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, bool promisc)
  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
  39         dev_sw_netstats_rx_add(brdev, skb->len);
  40
  41         vg = br_vlan_group_rcu(br);
  42
  43         /* Reset the offload_fwd_mark because there could be a stacked
  44          * bridge above, and it should not think this bridge it doing
  45          * that bridge's work forwarding out its ports.
  46          */
  47         br_switchdev_frame_unmark(skb);
  48
  49         /* Bridge is just like any other port.  Make sure the
  50          * packet is allowed except in promisc mode when someone
  51          * may be running packet capture.
  52          */
  53         if (!(brdev->flags & IFF_PROMISC) &&
  54             !br_allowed_egress(vg, skb)) {
  55                 kfree_skb(skb);
  56                 return NET_RX_DROP;
  57         }
  58
  59         indev = skb->dev;
  60         skb->dev = brdev;
  61         skb = br_handle_vlan(br, NULL, vg, skb);
  62         if (!skb)
  63                 return NET_RX_DROP;
  64         /* update the multicast stats if the packet is IGMP/MLD */
  65         br_multicast_count(br, NULL, skb, br_multicast_igmp_type(skb),
  66                            BR_MCAST_DIR_TX);
  67
  68         BR_INPUT_SKB_CB(skb)->promisc = promisc;
  69
  70         return NF_HOOK(NFPROTO_BRIDGE, NF_BR_LOCAL_IN,
  71                        dev_net(indev), NULL, skb, indev, NULL,
  72                        br_netif_receive_skb);
  73 }
  74
  75 /* note: already called with rcu_read_lock */
  76 int br_handle_frame_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
  77 {
  78         struct net_bridge_port *p = br_port_get_rcu(skb->dev);
  79         enum br_pkt_type pkt_type = BR_PKT_UNICAST;
  80         struct net_bridge_fdb_entry *dst = NULL;
  81         struct net_bridge_mcast_port *pmctx;
  82         struct net_bridge_mdb_entry *mdst;
  83         bool local_rcv, mcast_hit = false;
  84         struct net_bridge_mcast *brmctx;
  85         struct net_bridge_vlan *vlan;
  86         struct net_bridge *br;
  87         bool promisc;
  88         u16 vid = 0;
  89         u8 state;
  90
  91         if (!p)
  92                 goto drop;
  93
  94         br = p->br;
  95
  96         if (br_mst_is_enabled(br)) {
  97                 state = BR_STATE_FORWARDING;
  98         } else {
  99                 if (p->state == BR_STATE_DISABLED)
 100                         goto drop;
 101
 102                 state = p->state;
 103         }
 104
 105         brmctx = &p->br->multicast_ctx;
 106         pmctx = &p->multicast_ctx;
 107         if (!br_allowed_ingress(p->br, nbp_vlan_group_rcu(p), skb, &vid,
 108                                 &state, &vlan))
 109                 goto out;
 110
 111         if (p->flags & BR_PORT_LOCKED) {
 112                 struct net_bridge_fdb_entry *fdb_src =
 113                         br_fdb_find_rcu(br, eth_hdr(skb)->h_source, vid);
 114
 115                 if (!fdb_src) {
 116                         /* FDB miss. Create locked FDB entry if MAB is enabled
 117                          * and drop the packet.
 118                          */
 119                         if (p->flags & BR_PORT_MAB)
 120                                 br_fdb_update(br, p, eth_hdr(skb)->h_source,
 121                                               vid, BIT(BR_FDB_LOCKED));
 122                         goto drop;
 123                 } else if (READ_ONCE(fdb_src->dst) != p ||
 124                            test_bit(BR_FDB_LOCAL, &fdb_src->flags)) {
 125                         /* FDB mismatch. Drop the packet without roaming. */
 126                         goto drop;
 127                 } else if (test_bit(BR_FDB_LOCKED, &fdb_src->flags)) {
 128                         /* FDB match, but entry is locked. Refresh it and drop
 129                          * the packet.
 130                          */
 131                         br_fdb_update(br, p, eth_hdr(skb)->h_source, vid,
 132                                       BIT(BR_FDB_LOCKED));
 133                         goto drop;
 134                 }
 135         }
 136
 137         nbp_switchdev_frame_mark(p, skb);
 138
 139         /* insert into forwarding database after filtering to avoid spoofing */
 140         if (p->flags & BR_LEARNING)
 141                 br_fdb_update(br, p, eth_hdr(skb)->h_source, vid, 0);
 142
 143         promisc = !!(br->dev->flags & IFF_PROMISC);
 144         local_rcv = promisc;
 145
 146         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest)) {
 147                 /* by definition the broadcast is also a multicast address */
 148                 if (is_broadcast_ether_addr(eth_hdr(skb)->h_dest)) {
 149                         pkt_type = BR_PKT_BROADCAST;
 150                         local_rcv = true;
 151                 } else {
 152                         pkt_type = BR_PKT_MULTICAST;
 153                         if (br_multicast_rcv(&brmctx, &pmctx, vlan, skb, vid))
 154                                 goto drop;
 155                 }
 156         }
 157
 158         if (state == BR_STATE_LEARNING)
 159                 goto drop;
 160
 161         BR_INPUT_SKB_CB(skb)->brdev = br->dev;
 162         BR_INPUT_SKB_CB(skb)->src_port_isolated = !!(p->flags & BR_ISOLATED);
 163
 164         if (IS_ENABLED(CONFIG_INET) &&
 165             (skb->protocol == htons(ETH_P_ARP) ||
 166              skb->protocol == htons(ETH_P_RARP))) {
 167                 br_do_proxy_suppress_arp(skb, br, vid, p);
 168         } else if (IS_ENABLED(CONFIG_IPV6) &&
 169                    skb->protocol == htons(ETH_P_IPV6) &&
 170                    br_opt_get(br, BROPT_NEIGH_SUPPRESS_ENABLED) &&
 171                    pskb_may_pull(skb, sizeof(struct ipv6hdr) +
 172                                  sizeof(struct nd_msg)) &&
 173                    ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
 174                         struct nd_msg *msg, _msg;
 175
 176                         msg = br_is_nd_neigh_msg(skb, &_msg);
 177                         if (msg)
 178                                 br_do_suppress_nd(skb, br, vid, p, msg);
 179         }
 180
 181         switch (pkt_type) {
 182         case BR_PKT_MULTICAST:
 183                 mdst = br_mdb_entry_skb_get(brmctx, skb, vid);
 184                 if ((mdst || BR_INPUT_SKB_CB_MROUTERS_ONLY(skb)) &&
 185                     br_multicast_querier_exists(brmctx, eth_hdr(skb), mdst)) {
 186                         if ((mdst && mdst->host_joined) ||
 187                             br_multicast_is_router(brmctx, skb)) {
 188                                 local_rcv = true;
 189                                 DEV_STATS_INC(br->dev, multicast);
 190                         }
 191                         mcast_hit = true;
 192                 } else {
 193                         local_rcv = true;
 194                         DEV_STATS_INC(br->dev, multicast);
 195                 }
 196                 break;
 197         case BR_PKT_UNICAST:
 198                 dst = br_fdb_find_rcu(br, eth_hdr(skb)->h_dest, vid);
 199                 break;
 200         default:
 201                 break;
 202         }
 203
 204         if (dst) {
 205                 unsigned long now = jiffies;
 206
 207                 if (test_bit(BR_FDB_LOCAL, &dst->flags))
 208                         return br_pass_frame_up(skb, false);
 209
 210                 if (now != dst->used)
 211                         dst->used = now;
 212                 br_forward(dst->dst, skb, local_rcv, false);
 213         } else {
 214                 if (!mcast_hit)
 215                         br_flood(br, skb, pkt_type, local_rcv, false, vid);
 216                 else
 217                         br_multicast_flood(mdst, skb, brmctx, local_rcv, false);
 218         }
 219
 220         if (local_rcv)
 221                 return br_pass_frame_up(skb, promisc);
 222
 223 out:
 224         return 0;
 225 drop:
 226         kfree_skb(skb);
 227         goto out;
 228 }
 229 EXPORT_SYMBOL_GPL(br_handle_frame_finish);
 230
 231 static void __br_handle_local_finish(struct sk_buff *skb)
 232 {
 233         struct net_bridge_port *p = br_port_get_rcu(skb->dev);
 234         u16 vid = 0;
 235
 236         /* check if vlan is allowed, to avoid spoofing */
 237         if ((p->flags & BR_LEARNING) &&
 238             nbp_state_should_learn(p) &&
 239             !br_opt_get(p->br, BROPT_NO_LL_LEARN) &&
 240             br_should_learn(p, skb, &vid))
 241                 br_fdb_update(p->br, p, eth_hdr(skb)->h_source, vid, 0);
 242 }
 243
 244 /* note: already called with rcu_read_lock */
 245 static int br_handle_local_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
 246 {
 247         __br_handle_local_finish(skb);
 248
 249         /* return 1 to signal the okfn() was called so it's ok to use the skb */
 250         return 1;
 251 }
 252
 253 static int nf_hook_bridge_pre(struct sk_buff *skb, struct sk_buff **pskb)
 254 {
 255 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE
 256         struct nf_hook_entries *e = NULL;
 257         struct nf_hook_state state;
 258         unsigned int verdict, i;
 259         struct net *net;
 260         int ret;
 261
 262         net = dev_net(skb->dev);
 263 #ifdef HAVE_JUMP_LABEL
 264         if (!static_key_false(&nf_hooks_needed[NFPROTO_BRIDGE][NF_BR_PRE_ROUTING]))
 265                 goto frame_finish;
 266 #endif
 267
 268         e = rcu_dereference(net->nf.hooks_bridge[NF_BR_PRE_ROUTING]);
 269         if (!e)
 270                 goto frame_finish;
 271
 272         nf_hook_state_init(&state, NF_BR_PRE_ROUTING,
 273                            NFPROTO_BRIDGE, skb->dev, NULL, NULL,
 274                            net, br_handle_frame_finish);
 275
 276         for (i = 0; i < e->num_hook_entries; i++) {
 277                 verdict = nf_hook_entry_hookfn(&e->hooks[i], skb, &state);
 278                 switch (verdict & NF_VERDICT_MASK) {
 279                 case NF_ACCEPT:
 280                         if (BR_INPUT_SKB_CB(skb)->br_netfilter_broute) {
 281                                 *pskb = skb;
 282                                 return RX_HANDLER_PASS;
 283                         }
 284                         break;
 285                 case NF_DROP:
 286                         kfree_skb(skb);
 287                         return RX_HANDLER_CONSUMED;
 288                 case NF_QUEUE:
 289                         ret = nf_queue(skb, &state, i, verdict);
 290                         if (ret == 1)
 291                                 continue;
 292                         return RX_HANDLER_CONSUMED;
 293                 default: /* STOLEN */
 294                         return RX_HANDLER_CONSUMED;
 295                 }
 296         }
 297 frame_finish:
 298         net = dev_net(skb->dev);
 299         br_handle_frame_finish(net, NULL, skb);
 300 #else
 301         br_handle_frame_finish(dev_net(skb->dev), NULL, skb);
 302 #endif
 303         return RX_HANDLER_CONSUMED;
 304 }
 305
 306 /* Return 0 if the frame was not processed otherwise 1
 307  * note: already called with rcu_read_lock
 308  */
 309 static int br_process_frame_type(struct net_bridge_port *p,
 310                                  struct sk_buff *skb)
 311 {
 312         struct br_frame_type *tmp;
 313
 314         hlist_for_each_entry_rcu(tmp, &p->br->frame_type_list, list)
 315                 if (unlikely(tmp->type == skb->protocol))
 316                         return tmp->frame_handler(p, skb);
 317
 318         return 0;
 319 }
 320
 321 /*
 322  * Return NULL if skb is handled
 323  * note: already called with rcu_read_lock
 324  */
 325 static rx_handler_result_t br_handle_frame(struct sk_buff **pskb)
 326 {
 327         struct net_bridge_port *p;
 328         struct sk_buff *skb = *pskb;
 329         const unsigned char *dest = eth_hdr(skb)->h_dest;
 330
 331         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
 332                 return RX_HANDLER_PASS;
 333
 334         if (!is_valid_ether_addr(eth_hdr(skb)->h_source))
 335                 goto drop;
 336
 337         skb = skb_share_check(skb, GFP_ATOMIC);
 338         if (!skb)
 339                 return RX_HANDLER_CONSUMED;
 340
 341         memset(skb->cb, 0, sizeof(struct br_input_skb_cb));
 342         br_tc_skb_miss_set(skb, false);
 343
 344         p = br_port_get_rcu(skb->dev);
 345         if (p->flags & BR_VLAN_TUNNEL)
 346                 br_handle_ingress_vlan_tunnel(skb, p, nbp_vlan_group_rcu(p));
 347
 348         if (unlikely(is_link_local_ether_addr(dest))) {
 349                 u16 fwd_mask = p->br->group_fwd_mask_required;
 350
 351                 /*
 352                  * See IEEE 802.1D Table 7-10 Reserved addresses
 353                  *
 354                  * Assignment                           Value
 355                  * Bridge Group Address         01-80-C2-00-00-00
 356                  * (MAC Control) 802.3          01-80-C2-00-00-01
 357                  * (Link Aggregation) 802.3     01-80-C2-00-00-02
 358                  * 802.1X PAE address           01-80-C2-00-00-03
 359                  *
 360                  * 802.1AB LLDP                 01-80-C2-00-00-0E
 361                  *
 362                  * Others reserved for future standardization
 363                  */
 364                 fwd_mask |= p->group_fwd_mask;
 365                 switch (dest[5]) {
 366                 case 0x00:      /* Bridge Group Address */
 367                         /* If STP is turned off,
 368                            then must forward to keep loop detection */
 369                         if (p->br->stp_enabled == BR_NO_STP ||
 370                             fwd_mask & (1u << dest[5]))
 371                                 goto forward;
 372                         *pskb = skb;
 373                         __br_handle_local_finish(skb);
 374                         return RX_HANDLER_PASS;
 375
 376                 case 0x01:      /* IEEE MAC (Pause) */
 377                         goto drop;
 378
 379                 case 0x0E:      /* 802.1AB LLDP */
 380                         fwd_mask |= p->br->group_fwd_mask;
 381                         if (fwd_mask & (1u << dest[5]))
 382                                 goto forward;
 383                         *pskb = skb;
 384                         __br_handle_local_finish(skb);
 385                         return RX_HANDLER_PASS;
 386
 387                 default:
 388                         /* Allow selective forwarding for most other protocols */
 389                         fwd_mask |= p->br->group_fwd_mask;
 390                         if (fwd_mask & (1u << dest[5]))
 391                                 goto forward;
 392                 }
 393
 394                 BR_INPUT_SKB_CB(skb)->promisc = false;
 395
 396                 /* The else clause should be hit when nf_hook():
 397                  *   - returns < 0 (drop/error)
 398                  *   - returns = 0 (stolen/nf_queue)
 399                  * Thus return 1 from the okfn() to signal the skb is ok to pass
 400                  */
 401                 if (NF_HOOK(NFPROTO_BRIDGE, NF_BR_LOCAL_IN,
 402                             dev_net(skb->dev), NULL, skb, skb->dev, NULL,
 403                             br_handle_local_finish) == 1) {
 404                         return RX_HANDLER_PASS;
 405                 } else {
 406                         return RX_HANDLER_CONSUMED;
 407                 }
 408         }
 409
 410         if (unlikely(br_process_frame_type(p, skb)))
 411                 return RX_HANDLER_PASS;
 412
 413 forward:
 414         if (br_mst_is_enabled(p->br))
 415                 goto defer_stp_filtering;
 416
 417         switch (p->state) {
 418         case BR_STATE_FORWARDING:
 419         case BR_STATE_LEARNING:
 420 defer_stp_filtering:
 421                 if (ether_addr_equal(p->br->dev->dev_addr, dest))
 422                         skb->pkt_type = PACKET_HOST;
 423
 424                 return nf_hook_bridge_pre(skb, pskb);
 425         default:
 426 drop:
 427                 kfree_skb(skb);
 428         }
 429         return RX_HANDLER_CONSUMED;
 430 }
 431
 432 /* This function has no purpose other than to appease the br_port_get_rcu/rtnl
 433  * helpers which identify bridged ports according to the rx_handler installed
 434  * on them (so there _needs_ to be a bridge rx_handler even if we don't need it
 435  * to do anything useful). This bridge won't support traffic to/from the stack,
 436  * but only hardware bridging. So return RX_HANDLER_PASS so we don't steal
 437  * frames from the ETH_P_XDSA packet_type handler.
 438  */
 439 static rx_handler_result_t br_handle_frame_dummy(struct sk_buff **pskb)
 440 {
 441         return RX_HANDLER_PASS;
 442 }
 443
 444 rx_handler_func_t *br_get_rx_handler(const struct net_device *dev)
 445 {
 446         if (netdev_uses_dsa(dev))
 447                 return br_handle_frame_dummy;
 448
 449         return br_handle_frame;
 450 }
 451
 452 void br_add_frame(struct net_bridge *br, struct br_frame_type *ft)
 453 {
 454         hlist_add_head_rcu(&ft->list, &br->frame_type_list);
 455 }
 456
 457 void br_del_frame(struct net_bridge *br, struct br_frame_type *ft)
 458 {
 459         struct br_frame_type *tmp;
 460
 461         hlist_for_each_entry(tmp, &br->frame_type_list, list)
 462                 if (ft == tmp) {
 463                         hlist_del_rcu(&ft->list);
 464                         return;
 465                 }
 466 }