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