xref: /linux/net/bridge/br_input.c (revision 0c8a32eed1625a65798286fb73fea8710a908545)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *	Handle incoming frames
4  *	Linux ethernet bridge
5  *
6  *	Authors:
7  *	Lennert Buytenhek		<buytenh@gnu.org>
8  */
9 
10 #include <linux/slab.h>
11 #include <linux/kernel.h>
12 #include <linux/netdevice.h>
13 #include <linux/etherdevice.h>
14 #include <linux/netfilter_bridge.h>
15 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE
16 #include <net/netfilter/nf_queue.h>
17 #endif
18 #include <linux/neighbour.h>
19 #include <net/arp.h>
20 #include <net/dsa.h>
21 #include <linux/export.h>
22 #include <linux/rculist.h>
23 #include "br_private.h"
24 #include "br_private_tunnel.h"
25 
26 static int
27 br_netif_receive_skb(struct net *net, struct sock *sk, struct sk_buff *skb)
28 {
29 	br_drop_fake_rtable(skb);
30 	return netif_receive_skb(skb);
31 }
32 
33 static int br_pass_frame_up(struct sk_buff *skb)
34 {
35 	struct net_device *indev, *brdev = BR_INPUT_SKB_CB(skb)->brdev;
36 	struct net_bridge *br = netdev_priv(brdev);
37 	struct net_bridge_vlan_group *vg;
38 
39 	dev_sw_netstats_rx_add(brdev, skb->len);
40 
41 	vg = br_vlan_group_rcu(br);
42 	/* Bridge is just like any other port.  Make sure the
43 	 * packet is allowed except in promisc modue when someone
44 	 * may be running packet capture.
45 	 */
46 	if (!(brdev->flags & IFF_PROMISC) &&
47 	    !br_allowed_egress(vg, skb)) {
48 		kfree_skb(skb);
49 		return NET_RX_DROP;
50 	}
51 
52 	indev = skb->dev;
53 	skb->dev = brdev;
54 	skb = br_handle_vlan(br, NULL, vg, skb);
55 	if (!skb)
56 		return NET_RX_DROP;
57 	/* update the multicast stats if the packet is IGMP/MLD */
58 	br_multicast_count(br, NULL, skb, br_multicast_igmp_type(skb),
59 			   BR_MCAST_DIR_TX);
60 
61 	return NF_HOOK(NFPROTO_BRIDGE, NF_BR_LOCAL_IN,
62 		       dev_net(indev), NULL, skb, indev, NULL,
63 		       br_netif_receive_skb);
64 }
65 
66 /* note: already called with rcu_read_lock */
67 int br_handle_frame_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
68 {
69 	struct net_bridge_port *p = br_port_get_rcu(skb->dev);
70 	enum br_pkt_type pkt_type = BR_PKT_UNICAST;
71 	struct net_bridge_fdb_entry *dst = NULL;
72 	struct net_bridge_mdb_entry *mdst;
73 	bool local_rcv, mcast_hit = false;
74 	struct net_bridge *br;
75 	u16 vid = 0;
76 	u8 state;
77 
78 	if (!p || p->state == BR_STATE_DISABLED)
79 		goto drop;
80 
81 	state = p->state;
82 	if (!br_allowed_ingress(p->br, nbp_vlan_group_rcu(p), skb, &vid,
83 				&state))
84 		goto out;
85 
86 	nbp_switchdev_frame_mark(p, skb);
87 
88 	/* insert into forwarding database after filtering to avoid spoofing */
89 	br = p->br;
90 	if (p->flags & BR_LEARNING)
91 		br_fdb_update(br, p, eth_hdr(skb)->h_source, vid, 0);
92 
93 	local_rcv = !!(br->dev->flags & IFF_PROMISC);
94 	if (is_multicast_ether_addr(eth_hdr(skb)->h_dest)) {
95 		/* by definition the broadcast is also a multicast address */
96 		if (is_broadcast_ether_addr(eth_hdr(skb)->h_dest)) {
97 			pkt_type = BR_PKT_BROADCAST;
98 			local_rcv = true;
99 		} else {
100 			pkt_type = BR_PKT_MULTICAST;
101 			if (br_multicast_rcv(br, p, skb, vid))
102 				goto drop;
103 		}
104 	}
105 
106 	if (state == BR_STATE_LEARNING)
107 		goto drop;
108 
109 	BR_INPUT_SKB_CB(skb)->brdev = br->dev;
110 	BR_INPUT_SKB_CB(skb)->src_port_isolated = !!(p->flags & BR_ISOLATED);
111 
112 	if (IS_ENABLED(CONFIG_INET) &&
113 	    (skb->protocol == htons(ETH_P_ARP) ||
114 	     skb->protocol == htons(ETH_P_RARP))) {
115 		br_do_proxy_suppress_arp(skb, br, vid, p);
116 	} else if (IS_ENABLED(CONFIG_IPV6) &&
117 		   skb->protocol == htons(ETH_P_IPV6) &&
118 		   br_opt_get(br, BROPT_NEIGH_SUPPRESS_ENABLED) &&
119 		   pskb_may_pull(skb, sizeof(struct ipv6hdr) +
120 				 sizeof(struct nd_msg)) &&
121 		   ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
122 			struct nd_msg *msg, _msg;
123 
124 			msg = br_is_nd_neigh_msg(skb, &_msg);
125 			if (msg)
126 				br_do_suppress_nd(skb, br, vid, p, msg);
127 	}
128 
129 	switch (pkt_type) {
130 	case BR_PKT_MULTICAST:
131 		mdst = br_mdb_get(br, skb, vid);
132 		if ((mdst || BR_INPUT_SKB_CB_MROUTERS_ONLY(skb)) &&
133 		    br_multicast_querier_exists(br, eth_hdr(skb), mdst)) {
134 			if ((mdst && mdst->host_joined) ||
135 			    br_multicast_is_router(br)) {
136 				local_rcv = true;
137 				br->dev->stats.multicast++;
138 			}
139 			mcast_hit = true;
140 		} else {
141 			local_rcv = true;
142 			br->dev->stats.multicast++;
143 		}
144 		break;
145 	case BR_PKT_UNICAST:
146 		dst = br_fdb_find_rcu(br, eth_hdr(skb)->h_dest, vid);
147 	default:
148 		break;
149 	}
150 
151 	if (dst) {
152 		unsigned long now = jiffies;
153 
154 		if (test_bit(BR_FDB_LOCAL, &dst->flags))
155 			return br_pass_frame_up(skb);
156 
157 		if (now != dst->used)
158 			dst->used = now;
159 		br_forward(dst->dst, skb, local_rcv, false);
160 	} else {
161 		if (!mcast_hit)
162 			br_flood(br, skb, pkt_type, local_rcv, false);
163 		else
164 			br_multicast_flood(mdst, skb, local_rcv, false);
165 	}
166 
167 	if (local_rcv)
168 		return br_pass_frame_up(skb);
169 
170 out:
171 	return 0;
172 drop:
173 	kfree_skb(skb);
174 	goto out;
175 }
176 EXPORT_SYMBOL_GPL(br_handle_frame_finish);
177 
178 static void __br_handle_local_finish(struct sk_buff *skb)
179 {
180 	struct net_bridge_port *p = br_port_get_rcu(skb->dev);
181 	u16 vid = 0;
182 
183 	/* check if vlan is allowed, to avoid spoofing */
184 	if ((p->flags & BR_LEARNING) &&
185 	    nbp_state_should_learn(p) &&
186 	    !br_opt_get(p->br, BROPT_NO_LL_LEARN) &&
187 	    br_should_learn(p, skb, &vid))
188 		br_fdb_update(p->br, p, eth_hdr(skb)->h_source, vid, 0);
189 }
190 
191 /* note: already called with rcu_read_lock */
192 static int br_handle_local_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
193 {
194 	__br_handle_local_finish(skb);
195 
196 	/* return 1 to signal the okfn() was called so it's ok to use the skb */
197 	return 1;
198 }
199 
200 static int nf_hook_bridge_pre(struct sk_buff *skb, struct sk_buff **pskb)
201 {
202 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE
203 	struct nf_hook_entries *e = NULL;
204 	struct nf_hook_state state;
205 	unsigned int verdict, i;
206 	struct net *net;
207 	int ret;
208 
209 	net = dev_net(skb->dev);
210 #ifdef HAVE_JUMP_LABEL
211 	if (!static_key_false(&nf_hooks_needed[NFPROTO_BRIDGE][NF_BR_PRE_ROUTING]))
212 		goto frame_finish;
213 #endif
214 
215 	e = rcu_dereference(net->nf.hooks_bridge[NF_BR_PRE_ROUTING]);
216 	if (!e)
217 		goto frame_finish;
218 
219 	nf_hook_state_init(&state, NF_BR_PRE_ROUTING,
220 			   NFPROTO_BRIDGE, skb->dev, NULL, NULL,
221 			   net, br_handle_frame_finish);
222 
223 	for (i = 0; i < e->num_hook_entries; i++) {
224 		verdict = nf_hook_entry_hookfn(&e->hooks[i], skb, &state);
225 		switch (verdict & NF_VERDICT_MASK) {
226 		case NF_ACCEPT:
227 			if (BR_INPUT_SKB_CB(skb)->br_netfilter_broute) {
228 				*pskb = skb;
229 				return RX_HANDLER_PASS;
230 			}
231 			break;
232 		case NF_DROP:
233 			kfree_skb(skb);
234 			return RX_HANDLER_CONSUMED;
235 		case NF_QUEUE:
236 			ret = nf_queue(skb, &state, i, verdict);
237 			if (ret == 1)
238 				continue;
239 			return RX_HANDLER_CONSUMED;
240 		default: /* STOLEN */
241 			return RX_HANDLER_CONSUMED;
242 		}
243 	}
244 frame_finish:
245 	net = dev_net(skb->dev);
246 	br_handle_frame_finish(net, NULL, skb);
247 #else
248 	br_handle_frame_finish(dev_net(skb->dev), NULL, skb);
249 #endif
250 	return RX_HANDLER_CONSUMED;
251 }
252 
253 /* Return 0 if the frame was not processed otherwise 1
254  * note: already called with rcu_read_lock
255  */
256 static int br_process_frame_type(struct net_bridge_port *p,
257 				 struct sk_buff *skb)
258 {
259 	struct br_frame_type *tmp;
260 
261 	hlist_for_each_entry_rcu(tmp, &p->br->frame_type_list, list)
262 		if (unlikely(tmp->type == skb->protocol))
263 			return tmp->frame_handler(p, skb);
264 
265 	return 0;
266 }
267 
268 /*
269  * Return NULL if skb is handled
270  * note: already called with rcu_read_lock
271  */
272 static rx_handler_result_t br_handle_frame(struct sk_buff **pskb)
273 {
274 	struct net_bridge_port *p;
275 	struct sk_buff *skb = *pskb;
276 	const unsigned char *dest = eth_hdr(skb)->h_dest;
277 
278 	if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
279 		return RX_HANDLER_PASS;
280 
281 	if (!is_valid_ether_addr(eth_hdr(skb)->h_source))
282 		goto drop;
283 
284 	skb = skb_share_check(skb, GFP_ATOMIC);
285 	if (!skb)
286 		return RX_HANDLER_CONSUMED;
287 
288 	memset(skb->cb, 0, sizeof(struct br_input_skb_cb));
289 
290 	p = br_port_get_rcu(skb->dev);
291 	if (p->flags & BR_VLAN_TUNNEL) {
292 		if (br_handle_ingress_vlan_tunnel(skb, p,
293 						  nbp_vlan_group_rcu(p)))
294 			goto drop;
295 	}
296 
297 	if (unlikely(is_link_local_ether_addr(dest))) {
298 		u16 fwd_mask = p->br->group_fwd_mask_required;
299 
300 		/*
301 		 * See IEEE 802.1D Table 7-10 Reserved addresses
302 		 *
303 		 * Assignment		 		Value
304 		 * Bridge Group Address		01-80-C2-00-00-00
305 		 * (MAC Control) 802.3		01-80-C2-00-00-01
306 		 * (Link Aggregation) 802.3	01-80-C2-00-00-02
307 		 * 802.1X PAE address		01-80-C2-00-00-03
308 		 *
309 		 * 802.1AB LLDP 		01-80-C2-00-00-0E
310 		 *
311 		 * Others reserved for future standardization
312 		 */
313 		fwd_mask |= p->group_fwd_mask;
314 		switch (dest[5]) {
315 		case 0x00:	/* Bridge Group Address */
316 			/* If STP is turned off,
317 			   then must forward to keep loop detection */
318 			if (p->br->stp_enabled == BR_NO_STP ||
319 			    fwd_mask & (1u << dest[5]))
320 				goto forward;
321 			*pskb = skb;
322 			__br_handle_local_finish(skb);
323 			return RX_HANDLER_PASS;
324 
325 		case 0x01:	/* IEEE MAC (Pause) */
326 			goto drop;
327 
328 		case 0x0E:	/* 802.1AB LLDP */
329 			fwd_mask |= p->br->group_fwd_mask;
330 			if (fwd_mask & (1u << dest[5]))
331 				goto forward;
332 			*pskb = skb;
333 			__br_handle_local_finish(skb);
334 			return RX_HANDLER_PASS;
335 
336 		default:
337 			/* Allow selective forwarding for most other protocols */
338 			fwd_mask |= p->br->group_fwd_mask;
339 			if (fwd_mask & (1u << dest[5]))
340 				goto forward;
341 		}
342 
343 		/* The else clause should be hit when nf_hook():
344 		 *   - returns < 0 (drop/error)
345 		 *   - returns = 0 (stolen/nf_queue)
346 		 * Thus return 1 from the okfn() to signal the skb is ok to pass
347 		 */
348 		if (NF_HOOK(NFPROTO_BRIDGE, NF_BR_LOCAL_IN,
349 			    dev_net(skb->dev), NULL, skb, skb->dev, NULL,
350 			    br_handle_local_finish) == 1) {
351 			return RX_HANDLER_PASS;
352 		} else {
353 			return RX_HANDLER_CONSUMED;
354 		}
355 	}
356 
357 	if (unlikely(br_process_frame_type(p, skb)))
358 		return RX_HANDLER_PASS;
359 
360 forward:
361 	switch (p->state) {
362 	case BR_STATE_FORWARDING:
363 	case BR_STATE_LEARNING:
364 		if (ether_addr_equal(p->br->dev->dev_addr, dest))
365 			skb->pkt_type = PACKET_HOST;
366 
367 		return nf_hook_bridge_pre(skb, pskb);
368 	default:
369 drop:
370 		kfree_skb(skb);
371 	}
372 	return RX_HANDLER_CONSUMED;
373 }
374 
375 /* This function has no purpose other than to appease the br_port_get_rcu/rtnl
376  * helpers which identify bridged ports according to the rx_handler installed
377  * on them (so there _needs_ to be a bridge rx_handler even if we don't need it
378  * to do anything useful). This bridge won't support traffic to/from the stack,
379  * but only hardware bridging. So return RX_HANDLER_PASS so we don't steal
380  * frames from the ETH_P_XDSA packet_type handler.
381  */
382 static rx_handler_result_t br_handle_frame_dummy(struct sk_buff **pskb)
383 {
384 	return RX_HANDLER_PASS;
385 }
386 
387 rx_handler_func_t *br_get_rx_handler(const struct net_device *dev)
388 {
389 	if (netdev_uses_dsa(dev))
390 		return br_handle_frame_dummy;
391 
392 	return br_handle_frame;
393 }
394 
395 void br_add_frame(struct net_bridge *br, struct br_frame_type *ft)
396 {
397 	hlist_add_head_rcu(&ft->list, &br->frame_type_list);
398 }
399 
400 void br_del_frame(struct net_bridge *br, struct br_frame_type *ft)
401 {
402 	struct br_frame_type *tmp;
403 
404 	hlist_for_each_entry(tmp, &br->frame_type_list, list)
405 		if (ft == tmp) {
406 			hlist_del_rcu(&ft->list);
407 			return;
408 		}
409 }
410