_frame_up(struct sk_buff *skb)
  34 {
  35         struct net_device *indev, *brdev = BR_INPUT_SKB_CB(skb)->brdev;
  36         struct net_bridge *br = netdev_priv(brdev);
  37         struct net_bridge_vlan_group *vg;
  38         struct pcpu_sw_netstats *brstats = this_cpu_ptr(br->stats);
  39
  40         u64_stats_update_begin(&brstats->syncp);
  41         brstats->rx_packets++;
  42         brstats->rx_bytes += skb->len;
  43         u64_stats_update_end(&brstats->syncp);
  44
  45         vg = br_vlan_group_rcu(br);
  46         /* Bridge is just like any other port.  Make sure the
  47          * packet is allowed except in promisc modue when someone
  48          * may be running packet capture.
  49          */
  50         if (!(brdev->flags & IFF_PROMISC) &&
  51             !br_allowed_egress(vg, skb)) {
  52                 kfree_skb(skb);
  53                 return NET_RX_DROP;
  54         }
  55
  56         indev = skb->dev;
  57         skb->dev = brdev;
  58         skb = br_handle_vlan(br, NULL, vg, skb);
  59         if (!skb)
  60                 return NET_RX_DROP;
  61         /* update the multicast stats if the packet is IGMP/MLD */
  62         br_multicast_count(br, NULL, skb, br_multicast_igmp_type(skb),
  63                            BR_MCAST_DIR_TX);
  64
  65         return NF_HOOK(NFPROTO_BRIDGE, NF_BR_LOCAL_IN,
  66                        dev_net(indev), NULL, skb, indev, NULL,
  67                        br_netif_receive_skb);
  68 }
  69
  70 /* note: already called with rcu_read_lock */
  71 int br_handle_frame_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
  72 {
  73         struct net_bridge_port *p = br_port_get_rcu(skb->dev);
  74         enum br_pkt_type pkt_type = BR_PKT_UNICAST;
  75         struct net_bridge_fdb_entry *dst = NULL;
  76         struct net_bridge_mdb_entry *mdst;
  77         bool local_rcv, mcast_hit = false;
  78         struct net_bridge *br;
  79         u16 vid = 0;
  80         u8 state;
  81
  82         if (!p || p->state == BR_STATE_DISABLED)
  83                 goto drop;
  84
  85         state = p->state;
  86         if (!br_allowed_ingress(p->br, nbp_vlan_group_rcu(p), skb, &vid,
  87                                 &state))
  88                 goto out;
  89
  90         nbp_switchdev_frame_mark(p, skb);
  91
  92         /* insert into forwarding database after filtering to avoid spoofing */
  93         br = p->br;
  94         if (p->flags & BR_LEARNING)
  95                 br_fdb_update(br, p, eth_hdr(skb)->h_source, vid, 0);
  96
  97         local_rcv = !!(br->dev->flags & IFF_PROMISC);
  98         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest)) {
  99                 /* by definition the broadcast is also a multicast address */
 100                 if (is_broadcast_ether_addr(eth_hdr(skb)->h_dest)) {
 101                         pkt_type = BR_PKT_BROADCAST;
 102                         local_rcv = true;
 103                 } else {
 104                         pkt_type = BR_PKT_MULTICAST;
 105                         if (br_multicast_rcv(br, p, skb, vid))
 106                                 goto drop;
 107                 }
 108         }
 109
 110         if (state == BR_STATE_LEARNING)
 111                 goto drop;
 112
 113         BR_INPUT_SKB_CB(skb)->brdev = br->dev;
 114         BR_INPUT_SKB_CB(skb)->src_port_isolated = !!(p->flags & BR_ISOLATED);
 115
 116         if (IS_ENABLED(CONFIG_INET) &&
 117             (skb->protocol == htons(ETH_P_ARP) ||
 118              skb->protocol == htons(ETH_P_RARP))) {
 119                 br_do_proxy_suppress_arp(skb, br, vid, p);
 120         } else if (IS_ENABLED(CONFIG_IPV6) &&
 121                    skb->protocol == htons(ETH_P_IPV6) &&
 122                    br_opt_get(br, BROPT_NEIGH_SUPPRESS_ENABLED) &&
 123                    pskb_may_pull(skb, sizeof(struct ipv6hdr) +
 124                                  sizeof(struct nd_msg)) &&
 125                    ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
 126                         struct nd_msg *msg, _msg;
 127
 128                         msg = br_is_nd_neigh_msg(skb, &_msg);
 129                         if (msg)
 130                                 br_do_suppress_nd(skb, br, vid, p, msg);
 131         }
 132
 133         switch (pkt_type) {
 134         case BR_PKT_MULTICAST:
 135                 mdst = br_mdb_get(br, skb, vid);
 136                 if ((mdst || BR_INPUT_SKB_CB_MROUTERS_ONLY(skb)) &&
 137                     br_multicast_querier_exists(br, eth_hdr(skb), mdst)) {
 138                         if ((mdst && mdst->host_joined) ||
 139                             br_multicast_is_router(br)) {
 140                                 local_rcv = true;
 141                                 br->dev->stats.multicast++;
 142                         }
 143                         mcast_hit = true;
 144                 } else {
 145                         local_rcv = true;
 146                         br->dev->stats.multicast++;
 147                 }
 148                 break;
 149         case BR_PKT_UNICAST:
 150                 dst = br_fdb_find_rcu(br, eth_hdr(skb)->h_dest, vid);
 151         default:
 152                 break;
 153         }
 154
 155         if (dst) {
 156                 unsigned long now = jiffies;
 157
 158                 if (test_bit(BR_FDB_LOCAL, &dst->flags))
 159                         return br_pass_frame_up(skb);
 160
 161                 if (now != dst->used)
 162                         dst->used = now;
 163                 br_forward(dst->dst, skb, local_rcv, false);
 164         } else {
 165                 if (!mcast_hit)
 166                         br_flood(br, skb, pkt_type, local_rcv, false);
 167                 else
 168                         br_multicast_flood(mdst, skb, local_rcv, false);
 169         }
 170
 171         if (local_rcv)
 172                 return br_pass_frame_up(skb);
 173
 174 out:
 175         return 0;
 176 drop:
 177         kfree_skb(skb);
 178         goto out;
 179 }
 180 EXPORT_SYMBOL_GPL(br_handle_frame_finish);
 181
 182 static void __br_handle_local_finish(struct sk_buff *skb)
 183 {
 184         struct net_bridge_port *p = br_port_get_rcu(skb->dev);
 185         u16 vid = 0;
 186
 187         /* check if vlan is allowed, to avoid spoofing */
 188         if ((p->flags & BR_LEARNING) &&
 189             nbp_state_should_learn(p) &&
 190             !br_opt_get(p->br, BROPT_NO_LL_LEARN) &&
 191             br_should_learn(p, skb, &vid))
 192                 br_fdb_update(p->br, p, eth_hdr(skb)->h_source, vid, 0);
 193 }
 194
 195 /* note: already called with rcu_read_lock */
 196 static int br_handle_local_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
 197 {
 198         __br_handle_local_finish(skb);
 199
 200         /* return 1 to signal the okfn() was called so it's ok to use the skb */
 201         return 1;
 202 }
 203
 204 static int nf_hook_bridge_pre(struct sk_buff *skb, struct sk_buff **pskb)
 205 {
 206 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE
 207         struct nf_hook_entries *e = NULL;
 208         struct nf_hook_state state;
 209         unsigned int verdict, i;
 210         struct net *net;
 211         int ret;
 212
 213         net = dev_net(skb->dev);
 214 #ifdef HAVE_JUMP_LABEL
 215         if (!static_key_false(&nf_hooks_needed[NFPROTO_BRIDGE][NF_BR_PRE_ROUTING]))
 216                 goto frame_finish;
 217 #endif
 218
 219         e = rcu_dereference(net->nf.hooks_bridge[NF_BR_PRE_ROUTING]);
 220         if (!e)
 221                 goto frame_finish;
 222
 223         nf_hook_state_init(&state, NF_BR_PRE_ROUTING,
 224                            NFPROTO_BRIDGE, skb->dev, NULL, NULL,
 225                            net, br_handle_frame_finish);
 226
 227         for (i = 0; i < e->num_hook_entries; i++) {
 228                 verdict = nf_hook_entry_hookfn(&e->hooks[i], skb, &state);
 229                 switch (verdict & NF_VERDICT_MASK) {
 230                 case NF_ACCEPT:
 231                         if (BR_INPUT_SKB_CB(skb)->br_netfilter_broute) {
 232                                 *pskb = skb;
 233                                 return RX_HANDLER_PASS;
 234                         }
 235                         break;
 236                 case NF_DROP:
 237                         kfree_skb(skb);
 238                         return RX_HANDLER_CONSUMED;
 239                 case NF_QUEUE:
 240                         ret = nf_queue(skb, &state, i, verdict);
 241                         if (ret == 1)
 242                                 continue;
 243                         return RX_HANDLER_CONSUMED;
 244                 default: /* STOLEN */
 245                         return RX_HANDLER_CONSUMED;
 246                 }
 247         }
 248 frame_finish:
 249         net = dev_net(skb->dev);
 250         br_handle_frame_finish(net, NULL, skb);
 251 #else
 252         br_handle_frame_finish(dev_net(skb->dev), NULL, skb);
 253 #endif
 254         return RX_HANDLER_CONSUMED;
 255 }
 256
 257 /* Return 0 if the frame was not processed otherwise 1
 258  * note: already called with rcu_read_lock
 259  */
 260 static int br_process_frame_type(struct net_bridge_port *p,
 261                                  struct sk_buff *skb)
 262 {
 263         struct br_frame_type *tmp;
 264
 265         hlist_for_each_entry_rcu(tmp, &p->br->frame_type_list, list)
 266                 if (unlikely(tmp->type == skb->protocol))
 267                         return tmp->frame_handler(p, skb);
 268
 269         return 0;
 270 }
 271
 272 /*
 273  * Return NULL if skb is handled
 274  * note: already called with rcu_read_lock
 275  */
 276 static rx_handler_result_t br_handle_frame(struct sk_buff **pskb)
 277 {
 278         struct net_bridge_port *p;
 279         struct sk_buff *skb = *pskb;
 280         const unsigned char *dest = eth_hdr(skb)->h_dest;
 281
 282         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
 283                 return RX_HANDLER_PASS;
 284
 285         if (!is_valid_ether_addr(eth_hdr(skb)->h_source))
 286                 goto drop;
 287
 288         skb = skb_share_check(skb, GFP_ATOMIC);
 289         if (!skb)
 290                 return RX_HANDLER_CONSUMED;
 291
 292         memset(skb->cb, 0, sizeof(struct br_input_skb_cb));
 293
 294         p = br_port_get_rcu(skb->dev);
 295         if (p->flags & BR_VLAN_TUNNEL) {
 296                 if (br_handle_ingress_vlan_tunnel(skb, p,
 297                                                   nbp_vlan_group_rcu(p)))
 298                         goto drop;
 299         }
 300
 301         if (unlikely(is_link_local_ether_addr(dest))) {
 302                 u16 fwd_mask = p->br->group_fwd_mask_required;
 303
 304                 /*
 305                  * See IEEE 802.1D Table 7-10 Reserved addresses
 306                  *
 307                  * Assignment                           Value
 308                  * Bridge Group Address         01-80-C2-00-00-00
 309                  * (MAC Control) 802.3          01-80-C2-00-00-01
 310                  * (Link Aggregation) 802.3     01-80-C2-00-00-02
 311                  * 802.1X PAE address           01-80-C2-00-00-03
 312                  *
 313                  * 802.1AB LLDP                 01-80-C2-00-00-0E
 314                  *
 315                  * Others reserved for future standardization
 316                  */
 317                 fwd_mask |= p->group_fwd_mask;
 318                 switch (dest[5]) {
 319                 case 0x00:      /* Bridge Group Address */
 320                         /* If STP is turned off,
 321                            then must forward to keep loop detection */
 322                         if (p->br->stp_enabled == BR_NO_STP ||
 323                             fwd_mask & (1u << dest[5]))
 324                                 goto forward;
 325                         *pskb = skb;
 326                         __br_handle_local_finish(skb);
 327                         return RX_HANDLER_PASS;
 328
 329                 case 0x01:      /* IEEE MAC (Pause) */
 330                         goto drop;
 331
 332                 case 0x0E:      /* 802.1AB LLDP */
 333                         fwd_mask |= p->br->group_fwd_mask;
 334                         if (fwd_mask & (1u << dest[5]))
 335                                 goto forward;
 336                         *pskb = skb;
 337                         __br_handle_local_finish(skb);
 338                         return RX_HANDLER_PASS;
 339
 340                 default:
 341                         /* Allow selective forwarding for most other protocols */
 342                         fwd_mask |= p->br->group_fwd_mask;
 343                         if (fwd_mask & (1u << dest[5]))
 344                                 goto forward;
 345                 }
 346
 347                 /* The else clause should be hit when nf_hook():
 348                  *   - returns < 0 (drop/error)
 349                  *   - returns = 0 (stolen/nf_queue)
 350                  * Thus return 1 from the okfn() to signal the skb is ok to pass
 351                  */
 352                 if (NF_HOOK(NFPROTO_BRIDGE, NF_BR_LOCAL_IN,
 353                             dev_net(skb->dev), NULL, skb, skb->dev, NULL,
 354                             br_handle_local_finish) == 1) {
 355                         return RX_HANDLER_PASS;
 356                 } else {
 357                         return RX_HANDLER_CONSUMED;
 358                 }
 359         }
 360
 361         if (unlikely(br_process_frame_type(p, skb)))
 362                 return RX_HANDLER_PASS;
 363
 364 forward:
 365         switch (p->state) {
 366         case BR_STATE_FORWARDING:
 367         case BR_STATE_LEARNING:
 368                 if (ether_addr_equal(p->br->dev->dev_addr, dest))
 369                         skb->pkt_type = PACKET_HOST;
 370
 371                 return nf_hook_bridge_pre(skb, pskb);
 372         default:
 373 drop:
 374                 kfree_skb(skb);
 375         }
 376         return RX_HANDLER_CONSUMED;
 377 }
 378
 379 /* This function has no purpose other than to appease the br_port_get_rcu/rtnl
 380  * helpers which identify bridged ports according to the rx_handler installed
 381  * on them (so there _needs_ to be a bridge rx_handler even if we don't need it
 382  * to do anything useful). This bridge won't support traffic to/from the stack,
 383  * but only hardware bridging. So return RX_HANDLER_PASS so we don't steal
 384  * frames from the ETH_P_XDSA packet_type handler.
 385  */
 386 static rx_handler_result_t br_handle_frame_dummy(struct sk_buff **pskb)
 387 {
 388         return RX_HANDLER_PASS;
 389 }
 390
 391 rx_handler_func_t *br_get_rx_handler(const struct net_device *dev)
 392 {
 393         if (netdev_uses_dsa(dev))
 394                 return br_handle_frame_dummy;
 395
 396         return br_handle_frame;
 397 }
 398
 399 void br_add_frame(struct net_bridge *br, struct br_frame_type *ft)
 400 {
 401         hlist_add_head_rcu(&ft->list, &br->frame_type_list);
 402 }
 403
 404 void br_del_frame(struct net_bridge *br, struct br_frame_type *ft)
 405 {
 406         struct br_frame_type *tmp;
 407
 408         hlist_for_each_entry(tmp, &br->frame_type_list, list)
 409                 if (ft == tmp) {
 410                         hlist_del_rcu(&ft->list);
 411                         return;
 412                 }
 413 }