xref: /linux/net/bridge/br_input.c (revision 3a755ebcc2557e22b895b8976257f682c653db1d)
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 
43 	/* Reset the offload_fwd_mark because there could be a stacked
44 	 * bridge above, and it should not think this bridge it doing
45 	 * that bridge's work forwarding out its ports.
46 	 */
47 	br_switchdev_frame_unmark(skb);
48 
49 	/* Bridge is just like any other port.  Make sure the
50 	 * packet is allowed except in promisc mode when someone
51 	 * may be running packet capture.
52 	 */
53 	if (!(brdev->flags & IFF_PROMISC) &&
54 	    !br_allowed_egress(vg, skb)) {
55 		kfree_skb(skb);
56 		return NET_RX_DROP;
57 	}
58 
59 	indev = skb->dev;
60 	skb->dev = brdev;
61 	skb = br_handle_vlan(br, NULL, vg, skb);
62 	if (!skb)
63 		return NET_RX_DROP;
64 	/* update the multicast stats if the packet is IGMP/MLD */
65 	br_multicast_count(br, NULL, skb, br_multicast_igmp_type(skb),
66 			   BR_MCAST_DIR_TX);
67 
68 	return NF_HOOK(NFPROTO_BRIDGE, NF_BR_LOCAL_IN,
69 		       dev_net(indev), NULL, skb, indev, NULL,
70 		       br_netif_receive_skb);
71 }
72 
73 /* note: already called with rcu_read_lock */
74 int br_handle_frame_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
75 {
76 	struct net_bridge_port *p = br_port_get_rcu(skb->dev);
77 	enum br_pkt_type pkt_type = BR_PKT_UNICAST;
78 	struct net_bridge_fdb_entry *dst = NULL;
79 	struct net_bridge_mcast_port *pmctx;
80 	struct net_bridge_mdb_entry *mdst;
81 	bool local_rcv, mcast_hit = false;
82 	struct net_bridge_mcast *brmctx;
83 	struct net_bridge_vlan *vlan;
84 	struct net_bridge *br;
85 	u16 vid = 0;
86 	u8 state;
87 
88 	if (!p)
89 		goto drop;
90 
91 	br = p->br;
92 
93 	if (br_mst_is_enabled(br)) {
94 		state = BR_STATE_FORWARDING;
95 	} else {
96 		if (p->state == BR_STATE_DISABLED)
97 			goto drop;
98 
99 		state = p->state;
100 	}
101 
102 	brmctx = &p->br->multicast_ctx;
103 	pmctx = &p->multicast_ctx;
104 	if (!br_allowed_ingress(p->br, nbp_vlan_group_rcu(p), skb, &vid,
105 				&state, &vlan))
106 		goto out;
107 
108 	if (p->flags & BR_PORT_LOCKED) {
109 		struct net_bridge_fdb_entry *fdb_src =
110 			br_fdb_find_rcu(br, eth_hdr(skb)->h_source, vid);
111 
112 		if (!fdb_src || READ_ONCE(fdb_src->dst) != p ||
113 		    test_bit(BR_FDB_LOCAL, &fdb_src->flags))
114 			goto drop;
115 	}
116 
117 	nbp_switchdev_frame_mark(p, skb);
118 
119 	/* insert into forwarding database after filtering to avoid spoofing */
120 	if (p->flags & BR_LEARNING)
121 		br_fdb_update(br, p, eth_hdr(skb)->h_source, vid, 0);
122 
123 	local_rcv = !!(br->dev->flags & IFF_PROMISC);
124 	if (is_multicast_ether_addr(eth_hdr(skb)->h_dest)) {
125 		/* by definition the broadcast is also a multicast address */
126 		if (is_broadcast_ether_addr(eth_hdr(skb)->h_dest)) {
127 			pkt_type = BR_PKT_BROADCAST;
128 			local_rcv = true;
129 		} else {
130 			pkt_type = BR_PKT_MULTICAST;
131 			if (br_multicast_rcv(&brmctx, &pmctx, vlan, skb, vid))
132 				goto drop;
133 		}
134 	}
135 
136 	if (state == BR_STATE_LEARNING)
137 		goto drop;
138 
139 	BR_INPUT_SKB_CB(skb)->brdev = br->dev;
140 	BR_INPUT_SKB_CB(skb)->src_port_isolated = !!(p->flags & BR_ISOLATED);
141 
142 	if (IS_ENABLED(CONFIG_INET) &&
143 	    (skb->protocol == htons(ETH_P_ARP) ||
144 	     skb->protocol == htons(ETH_P_RARP))) {
145 		br_do_proxy_suppress_arp(skb, br, vid, p);
146 	} else if (IS_ENABLED(CONFIG_IPV6) &&
147 		   skb->protocol == htons(ETH_P_IPV6) &&
148 		   br_opt_get(br, BROPT_NEIGH_SUPPRESS_ENABLED) &&
149 		   pskb_may_pull(skb, sizeof(struct ipv6hdr) +
150 				 sizeof(struct nd_msg)) &&
151 		   ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
152 			struct nd_msg *msg, _msg;
153 
154 			msg = br_is_nd_neigh_msg(skb, &_msg);
155 			if (msg)
156 				br_do_suppress_nd(skb, br, vid, p, msg);
157 	}
158 
159 	switch (pkt_type) {
160 	case BR_PKT_MULTICAST:
161 		mdst = br_mdb_get(brmctx, skb, vid);
162 		if ((mdst || BR_INPUT_SKB_CB_MROUTERS_ONLY(skb)) &&
163 		    br_multicast_querier_exists(brmctx, eth_hdr(skb), mdst)) {
164 			if ((mdst && mdst->host_joined) ||
165 			    br_multicast_is_router(brmctx, skb)) {
166 				local_rcv = true;
167 				br->dev->stats.multicast++;
168 			}
169 			mcast_hit = true;
170 		} else {
171 			local_rcv = true;
172 			br->dev->stats.multicast++;
173 		}
174 		break;
175 	case BR_PKT_UNICAST:
176 		dst = br_fdb_find_rcu(br, eth_hdr(skb)->h_dest, vid);
177 		break;
178 	default:
179 		break;
180 	}
181 
182 	if (dst) {
183 		unsigned long now = jiffies;
184 
185 		if (test_bit(BR_FDB_LOCAL, &dst->flags))
186 			return br_pass_frame_up(skb);
187 
188 		if (now != dst->used)
189 			dst->used = now;
190 		br_forward(dst->dst, skb, local_rcv, false);
191 	} else {
192 		if (!mcast_hit)
193 			br_flood(br, skb, pkt_type, local_rcv, false);
194 		else
195 			br_multicast_flood(mdst, skb, brmctx, local_rcv, false);
196 	}
197 
198 	if (local_rcv)
199 		return br_pass_frame_up(skb);
200 
201 out:
202 	return 0;
203 drop:
204 	kfree_skb(skb);
205 	goto out;
206 }
207 EXPORT_SYMBOL_GPL(br_handle_frame_finish);
208 
209 static void __br_handle_local_finish(struct sk_buff *skb)
210 {
211 	struct net_bridge_port *p = br_port_get_rcu(skb->dev);
212 	u16 vid = 0;
213 
214 	/* check if vlan is allowed, to avoid spoofing */
215 	if ((p->flags & BR_LEARNING) &&
216 	    nbp_state_should_learn(p) &&
217 	    !br_opt_get(p->br, BROPT_NO_LL_LEARN) &&
218 	    br_should_learn(p, skb, &vid))
219 		br_fdb_update(p->br, p, eth_hdr(skb)->h_source, vid, 0);
220 }
221 
222 /* note: already called with rcu_read_lock */
223 static int br_handle_local_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
224 {
225 	__br_handle_local_finish(skb);
226 
227 	/* return 1 to signal the okfn() was called so it's ok to use the skb */
228 	return 1;
229 }
230 
231 static int nf_hook_bridge_pre(struct sk_buff *skb, struct sk_buff **pskb)
232 {
233 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE
234 	struct nf_hook_entries *e = NULL;
235 	struct nf_hook_state state;
236 	unsigned int verdict, i;
237 	struct net *net;
238 	int ret;
239 
240 	net = dev_net(skb->dev);
241 #ifdef HAVE_JUMP_LABEL
242 	if (!static_key_false(&nf_hooks_needed[NFPROTO_BRIDGE][NF_BR_PRE_ROUTING]))
243 		goto frame_finish;
244 #endif
245 
246 	e = rcu_dereference(net->nf.hooks_bridge[NF_BR_PRE_ROUTING]);
247 	if (!e)
248 		goto frame_finish;
249 
250 	nf_hook_state_init(&state, NF_BR_PRE_ROUTING,
251 			   NFPROTO_BRIDGE, skb->dev, NULL, NULL,
252 			   net, br_handle_frame_finish);
253 
254 	for (i = 0; i < e->num_hook_entries; i++) {
255 		verdict = nf_hook_entry_hookfn(&e->hooks[i], skb, &state);
256 		switch (verdict & NF_VERDICT_MASK) {
257 		case NF_ACCEPT:
258 			if (BR_INPUT_SKB_CB(skb)->br_netfilter_broute) {
259 				*pskb = skb;
260 				return RX_HANDLER_PASS;
261 			}
262 			break;
263 		case NF_DROP:
264 			kfree_skb(skb);
265 			return RX_HANDLER_CONSUMED;
266 		case NF_QUEUE:
267 			ret = nf_queue(skb, &state, i, verdict);
268 			if (ret == 1)
269 				continue;
270 			return RX_HANDLER_CONSUMED;
271 		default: /* STOLEN */
272 			return RX_HANDLER_CONSUMED;
273 		}
274 	}
275 frame_finish:
276 	net = dev_net(skb->dev);
277 	br_handle_frame_finish(net, NULL, skb);
278 #else
279 	br_handle_frame_finish(dev_net(skb->dev), NULL, skb);
280 #endif
281 	return RX_HANDLER_CONSUMED;
282 }
283 
284 /* Return 0 if the frame was not processed otherwise 1
285  * note: already called with rcu_read_lock
286  */
287 static int br_process_frame_type(struct net_bridge_port *p,
288 				 struct sk_buff *skb)
289 {
290 	struct br_frame_type *tmp;
291 
292 	hlist_for_each_entry_rcu(tmp, &p->br->frame_type_list, list)
293 		if (unlikely(tmp->type == skb->protocol))
294 			return tmp->frame_handler(p, skb);
295 
296 	return 0;
297 }
298 
299 /*
300  * Return NULL if skb is handled
301  * note: already called with rcu_read_lock
302  */
303 static rx_handler_result_t br_handle_frame(struct sk_buff **pskb)
304 {
305 	struct net_bridge_port *p;
306 	struct sk_buff *skb = *pskb;
307 	const unsigned char *dest = eth_hdr(skb)->h_dest;
308 
309 	if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
310 		return RX_HANDLER_PASS;
311 
312 	if (!is_valid_ether_addr(eth_hdr(skb)->h_source))
313 		goto drop;
314 
315 	skb = skb_share_check(skb, GFP_ATOMIC);
316 	if (!skb)
317 		return RX_HANDLER_CONSUMED;
318 
319 	memset(skb->cb, 0, sizeof(struct br_input_skb_cb));
320 
321 	p = br_port_get_rcu(skb->dev);
322 	if (p->flags & BR_VLAN_TUNNEL)
323 		br_handle_ingress_vlan_tunnel(skb, p, nbp_vlan_group_rcu(p));
324 
325 	if (unlikely(is_link_local_ether_addr(dest))) {
326 		u16 fwd_mask = p->br->group_fwd_mask_required;
327 
328 		/*
329 		 * See IEEE 802.1D Table 7-10 Reserved addresses
330 		 *
331 		 * Assignment		 		Value
332 		 * Bridge Group Address		01-80-C2-00-00-00
333 		 * (MAC Control) 802.3		01-80-C2-00-00-01
334 		 * (Link Aggregation) 802.3	01-80-C2-00-00-02
335 		 * 802.1X PAE address		01-80-C2-00-00-03
336 		 *
337 		 * 802.1AB LLDP 		01-80-C2-00-00-0E
338 		 *
339 		 * Others reserved for future standardization
340 		 */
341 		fwd_mask |= p->group_fwd_mask;
342 		switch (dest[5]) {
343 		case 0x00:	/* Bridge Group Address */
344 			/* If STP is turned off,
345 			   then must forward to keep loop detection */
346 			if (p->br->stp_enabled == BR_NO_STP ||
347 			    fwd_mask & (1u << dest[5]))
348 				goto forward;
349 			*pskb = skb;
350 			__br_handle_local_finish(skb);
351 			return RX_HANDLER_PASS;
352 
353 		case 0x01:	/* IEEE MAC (Pause) */
354 			goto drop;
355 
356 		case 0x0E:	/* 802.1AB LLDP */
357 			fwd_mask |= p->br->group_fwd_mask;
358 			if (fwd_mask & (1u << dest[5]))
359 				goto forward;
360 			*pskb = skb;
361 			__br_handle_local_finish(skb);
362 			return RX_HANDLER_PASS;
363 
364 		default:
365 			/* Allow selective forwarding for most other protocols */
366 			fwd_mask |= p->br->group_fwd_mask;
367 			if (fwd_mask & (1u << dest[5]))
368 				goto forward;
369 		}
370 
371 		/* The else clause should be hit when nf_hook():
372 		 *   - returns < 0 (drop/error)
373 		 *   - returns = 0 (stolen/nf_queue)
374 		 * Thus return 1 from the okfn() to signal the skb is ok to pass
375 		 */
376 		if (NF_HOOK(NFPROTO_BRIDGE, NF_BR_LOCAL_IN,
377 			    dev_net(skb->dev), NULL, skb, skb->dev, NULL,
378 			    br_handle_local_finish) == 1) {
379 			return RX_HANDLER_PASS;
380 		} else {
381 			return RX_HANDLER_CONSUMED;
382 		}
383 	}
384 
385 	if (unlikely(br_process_frame_type(p, skb)))
386 		return RX_HANDLER_PASS;
387 
388 forward:
389 	if (br_mst_is_enabled(p->br))
390 		goto defer_stp_filtering;
391 
392 	switch (p->state) {
393 	case BR_STATE_FORWARDING:
394 	case BR_STATE_LEARNING:
395 defer_stp_filtering:
396 		if (ether_addr_equal(p->br->dev->dev_addr, dest))
397 			skb->pkt_type = PACKET_HOST;
398 
399 		return nf_hook_bridge_pre(skb, pskb);
400 	default:
401 drop:
402 		kfree_skb(skb);
403 	}
404 	return RX_HANDLER_CONSUMED;
405 }
406 
407 /* This function has no purpose other than to appease the br_port_get_rcu/rtnl
408  * helpers which identify bridged ports according to the rx_handler installed
409  * on them (so there _needs_ to be a bridge rx_handler even if we don't need it
410  * to do anything useful). This bridge won't support traffic to/from the stack,
411  * but only hardware bridging. So return RX_HANDLER_PASS so we don't steal
412  * frames from the ETH_P_XDSA packet_type handler.
413  */
414 static rx_handler_result_t br_handle_frame_dummy(struct sk_buff **pskb)
415 {
416 	return RX_HANDLER_PASS;
417 }
418 
419 rx_handler_func_t *br_get_rx_handler(const struct net_device *dev)
420 {
421 	if (netdev_uses_dsa(dev))
422 		return br_handle_frame_dummy;
423 
424 	return br_handle_frame;
425 }
426 
427 void br_add_frame(struct net_bridge *br, struct br_frame_type *ft)
428 {
429 	hlist_add_head_rcu(&ft->list, &br->frame_type_list);
430 }
431 
432 void br_del_frame(struct net_bridge *br, struct br_frame_type *ft)
433 {
434 	struct br_frame_type *tmp;
435 
436 	hlist_for_each_entry(tmp, &br->frame_type_list, list)
437 		if (ft == tmp) {
438 			hlist_del_rcu(&ft->list);
439 			return;
440 		}
441 }
442