xref: /openbmc/linux/net/bridge/br_forward.c (revision 5a170e9e)
1 /*
2  *	Forwarding decision
3  *	Linux ethernet bridge
4  *
5  *	Authors:
6  *	Lennert Buytenhek		<buytenh@gnu.org>
7  *
8  *	This program is free software; you can redistribute it and/or
9  *	modify it under the terms of the GNU General Public License
10  *	as published by the Free Software Foundation; either version
11  *	2 of the License, or (at your option) any later version.
12  */
13 
14 #include <linux/err.h>
15 #include <linux/slab.h>
16 #include <linux/kernel.h>
17 #include <linux/netdevice.h>
18 #include <linux/netpoll.h>
19 #include <linux/skbuff.h>
20 #include <linux/if_vlan.h>
21 #include <linux/netfilter_bridge.h>
22 #include "br_private.h"
23 
24 /* Don't forward packets to originating port or forwarding disabled */
25 static inline int should_deliver(const struct net_bridge_port *p,
26 				 const struct sk_buff *skb)
27 {
28 	struct net_bridge_vlan_group *vg;
29 
30 	vg = nbp_vlan_group_rcu(p);
31 	return ((p->flags & BR_HAIRPIN_MODE) || skb->dev != p->dev) &&
32 		br_allowed_egress(vg, skb) && p->state == BR_STATE_FORWARDING &&
33 		nbp_switchdev_allowed_egress(p, skb) &&
34 		!br_skb_isolated(p, skb);
35 }
36 
37 int br_dev_queue_push_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
38 {
39 	if (!is_skb_forwardable(skb->dev, skb))
40 		goto drop;
41 
42 	skb_push(skb, ETH_HLEN);
43 	br_drop_fake_rtable(skb);
44 
45 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
46 	    (skb->protocol == htons(ETH_P_8021Q) ||
47 	     skb->protocol == htons(ETH_P_8021AD))) {
48 		int depth;
49 
50 		if (!__vlan_get_protocol(skb, skb->protocol, &depth))
51 			goto drop;
52 
53 		skb_set_network_header(skb, depth);
54 	}
55 
56 	dev_queue_xmit(skb);
57 
58 	return 0;
59 
60 drop:
61 	kfree_skb(skb);
62 	return 0;
63 }
64 EXPORT_SYMBOL_GPL(br_dev_queue_push_xmit);
65 
66 int br_forward_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
67 {
68 	return NF_HOOK(NFPROTO_BRIDGE, NF_BR_POST_ROUTING,
69 		       net, sk, skb, NULL, skb->dev,
70 		       br_dev_queue_push_xmit);
71 
72 }
73 EXPORT_SYMBOL_GPL(br_forward_finish);
74 
75 static void __br_forward(const struct net_bridge_port *to,
76 			 struct sk_buff *skb, bool local_orig)
77 {
78 	struct net_bridge_vlan_group *vg;
79 	struct net_device *indev;
80 	struct net *net;
81 	int br_hook;
82 
83 	vg = nbp_vlan_group_rcu(to);
84 	skb = br_handle_vlan(to->br, to, vg, skb);
85 	if (!skb)
86 		return;
87 
88 	indev = skb->dev;
89 	skb->dev = to->dev;
90 	if (!local_orig) {
91 		if (skb_warn_if_lro(skb)) {
92 			kfree_skb(skb);
93 			return;
94 		}
95 		br_hook = NF_BR_FORWARD;
96 		skb_forward_csum(skb);
97 		net = dev_net(indev);
98 	} else {
99 		if (unlikely(netpoll_tx_running(to->br->dev))) {
100 			if (!is_skb_forwardable(skb->dev, skb)) {
101 				kfree_skb(skb);
102 			} else {
103 				skb_push(skb, ETH_HLEN);
104 				br_netpoll_send_skb(to, skb);
105 			}
106 			return;
107 		}
108 		br_hook = NF_BR_LOCAL_OUT;
109 		net = dev_net(skb->dev);
110 		indev = NULL;
111 	}
112 
113 	NF_HOOK(NFPROTO_BRIDGE, br_hook,
114 		net, NULL, skb, indev, skb->dev,
115 		br_forward_finish);
116 }
117 
118 static int deliver_clone(const struct net_bridge_port *prev,
119 			 struct sk_buff *skb, bool local_orig)
120 {
121 	struct net_device *dev = BR_INPUT_SKB_CB(skb)->brdev;
122 
123 	skb = skb_clone(skb, GFP_ATOMIC);
124 	if (!skb) {
125 		dev->stats.tx_dropped++;
126 		return -ENOMEM;
127 	}
128 
129 	__br_forward(prev, skb, local_orig);
130 	return 0;
131 }
132 
133 /**
134  * br_forward - forward a packet to a specific port
135  * @to: destination port
136  * @skb: packet being forwarded
137  * @local_rcv: packet will be received locally after forwarding
138  * @local_orig: packet is locally originated
139  *
140  * Should be called with rcu_read_lock.
141  */
142 void br_forward(const struct net_bridge_port *to,
143 		struct sk_buff *skb, bool local_rcv, bool local_orig)
144 {
145 	if (unlikely(!to))
146 		goto out;
147 
148 	/* redirect to backup link if the destination port is down */
149 	if (rcu_access_pointer(to->backup_port) && !netif_carrier_ok(to->dev)) {
150 		struct net_bridge_port *backup_port;
151 
152 		backup_port = rcu_dereference(to->backup_port);
153 		if (unlikely(!backup_port))
154 			goto out;
155 		to = backup_port;
156 	}
157 
158 	if (should_deliver(to, skb)) {
159 		if (local_rcv)
160 			deliver_clone(to, skb, local_orig);
161 		else
162 			__br_forward(to, skb, local_orig);
163 		return;
164 	}
165 
166 out:
167 	if (!local_rcv)
168 		kfree_skb(skb);
169 }
170 EXPORT_SYMBOL_GPL(br_forward);
171 
172 static struct net_bridge_port *maybe_deliver(
173 	struct net_bridge_port *prev, struct net_bridge_port *p,
174 	struct sk_buff *skb, bool local_orig)
175 {
176 	int err;
177 
178 	if (!should_deliver(p, skb))
179 		return prev;
180 
181 	if (!prev)
182 		goto out;
183 
184 	err = deliver_clone(prev, skb, local_orig);
185 	if (err)
186 		return ERR_PTR(err);
187 
188 out:
189 	return p;
190 }
191 
192 /* called under rcu_read_lock */
193 void br_flood(struct net_bridge *br, struct sk_buff *skb,
194 	      enum br_pkt_type pkt_type, bool local_rcv, bool local_orig)
195 {
196 	u8 igmp_type = br_multicast_igmp_type(skb);
197 	struct net_bridge_port *prev = NULL;
198 	struct net_bridge_port *p;
199 
200 	list_for_each_entry_rcu(p, &br->port_list, list) {
201 		/* Do not flood unicast traffic to ports that turn it off, nor
202 		 * other traffic if flood off, except for traffic we originate
203 		 */
204 		switch (pkt_type) {
205 		case BR_PKT_UNICAST:
206 			if (!(p->flags & BR_FLOOD))
207 				continue;
208 			break;
209 		case BR_PKT_MULTICAST:
210 			if (!(p->flags & BR_MCAST_FLOOD) && skb->dev != br->dev)
211 				continue;
212 			break;
213 		case BR_PKT_BROADCAST:
214 			if (!(p->flags & BR_BCAST_FLOOD) && skb->dev != br->dev)
215 				continue;
216 			break;
217 		}
218 
219 		/* Do not flood to ports that enable proxy ARP */
220 		if (p->flags & BR_PROXYARP)
221 			continue;
222 		if ((p->flags & (BR_PROXYARP_WIFI | BR_NEIGH_SUPPRESS)) &&
223 		    BR_INPUT_SKB_CB(skb)->proxyarp_replied)
224 			continue;
225 
226 		prev = maybe_deliver(prev, p, skb, local_orig);
227 		if (IS_ERR(prev))
228 			goto out;
229 		if (prev == p)
230 			br_multicast_count(p->br, p, skb, igmp_type,
231 					   BR_MCAST_DIR_TX);
232 	}
233 
234 	if (!prev)
235 		goto out;
236 
237 	if (local_rcv)
238 		deliver_clone(prev, skb, local_orig);
239 	else
240 		__br_forward(prev, skb, local_orig);
241 	return;
242 
243 out:
244 	if (!local_rcv)
245 		kfree_skb(skb);
246 }
247 
248 #ifdef CONFIG_BRIDGE_IGMP_SNOOPING
249 static void maybe_deliver_addr(struct net_bridge_port *p, struct sk_buff *skb,
250 			       const unsigned char *addr, bool local_orig)
251 {
252 	struct net_device *dev = BR_INPUT_SKB_CB(skb)->brdev;
253 	const unsigned char *src = eth_hdr(skb)->h_source;
254 
255 	if (!should_deliver(p, skb))
256 		return;
257 
258 	/* Even with hairpin, no soliloquies - prevent breaking IPv6 DAD */
259 	if (skb->dev == p->dev && ether_addr_equal(src, addr))
260 		return;
261 
262 	skb = skb_copy(skb, GFP_ATOMIC);
263 	if (!skb) {
264 		dev->stats.tx_dropped++;
265 		return;
266 	}
267 
268 	if (!is_broadcast_ether_addr(addr))
269 		memcpy(eth_hdr(skb)->h_dest, addr, ETH_ALEN);
270 
271 	__br_forward(p, skb, local_orig);
272 }
273 
274 /* called with rcu_read_lock */
275 void br_multicast_flood(struct net_bridge_mdb_entry *mdst,
276 			struct sk_buff *skb,
277 			bool local_rcv, bool local_orig)
278 {
279 	struct net_device *dev = BR_INPUT_SKB_CB(skb)->brdev;
280 	u8 igmp_type = br_multicast_igmp_type(skb);
281 	struct net_bridge *br = netdev_priv(dev);
282 	struct net_bridge_port *prev = NULL;
283 	struct net_bridge_port_group *p;
284 	struct hlist_node *rp;
285 
286 	rp = rcu_dereference(hlist_first_rcu(&br->router_list));
287 	p = mdst ? rcu_dereference(mdst->ports) : NULL;
288 	while (p || rp) {
289 		struct net_bridge_port *port, *lport, *rport;
290 
291 		lport = p ? p->port : NULL;
292 		rport = hlist_entry_safe(rp, struct net_bridge_port, rlist);
293 
294 		if ((unsigned long)lport > (unsigned long)rport) {
295 			port = lport;
296 
297 			if (port->flags & BR_MULTICAST_TO_UNICAST) {
298 				maybe_deliver_addr(lport, skb, p->eth_addr,
299 						   local_orig);
300 				goto delivered;
301 			}
302 		} else {
303 			port = rport;
304 		}
305 
306 		prev = maybe_deliver(prev, port, skb, local_orig);
307 delivered:
308 		if (IS_ERR(prev))
309 			goto out;
310 		if (prev == port)
311 			br_multicast_count(port->br, port, skb, igmp_type,
312 					   BR_MCAST_DIR_TX);
313 
314 		if ((unsigned long)lport >= (unsigned long)port)
315 			p = rcu_dereference(p->next);
316 		if ((unsigned long)rport >= (unsigned long)port)
317 			rp = rcu_dereference(hlist_next_rcu(rp));
318 	}
319 
320 	if (!prev)
321 		goto out;
322 
323 	if (local_rcv)
324 		deliver_clone(prev, skb, local_orig);
325 	else
326 		__br_forward(prev, skb, local_orig);
327 	return;
328 
329 out:
330 	if (!local_rcv)
331 		kfree_skb(skb);
332 }
333 #endif
334