xref: /openbmc/linux/net/bridge/br_input.c (revision babbdf5b)
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 mode 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 		break;
148 	default:
149 		break;
150 	}
151 
152 	if (dst) {
153 		unsigned long now = jiffies;
154 
155 		if (test_bit(BR_FDB_LOCAL, &dst->flags))
156 			return br_pass_frame_up(skb);
157 
158 		if (now != dst->used)
159 			dst->used = now;
160 		br_forward(dst->dst, skb, local_rcv, false);
161 	} else {
162 		if (!mcast_hit)
163 			br_flood(br, skb, pkt_type, local_rcv, false);
164 		else
165 			br_multicast_flood(mdst, skb, local_rcv, false);
166 	}
167 
168 	if (local_rcv)
169 		return br_pass_frame_up(skb);
170 
171 out:
172 	return 0;
173 drop:
174 	kfree_skb(skb);
175 	goto out;
176 }
177 EXPORT_SYMBOL_GPL(br_handle_frame_finish);
178 
179 static void __br_handle_local_finish(struct sk_buff *skb)
180 {
181 	struct net_bridge_port *p = br_port_get_rcu(skb->dev);
182 	u16 vid = 0;
183 
184 	/* check if vlan is allowed, to avoid spoofing */
185 	if ((p->flags & BR_LEARNING) &&
186 	    nbp_state_should_learn(p) &&
187 	    !br_opt_get(p->br, BROPT_NO_LL_LEARN) &&
188 	    br_should_learn(p, skb, &vid))
189 		br_fdb_update(p->br, p, eth_hdr(skb)->h_source, vid, 0);
190 }
191 
192 /* note: already called with rcu_read_lock */
193 static int br_handle_local_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
194 {
195 	__br_handle_local_finish(skb);
196 
197 	/* return 1 to signal the okfn() was called so it's ok to use the skb */
198 	return 1;
199 }
200 
201 static int nf_hook_bridge_pre(struct sk_buff *skb, struct sk_buff **pskb)
202 {
203 #ifdef CONFIG_NETFILTER_FAMILY_BRIDGE
204 	struct nf_hook_entries *e = NULL;
205 	struct nf_hook_state state;
206 	unsigned int verdict, i;
207 	struct net *net;
208 	int ret;
209 
210 	net = dev_net(skb->dev);
211 #ifdef HAVE_JUMP_LABEL
212 	if (!static_key_false(&nf_hooks_needed[NFPROTO_BRIDGE][NF_BR_PRE_ROUTING]))
213 		goto frame_finish;
214 #endif
215 
216 	e = rcu_dereference(net->nf.hooks_bridge[NF_BR_PRE_ROUTING]);
217 	if (!e)
218 		goto frame_finish;
219 
220 	nf_hook_state_init(&state, NF_BR_PRE_ROUTING,
221 			   NFPROTO_BRIDGE, skb->dev, NULL, NULL,
222 			   net, br_handle_frame_finish);
223 
224 	for (i = 0; i < e->num_hook_entries; i++) {
225 		verdict = nf_hook_entry_hookfn(&e->hooks[i], skb, &state);
226 		switch (verdict & NF_VERDICT_MASK) {
227 		case NF_ACCEPT:
228 			if (BR_INPUT_SKB_CB(skb)->br_netfilter_broute) {
229 				*pskb = skb;
230 				return RX_HANDLER_PASS;
231 			}
232 			break;
233 		case NF_DROP:
234 			kfree_skb(skb);
235 			return RX_HANDLER_CONSUMED;
236 		case NF_QUEUE:
237 			ret = nf_queue(skb, &state, i, verdict);
238 			if (ret == 1)
239 				continue;
240 			return RX_HANDLER_CONSUMED;
241 		default: /* STOLEN */
242 			return RX_HANDLER_CONSUMED;
243 		}
244 	}
245 frame_finish:
246 	net = dev_net(skb->dev);
247 	br_handle_frame_finish(net, NULL, skb);
248 #else
249 	br_handle_frame_finish(dev_net(skb->dev), NULL, skb);
250 #endif
251 	return RX_HANDLER_CONSUMED;
252 }
253 
254 /* Return 0 if the frame was not processed otherwise 1
255  * note: already called with rcu_read_lock
256  */
257 static int br_process_frame_type(struct net_bridge_port *p,
258 				 struct sk_buff *skb)
259 {
260 	struct br_frame_type *tmp;
261 
262 	hlist_for_each_entry_rcu(tmp, &p->br->frame_type_list, list)
263 		if (unlikely(tmp->type == skb->protocol))
264 			return tmp->frame_handler(p, skb);
265 
266 	return 0;
267 }
268 
269 /*
270  * Return NULL if skb is handled
271  * note: already called with rcu_read_lock
272  */
273 static rx_handler_result_t br_handle_frame(struct sk_buff **pskb)
274 {
275 	struct net_bridge_port *p;
276 	struct sk_buff *skb = *pskb;
277 	const unsigned char *dest = eth_hdr(skb)->h_dest;
278 
279 	if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
280 		return RX_HANDLER_PASS;
281 
282 	if (!is_valid_ether_addr(eth_hdr(skb)->h_source))
283 		goto drop;
284 
285 	skb = skb_share_check(skb, GFP_ATOMIC);
286 	if (!skb)
287 		return RX_HANDLER_CONSUMED;
288 
289 	memset(skb->cb, 0, sizeof(struct br_input_skb_cb));
290 
291 	p = br_port_get_rcu(skb->dev);
292 	if (p->flags & BR_VLAN_TUNNEL) {
293 		if (br_handle_ingress_vlan_tunnel(skb, p,
294 						  nbp_vlan_group_rcu(p)))
295 			goto drop;
296 	}
297 
298 	if (unlikely(is_link_local_ether_addr(dest))) {
299 		u16 fwd_mask = p->br->group_fwd_mask_required;
300 
301 		/*
302 		 * See IEEE 802.1D Table 7-10 Reserved addresses
303 		 *
304 		 * Assignment		 		Value
305 		 * Bridge Group Address		01-80-C2-00-00-00
306 		 * (MAC Control) 802.3		01-80-C2-00-00-01
307 		 * (Link Aggregation) 802.3	01-80-C2-00-00-02
308 		 * 802.1X PAE address		01-80-C2-00-00-03
309 		 *
310 		 * 802.1AB LLDP 		01-80-C2-00-00-0E
311 		 *
312 		 * Others reserved for future standardization
313 		 */
314 		fwd_mask |= p->group_fwd_mask;
315 		switch (dest[5]) {
316 		case 0x00:	/* Bridge Group Address */
317 			/* If STP is turned off,
318 			   then must forward to keep loop detection */
319 			if (p->br->stp_enabled == BR_NO_STP ||
320 			    fwd_mask & (1u << dest[5]))
321 				goto forward;
322 			*pskb = skb;
323 			__br_handle_local_finish(skb);
324 			return RX_HANDLER_PASS;
325 
326 		case 0x01:	/* IEEE MAC (Pause) */
327 			goto drop;
328 
329 		case 0x0E:	/* 802.1AB LLDP */
330 			fwd_mask |= p->br->group_fwd_mask;
331 			if (fwd_mask & (1u << dest[5]))
332 				goto forward;
333 			*pskb = skb;
334 			__br_handle_local_finish(skb);
335 			return RX_HANDLER_PASS;
336 
337 		default:
338 			/* Allow selective forwarding for most other protocols */
339 			fwd_mask |= p->br->group_fwd_mask;
340 			if (fwd_mask & (1u << dest[5]))
341 				goto forward;
342 		}
343 
344 		/* The else clause should be hit when nf_hook():
345 		 *   - returns < 0 (drop/error)
346 		 *   - returns = 0 (stolen/nf_queue)
347 		 * Thus return 1 from the okfn() to signal the skb is ok to pass
348 		 */
349 		if (NF_HOOK(NFPROTO_BRIDGE, NF_BR_LOCAL_IN,
350 			    dev_net(skb->dev), NULL, skb, skb->dev, NULL,
351 			    br_handle_local_finish) == 1) {
352 			return RX_HANDLER_PASS;
353 		} else {
354 			return RX_HANDLER_CONSUMED;
355 		}
356 	}
357 
358 	if (unlikely(br_process_frame_type(p, skb)))
359 		return RX_HANDLER_PASS;
360 
361 forward:
362 	switch (p->state) {
363 	case BR_STATE_FORWARDING:
364 	case BR_STATE_LEARNING:
365 		if (ether_addr_equal(p->br->dev->dev_addr, dest))
366 			skb->pkt_type = PACKET_HOST;
367 
368 		return nf_hook_bridge_pre(skb, pskb);
369 	default:
370 drop:
371 		kfree_skb(skb);
372 	}
373 	return RX_HANDLER_CONSUMED;
374 }
375 
376 /* This function has no purpose other than to appease the br_port_get_rcu/rtnl
377  * helpers which identify bridged ports according to the rx_handler installed
378  * on them (so there _needs_ to be a bridge rx_handler even if we don't need it
379  * to do anything useful). This bridge won't support traffic to/from the stack,
380  * but only hardware bridging. So return RX_HANDLER_PASS so we don't steal
381  * frames from the ETH_P_XDSA packet_type handler.
382  */
383 static rx_handler_result_t br_handle_frame_dummy(struct sk_buff **pskb)
384 {
385 	return RX_HANDLER_PASS;
386 }
387 
388 rx_handler_func_t *br_get_rx_handler(const struct net_device *dev)
389 {
390 	if (netdev_uses_dsa(dev))
391 		return br_handle_frame_dummy;
392 
393 	return br_handle_frame;
394 }
395 
396 void br_add_frame(struct net_bridge *br, struct br_frame_type *ft)
397 {
398 	hlist_add_head_rcu(&ft->list, &br->frame_type_list);
399 }
400 
401 void br_del_frame(struct net_bridge *br, struct br_frame_type *ft)
402 {
403 	struct br_frame_type *tmp;
404 
405 	hlist_for_each_entry(tmp, &br->frame_type_list, list)
406 		if (ft == tmp) {
407 			hlist_del_rcu(&ft->list);
408 			return;
409 		}
410 }
411