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 <linux/export.h>
  21 #include <linux/rculist.h>
  22 #include "br_private.h"
  23 #include "br_private_tunnel.h"
  24
  25 static int
  26 br_netif_receive_skb(struct net *net, struct sock *sk, struct sk_buff *skb)
  27 {
  28         br_drop_fake_rtable(skb);
  29         return netif_receive_skb(skb);
  30 }
  31
  32 static int br_pass_frame_up(struct sk_buff *skb)
  33 {
  34         struct net_device *indev, *brdev = BR_INPUT_SKB_CB(skb)->brdev;
  35         struct net_bridge *br = netdev_priv(brdev);
  36         struct net_bridge_vlan_group *vg;
  37         struct pcpu_sw_netstats *brstats = this_cpu_ptr(br->stats);
  38
  39         u64_stats_update_begin(&brstats->syncp);
  40         brstats->rx_packets++;
  41         brstats->rx_bytes += skb->len;
  42         u64_stats_update_end(&brstats->syncp);
  43
  44         vg = br_vlan_group_rcu(br);
  45
  46         /* Reset the offload_fwd_mark because there could be a stacked
  47          * bridge above, and it should not think this bridge it doing
  48          * that bridge's work forwarding out its ports.
  49          */
  50         br_switchdev_frame_unmark(skb);
  51
  52         /* Bridge is just like any other port.  Make sure the
  53          * packet is allowed except in promisc modue when someone
  54          * may be running packet capture.
  55          */
  56         if (!(brdev->flags & IFF_PROMISC) &&
  57             !br_allowed_egress(vg, skb)) {
  58                 kfree_skb(skb);
  59                 return NET_RX_DROP;
  60         }
  61
  62         indev = skb->dev;
  63         skb->dev = brdev;
  64         skb = br_handle_vlan(br, NULL, vg, skb);
  65         if (!skb)
  66                 return NET_RX_DROP;
  67         /* update the multicast stats if the packet is IGMP/MLD */
  68         br_multicast_count(br, NULL, skb, br_multicast_igmp_type(skb),
  69                            BR_MCAST_DIR_TX);
  70
  71         return NF_HOOK(NFPROTO_BRIDGE, NF_BR_LOCAL_IN,
  72                        dev_net(indev), NULL, skb, indev, NULL,
  73                        br_netif_receive_skb);
  74 }
  75
  76 /* note: already called with rcu_read_lock */
  77 int br_handle_frame_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
  78 {
  79         struct net_bridge_port *p = br_port_get_rcu(skb->dev);
  80         enum br_pkt_type pkt_type = BR_PKT_UNICAST;
  81         struct net_bridge_fdb_entry *dst = NULL;
  82         struct net_bridge_mdb_entry *mdst;
  83         bool local_rcv, mcast_hit = false;
  84         struct net_bridge *br;
  85         u16 vid = 0;
  86
  87         if (!p || p->state == BR_STATE_DISABLED)
  88                 goto drop;
  89
  90         if (!br_allowed_ingress(p->br, nbp_vlan_group_rcu(p), skb, &vid))
  91                 goto out;
  92
  93         nbp_switchdev_frame_mark(p, skb);
  94
  95         /* insert into forwarding database after filtering to avoid spoofing */
  96         br = p->br;
  97         if (p->flags & BR_LEARNING)
  98                 br_fdb_update(br, p, eth_hdr(skb)->h_source, vid, false);
  99
 100         local_rcv = !!(br->dev->flags & IFF_PROMISC);
 101         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest)) {
 102                 /* by definition the broadcast is also a multicast address */
 103                 if (is_broadcast_ether_addr(eth_hdr(skb)->h_dest)) {
 104                         pkt_type = BR_PKT_BROADCAST;
 105                         local_rcv = true;
 106                 } else {
 107                         pkt_type = BR_PKT_MULTICAST;
 108                         if (br_multicast_rcv(br, p, skb, vid))
 109                                 goto drop;
 110                 }
 111         }
 112
 113         if (p->state == BR_STATE_LEARNING)
 114                 goto drop;
 115
 116         BR_INPUT_SKB_CB(skb)->brdev = br->dev;
 117         BR_INPUT_SKB_CB(skb)->src_port_isolated = !!(p->flags & BR_ISOLATED);
 118
 119         if (IS_ENABLED(CONFIG_INET) &&
 120             (skb->protocol == htons(ETH_P_ARP) ||
 121              skb->protocol == htons(ETH_P_RARP))) {
 122                 br_do_proxy_suppress_arp(skb, br, vid, p);
 123         } else if (IS_ENABLED(CONFIG_IPV6) &&
 124                    skb->protocol == htons(ETH_P_IPV6) &&
 125                    br_opt_get(br, BROPT_NEIGH_SUPPRESS_ENABLED) &&
 126                    pskb_may_pull(skb, sizeof(struct ipv6hdr) +
 127                                  sizeof(struct nd_msg)) &&
 128                    ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
 129                         struct nd_msg *msg, _msg;
 130
 131                         msg = br_is_nd_neigh_msg(skb, &_msg);
 132                         if (msg)
 133                                 br_do_suppress_nd(skb, br, vid, p, msg);
 134         }
 135
 136         switch (pkt_type) {
 137         case BR_PKT_MULTICAST:
 138                 mdst = br_mdb_get(br, skb, vid);
 139                 if ((mdst || BR_INPUT_SKB_CB_MROUTERS_ONLY(skb)) &&
 140                     br_multicast_querier_exists(br, eth_hdr(skb))) {
 141                         if ((mdst && mdst->host_joined) ||
 142                             br_multicast_is_router(br)) {
 143                                 local_rcv = true;
 144                                 br->dev->stats.multicast++;
 145                         }
 146                         mcast_hit = true;
 147                 } else {
 148                         local_rcv = true;
 149                         br->dev->stats.multicast++;
 150                 }
 151                 break;
 152         case BR_PKT_UNICAST:
 153                 dst = br_fdb_find_rcu(br, eth_hdr(skb)->h_dest, vid);
 154         default:
 155                 break;
 156         }
 157
 158         if (dst) {
 159                 unsigned long now = jiffies;
 160
 161                 if (dst->is_local)
 162                         return br_pass_frame_up(skb);
 163
 164                 if (now != dst->used)
 165                         dst->used = now;
 166                 br_forward(dst->dst, skb, local_rcv, false);
 167         } else {
 168                 if (!mcast_hit)
 169                         br_flood(br, skb, pkt_type, local_rcv, false);
 170                 else
 171                         br_multicast_flood(mdst, skb, local_rcv, false);
 172         }
 173
 174         if (local_rcv)
 175                 return br_pass_frame_up(skb);
 176
 177 out:
 178         return 0;
 179 drop:
 180         kfree_skb(skb);
 181         goto out;
 182 }
 183 EXPORT_SYMBOL_GPL(br_handle_frame_finish);
 184
 185 static void __br_handle_local_finish(struct sk_buff *skb)
 186 {
 187         struct net_bridge_port *p = br_port_get_rcu(skb->dev);
 188         u16 vid = 0;
 189
 190         /* check if vlan is allowed, to avoid spoofing */
 191         if ((p->flags & BR_LEARNING) &&
 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, false);
 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 /*
 260  * Return NULL if skb is handled
 261  * note: already called with rcu_read_lock
 262  */
 263 rx_handler_result_t br_handle_frame(struct sk_buff **pskb)
 264 {
 265         struct net_bridge_port *p;
 266         struct sk_buff *skb = *pskb;
 267         const unsigned char *dest = eth_hdr(skb)->h_dest;
 268
 269         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
 270                 return RX_HANDLER_PASS;
 271
 272         if (!is_valid_ether_addr(eth_hdr(skb)->h_source))
 273                 goto drop;
 274
 275         skb = skb_share_check(skb, GFP_ATOMIC);
 276         if (!skb)
 277                 return RX_HANDLER_CONSUMED;
 278
 279         memset(skb->cb, 0, sizeof(struct br_input_skb_cb));
 280
 281         p = br_port_get_rcu(skb->dev);
 282         if (p->flags & BR_VLAN_TUNNEL) {
 283                 if (br_handle_ingress_vlan_tunnel(skb, p,
 284                                                   nbp_vlan_group_rcu(p)))
 285                         goto drop;
 286         }
 287
 288         if (unlikely(is_link_local_ether_addr(dest))) {
 289                 u16 fwd_mask = p->br->group_fwd_mask_required;
 290
 291                 /*
 292                  * See IEEE 802.1D Table 7-10 Reserved addresses
 293                  *
 294                  * Assignment                           Value
 295                  * Bridge Group Address         01-80-C2-00-00-00
 296                  * (MAC Control) 802.3          01-80-C2-00-00-01
 297                  * (Link Aggregation) 802.3     01-80-C2-00-00-02
 298                  * 802.1X PAE address           01-80-C2-00-00-03
 299                  *
 300                  * 802.1AB LLDP                 01-80-C2-00-00-0E
 301                  *
 302                  * Others reserved for future standardization
 303                  */
 304                 fwd_mask |= p->group_fwd_mask;
 305                 switch (dest[5]) {
 306                 case 0x00:      /* Bridge Group Address */
 307                         /* If STP is turned off,
 308                            then must forward to keep loop detection */
 309                         if (p->br->stp_enabled == BR_NO_STP ||
 310                             fwd_mask & (1u << dest[5]))
 311                                 goto forward;
 312                         *pskb = skb;
 313                         __br_handle_local_finish(skb);
 314                         return RX_HANDLER_PASS;
 315
 316                 case 0x01:      /* IEEE MAC (Pause) */
 317                         goto drop;
 318
 319                 case 0x0E:      /* 802.1AB LLDP */
 320                         fwd_mask |= p->br->group_fwd_mask;
 321                         if (fwd_mask & (1u << dest[5]))
 322                                 goto forward;
 323                         *pskb = skb;
 324                         __br_handle_local_finish(skb);
 325                         return RX_HANDLER_PASS;
 326
 327                 default:
 328                         /* Allow selective forwarding for most other protocols */
 329                         fwd_mask |= p->br->group_fwd_mask;
 330                         if (fwd_mask & (1u << dest[5]))
 331                                 goto forward;
 332                 }
 333
 334                 /* The else clause should be hit when nf_hook():
 335                  *   - returns < 0 (drop/error)
 336                  *   - returns = 0 (stolen/nf_queue)
 337                  * Thus return 1 from the okfn() to signal the skb is ok to pass
 338                  */
 339                 if (NF_HOOK(NFPROTO_BRIDGE, NF_BR_LOCAL_IN,
 340                             dev_net(skb->dev), NULL, skb, skb->dev, NULL,
 341                             br_handle_local_finish) == 1) {
 342                         return RX_HANDLER_PASS;
 343                 } else {
 344                         return RX_HANDLER_CONSUMED;
 345                 }
 346         }
 347
 348 forward:
 349         switch (p->state) {
 350         case BR_STATE_FORWARDING:
 351         case BR_STATE_LEARNING:
 352                 if (ether_addr_equal(p->br->dev->dev_addr, dest))
 353                         skb->pkt_type = PACKET_HOST;
 354
 355                 return nf_hook_bridge_pre(skb, pskb);
 356         default:
 357 drop:
 358                 kfree_skb(skb);
 359         }
 360         return RX_HANDLER_CONSUMED;
 361 }