xref: /linux/drivers/net/ipvlan/ipvlan_core.c (revision d96fc832bcb6269d96e33d506f33033d7ed08598)
1 /* Copyright (c) 2014 Mahesh Bandewar <maheshb@google.com>
2  *
3  * This program is free software; you can redistribute it and/or
4  * modify it under the terms of the GNU General Public License as
5  * published by the Free Software Foundation; either version 2 of
6  * the License, or (at your option) any later version.
7  *
8  */
9 
10 #include "ipvlan.h"
11 
12 static u32 ipvlan_jhash_secret __read_mostly;
13 
14 void ipvlan_init_secret(void)
15 {
16 	net_get_random_once(&ipvlan_jhash_secret, sizeof(ipvlan_jhash_secret));
17 }
18 
19 void ipvlan_count_rx(const struct ipvl_dev *ipvlan,
20 			    unsigned int len, bool success, bool mcast)
21 {
22 	if (likely(success)) {
23 		struct ipvl_pcpu_stats *pcptr;
24 
25 		pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
26 		u64_stats_update_begin(&pcptr->syncp);
27 		pcptr->rx_pkts++;
28 		pcptr->rx_bytes += len;
29 		if (mcast)
30 			pcptr->rx_mcast++;
31 		u64_stats_update_end(&pcptr->syncp);
32 	} else {
33 		this_cpu_inc(ipvlan->pcpu_stats->rx_errs);
34 	}
35 }
36 EXPORT_SYMBOL_GPL(ipvlan_count_rx);
37 
38 #if IS_ENABLED(CONFIG_IPV6)
39 static u8 ipvlan_get_v6_hash(const void *iaddr)
40 {
41 	const struct in6_addr *ip6_addr = iaddr;
42 
43 	return __ipv6_addr_jhash(ip6_addr, ipvlan_jhash_secret) &
44 	       IPVLAN_HASH_MASK;
45 }
46 #else
47 static u8 ipvlan_get_v6_hash(const void *iaddr)
48 {
49 	return 0;
50 }
51 #endif
52 
53 static u8 ipvlan_get_v4_hash(const void *iaddr)
54 {
55 	const struct in_addr *ip4_addr = iaddr;
56 
57 	return jhash_1word(ip4_addr->s_addr, ipvlan_jhash_secret) &
58 	       IPVLAN_HASH_MASK;
59 }
60 
61 static bool addr_equal(bool is_v6, struct ipvl_addr *addr, const void *iaddr)
62 {
63 	if (!is_v6 && addr->atype == IPVL_IPV4) {
64 		struct in_addr *i4addr = (struct in_addr *)iaddr;
65 
66 		return addr->ip4addr.s_addr == i4addr->s_addr;
67 #if IS_ENABLED(CONFIG_IPV6)
68 	} else if (is_v6 && addr->atype == IPVL_IPV6) {
69 		struct in6_addr *i6addr = (struct in6_addr *)iaddr;
70 
71 		return ipv6_addr_equal(&addr->ip6addr, i6addr);
72 #endif
73 	}
74 
75 	return false;
76 }
77 
78 static struct ipvl_addr *ipvlan_ht_addr_lookup(const struct ipvl_port *port,
79 					       const void *iaddr, bool is_v6)
80 {
81 	struct ipvl_addr *addr;
82 	u8 hash;
83 
84 	hash = is_v6 ? ipvlan_get_v6_hash(iaddr) :
85 	       ipvlan_get_v4_hash(iaddr);
86 	hlist_for_each_entry_rcu(addr, &port->hlhead[hash], hlnode)
87 		if (addr_equal(is_v6, addr, iaddr))
88 			return addr;
89 	return NULL;
90 }
91 
92 void ipvlan_ht_addr_add(struct ipvl_dev *ipvlan, struct ipvl_addr *addr)
93 {
94 	struct ipvl_port *port = ipvlan->port;
95 	u8 hash;
96 
97 	hash = (addr->atype == IPVL_IPV6) ?
98 	       ipvlan_get_v6_hash(&addr->ip6addr) :
99 	       ipvlan_get_v4_hash(&addr->ip4addr);
100 	if (hlist_unhashed(&addr->hlnode))
101 		hlist_add_head_rcu(&addr->hlnode, &port->hlhead[hash]);
102 }
103 
104 void ipvlan_ht_addr_del(struct ipvl_addr *addr)
105 {
106 	hlist_del_init_rcu(&addr->hlnode);
107 }
108 
109 struct ipvl_addr *ipvlan_find_addr(const struct ipvl_dev *ipvlan,
110 				   const void *iaddr, bool is_v6)
111 {
112 	struct ipvl_addr *addr;
113 
114 	list_for_each_entry(addr, &ipvlan->addrs, anode)
115 		if (addr_equal(is_v6, addr, iaddr))
116 			return addr;
117 	return NULL;
118 }
119 
120 bool ipvlan_addr_busy(struct ipvl_port *port, void *iaddr, bool is_v6)
121 {
122 	struct ipvl_dev *ipvlan;
123 
124 	ASSERT_RTNL();
125 
126 	list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
127 		if (ipvlan_find_addr(ipvlan, iaddr, is_v6))
128 			return true;
129 	}
130 	return false;
131 }
132 
133 static void *ipvlan_get_L3_hdr(struct ipvl_port *port, struct sk_buff *skb, int *type)
134 {
135 	void *lyr3h = NULL;
136 
137 	switch (skb->protocol) {
138 	case htons(ETH_P_ARP): {
139 		struct arphdr *arph;
140 
141 		if (unlikely(!pskb_may_pull(skb, arp_hdr_len(port->dev))))
142 			return NULL;
143 
144 		arph = arp_hdr(skb);
145 		*type = IPVL_ARP;
146 		lyr3h = arph;
147 		break;
148 	}
149 	case htons(ETH_P_IP): {
150 		u32 pktlen;
151 		struct iphdr *ip4h;
152 
153 		if (unlikely(!pskb_may_pull(skb, sizeof(*ip4h))))
154 			return NULL;
155 
156 		ip4h = ip_hdr(skb);
157 		pktlen = ntohs(ip4h->tot_len);
158 		if (ip4h->ihl < 5 || ip4h->version != 4)
159 			return NULL;
160 		if (skb->len < pktlen || pktlen < (ip4h->ihl * 4))
161 			return NULL;
162 
163 		*type = IPVL_IPV4;
164 		lyr3h = ip4h;
165 		break;
166 	}
167 #if IS_ENABLED(CONFIG_IPV6)
168 	case htons(ETH_P_IPV6): {
169 		struct ipv6hdr *ip6h;
170 
171 		if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h))))
172 			return NULL;
173 
174 		ip6h = ipv6_hdr(skb);
175 		if (ip6h->version != 6)
176 			return NULL;
177 
178 		*type = IPVL_IPV6;
179 		lyr3h = ip6h;
180 		/* Only Neighbour Solicitation pkts need different treatment */
181 		if (ipv6_addr_any(&ip6h->saddr) &&
182 		    ip6h->nexthdr == NEXTHDR_ICMP) {
183 			struct icmp6hdr	*icmph;
184 
185 			if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h) + sizeof(*icmph))))
186 				return NULL;
187 
188 			ip6h = ipv6_hdr(skb);
189 			icmph = (struct icmp6hdr *)(ip6h + 1);
190 
191 			if (icmph->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
192 				/* Need to access the ipv6 address in body */
193 				if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h) + sizeof(*icmph)
194 						+ sizeof(struct in6_addr))))
195 					return NULL;
196 
197 				ip6h = ipv6_hdr(skb);
198 				icmph = (struct icmp6hdr *)(ip6h + 1);
199 			}
200 
201 			*type = IPVL_ICMPV6;
202 			lyr3h = icmph;
203 		}
204 		break;
205 	}
206 #endif
207 	default:
208 		return NULL;
209 	}
210 
211 	return lyr3h;
212 }
213 
214 unsigned int ipvlan_mac_hash(const unsigned char *addr)
215 {
216 	u32 hash = jhash_1word(__get_unaligned_cpu32(addr+2),
217 			       ipvlan_jhash_secret);
218 
219 	return hash & IPVLAN_MAC_FILTER_MASK;
220 }
221 
222 void ipvlan_process_multicast(struct work_struct *work)
223 {
224 	struct ipvl_port *port = container_of(work, struct ipvl_port, wq);
225 	struct ethhdr *ethh;
226 	struct ipvl_dev *ipvlan;
227 	struct sk_buff *skb, *nskb;
228 	struct sk_buff_head list;
229 	unsigned int len;
230 	unsigned int mac_hash;
231 	int ret;
232 	u8 pkt_type;
233 	bool tx_pkt;
234 
235 	__skb_queue_head_init(&list);
236 
237 	spin_lock_bh(&port->backlog.lock);
238 	skb_queue_splice_tail_init(&port->backlog, &list);
239 	spin_unlock_bh(&port->backlog.lock);
240 
241 	while ((skb = __skb_dequeue(&list)) != NULL) {
242 		struct net_device *dev = skb->dev;
243 		bool consumed = false;
244 
245 		ethh = eth_hdr(skb);
246 		tx_pkt = IPVL_SKB_CB(skb)->tx_pkt;
247 		mac_hash = ipvlan_mac_hash(ethh->h_dest);
248 
249 		if (ether_addr_equal(ethh->h_dest, port->dev->broadcast))
250 			pkt_type = PACKET_BROADCAST;
251 		else
252 			pkt_type = PACKET_MULTICAST;
253 
254 		rcu_read_lock();
255 		list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
256 			if (tx_pkt && (ipvlan->dev == skb->dev))
257 				continue;
258 			if (!test_bit(mac_hash, ipvlan->mac_filters))
259 				continue;
260 			if (!(ipvlan->dev->flags & IFF_UP))
261 				continue;
262 			ret = NET_RX_DROP;
263 			len = skb->len + ETH_HLEN;
264 			nskb = skb_clone(skb, GFP_ATOMIC);
265 			local_bh_disable();
266 			if (nskb) {
267 				consumed = true;
268 				nskb->pkt_type = pkt_type;
269 				nskb->dev = ipvlan->dev;
270 				if (tx_pkt)
271 					ret = dev_forward_skb(ipvlan->dev, nskb);
272 				else
273 					ret = netif_rx(nskb);
274 			}
275 			ipvlan_count_rx(ipvlan, len, ret == NET_RX_SUCCESS, true);
276 			local_bh_enable();
277 		}
278 		rcu_read_unlock();
279 
280 		if (tx_pkt) {
281 			/* If the packet originated here, send it out. */
282 			skb->dev = port->dev;
283 			skb->pkt_type = pkt_type;
284 			dev_queue_xmit(skb);
285 		} else {
286 			if (consumed)
287 				consume_skb(skb);
288 			else
289 				kfree_skb(skb);
290 		}
291 		if (dev)
292 			dev_put(dev);
293 	}
294 }
295 
296 static void ipvlan_skb_crossing_ns(struct sk_buff *skb, struct net_device *dev)
297 {
298 	bool xnet = true;
299 
300 	if (dev)
301 		xnet = !net_eq(dev_net(skb->dev), dev_net(dev));
302 
303 	skb_scrub_packet(skb, xnet);
304 	if (dev)
305 		skb->dev = dev;
306 }
307 
308 static int ipvlan_rcv_frame(struct ipvl_addr *addr, struct sk_buff **pskb,
309 			    bool local)
310 {
311 	struct ipvl_dev *ipvlan = addr->master;
312 	struct net_device *dev = ipvlan->dev;
313 	unsigned int len;
314 	rx_handler_result_t ret = RX_HANDLER_CONSUMED;
315 	bool success = false;
316 	struct sk_buff *skb = *pskb;
317 
318 	len = skb->len + ETH_HLEN;
319 	/* Only packets exchanged between two local slaves need to have
320 	 * device-up check as well as skb-share check.
321 	 */
322 	if (local) {
323 		if (unlikely(!(dev->flags & IFF_UP))) {
324 			kfree_skb(skb);
325 			goto out;
326 		}
327 
328 		skb = skb_share_check(skb, GFP_ATOMIC);
329 		if (!skb)
330 			goto out;
331 
332 		*pskb = skb;
333 	}
334 
335 	if (local) {
336 		skb->pkt_type = PACKET_HOST;
337 		if (dev_forward_skb(ipvlan->dev, skb) == NET_RX_SUCCESS)
338 			success = true;
339 	} else {
340 		skb->dev = dev;
341 		ret = RX_HANDLER_ANOTHER;
342 		success = true;
343 	}
344 
345 out:
346 	ipvlan_count_rx(ipvlan, len, success, false);
347 	return ret;
348 }
349 
350 static struct ipvl_addr *ipvlan_addr_lookup(struct ipvl_port *port,
351 					    void *lyr3h, int addr_type,
352 					    bool use_dest)
353 {
354 	struct ipvl_addr *addr = NULL;
355 
356 	switch (addr_type) {
357 #if IS_ENABLED(CONFIG_IPV6)
358 	case IPVL_IPV6: {
359 		struct ipv6hdr *ip6h;
360 		struct in6_addr *i6addr;
361 
362 		ip6h = (struct ipv6hdr *)lyr3h;
363 		i6addr = use_dest ? &ip6h->daddr : &ip6h->saddr;
364 		addr = ipvlan_ht_addr_lookup(port, i6addr, true);
365 		break;
366 	}
367 	case IPVL_ICMPV6: {
368 		struct nd_msg *ndmh;
369 		struct in6_addr *i6addr;
370 
371 		/* Make sure that the NeighborSolicitation ICMPv6 packets
372 		 * are handled to avoid DAD issue.
373 		 */
374 		ndmh = (struct nd_msg *)lyr3h;
375 		if (ndmh->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
376 			i6addr = &ndmh->target;
377 			addr = ipvlan_ht_addr_lookup(port, i6addr, true);
378 		}
379 		break;
380 	}
381 #endif
382 	case IPVL_IPV4: {
383 		struct iphdr *ip4h;
384 		__be32 *i4addr;
385 
386 		ip4h = (struct iphdr *)lyr3h;
387 		i4addr = use_dest ? &ip4h->daddr : &ip4h->saddr;
388 		addr = ipvlan_ht_addr_lookup(port, i4addr, false);
389 		break;
390 	}
391 	case IPVL_ARP: {
392 		struct arphdr *arph;
393 		unsigned char *arp_ptr;
394 		__be32 dip;
395 
396 		arph = (struct arphdr *)lyr3h;
397 		arp_ptr = (unsigned char *)(arph + 1);
398 		if (use_dest)
399 			arp_ptr += (2 * port->dev->addr_len) + 4;
400 		else
401 			arp_ptr += port->dev->addr_len;
402 
403 		memcpy(&dip, arp_ptr, 4);
404 		addr = ipvlan_ht_addr_lookup(port, &dip, false);
405 		break;
406 	}
407 	}
408 
409 	return addr;
410 }
411 
412 static int ipvlan_process_v4_outbound(struct sk_buff *skb)
413 {
414 	const struct iphdr *ip4h = ip_hdr(skb);
415 	struct net_device *dev = skb->dev;
416 	struct net *net = dev_net(dev);
417 	struct rtable *rt;
418 	int err, ret = NET_XMIT_DROP;
419 	struct flowi4 fl4 = {
420 		.flowi4_oif = dev->ifindex,
421 		.flowi4_tos = RT_TOS(ip4h->tos),
422 		.flowi4_flags = FLOWI_FLAG_ANYSRC,
423 		.flowi4_mark = skb->mark,
424 		.daddr = ip4h->daddr,
425 		.saddr = ip4h->saddr,
426 	};
427 
428 	rt = ip_route_output_flow(net, &fl4, NULL);
429 	if (IS_ERR(rt))
430 		goto err;
431 
432 	if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
433 		ip_rt_put(rt);
434 		goto err;
435 	}
436 	skb_dst_set(skb, &rt->dst);
437 	err = ip_local_out(net, skb->sk, skb);
438 	if (unlikely(net_xmit_eval(err)))
439 		dev->stats.tx_errors++;
440 	else
441 		ret = NET_XMIT_SUCCESS;
442 	goto out;
443 err:
444 	dev->stats.tx_errors++;
445 	kfree_skb(skb);
446 out:
447 	return ret;
448 }
449 
450 #if IS_ENABLED(CONFIG_IPV6)
451 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
452 {
453 	const struct ipv6hdr *ip6h = ipv6_hdr(skb);
454 	struct net_device *dev = skb->dev;
455 	struct net *net = dev_net(dev);
456 	struct dst_entry *dst;
457 	int err, ret = NET_XMIT_DROP;
458 	struct flowi6 fl6 = {
459 		.flowi6_oif = dev->ifindex,
460 		.daddr = ip6h->daddr,
461 		.saddr = ip6h->saddr,
462 		.flowi6_flags = FLOWI_FLAG_ANYSRC,
463 		.flowlabel = ip6_flowinfo(ip6h),
464 		.flowi6_mark = skb->mark,
465 		.flowi6_proto = ip6h->nexthdr,
466 	};
467 
468 	dst = ip6_route_output(net, NULL, &fl6);
469 	if (dst->error) {
470 		ret = dst->error;
471 		dst_release(dst);
472 		goto err;
473 	}
474 	skb_dst_set(skb, dst);
475 	err = ip6_local_out(net, skb->sk, skb);
476 	if (unlikely(net_xmit_eval(err)))
477 		dev->stats.tx_errors++;
478 	else
479 		ret = NET_XMIT_SUCCESS;
480 	goto out;
481 err:
482 	dev->stats.tx_errors++;
483 	kfree_skb(skb);
484 out:
485 	return ret;
486 }
487 #else
488 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
489 {
490 	return NET_XMIT_DROP;
491 }
492 #endif
493 
494 static int ipvlan_process_outbound(struct sk_buff *skb)
495 {
496 	struct ethhdr *ethh = eth_hdr(skb);
497 	int ret = NET_XMIT_DROP;
498 
499 	/* In this mode we dont care about multicast and broadcast traffic */
500 	if (is_multicast_ether_addr(ethh->h_dest)) {
501 		pr_warn_ratelimited("Dropped {multi|broad}cast of type= [%x]\n",
502 				    ntohs(skb->protocol));
503 		kfree_skb(skb);
504 		goto out;
505 	}
506 
507 	/* The ipvlan is a pseudo-L2 device, so the packets that we receive
508 	 * will have L2; which need to discarded and processed further
509 	 * in the net-ns of the main-device.
510 	 */
511 	if (skb_mac_header_was_set(skb)) {
512 		skb_pull(skb, sizeof(*ethh));
513 		skb->mac_header = (typeof(skb->mac_header))~0U;
514 		skb_reset_network_header(skb);
515 	}
516 
517 	if (skb->protocol == htons(ETH_P_IPV6))
518 		ret = ipvlan_process_v6_outbound(skb);
519 	else if (skb->protocol == htons(ETH_P_IP))
520 		ret = ipvlan_process_v4_outbound(skb);
521 	else {
522 		pr_warn_ratelimited("Dropped outbound packet type=%x\n",
523 				    ntohs(skb->protocol));
524 		kfree_skb(skb);
525 	}
526 out:
527 	return ret;
528 }
529 
530 static void ipvlan_multicast_enqueue(struct ipvl_port *port,
531 				     struct sk_buff *skb, bool tx_pkt)
532 {
533 	if (skb->protocol == htons(ETH_P_PAUSE)) {
534 		kfree_skb(skb);
535 		return;
536 	}
537 
538 	/* Record that the deferred packet is from TX or RX path. By
539 	 * looking at mac-addresses on packet will lead to erronus decisions.
540 	 * (This would be true for a loopback-mode on master device or a
541 	 * hair-pin mode of the switch.)
542 	 */
543 	IPVL_SKB_CB(skb)->tx_pkt = tx_pkt;
544 
545 	spin_lock(&port->backlog.lock);
546 	if (skb_queue_len(&port->backlog) < IPVLAN_QBACKLOG_LIMIT) {
547 		if (skb->dev)
548 			dev_hold(skb->dev);
549 		__skb_queue_tail(&port->backlog, skb);
550 		spin_unlock(&port->backlog.lock);
551 		schedule_work(&port->wq);
552 	} else {
553 		spin_unlock(&port->backlog.lock);
554 		atomic_long_inc(&skb->dev->rx_dropped);
555 		kfree_skb(skb);
556 	}
557 }
558 
559 static int ipvlan_xmit_mode_l3(struct sk_buff *skb, struct net_device *dev)
560 {
561 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
562 	void *lyr3h;
563 	struct ipvl_addr *addr;
564 	int addr_type;
565 
566 	lyr3h = ipvlan_get_L3_hdr(ipvlan->port, skb, &addr_type);
567 	if (!lyr3h)
568 		goto out;
569 
570 	if (!ipvlan_is_vepa(ipvlan->port)) {
571 		addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
572 		if (addr) {
573 			if (ipvlan_is_private(ipvlan->port)) {
574 				consume_skb(skb);
575 				return NET_XMIT_DROP;
576 			}
577 			return ipvlan_rcv_frame(addr, &skb, true);
578 		}
579 	}
580 out:
581 	ipvlan_skb_crossing_ns(skb, ipvlan->phy_dev);
582 	return ipvlan_process_outbound(skb);
583 }
584 
585 static int ipvlan_xmit_mode_l2(struct sk_buff *skb, struct net_device *dev)
586 {
587 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
588 	struct ethhdr *eth = eth_hdr(skb);
589 	struct ipvl_addr *addr;
590 	void *lyr3h;
591 	int addr_type;
592 
593 	if (!ipvlan_is_vepa(ipvlan->port) &&
594 	    ether_addr_equal(eth->h_dest, eth->h_source)) {
595 		lyr3h = ipvlan_get_L3_hdr(ipvlan->port, skb, &addr_type);
596 		if (lyr3h) {
597 			addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
598 			if (addr) {
599 				if (ipvlan_is_private(ipvlan->port)) {
600 					consume_skb(skb);
601 					return NET_XMIT_DROP;
602 				}
603 				return ipvlan_rcv_frame(addr, &skb, true);
604 			}
605 		}
606 		skb = skb_share_check(skb, GFP_ATOMIC);
607 		if (!skb)
608 			return NET_XMIT_DROP;
609 
610 		/* Packet definitely does not belong to any of the
611 		 * virtual devices, but the dest is local. So forward
612 		 * the skb for the main-dev. At the RX side we just return
613 		 * RX_PASS for it to be processed further on the stack.
614 		 */
615 		return dev_forward_skb(ipvlan->phy_dev, skb);
616 
617 	} else if (is_multicast_ether_addr(eth->h_dest)) {
618 		ipvlan_skb_crossing_ns(skb, NULL);
619 		ipvlan_multicast_enqueue(ipvlan->port, skb, true);
620 		return NET_XMIT_SUCCESS;
621 	}
622 
623 	skb->dev = ipvlan->phy_dev;
624 	return dev_queue_xmit(skb);
625 }
626 
627 int ipvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
628 {
629 	struct ipvl_dev *ipvlan = netdev_priv(dev);
630 	struct ipvl_port *port = ipvlan_port_get_rcu_bh(ipvlan->phy_dev);
631 
632 	if (!port)
633 		goto out;
634 
635 	if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
636 		goto out;
637 
638 	switch(port->mode) {
639 	case IPVLAN_MODE_L2:
640 		return ipvlan_xmit_mode_l2(skb, dev);
641 	case IPVLAN_MODE_L3:
642 	case IPVLAN_MODE_L3S:
643 		return ipvlan_xmit_mode_l3(skb, dev);
644 	}
645 
646 	/* Should not reach here */
647 	WARN_ONCE(true, "ipvlan_queue_xmit() called for mode = [%hx]\n",
648 			  port->mode);
649 out:
650 	kfree_skb(skb);
651 	return NET_XMIT_DROP;
652 }
653 
654 static bool ipvlan_external_frame(struct sk_buff *skb, struct ipvl_port *port)
655 {
656 	struct ethhdr *eth = eth_hdr(skb);
657 	struct ipvl_addr *addr;
658 	void *lyr3h;
659 	int addr_type;
660 
661 	if (ether_addr_equal(eth->h_source, skb->dev->dev_addr)) {
662 		lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
663 		if (!lyr3h)
664 			return true;
665 
666 		addr = ipvlan_addr_lookup(port, lyr3h, addr_type, false);
667 		if (addr)
668 			return false;
669 	}
670 
671 	return true;
672 }
673 
674 static rx_handler_result_t ipvlan_handle_mode_l3(struct sk_buff **pskb,
675 						 struct ipvl_port *port)
676 {
677 	void *lyr3h;
678 	int addr_type;
679 	struct ipvl_addr *addr;
680 	struct sk_buff *skb = *pskb;
681 	rx_handler_result_t ret = RX_HANDLER_PASS;
682 
683 	lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
684 	if (!lyr3h)
685 		goto out;
686 
687 	addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
688 	if (addr)
689 		ret = ipvlan_rcv_frame(addr, pskb, false);
690 
691 out:
692 	return ret;
693 }
694 
695 static rx_handler_result_t ipvlan_handle_mode_l2(struct sk_buff **pskb,
696 						 struct ipvl_port *port)
697 {
698 	struct sk_buff *skb = *pskb;
699 	struct ethhdr *eth = eth_hdr(skb);
700 	rx_handler_result_t ret = RX_HANDLER_PASS;
701 
702 	if (is_multicast_ether_addr(eth->h_dest)) {
703 		if (ipvlan_external_frame(skb, port)) {
704 			struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
705 
706 			/* External frames are queued for device local
707 			 * distribution, but a copy is given to master
708 			 * straight away to avoid sending duplicates later
709 			 * when work-queue processes this frame. This is
710 			 * achieved by returning RX_HANDLER_PASS.
711 			 */
712 			if (nskb) {
713 				ipvlan_skb_crossing_ns(nskb, NULL);
714 				ipvlan_multicast_enqueue(port, nskb, false);
715 			}
716 		}
717 	} else {
718 		/* Perform like l3 mode for non-multicast packet */
719 		ret = ipvlan_handle_mode_l3(pskb, port);
720 	}
721 
722 	return ret;
723 }
724 
725 rx_handler_result_t ipvlan_handle_frame(struct sk_buff **pskb)
726 {
727 	struct sk_buff *skb = *pskb;
728 	struct ipvl_port *port = ipvlan_port_get_rcu(skb->dev);
729 
730 	if (!port)
731 		return RX_HANDLER_PASS;
732 
733 	switch (port->mode) {
734 	case IPVLAN_MODE_L2:
735 		return ipvlan_handle_mode_l2(pskb, port);
736 	case IPVLAN_MODE_L3:
737 		return ipvlan_handle_mode_l3(pskb, port);
738 	case IPVLAN_MODE_L3S:
739 		return RX_HANDLER_PASS;
740 	}
741 
742 	/* Should not reach here */
743 	WARN_ONCE(true, "ipvlan_handle_frame() called for mode = [%hx]\n",
744 			  port->mode);
745 	kfree_skb(skb);
746 	return RX_HANDLER_CONSUMED;
747 }
748 
749 static struct ipvl_addr *ipvlan_skb_to_addr(struct sk_buff *skb,
750 					    struct net_device *dev)
751 {
752 	struct ipvl_addr *addr = NULL;
753 	struct ipvl_port *port;
754 	void *lyr3h;
755 	int addr_type;
756 
757 	if (!dev || !netif_is_ipvlan_port(dev))
758 		goto out;
759 
760 	port = ipvlan_port_get_rcu(dev);
761 	if (!port || port->mode != IPVLAN_MODE_L3S)
762 		goto out;
763 
764 	lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
765 	if (!lyr3h)
766 		goto out;
767 
768 	addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
769 out:
770 	return addr;
771 }
772 
773 struct sk_buff *ipvlan_l3_rcv(struct net_device *dev, struct sk_buff *skb,
774 			      u16 proto)
775 {
776 	struct ipvl_addr *addr;
777 	struct net_device *sdev;
778 
779 	addr = ipvlan_skb_to_addr(skb, dev);
780 	if (!addr)
781 		goto out;
782 
783 	sdev = addr->master->dev;
784 	switch (proto) {
785 	case AF_INET:
786 	{
787 		int err;
788 		struct iphdr *ip4h = ip_hdr(skb);
789 
790 		err = ip_route_input_noref(skb, ip4h->daddr, ip4h->saddr,
791 					   ip4h->tos, sdev);
792 		if (unlikely(err))
793 			goto out;
794 		break;
795 	}
796 #if IS_ENABLED(CONFIG_IPV6)
797 	case AF_INET6:
798 	{
799 		struct dst_entry *dst;
800 		struct ipv6hdr *ip6h = ipv6_hdr(skb);
801 		int flags = RT6_LOOKUP_F_HAS_SADDR;
802 		struct flowi6 fl6 = {
803 			.flowi6_iif   = sdev->ifindex,
804 			.daddr        = ip6h->daddr,
805 			.saddr        = ip6h->saddr,
806 			.flowlabel    = ip6_flowinfo(ip6h),
807 			.flowi6_mark  = skb->mark,
808 			.flowi6_proto = ip6h->nexthdr,
809 		};
810 
811 		skb_dst_drop(skb);
812 		dst = ip6_route_input_lookup(dev_net(sdev), sdev, &fl6, flags);
813 		skb_dst_set(skb, dst);
814 		break;
815 	}
816 #endif
817 	default:
818 		break;
819 	}
820 
821 out:
822 	return skb;
823 }
824 
825 unsigned int ipvlan_nf_input(void *priv, struct sk_buff *skb,
826 			     const struct nf_hook_state *state)
827 {
828 	struct ipvl_addr *addr;
829 	unsigned int len;
830 
831 	addr = ipvlan_skb_to_addr(skb, skb->dev);
832 	if (!addr)
833 		goto out;
834 
835 	skb->dev = addr->master->dev;
836 	len = skb->len + ETH_HLEN;
837 	ipvlan_count_rx(addr->master, len, true, false);
838 out:
839 	return NF_ACCEPT;
840 }
841