net/bridge/br_input.c

0001 // SPDX-License-Identifier: GPL-2.0-or-later
0002 /*
0003  *  Handle incoming frames
0004  *  Linux ethernet bridge
0005  *
0006  *  Authors:
0007  *  Lennert Buytenhek       <buytenh@gnu.org>
0008  */
0009
0010 #include <linux/slab.h>
0011 #include <linux/kernel.h>
0012 #include <linux/netdevice.h>
0013 #include <linux/etherdevice.h>
0014 #include <linux/netfilter_bridge.h>
0015 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE
0016 #include <net/netfilter/nf_queue.h>
0017 #endif
0018 #include <linux/neighbour.h>
0019 #include <net/arp.h>
0020 #include <net/dsa.h>
0021 #include <linux/export.h>
0022 #include <linux/rculist.h>
0023 #include "br_private.h"
0024 #include "br_private_tunnel.h"
0025
0026 static int
0027 br_netif_receive_skb(struct net *net, struct sock *sk, struct sk_buff *skb)
0028 {
0029     br_drop_fake_rtable(skb);
0030     return netif_receive_skb(skb);
0031 }
0032
0033 static int br_pass_frame_up(struct sk_buff *skb)
0034 {
0035     struct net_device *indev, *brdev = BR_INPUT_SKB_CB(skb)->brdev;
0036     struct net_bridge *br = netdev_priv(brdev);
0037     struct net_bridge_vlan_group *vg;
0038
0039     dev_sw_netstats_rx_add(brdev, skb->len);
0040
0041     vg = br_vlan_group_rcu(br);
0042
0043     /* Reset the offload_fwd_mark because there could be a stacked
0044      * bridge above, and it should not think this bridge it doing
0045      * that bridge's work forwarding out its ports.
0046      */
0047     br_switchdev_frame_unmark(skb);
0048
0049     /* Bridge is just like any other port.  Make sure the
0050      * packet is allowed except in promisc mode when someone
0051      * may be running packet capture.
0052      */
0053     if (!(brdev->flags & IFF_PROMISC) &&
0054         !br_allowed_egress(vg, skb)) {
0055         kfree_skb(skb);
0056         return NET_RX_DROP;
0057     }
0058
0059     indev = skb->dev;
0060     skb->dev = brdev;
0061     skb = br_handle_vlan(br, NULL, vg, skb);
0062     if (!skb)
0063         return NET_RX_DROP;
0064     /* update the multicast stats if the packet is IGMP/MLD */
0065     br_multicast_count(br, NULL, skb, br_multicast_igmp_type(skb),
0066                BR_MCAST_DIR_TX);
0067
0068     return NF_HOOK(NFPROTO_BRIDGE, NF_BR_LOCAL_IN,
0069                dev_net(indev), NULL, skb, indev, NULL,
0070                br_netif_receive_skb);
0071 }
0072
0073 /* note: already called with rcu_read_lock */
0074 int br_handle_frame_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
0075 {
0076     struct net_bridge_port *p = br_port_get_rcu(skb->dev);
0077     enum br_pkt_type pkt_type = BR_PKT_UNICAST;
0078     struct net_bridge_fdb_entry *dst = NULL;
0079     struct net_bridge_mcast_port *pmctx;
0080     struct net_bridge_mdb_entry *mdst;
0081     bool local_rcv, mcast_hit = false;
0082     struct net_bridge_mcast *brmctx;
0083     struct net_bridge_vlan *vlan;
0084     struct net_bridge *br;
0085     u16 vid = 0;
0086     u8 state;
0087
0088     if (!p)
0089         goto drop;
0090
0091     br = p->br;
0092
0093     if (br_mst_is_enabled(br)) {
0094         state = BR_STATE_FORWARDING;
0095     } else {
0096         if (p->state == BR_STATE_DISABLED)
0097             goto drop;
0098
0099         state = p->state;
0100     }
0101
0102     brmctx = &p->br->multicast_ctx;
0103     pmctx = &p->multicast_ctx;
0104     if (!br_allowed_ingress(p->br, nbp_vlan_group_rcu(p), skb, &vid,
0105                 &state, &vlan))
0106         goto out;
0107
0108     if (p->flags & BR_PORT_LOCKED) {
0109         struct net_bridge_fdb_entry *fdb_src =
0110             br_fdb_find_rcu(br, eth_hdr(skb)->h_source, vid);
0111
0112         if (!fdb_src || READ_ONCE(fdb_src->dst) != p ||
0113             test_bit(BR_FDB_LOCAL, &fdb_src->flags))
0114             goto drop;
0115     }
0116
0117     nbp_switchdev_frame_mark(p, skb);
0118
0119     /* insert into forwarding database after filtering to avoid spoofing */
0120     if (p->flags & BR_LEARNING)
0121         br_fdb_update(br, p, eth_hdr(skb)->h_source, vid, 0);
0122
0123     local_rcv = !!(br->dev->flags & IFF_PROMISC);
0124     if (is_multicast_ether_addr(eth_hdr(skb)->h_dest)) {
0125         /* by definition the broadcast is also a multicast address */
0126         if (is_broadcast_ether_addr(eth_hdr(skb)->h_dest)) {
0127             pkt_type = BR_PKT_BROADCAST;
0128             local_rcv = true;
0129         } else {
0130             pkt_type = BR_PKT_MULTICAST;
0131             if (br_multicast_rcv(&brmctx, &pmctx, vlan, skb, vid))
0132                 goto drop;
0133         }
0134     }
0135
0136     if (state == BR_STATE_LEARNING)
0137         goto drop;
0138
0139     BR_INPUT_SKB_CB(skb)->brdev = br->dev;
0140     BR_INPUT_SKB_CB(skb)->src_port_isolated = !!(p->flags & BR_ISOLATED);
0141
0142     if (IS_ENABLED(CONFIG_INET) &&
0143         (skb->protocol == htons(ETH_P_ARP) ||
0144          skb->protocol == htons(ETH_P_RARP))) {
0145         br_do_proxy_suppress_arp(skb, br, vid, p);
0146     } else if (IS_ENABLED(CONFIG_IPV6) &&
0147            skb->protocol == htons(ETH_P_IPV6) &&
0148            br_opt_get(br, BROPT_NEIGH_SUPPRESS_ENABLED) &&
0149            pskb_may_pull(skb, sizeof(struct ipv6hdr) +
0150                  sizeof(struct nd_msg)) &&
0151            ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
0152             struct nd_msg *msg, _msg;
0153
0154             msg = br_is_nd_neigh_msg(skb, &_msg);
0155             if (msg)
0156                 br_do_suppress_nd(skb, br, vid, p, msg);
0157     }
0158
0159     switch (pkt_type) {
0160     case BR_PKT_MULTICAST:
0161         mdst = br_mdb_get(brmctx, skb, vid);
0162         if ((mdst || BR_INPUT_SKB_CB_MROUTERS_ONLY(skb)) &&
0163             br_multicast_querier_exists(brmctx, eth_hdr(skb), mdst)) {
0164             if ((mdst && mdst->host_joined) ||
0165                 br_multicast_is_router(brmctx, skb)) {
0166                 local_rcv = true;
0167                 br->dev->stats.multicast++;
0168             }
0169             mcast_hit = true;
0170         } else {
0171             local_rcv = true;
0172             br->dev->stats.multicast++;
0173         }
0174         break;
0175     case BR_PKT_UNICAST:
0176         dst = br_fdb_find_rcu(br, eth_hdr(skb)->h_dest, vid);
0177         break;
0178     default:
0179         break;
0180     }
0181
0182     if (dst) {
0183         unsigned long now = jiffies;
0184
0185         if (test_bit(BR_FDB_LOCAL, &dst->flags))
0186             return br_pass_frame_up(skb);
0187
0188         if (now != dst->used)
0189             dst->used = now;
0190         br_forward(dst->dst, skb, local_rcv, false);
0191     } else {
0192         if (!mcast_hit)
0193             br_flood(br, skb, pkt_type, local_rcv, false);
0194         else
0195             br_multicast_flood(mdst, skb, brmctx, local_rcv, false);
0196     }
0197
0198     if (local_rcv)
0199         return br_pass_frame_up(skb);
0200
0201 out:
0202     return 0;
0203 drop:
0204     kfree_skb(skb);
0205     goto out;
0206 }
0207 EXPORT_SYMBOL_GPL(br_handle_frame_finish);
0208
0209 static void __br_handle_local_finish(struct sk_buff *skb)
0210 {
0211     struct net_bridge_port *p = br_port_get_rcu(skb->dev);
0212     u16 vid = 0;
0213
0214     /* check if vlan is allowed, to avoid spoofing */
0215     if ((p->flags & BR_LEARNING) &&
0216         nbp_state_should_learn(p) &&
0217         !br_opt_get(p->br, BROPT_NO_LL_LEARN) &&
0218         br_should_learn(p, skb, &vid))
0219         br_fdb_update(p->br, p, eth_hdr(skb)->h_source, vid, 0);
0220 }
0221
0222 /* note: already called with rcu_read_lock */
0223 static int br_handle_local_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
0224 {
0225     __br_handle_local_finish(skb);
0226
0227     /* return 1 to signal the okfn() was called so it's ok to use the skb */
0228     return 1;
0229 }
0230
0231 static int nf_hook_bridge_pre(struct sk_buff *skb, struct sk_buff **pskb)
0232 {
0233 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE
0234     struct nf_hook_entries *e = NULL;
0235     struct nf_hook_state state;
0236     unsigned int verdict, i;
0237     struct net *net;
0238     int ret;
0239
0240     net = dev_net(skb->dev);
0241 #ifdef HAVE_JUMP_LABEL
0242     if (!static_key_false(&nf_hooks_needed[NFPROTO_BRIDGE][NF_BR_PRE_ROUTING]))
0243         goto frame_finish;
0244 #endif
0245
0246     e = rcu_dereference(net->nf.hooks_bridge[NF_BR_PRE_ROUTING]);
0247     if (!e)
0248         goto frame_finish;
0249
0250     nf_hook_state_init(&state, NF_BR_PRE_ROUTING,
0251                NFPROTO_BRIDGE, skb->dev, NULL, NULL,
0252                net, br_handle_frame_finish);
0253
0254     for (i = 0; i < e->num_hook_entries; i++) {
0255         verdict = nf_hook_entry_hookfn(&e->hooks[i], skb, &state);
0256         switch (verdict & NF_VERDICT_MASK) {
0257         case NF_ACCEPT:
0258             if (BR_INPUT_SKB_CB(skb)->br_netfilter_broute) {
0259                 *pskb = skb;
0260                 return RX_HANDLER_PASS;
0261             }
0262             break;
0263         case NF_DROP:
0264             kfree_skb(skb);
0265             return RX_HANDLER_CONSUMED;
0266         case NF_QUEUE:
0267             ret = nf_queue(skb, &state, i, verdict);
0268             if (ret == 1)
0269                 continue;
0270             return RX_HANDLER_CONSUMED;
0271         default: /* STOLEN */
0272             return RX_HANDLER_CONSUMED;
0273         }
0274     }
0275 frame_finish:
0276     net = dev_net(skb->dev);
0277     br_handle_frame_finish(net, NULL, skb);
0278 #else
0279     br_handle_frame_finish(dev_net(skb->dev), NULL, skb);
0280 #endif
0281     return RX_HANDLER_CONSUMED;
0282 }
0283
0284 /* Return 0 if the frame was not processed otherwise 1
0285  * note: already called with rcu_read_lock
0286  */
0287 static int br_process_frame_type(struct net_bridge_port *p,
0288                  struct sk_buff *skb)
0289 {
0290     struct br_frame_type *tmp;
0291
0292     hlist_for_each_entry_rcu(tmp, &p->br->frame_type_list, list)
0293         if (unlikely(tmp->type == skb->protocol))
0294             return tmp->frame_handler(p, skb);
0295
0296     return 0;
0297 }
0298
0299 /*
0300  * Return NULL if skb is handled
0301  * note: already called with rcu_read_lock
0302  */
0303 static rx_handler_result_t br_handle_frame(struct sk_buff **pskb)
0304 {
0305     struct net_bridge_port *p;
0306     struct sk_buff *skb = *pskb;
0307     const unsigned char *dest = eth_hdr(skb)->h_dest;
0308
0309     if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
0310         return RX_HANDLER_PASS;
0311
0312     if (!is_valid_ether_addr(eth_hdr(skb)->h_source))
0313         goto drop;
0314
0315     skb = skb_share_check(skb, GFP_ATOMIC);
0316     if (!skb)
0317         return RX_HANDLER_CONSUMED;
0318
0319     memset(skb->cb, 0, sizeof(struct br_input_skb_cb));
0320
0321     p = br_port_get_rcu(skb->dev);
0322     if (p->flags & BR_VLAN_TUNNEL)
0323         br_handle_ingress_vlan_tunnel(skb, p, nbp_vlan_group_rcu(p));
0324
0325     if (unlikely(is_link_local_ether_addr(dest))) {
0326         u16 fwd_mask = p->br->group_fwd_mask_required;
0327
0328         /*
0329          * See IEEE 802.1D Table 7-10 Reserved addresses
0330          *
0331          * Assignment               Value
0332          * Bridge Group Address     01-80-C2-00-00-00
0333          * (MAC Control) 802.3      01-80-C2-00-00-01
0334          * (Link Aggregation) 802.3 01-80-C2-00-00-02
0335          * 802.1X PAE address       01-80-C2-00-00-03
0336          *
0337          * 802.1AB LLDP         01-80-C2-00-00-0E
0338          *
0339          * Others reserved for future standardization
0340          */
0341         fwd_mask |= p->group_fwd_mask;
0342         switch (dest[5]) {
0343         case 0x00:  /* Bridge Group Address */
0344             /* If STP is turned off,
0345                then must forward to keep loop detection */
0346             if (p->br->stp_enabled == BR_NO_STP ||
0347                 fwd_mask & (1u << dest[5]))
0348                 goto forward;
0349             *pskb = skb;
0350             __br_handle_local_finish(skb);
0351             return RX_HANDLER_PASS;
0352
0353         case 0x01:  /* IEEE MAC (Pause) */
0354             goto drop;
0355
0356         case 0x0E:  /* 802.1AB LLDP */
0357             fwd_mask |= p->br->group_fwd_mask;
0358             if (fwd_mask & (1u << dest[5]))
0359                 goto forward;
0360             *pskb = skb;
0361             __br_handle_local_finish(skb);
0362             return RX_HANDLER_PASS;
0363
0364         default:
0365             /* Allow selective forwarding for most other protocols */
0366             fwd_mask |= p->br->group_fwd_mask;
0367             if (fwd_mask & (1u << dest[5]))
0368                 goto forward;
0369         }
0370
0371         /* The else clause should be hit when nf_hook():
0372          *   - returns < 0 (drop/error)
0373          *   - returns = 0 (stolen/nf_queue)
0374          * Thus return 1 from the okfn() to signal the skb is ok to pass
0375          */
0376         if (NF_HOOK(NFPROTO_BRIDGE, NF_BR_LOCAL_IN,
0377                 dev_net(skb->dev), NULL, skb, skb->dev, NULL,
0378                 br_handle_local_finish) == 1) {
0379             return RX_HANDLER_PASS;
0380         } else {
0381             return RX_HANDLER_CONSUMED;
0382         }
0383     }
0384
0385     if (unlikely(br_process_frame_type(p, skb)))
0386         return RX_HANDLER_PASS;
0387
0388 forward:
0389     if (br_mst_is_enabled(p->br))
0390         goto defer_stp_filtering;
0391
0392     switch (p->state) {
0393     case BR_STATE_FORWARDING:
0394     case BR_STATE_LEARNING:
0395 defer_stp_filtering:
0396         if (ether_addr_equal(p->br->dev->dev_addr, dest))
0397             skb->pkt_type = PACKET_HOST;
0398
0399         return nf_hook_bridge_pre(skb, pskb);
0400     default:
0401 drop:
0402         kfree_skb(skb);
0403     }
0404     return RX_HANDLER_CONSUMED;
0405 }
0406
0407 /* This function has no purpose other than to appease the br_port_get_rcu/rtnl
0408  * helpers which identify bridged ports according to the rx_handler installed
0409  * on them (so there _needs_ to be a bridge rx_handler even if we don't need it
0410  * to do anything useful). This bridge won't support traffic to/from the stack,
0411  * but only hardware bridging. So return RX_HANDLER_PASS so we don't steal
0412  * frames from the ETH_P_XDSA packet_type handler.
0413  */
0414 static rx_handler_result_t br_handle_frame_dummy(struct sk_buff **pskb)
0415 {
0416     return RX_HANDLER_PASS;
0417 }
0418
0419 rx_handler_func_t *br_get_rx_handler(const struct net_device *dev)
0420 {
0421     if (netdev_uses_dsa(dev))
0422         return br_handle_frame_dummy;
0423
0424     return br_handle_frame;
0425 }
0426
0427 void br_add_frame(struct net_bridge *br, struct br_frame_type *ft)
0428 {
0429     hlist_add_head_rcu(&ft->list, &br->frame_type_list);
0430 }
0431
0432 void br_del_frame(struct net_bridge *br, struct br_frame_type *ft)
0433 {
0434     struct br_frame_type *tmp;
0435
0436     hlist_for_each_entry(tmp, &br->frame_type_list, list)
0437         if (ft == tmp) {
0438             hlist_del_rcu(&ft->list);
0439             return;
0440         }
0441 }