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