net/bridge/br_forward.c

0001 // SPDX-License-Identifier: GPL-2.0-or-later
0002 /*
0003  *  Forwarding decision
0004  *  Linux ethernet bridge
0005  *
0006  *  Authors:
0007  *  Lennert Buytenhek       <buytenh@gnu.org>
0008  */
0009
0010 #include <linux/err.h>
0011 #include <linux/slab.h>
0012 #include <linux/kernel.h>
0013 #include <linux/netdevice.h>
0014 #include <linux/netpoll.h>
0015 #include <linux/skbuff.h>
0016 #include <linux/if_vlan.h>
0017 #include <linux/netfilter_bridge.h>
0018 #include "br_private.h"
0019
0020 /* Don't forward packets to originating port or forwarding disabled */
0021 static inline int should_deliver(const struct net_bridge_port *p,
0022                  const struct sk_buff *skb)
0023 {
0024     struct net_bridge_vlan_group *vg;
0025
0026     vg = nbp_vlan_group_rcu(p);
0027     return ((p->flags & BR_HAIRPIN_MODE) || skb->dev != p->dev) &&
0028         p->state == BR_STATE_FORWARDING && br_allowed_egress(vg, skb) &&
0029         nbp_switchdev_allowed_egress(p, skb) &&
0030         !br_skb_isolated(p, skb);
0031 }
0032
0033 int br_dev_queue_push_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
0034 {
0035     skb_push(skb, ETH_HLEN);
0036     if (!is_skb_forwardable(skb->dev, skb))
0037         goto drop;
0038
0039     br_drop_fake_rtable(skb);
0040
0041     if (skb->ip_summed == CHECKSUM_PARTIAL &&
0042         eth_type_vlan(skb->protocol)) {
0043         int depth;
0044
0045         if (!__vlan_get_protocol(skb, skb->protocol, &depth))
0046             goto drop;
0047
0048         skb_set_network_header(skb, depth);
0049     }
0050
0051     br_switchdev_frame_set_offload_fwd_mark(skb);
0052
0053     dev_queue_xmit(skb);
0054
0055     return 0;
0056
0057 drop:
0058     kfree_skb(skb);
0059     return 0;
0060 }
0061 EXPORT_SYMBOL_GPL(br_dev_queue_push_xmit);
0062
0063 int br_forward_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
0064 {
0065     skb_clear_tstamp(skb);
0066     return NF_HOOK(NFPROTO_BRIDGE, NF_BR_POST_ROUTING,
0067                net, sk, skb, NULL, skb->dev,
0068                br_dev_queue_push_xmit);
0069
0070 }
0071 EXPORT_SYMBOL_GPL(br_forward_finish);
0072
0073 static void __br_forward(const struct net_bridge_port *to,
0074              struct sk_buff *skb, bool local_orig)
0075 {
0076     struct net_bridge_vlan_group *vg;
0077     struct net_device *indev;
0078     struct net *net;
0079     int br_hook;
0080
0081     /* Mark the skb for forwarding offload early so that br_handle_vlan()
0082      * can know whether to pop the VLAN header on egress or keep it.
0083      */
0084     nbp_switchdev_frame_mark_tx_fwd_offload(to, skb);
0085
0086     vg = nbp_vlan_group_rcu(to);
0087     skb = br_handle_vlan(to->br, to, vg, skb);
0088     if (!skb)
0089         return;
0090
0091     indev = skb->dev;
0092     skb->dev = to->dev;
0093     if (!local_orig) {
0094         if (skb_warn_if_lro(skb)) {
0095             kfree_skb(skb);
0096             return;
0097         }
0098         br_hook = NF_BR_FORWARD;
0099         skb_forward_csum(skb);
0100         net = dev_net(indev);
0101     } else {
0102         if (unlikely(netpoll_tx_running(to->br->dev))) {
0103             skb_push(skb, ETH_HLEN);
0104             if (!is_skb_forwardable(skb->dev, skb))
0105                 kfree_skb(skb);
0106             else
0107                 br_netpoll_send_skb(to, skb);
0108             return;
0109         }
0110         br_hook = NF_BR_LOCAL_OUT;
0111         net = dev_net(skb->dev);
0112         indev = NULL;
0113     }
0114
0115     NF_HOOK(NFPROTO_BRIDGE, br_hook,
0116         net, NULL, skb, indev, skb->dev,
0117         br_forward_finish);
0118 }
0119
0120 static int deliver_clone(const struct net_bridge_port *prev,
0121              struct sk_buff *skb, bool local_orig)
0122 {
0123     struct net_device *dev = BR_INPUT_SKB_CB(skb)->brdev;
0124
0125     skb = skb_clone(skb, GFP_ATOMIC);
0126     if (!skb) {
0127         dev->stats.tx_dropped++;
0128         return -ENOMEM;
0129     }
0130
0131     __br_forward(prev, skb, local_orig);
0132     return 0;
0133 }
0134
0135 /**
0136  * br_forward - forward a packet to a specific port
0137  * @to: destination port
0138  * @skb: packet being forwarded
0139  * @local_rcv: packet will be received locally after forwarding
0140  * @local_orig: packet is locally originated
0141  *
0142  * Should be called with rcu_read_lock.
0143  */
0144 void br_forward(const struct net_bridge_port *to,
0145         struct sk_buff *skb, bool local_rcv, bool local_orig)
0146 {
0147     if (unlikely(!to))
0148         goto out;
0149
0150     /* redirect to backup link if the destination port is down */
0151     if (rcu_access_pointer(to->backup_port) && !netif_carrier_ok(to->dev)) {
0152         struct net_bridge_port *backup_port;
0153
0154         backup_port = rcu_dereference(to->backup_port);
0155         if (unlikely(!backup_port))
0156             goto out;
0157         to = backup_port;
0158     }
0159
0160     if (should_deliver(to, skb)) {
0161         if (local_rcv)
0162             deliver_clone(to, skb, local_orig);
0163         else
0164             __br_forward(to, skb, local_orig);
0165         return;
0166     }
0167
0168 out:
0169     if (!local_rcv)
0170         kfree_skb(skb);
0171 }
0172 EXPORT_SYMBOL_GPL(br_forward);
0173
0174 static struct net_bridge_port *maybe_deliver(
0175     struct net_bridge_port *prev, struct net_bridge_port *p,
0176     struct sk_buff *skb, bool local_orig)
0177 {
0178     u8 igmp_type = br_multicast_igmp_type(skb);
0179     int err;
0180
0181     if (!should_deliver(p, skb))
0182         return prev;
0183
0184     nbp_switchdev_frame_mark_tx_fwd_to_hwdom(p, skb);
0185
0186     if (!prev)
0187         goto out;
0188
0189     err = deliver_clone(prev, skb, local_orig);
0190     if (err)
0191         return ERR_PTR(err);
0192 out:
0193     br_multicast_count(p->br, p, skb, igmp_type, BR_MCAST_DIR_TX);
0194
0195     return p;
0196 }
0197
0198 /* called under rcu_read_lock */
0199 void br_flood(struct net_bridge *br, struct sk_buff *skb,
0200           enum br_pkt_type pkt_type, bool local_rcv, bool local_orig)
0201 {
0202     struct net_bridge_port *prev = NULL;
0203     struct net_bridge_port *p;
0204
0205     list_for_each_entry_rcu(p, &br->port_list, list) {
0206         /* Do not flood unicast traffic to ports that turn it off, nor
0207          * other traffic if flood off, except for traffic we originate
0208          */
0209         switch (pkt_type) {
0210         case BR_PKT_UNICAST:
0211             if (!(p->flags & BR_FLOOD))
0212                 continue;
0213             break;
0214         case BR_PKT_MULTICAST:
0215             if (!(p->flags & BR_MCAST_FLOOD) && skb->dev != br->dev)
0216                 continue;
0217             break;
0218         case BR_PKT_BROADCAST:
0219             if (!(p->flags & BR_BCAST_FLOOD) && skb->dev != br->dev)
0220                 continue;
0221             break;
0222         }
0223
0224         /* Do not flood to ports that enable proxy ARP */
0225         if (p->flags & BR_PROXYARP)
0226             continue;
0227         if ((p->flags & (BR_PROXYARP_WIFI | BR_NEIGH_SUPPRESS)) &&
0228             BR_INPUT_SKB_CB(skb)->proxyarp_replied)
0229             continue;
0230
0231         prev = maybe_deliver(prev, p, skb, local_orig);
0232         if (IS_ERR(prev))
0233             goto out;
0234     }
0235
0236     if (!prev)
0237         goto out;
0238
0239     if (local_rcv)
0240         deliver_clone(prev, skb, local_orig);
0241     else
0242         __br_forward(prev, skb, local_orig);
0243     return;
0244
0245 out:
0246     if (!local_rcv)
0247         kfree_skb(skb);
0248 }
0249
0250 #ifdef CONFIG_BRIDGE_IGMP_SNOOPING
0251 static void maybe_deliver_addr(struct net_bridge_port *p, struct sk_buff *skb,
0252                    const unsigned char *addr, bool local_orig)
0253 {
0254     struct net_device *dev = BR_INPUT_SKB_CB(skb)->brdev;
0255     const unsigned char *src = eth_hdr(skb)->h_source;
0256
0257     if (!should_deliver(p, skb))
0258         return;
0259
0260     /* Even with hairpin, no soliloquies - prevent breaking IPv6 DAD */
0261     if (skb->dev == p->dev && ether_addr_equal(src, addr))
0262         return;
0263
0264     skb = skb_copy(skb, GFP_ATOMIC);
0265     if (!skb) {
0266         dev->stats.tx_dropped++;
0267         return;
0268     }
0269
0270     if (!is_broadcast_ether_addr(addr))
0271         memcpy(eth_hdr(skb)->h_dest, addr, ETH_ALEN);
0272
0273     __br_forward(p, skb, local_orig);
0274 }
0275
0276 /* called with rcu_read_lock */
0277 void br_multicast_flood(struct net_bridge_mdb_entry *mdst,
0278             struct sk_buff *skb,
0279             struct net_bridge_mcast *brmctx,
0280             bool local_rcv, bool local_orig)
0281 {
0282     struct net_bridge_port *prev = NULL;
0283     struct net_bridge_port_group *p;
0284     bool allow_mode_include = true;
0285     struct hlist_node *rp;
0286
0287     rp = br_multicast_get_first_rport_node(brmctx, skb);
0288
0289     if (mdst) {
0290         p = rcu_dereference(mdst->ports);
0291         if (br_multicast_should_handle_mode(brmctx, mdst->addr.proto) &&
0292             br_multicast_is_star_g(&mdst->addr))
0293             allow_mode_include = false;
0294     } else {
0295         p = NULL;
0296     }
0297
0298     while (p || rp) {
0299         struct net_bridge_port *port, *lport, *rport;
0300
0301         lport = p ? p->key.port : NULL;
0302         rport = br_multicast_rport_from_node_skb(rp, skb);
0303
0304         if ((unsigned long)lport > (unsigned long)rport) {
0305             port = lport;
0306
0307             if (port->flags & BR_MULTICAST_TO_UNICAST) {
0308                 maybe_deliver_addr(lport, skb, p->eth_addr,
0309                            local_orig);
0310                 goto delivered;
0311             }
0312             if ((!allow_mode_include &&
0313                  p->filter_mode == MCAST_INCLUDE) ||
0314                 (p->flags & MDB_PG_FLAGS_BLOCKED))
0315                 goto delivered;
0316         } else {
0317             port = rport;
0318         }
0319
0320         prev = maybe_deliver(prev, port, skb, local_orig);
0321         if (IS_ERR(prev))
0322             goto out;
0323 delivered:
0324         if ((unsigned long)lport >= (unsigned long)port)
0325             p = rcu_dereference(p->next);
0326         if ((unsigned long)rport >= (unsigned long)port)
0327             rp = rcu_dereference(hlist_next_rcu(rp));
0328     }
0329
0330     if (!prev)
0331         goto out;
0332
0333     if (local_rcv)
0334         deliver_clone(prev, skb, local_orig);
0335     else
0336         __br_forward(prev, skb, local_orig);
0337     return;
0338
0339 out:
0340     if (!local_rcv)
0341         kfree_skb(skb);
0342 }
0343 #endif