net/bridge/br_forward.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Forwarding decision
   4  *      Linux ethernet bridge
   5  *
   6  *      Authors:
   7  *      Lennert Buytenhek               <buytenh@gnu.org>
   8  */
   9
  10 #include <linux/err.h>
  11 #include <linux/slab.h>
  12 #include <linux/kernel.h>
  13 #include <linux/netdevice.h>
  14 #include <linux/netpoll.h>
  15 #include <linux/skbuff.h>
  16 #include <linux/if_vlan.h>
  17 #include <linux/netfilter_bridge.h>
  18 #include "br_private.h"
  19
  20 /* Don't forward packets to originating port or forwarding disabled */
  21 static inline int should_deliver(const struct net_bridge_port *p,
  22                                  const struct sk_buff *skb)
  23 {
  24         struct net_bridge_vlan_group *vg;
  25
  26         vg = nbp_vlan_group_rcu(p);
  27         return ((p->flags & BR_HAIRPIN_MODE) || skb->dev != p->dev) &&
  28                 p->state == BR_STATE_FORWARDING && br_allowed_egress(vg, skb) &&
  29                 nbp_switchdev_allowed_egress(p, skb) &&
  30                 !br_skb_isolated(p, skb);
  31 }
  32
  33 int br_dev_queue_push_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
  34 {
  35         skb_push(skb, ETH_HLEN);
  36         if (!is_skb_forwardable(skb->dev, skb))
  37                 goto drop;
  38
  39         br_drop_fake_rtable(skb);
  40
  41         if (skb->ip_summed == CHECKSUM_PARTIAL &&
  42             eth_type_vlan(skb->protocol)) {
  43                 int depth;
  44
  45                 if (!vlan_get_protocol_and_depth(skb, skb->protocol, &depth))
  46                         goto drop;
  47
  48                 skb_set_network_header(skb, depth);
  49         }
  50
  51         br_switchdev_frame_set_offload_fwd_mark(skb);
  52
  53         dev_queue_xmit(skb);
  54
  55         return 0;
  56
  57 drop:
  58         kfree_skb(skb);
  59         return 0;
  60 }
  61 EXPORT_SYMBOL_GPL(br_dev_queue_push_xmit);
  62
  63 int br_forward_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
  64 {
  65         skb_clear_tstamp(skb);
  66         return NF_HOOK(NFPROTO_BRIDGE, NF_BR_POST_ROUTING,
  67                        net, sk, skb, NULL, skb->dev,
  68                        br_dev_queue_push_xmit);
  69
  70 }
  71 EXPORT_SYMBOL_GPL(br_forward_finish);
  72
  73 static void __br_forward(const struct net_bridge_port *to,
  74                          struct sk_buff *skb, bool local_orig)
  75 {
  76         struct net_bridge_vlan_group *vg;
  77         struct net_device *indev;
  78         struct net *net;
  79         int br_hook;
  80
  81         /* Mark the skb for forwarding offload early so that br_handle_vlan()
  82          * can know whether to pop the VLAN header on egress or keep it.
  83          */
  84         nbp_switchdev_frame_mark_tx_fwd_offload(to, skb);
  85
  86         vg = nbp_vlan_group_rcu(to);
  87         skb = br_handle_vlan(to->br, to, vg, skb);
  88         if (!skb)
  89                 return;
  90
  91         indev = skb->dev;
  92         skb->dev = to->dev;
  93         if (!local_orig) {
  94                 if (skb_warn_if_lro(skb)) {
  95                         kfree_skb(skb);
  96                         return;
  97                 }
  98                 br_hook = NF_BR_FORWARD;
  99                 skb_forward_csum(skb);
 100                 net = dev_net(indev);
 101         } else {
 102                 if (unlikely(netpoll_tx_running(to->br->dev))) {
 103                         skb_push(skb, ETH_HLEN);
 104                         if (!is_skb_forwardable(skb->dev, skb))
 105                                 kfree_skb(skb);
 106                         else
 107                                 br_netpoll_send_skb(to, skb);
 108                         return;
 109                 }
 110                 br_hook = NF_BR_LOCAL_OUT;
 111                 net = dev_net(skb->dev);
 112                 indev = NULL;
 113         }
 114
 115         NF_HOOK(NFPROTO_BRIDGE, br_hook,
 116                 net, NULL, skb, indev, skb->dev,
 117                 br_forward_finish);
 118 }
 119
 120 static int deliver_clone(const struct net_bridge_port *prev,
 121                          struct sk_buff *skb, bool local_orig)
 122 {
 123         struct net_device *dev = BR_INPUT_SKB_CB(skb)->brdev;
 124
 125         skb = skb_clone(skb, GFP_ATOMIC);
 126         if (!skb) {
 127                 DEV_STATS_INC(dev, tx_dropped);
 128                 return -ENOMEM;
 129         }
 130
 131         __br_forward(prev, skb, local_orig);
 132         return 0;
 133 }
 134
 135 /**
 136  * br_forward - forward a packet to a specific port
 137  * @to: destination port
 138  * @skb: packet being forwarded
 139  * @local_rcv: packet will be received locally after forwarding
 140  * @local_orig: packet is locally originated
 141  *
 142  * Should be called with rcu_read_lock.
 143  */
 144 void br_forward(const struct net_bridge_port *to,
 145                 struct sk_buff *skb, bool local_rcv, bool local_orig)
 146 {
 147         if (unlikely(!to))
 148                 goto out;
 149
 150         /* redirect to backup link if the destination port is down */
 151         if (rcu_access_pointer(to->backup_port) && !netif_carrier_ok(to->dev)) {
 152                 struct net_bridge_port *backup_port;
 153
 154                 backup_port = rcu_dereference(to->backup_port);
 155                 if (unlikely(!backup_port))
 156                         goto out;
 157                 BR_INPUT_SKB_CB(skb)->backup_nhid = READ_ONCE(to->backup_nhid);
 158                 to = backup_port;
 159         }
 160
 161         if (should_deliver(to, skb)) {
 162                 if (local_rcv)
 163                         deliver_clone(to, skb, local_orig);
 164                 else
 165                         __br_forward(to, skb, local_orig);
 166                 return;
 167         }
 168
 169 out:
 170         if (!local_rcv)
 171                 kfree_skb(skb);
 172 }
 173 EXPORT_SYMBOL_GPL(br_forward);
 174
 175 static struct net_bridge_port *maybe_deliver(
 176         struct net_bridge_port *prev, struct net_bridge_port *p,
 177         struct sk_buff *skb, bool local_orig)
 178 {
 179         u8 igmp_type = br_multicast_igmp_type(skb);
 180         int err;
 181
 182         if (!should_deliver(p, skb))
 183                 return prev;
 184
 185         nbp_switchdev_frame_mark_tx_fwd_to_hwdom(p, skb);
 186
 187         if (!prev)
 188                 goto out;
 189
 190         err = deliver_clone(prev, skb, local_orig);
 191         if (err)
 192                 return ERR_PTR(err);
 193 out:
 194         br_multicast_count(p->br, p, skb, igmp_type, BR_MCAST_DIR_TX);
 195
 196         return p;
 197 }
 198
 199 /* called under rcu_read_lock */
 200 void br_flood(struct net_bridge *br, struct sk_buff *skb,
 201               enum br_pkt_type pkt_type, bool local_rcv, bool local_orig,
 202               u16 vid)
 203 {
 204         struct net_bridge_port *prev = NULL;
 205         struct net_bridge_port *p;
 206
 207         br_tc_skb_miss_set(skb, pkt_type != BR_PKT_BROADCAST);
 208
 209         list_for_each_entry_rcu(p, &br->port_list, list) {
 210                 /* Do not flood unicast traffic to ports that turn it off, nor
 211                  * other traffic if flood off, except for traffic we originate
 212                  */
 213                 switch (pkt_type) {
 214                 case BR_PKT_UNICAST:
 215                         if (!(p->flags & BR_FLOOD))
 216                                 continue;
 217                         break;
 218                 case BR_PKT_MULTICAST:
 219                         if (!(p->flags & BR_MCAST_FLOOD) && skb->dev != br->dev)
 220                                 continue;
 221                         break;
 222                 case BR_PKT_BROADCAST:
 223                         if (!(p->flags & BR_BCAST_FLOOD) && skb->dev != br->dev)
 224                                 continue;
 225                         break;
 226                 }
 227
 228                 /* Do not flood to ports that enable proxy ARP */
 229                 if (p->flags & BR_PROXYARP)
 230                         continue;
 231                 if (BR_INPUT_SKB_CB(skb)->proxyarp_replied &&
 232                     ((p->flags & BR_PROXYARP_WIFI) ||
 233                      br_is_neigh_suppress_enabled(p, vid)))
 234                         continue;
 235
 236                 prev = maybe_deliver(prev, p, skb, local_orig);
 237                 if (IS_ERR(prev))
 238                         goto out;
 239         }
 240
 241         if (!prev)
 242                 goto out;
 243
 244         if (local_rcv)
 245                 deliver_clone(prev, skb, local_orig);
 246         else
 247                 __br_forward(prev, skb, local_orig);
 248         return;
 249
 250 out:
 251         if (!local_rcv)
 252                 kfree_skb(skb);
 253 }
 254
 255 #ifdef CONFIG_BRIDGE_IGMP_SNOOPING
 256 static void maybe_deliver_addr(struct net_bridge_port *p, struct sk_buff *skb,
 257                                const unsigned char *addr, bool local_orig)
 258 {
 259         struct net_device *dev = BR_INPUT_SKB_CB(skb)->brdev;
 260         const unsigned char *src = eth_hdr(skb)->h_source;
 261
 262         if (!should_deliver(p, skb))
 263                 return;
 264
 265         /* Even with hairpin, no soliloquies - prevent breaking IPv6 DAD */
 266         if (skb->dev == p->dev && ether_addr_equal(src, addr))
 267                 return;
 268
 269         skb = skb_copy(skb, GFP_ATOMIC);
 270         if (!skb) {
 271                 DEV_STATS_INC(dev, tx_dropped);
 272                 return;
 273         }
 274
 275         if (!is_broadcast_ether_addr(addr))
 276                 memcpy(eth_hdr(skb)->h_dest, addr, ETH_ALEN);
 277
 278         __br_forward(p, skb, local_orig);
 279 }
 280
 281 /* called with rcu_read_lock */
 282 void br_multicast_flood(struct net_bridge_mdb_entry *mdst,
 283                         struct sk_buff *skb,
 284                         struct net_bridge_mcast *brmctx,
 285                         bool local_rcv, bool local_orig)
 286 {
 287         struct net_bridge_port *prev = NULL;
 288         struct net_bridge_port_group *p;
 289         bool allow_mode_include = true;
 290         struct hlist_node *rp;
 291
 292         rp = br_multicast_get_first_rport_node(brmctx, skb);
 293
 294         if (mdst) {
 295                 p = rcu_dereference(mdst->ports);
 296                 if (br_multicast_should_handle_mode(brmctx, mdst->addr.proto) &&
 297                     br_multicast_is_star_g(&mdst->addr))
 298                         allow_mode_include = false;
 299         } else {
 300                 p = NULL;
 301                 br_tc_skb_miss_set(skb, true);
 302         }
 303
 304         while (p || rp) {
 305                 struct net_bridge_port *port, *lport, *rport;
 306
 307                 lport = p ? p->key.port : NULL;
 308                 rport = br_multicast_rport_from_node_skb(rp, skb);
 309
 310                 if ((unsigned long)lport > (unsigned long)rport) {
 311                         port = lport;
 312
 313                         if (port->flags & BR_MULTICAST_TO_UNICAST) {
 314                                 maybe_deliver_addr(lport, skb, p->eth_addr,
 315                                                    local_orig);
 316                                 goto delivered;
 317                         }
 318                         if ((!allow_mode_include &&
 319                              p->filter_mode == MCAST_INCLUDE) ||
 320                             (p->flags & MDB_PG_FLAGS_BLOCKED))
 321                                 goto delivered;
 322                 } else {
 323                         port = rport;
 324                 }
 325
 326                 prev = maybe_deliver(prev, port, skb, local_orig);
 327                 if (IS_ERR(prev))
 328                         goto out;
 329 delivered:
 330                 if ((unsigned long)lport >= (unsigned long)port)
 331                         p = rcu_dereference(p->next);
 332                 if ((unsigned long)rport >= (unsigned long)port)
 333                         rp = rcu_dereference(hlist_next_rcu(rp));
 334         }
 335
 336         if (!prev)
 337                 goto out;
 338
 339         if (local_rcv)
 340                 deliver_clone(prev, skb, local_orig);
 341         else
 342                 __br_forward(prev, skb, local_orig);
 343         return;
 344
 345 out:
 346         if (!local_rcv)
 347                 kfree_skb(skb);
 348 }
 349 #endif