xref: /openbmc/linux/net/bridge/br_forward.c (revision 36de991e)
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(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->tstamp = 0;
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.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 		to = backup_port;
158 	}
159 
160 	if (should_deliver(to, skb)) {
161 		if (local_rcv)
162 			deliver_clone(to, skb, local_orig);
163 		else
164 			__br_forward(to, skb, local_orig);
165 		return;
166 	}
167 
168 out:
169 	if (!local_rcv)
170 		kfree_skb(skb);
171 }
172 EXPORT_SYMBOL_GPL(br_forward);
173 
174 static struct net_bridge_port *maybe_deliver(
175 	struct net_bridge_port *prev, struct net_bridge_port *p,
176 	struct sk_buff *skb, bool local_orig)
177 {
178 	u8 igmp_type = br_multicast_igmp_type(skb);
179 	int err;
180 
181 	if (!should_deliver(p, skb))
182 		return prev;
183 
184 	nbp_switchdev_frame_mark_tx_fwd_to_hwdom(p, skb);
185 
186 	if (!prev)
187 		goto out;
188 
189 	err = deliver_clone(prev, skb, local_orig);
190 	if (err)
191 		return ERR_PTR(err);
192 out:
193 	br_multicast_count(p->br, p, skb, igmp_type, BR_MCAST_DIR_TX);
194 
195 	return p;
196 }
197 
198 /* called under rcu_read_lock */
199 void br_flood(struct net_bridge *br, struct sk_buff *skb,
200 	      enum br_pkt_type pkt_type, bool local_rcv, bool local_orig)
201 {
202 	struct net_bridge_port *prev = NULL;
203 	struct net_bridge_port *p;
204 
205 	list_for_each_entry_rcu(p, &br->port_list, list) {
206 		/* Do not flood unicast traffic to ports that turn it off, nor
207 		 * other traffic if flood off, except for traffic we originate
208 		 */
209 		switch (pkt_type) {
210 		case BR_PKT_UNICAST:
211 			if (!(p->flags & BR_FLOOD))
212 				continue;
213 			break;
214 		case BR_PKT_MULTICAST:
215 			if (!(p->flags & BR_MCAST_FLOOD) && skb->dev != br->dev)
216 				continue;
217 			break;
218 		case BR_PKT_BROADCAST:
219 			if (!(p->flags & BR_BCAST_FLOOD) && skb->dev != br->dev)
220 				continue;
221 			break;
222 		}
223 
224 		/* Do not flood to ports that enable proxy ARP */
225 		if (p->flags & BR_PROXYARP)
226 			continue;
227 		if ((p->flags & (BR_PROXYARP_WIFI | BR_NEIGH_SUPPRESS)) &&
228 		    BR_INPUT_SKB_CB(skb)->proxyarp_replied)
229 			continue;
230 
231 		prev = maybe_deliver(prev, p, skb, local_orig);
232 		if (IS_ERR(prev))
233 			goto out;
234 	}
235 
236 	if (!prev)
237 		goto out;
238 
239 	if (local_rcv)
240 		deliver_clone(prev, skb, local_orig);
241 	else
242 		__br_forward(prev, skb, local_orig);
243 	return;
244 
245 out:
246 	if (!local_rcv)
247 		kfree_skb(skb);
248 }
249 
250 #ifdef CONFIG_BRIDGE_IGMP_SNOOPING
251 static void maybe_deliver_addr(struct net_bridge_port *p, struct sk_buff *skb,
252 			       const unsigned char *addr, bool local_orig)
253 {
254 	struct net_device *dev = BR_INPUT_SKB_CB(skb)->brdev;
255 	const unsigned char *src = eth_hdr(skb)->h_source;
256 
257 	if (!should_deliver(p, skb))
258 		return;
259 
260 	/* Even with hairpin, no soliloquies - prevent breaking IPv6 DAD */
261 	if (skb->dev == p->dev && ether_addr_equal(src, addr))
262 		return;
263 
264 	skb = skb_copy(skb, GFP_ATOMIC);
265 	if (!skb) {
266 		dev->stats.tx_dropped++;
267 		return;
268 	}
269 
270 	if (!is_broadcast_ether_addr(addr))
271 		memcpy(eth_hdr(skb)->h_dest, addr, ETH_ALEN);
272 
273 	__br_forward(p, skb, local_orig);
274 }
275 
276 /* called with rcu_read_lock */
277 void br_multicast_flood(struct net_bridge_mdb_entry *mdst,
278 			struct sk_buff *skb,
279 			struct net_bridge_mcast *brmctx,
280 			bool local_rcv, bool local_orig)
281 {
282 	struct net_bridge_port *prev = NULL;
283 	struct net_bridge_port_group *p;
284 	bool allow_mode_include = true;
285 	struct hlist_node *rp;
286 
287 	rp = br_multicast_get_first_rport_node(brmctx, skb);
288 
289 	if (mdst) {
290 		p = rcu_dereference(mdst->ports);
291 		if (br_multicast_should_handle_mode(brmctx, mdst->addr.proto) &&
292 		    br_multicast_is_star_g(&mdst->addr))
293 			allow_mode_include = false;
294 	} else {
295 		p = NULL;
296 	}
297 
298 	while (p || rp) {
299 		struct net_bridge_port *port, *lport, *rport;
300 
301 		lport = p ? p->key.port : NULL;
302 		rport = br_multicast_rport_from_node_skb(rp, skb);
303 
304 		if ((unsigned long)lport > (unsigned long)rport) {
305 			port = lport;
306 
307 			if (port->flags & BR_MULTICAST_TO_UNICAST) {
308 				maybe_deliver_addr(lport, skb, p->eth_addr,
309 						   local_orig);
310 				goto delivered;
311 			}
312 			if ((!allow_mode_include &&
313 			     p->filter_mode == MCAST_INCLUDE) ||
314 			    (p->flags & MDB_PG_FLAGS_BLOCKED))
315 				goto delivered;
316 		} else {
317 			port = rport;
318 		}
319 
320 		prev = maybe_deliver(prev, port, skb, local_orig);
321 		if (IS_ERR(prev))
322 			goto out;
323 delivered:
324 		if ((unsigned long)lport >= (unsigned long)port)
325 			p = rcu_dereference(p->next);
326 		if ((unsigned long)rport >= (unsigned long)port)
327 			rp = rcu_dereference(hlist_next_rcu(rp));
328 	}
329 
330 	if (!prev)
331 		goto out;
332 
333 	if (local_rcv)
334 		deliver_clone(prev, skb, local_orig);
335 	else
336 		__br_forward(prev, skb, local_orig);
337 	return;
338 
339 out:
340 	if (!local_rcv)
341 		kfree_skb(skb);
342 }
343 #endif
344