xref: /openbmc/linux/drivers/net/ipvlan/ipvlan_core.c (revision 680ef72a)
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 static u8 ipvlan_get_v6_hash(const void *iaddr)
39 {
40 	const struct in6_addr *ip6_addr = iaddr;
41 
42 	return __ipv6_addr_jhash(ip6_addr, ipvlan_jhash_secret) &
43 	       IPVLAN_HASH_MASK;
44 }
45 
46 static u8 ipvlan_get_v4_hash(const void *iaddr)
47 {
48 	const struct in_addr *ip4_addr = iaddr;
49 
50 	return jhash_1word(ip4_addr->s_addr, ipvlan_jhash_secret) &
51 	       IPVLAN_HASH_MASK;
52 }
53 
54 static struct ipvl_addr *ipvlan_ht_addr_lookup(const struct ipvl_port *port,
55 					       const void *iaddr, bool is_v6)
56 {
57 	struct ipvl_addr *addr;
58 	u8 hash;
59 
60 	hash = is_v6 ? ipvlan_get_v6_hash(iaddr) :
61 	       ipvlan_get_v4_hash(iaddr);
62 	hlist_for_each_entry_rcu(addr, &port->hlhead[hash], hlnode) {
63 		if (is_v6 && addr->atype == IPVL_IPV6 &&
64 		    ipv6_addr_equal(&addr->ip6addr, iaddr))
65 			return addr;
66 		else if (!is_v6 && addr->atype == IPVL_IPV4 &&
67 			 addr->ip4addr.s_addr ==
68 				((struct in_addr *)iaddr)->s_addr)
69 			return addr;
70 	}
71 	return NULL;
72 }
73 
74 void ipvlan_ht_addr_add(struct ipvl_dev *ipvlan, struct ipvl_addr *addr)
75 {
76 	struct ipvl_port *port = ipvlan->port;
77 	u8 hash;
78 
79 	hash = (addr->atype == IPVL_IPV6) ?
80 	       ipvlan_get_v6_hash(&addr->ip6addr) :
81 	       ipvlan_get_v4_hash(&addr->ip4addr);
82 	if (hlist_unhashed(&addr->hlnode))
83 		hlist_add_head_rcu(&addr->hlnode, &port->hlhead[hash]);
84 }
85 
86 void ipvlan_ht_addr_del(struct ipvl_addr *addr)
87 {
88 	hlist_del_init_rcu(&addr->hlnode);
89 }
90 
91 struct ipvl_addr *ipvlan_find_addr(const struct ipvl_dev *ipvlan,
92 				   const void *iaddr, bool is_v6)
93 {
94 	struct ipvl_addr *addr;
95 
96 	list_for_each_entry(addr, &ipvlan->addrs, anode) {
97 		if ((is_v6 && addr->atype == IPVL_IPV6 &&
98 		    ipv6_addr_equal(&addr->ip6addr, iaddr)) ||
99 		    (!is_v6 && addr->atype == IPVL_IPV4 &&
100 		    addr->ip4addr.s_addr == ((struct in_addr *)iaddr)->s_addr))
101 			return addr;
102 	}
103 	return NULL;
104 }
105 
106 bool ipvlan_addr_busy(struct ipvl_port *port, void *iaddr, bool is_v6)
107 {
108 	struct ipvl_dev *ipvlan;
109 
110 	ASSERT_RTNL();
111 
112 	list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
113 		if (ipvlan_find_addr(ipvlan, iaddr, is_v6))
114 			return true;
115 	}
116 	return false;
117 }
118 
119 static void *ipvlan_get_L3_hdr(struct ipvl_port *port, struct sk_buff *skb, int *type)
120 {
121 	void *lyr3h = NULL;
122 
123 	switch (skb->protocol) {
124 	case htons(ETH_P_ARP): {
125 		struct arphdr *arph;
126 
127 		if (unlikely(!pskb_may_pull(skb, arp_hdr_len(port->dev))))
128 			return NULL;
129 
130 		arph = arp_hdr(skb);
131 		*type = IPVL_ARP;
132 		lyr3h = arph;
133 		break;
134 	}
135 	case htons(ETH_P_IP): {
136 		u32 pktlen;
137 		struct iphdr *ip4h;
138 
139 		if (unlikely(!pskb_may_pull(skb, sizeof(*ip4h))))
140 			return NULL;
141 
142 		ip4h = ip_hdr(skb);
143 		pktlen = ntohs(ip4h->tot_len);
144 		if (ip4h->ihl < 5 || ip4h->version != 4)
145 			return NULL;
146 		if (skb->len < pktlen || pktlen < (ip4h->ihl * 4))
147 			return NULL;
148 
149 		*type = IPVL_IPV4;
150 		lyr3h = ip4h;
151 		break;
152 	}
153 	case htons(ETH_P_IPV6): {
154 		struct ipv6hdr *ip6h;
155 
156 		if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h))))
157 			return NULL;
158 
159 		ip6h = ipv6_hdr(skb);
160 		if (ip6h->version != 6)
161 			return NULL;
162 
163 		*type = IPVL_IPV6;
164 		lyr3h = ip6h;
165 		/* Only Neighbour Solicitation pkts need different treatment */
166 		if (ipv6_addr_any(&ip6h->saddr) &&
167 		    ip6h->nexthdr == NEXTHDR_ICMP) {
168 			struct icmp6hdr	*icmph;
169 
170 			if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h) + sizeof(*icmph))))
171 				return NULL;
172 
173 			ip6h = ipv6_hdr(skb);
174 			icmph = (struct icmp6hdr *)(ip6h + 1);
175 
176 			if (icmph->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
177 				/* Need to access the ipv6 address in body */
178 				if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h) + sizeof(*icmph)
179 						+ sizeof(struct in6_addr))))
180 					return NULL;
181 
182 				ip6h = ipv6_hdr(skb);
183 				icmph = (struct icmp6hdr *)(ip6h + 1);
184 			}
185 
186 			*type = IPVL_ICMPV6;
187 			lyr3h = icmph;
188 		}
189 		break;
190 	}
191 	default:
192 		return NULL;
193 	}
194 
195 	return lyr3h;
196 }
197 
198 unsigned int ipvlan_mac_hash(const unsigned char *addr)
199 {
200 	u32 hash = jhash_1word(__get_unaligned_cpu32(addr+2),
201 			       ipvlan_jhash_secret);
202 
203 	return hash & IPVLAN_MAC_FILTER_MASK;
204 }
205 
206 void ipvlan_process_multicast(struct work_struct *work)
207 {
208 	struct ipvl_port *port = container_of(work, struct ipvl_port, wq);
209 	struct ethhdr *ethh;
210 	struct ipvl_dev *ipvlan;
211 	struct sk_buff *skb, *nskb;
212 	struct sk_buff_head list;
213 	unsigned int len;
214 	unsigned int mac_hash;
215 	int ret;
216 	u8 pkt_type;
217 	bool tx_pkt;
218 
219 	__skb_queue_head_init(&list);
220 
221 	spin_lock_bh(&port->backlog.lock);
222 	skb_queue_splice_tail_init(&port->backlog, &list);
223 	spin_unlock_bh(&port->backlog.lock);
224 
225 	while ((skb = __skb_dequeue(&list)) != NULL) {
226 		struct net_device *dev = skb->dev;
227 		bool consumed = false;
228 
229 		ethh = eth_hdr(skb);
230 		tx_pkt = IPVL_SKB_CB(skb)->tx_pkt;
231 		mac_hash = ipvlan_mac_hash(ethh->h_dest);
232 
233 		if (ether_addr_equal(ethh->h_dest, port->dev->broadcast))
234 			pkt_type = PACKET_BROADCAST;
235 		else
236 			pkt_type = PACKET_MULTICAST;
237 
238 		rcu_read_lock();
239 		list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
240 			if (tx_pkt && (ipvlan->dev == skb->dev))
241 				continue;
242 			if (!test_bit(mac_hash, ipvlan->mac_filters))
243 				continue;
244 			if (!(ipvlan->dev->flags & IFF_UP))
245 				continue;
246 			ret = NET_RX_DROP;
247 			len = skb->len + ETH_HLEN;
248 			nskb = skb_clone(skb, GFP_ATOMIC);
249 			local_bh_disable();
250 			if (nskb) {
251 				consumed = true;
252 				nskb->pkt_type = pkt_type;
253 				nskb->dev = ipvlan->dev;
254 				if (tx_pkt)
255 					ret = dev_forward_skb(ipvlan->dev, nskb);
256 				else
257 					ret = netif_rx(nskb);
258 			}
259 			ipvlan_count_rx(ipvlan, len, ret == NET_RX_SUCCESS, true);
260 			local_bh_enable();
261 		}
262 		rcu_read_unlock();
263 
264 		if (tx_pkt) {
265 			/* If the packet originated here, send it out. */
266 			skb->dev = port->dev;
267 			skb->pkt_type = pkt_type;
268 			dev_queue_xmit(skb);
269 		} else {
270 			if (consumed)
271 				consume_skb(skb);
272 			else
273 				kfree_skb(skb);
274 		}
275 		if (dev)
276 			dev_put(dev);
277 	}
278 }
279 
280 static void ipvlan_skb_crossing_ns(struct sk_buff *skb, struct net_device *dev)
281 {
282 	bool xnet = true;
283 
284 	if (dev)
285 		xnet = !net_eq(dev_net(skb->dev), dev_net(dev));
286 
287 	skb_scrub_packet(skb, xnet);
288 	if (dev)
289 		skb->dev = dev;
290 }
291 
292 static int ipvlan_rcv_frame(struct ipvl_addr *addr, struct sk_buff **pskb,
293 			    bool local)
294 {
295 	struct ipvl_dev *ipvlan = addr->master;
296 	struct net_device *dev = ipvlan->dev;
297 	unsigned int len;
298 	rx_handler_result_t ret = RX_HANDLER_CONSUMED;
299 	bool success = false;
300 	struct sk_buff *skb = *pskb;
301 
302 	len = skb->len + ETH_HLEN;
303 	/* Only packets exchanged between two local slaves need to have
304 	 * device-up check as well as skb-share check.
305 	 */
306 	if (local) {
307 		if (unlikely(!(dev->flags & IFF_UP))) {
308 			kfree_skb(skb);
309 			goto out;
310 		}
311 
312 		skb = skb_share_check(skb, GFP_ATOMIC);
313 		if (!skb)
314 			goto out;
315 
316 		*pskb = skb;
317 	}
318 	ipvlan_skb_crossing_ns(skb, dev);
319 
320 	if (local) {
321 		skb->pkt_type = PACKET_HOST;
322 		if (dev_forward_skb(ipvlan->dev, skb) == NET_RX_SUCCESS)
323 			success = true;
324 	} else {
325 		ret = RX_HANDLER_ANOTHER;
326 		success = true;
327 	}
328 
329 out:
330 	ipvlan_count_rx(ipvlan, len, success, false);
331 	return ret;
332 }
333 
334 static struct ipvl_addr *ipvlan_addr_lookup(struct ipvl_port *port,
335 					    void *lyr3h, int addr_type,
336 					    bool use_dest)
337 {
338 	struct ipvl_addr *addr = NULL;
339 
340 	if (addr_type == IPVL_IPV6) {
341 		struct ipv6hdr *ip6h;
342 		struct in6_addr *i6addr;
343 
344 		ip6h = (struct ipv6hdr *)lyr3h;
345 		i6addr = use_dest ? &ip6h->daddr : &ip6h->saddr;
346 		addr = ipvlan_ht_addr_lookup(port, i6addr, true);
347 	} else if (addr_type == IPVL_ICMPV6) {
348 		struct nd_msg *ndmh;
349 		struct in6_addr *i6addr;
350 
351 		/* Make sure that the NeighborSolicitation ICMPv6 packets
352 		 * are handled to avoid DAD issue.
353 		 */
354 		ndmh = (struct nd_msg *)lyr3h;
355 		if (ndmh->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
356 			i6addr = &ndmh->target;
357 			addr = ipvlan_ht_addr_lookup(port, i6addr, true);
358 		}
359 	} else if (addr_type == IPVL_IPV4) {
360 		struct iphdr *ip4h;
361 		__be32 *i4addr;
362 
363 		ip4h = (struct iphdr *)lyr3h;
364 		i4addr = use_dest ? &ip4h->daddr : &ip4h->saddr;
365 		addr = ipvlan_ht_addr_lookup(port, i4addr, false);
366 	} else if (addr_type == IPVL_ARP) {
367 		struct arphdr *arph;
368 		unsigned char *arp_ptr;
369 		__be32 dip;
370 
371 		arph = (struct arphdr *)lyr3h;
372 		arp_ptr = (unsigned char *)(arph + 1);
373 		if (use_dest)
374 			arp_ptr += (2 * port->dev->addr_len) + 4;
375 		else
376 			arp_ptr += port->dev->addr_len;
377 
378 		memcpy(&dip, arp_ptr, 4);
379 		addr = ipvlan_ht_addr_lookup(port, &dip, false);
380 	}
381 
382 	return addr;
383 }
384 
385 static int ipvlan_process_v4_outbound(struct sk_buff *skb)
386 {
387 	const struct iphdr *ip4h = ip_hdr(skb);
388 	struct net_device *dev = skb->dev;
389 	struct net *net = dev_net(dev);
390 	struct rtable *rt;
391 	int err, ret = NET_XMIT_DROP;
392 	struct flowi4 fl4 = {
393 		.flowi4_oif = dev->ifindex,
394 		.flowi4_tos = RT_TOS(ip4h->tos),
395 		.flowi4_flags = FLOWI_FLAG_ANYSRC,
396 		.daddr = ip4h->daddr,
397 		.saddr = ip4h->saddr,
398 	};
399 
400 	rt = ip_route_output_flow(net, &fl4, NULL);
401 	if (IS_ERR(rt))
402 		goto err;
403 
404 	if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
405 		ip_rt_put(rt);
406 		goto err;
407 	}
408 	skb_dst_set(skb, &rt->dst);
409 	err = ip_local_out(net, skb->sk, skb);
410 	if (unlikely(net_xmit_eval(err)))
411 		dev->stats.tx_errors++;
412 	else
413 		ret = NET_XMIT_SUCCESS;
414 	goto out;
415 err:
416 	dev->stats.tx_errors++;
417 	kfree_skb(skb);
418 out:
419 	return ret;
420 }
421 
422 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
423 {
424 	const struct ipv6hdr *ip6h = ipv6_hdr(skb);
425 	struct net_device *dev = skb->dev;
426 	struct net *net = dev_net(dev);
427 	struct dst_entry *dst;
428 	int err, ret = NET_XMIT_DROP;
429 	struct flowi6 fl6 = {
430 		.flowi6_oif = dev->ifindex,
431 		.daddr = ip6h->daddr,
432 		.saddr = ip6h->saddr,
433 		.flowi6_flags = FLOWI_FLAG_ANYSRC,
434 		.flowlabel = ip6_flowinfo(ip6h),
435 		.flowi6_mark = skb->mark,
436 		.flowi6_proto = ip6h->nexthdr,
437 	};
438 
439 	dst = ip6_route_output(net, NULL, &fl6);
440 	if (dst->error) {
441 		ret = dst->error;
442 		dst_release(dst);
443 		goto err;
444 	}
445 	skb_dst_set(skb, dst);
446 	err = ip6_local_out(net, skb->sk, skb);
447 	if (unlikely(net_xmit_eval(err)))
448 		dev->stats.tx_errors++;
449 	else
450 		ret = NET_XMIT_SUCCESS;
451 	goto out;
452 err:
453 	dev->stats.tx_errors++;
454 	kfree_skb(skb);
455 out:
456 	return ret;
457 }
458 
459 static int ipvlan_process_outbound(struct sk_buff *skb)
460 {
461 	struct ethhdr *ethh = eth_hdr(skb);
462 	int ret = NET_XMIT_DROP;
463 
464 	/* In this mode we dont care about multicast and broadcast traffic */
465 	if (is_multicast_ether_addr(ethh->h_dest)) {
466 		pr_warn_ratelimited("Dropped {multi|broad}cast of type= [%x]\n",
467 				    ntohs(skb->protocol));
468 		kfree_skb(skb);
469 		goto out;
470 	}
471 
472 	/* The ipvlan is a pseudo-L2 device, so the packets that we receive
473 	 * will have L2; which need to discarded and processed further
474 	 * in the net-ns of the main-device.
475 	 */
476 	if (skb_mac_header_was_set(skb)) {
477 		skb_pull(skb, sizeof(*ethh));
478 		skb->mac_header = (typeof(skb->mac_header))~0U;
479 		skb_reset_network_header(skb);
480 	}
481 
482 	if (skb->protocol == htons(ETH_P_IPV6))
483 		ret = ipvlan_process_v6_outbound(skb);
484 	else if (skb->protocol == htons(ETH_P_IP))
485 		ret = ipvlan_process_v4_outbound(skb);
486 	else {
487 		pr_warn_ratelimited("Dropped outbound packet type=%x\n",
488 				    ntohs(skb->protocol));
489 		kfree_skb(skb);
490 	}
491 out:
492 	return ret;
493 }
494 
495 static void ipvlan_multicast_enqueue(struct ipvl_port *port,
496 				     struct sk_buff *skb, bool tx_pkt)
497 {
498 	if (skb->protocol == htons(ETH_P_PAUSE)) {
499 		kfree_skb(skb);
500 		return;
501 	}
502 
503 	/* Record that the deferred packet is from TX or RX path. By
504 	 * looking at mac-addresses on packet will lead to erronus decisions.
505 	 * (This would be true for a loopback-mode on master device or a
506 	 * hair-pin mode of the switch.)
507 	 */
508 	IPVL_SKB_CB(skb)->tx_pkt = tx_pkt;
509 
510 	spin_lock(&port->backlog.lock);
511 	if (skb_queue_len(&port->backlog) < IPVLAN_QBACKLOG_LIMIT) {
512 		if (skb->dev)
513 			dev_hold(skb->dev);
514 		__skb_queue_tail(&port->backlog, skb);
515 		spin_unlock(&port->backlog.lock);
516 		schedule_work(&port->wq);
517 	} else {
518 		spin_unlock(&port->backlog.lock);
519 		atomic_long_inc(&skb->dev->rx_dropped);
520 		kfree_skb(skb);
521 	}
522 }
523 
524 static int ipvlan_xmit_mode_l3(struct sk_buff *skb, struct net_device *dev)
525 {
526 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
527 	void *lyr3h;
528 	struct ipvl_addr *addr;
529 	int addr_type;
530 
531 	lyr3h = ipvlan_get_L3_hdr(ipvlan->port, skb, &addr_type);
532 	if (!lyr3h)
533 		goto out;
534 
535 	if (!ipvlan_is_vepa(ipvlan->port)) {
536 		addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
537 		if (addr) {
538 			if (ipvlan_is_private(ipvlan->port)) {
539 				consume_skb(skb);
540 				return NET_XMIT_DROP;
541 			}
542 			return ipvlan_rcv_frame(addr, &skb, true);
543 		}
544 	}
545 out:
546 	ipvlan_skb_crossing_ns(skb, ipvlan->phy_dev);
547 	return ipvlan_process_outbound(skb);
548 }
549 
550 static int ipvlan_xmit_mode_l2(struct sk_buff *skb, struct net_device *dev)
551 {
552 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
553 	struct ethhdr *eth = eth_hdr(skb);
554 	struct ipvl_addr *addr;
555 	void *lyr3h;
556 	int addr_type;
557 
558 	if (!ipvlan_is_vepa(ipvlan->port) &&
559 	    ether_addr_equal(eth->h_dest, eth->h_source)) {
560 		lyr3h = ipvlan_get_L3_hdr(ipvlan->port, skb, &addr_type);
561 		if (lyr3h) {
562 			addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
563 			if (addr) {
564 				if (ipvlan_is_private(ipvlan->port)) {
565 					consume_skb(skb);
566 					return NET_XMIT_DROP;
567 				}
568 				return ipvlan_rcv_frame(addr, &skb, true);
569 			}
570 		}
571 		skb = skb_share_check(skb, GFP_ATOMIC);
572 		if (!skb)
573 			return NET_XMIT_DROP;
574 
575 		/* Packet definitely does not belong to any of the
576 		 * virtual devices, but the dest is local. So forward
577 		 * the skb for the main-dev. At the RX side we just return
578 		 * RX_PASS for it to be processed further on the stack.
579 		 */
580 		return dev_forward_skb(ipvlan->phy_dev, skb);
581 
582 	} else if (is_multicast_ether_addr(eth->h_dest)) {
583 		ipvlan_skb_crossing_ns(skb, NULL);
584 		ipvlan_multicast_enqueue(ipvlan->port, skb, true);
585 		return NET_XMIT_SUCCESS;
586 	}
587 
588 	ipvlan_skb_crossing_ns(skb, ipvlan->phy_dev);
589 	return dev_queue_xmit(skb);
590 }
591 
592 int ipvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
593 {
594 	struct ipvl_dev *ipvlan = netdev_priv(dev);
595 	struct ipvl_port *port = ipvlan_port_get_rcu_bh(ipvlan->phy_dev);
596 
597 	if (!port)
598 		goto out;
599 
600 	if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
601 		goto out;
602 
603 	switch(port->mode) {
604 	case IPVLAN_MODE_L2:
605 		return ipvlan_xmit_mode_l2(skb, dev);
606 	case IPVLAN_MODE_L3:
607 	case IPVLAN_MODE_L3S:
608 		return ipvlan_xmit_mode_l3(skb, dev);
609 	}
610 
611 	/* Should not reach here */
612 	WARN_ONCE(true, "ipvlan_queue_xmit() called for mode = [%hx]\n",
613 			  port->mode);
614 out:
615 	kfree_skb(skb);
616 	return NET_XMIT_DROP;
617 }
618 
619 static bool ipvlan_external_frame(struct sk_buff *skb, struct ipvl_port *port)
620 {
621 	struct ethhdr *eth = eth_hdr(skb);
622 	struct ipvl_addr *addr;
623 	void *lyr3h;
624 	int addr_type;
625 
626 	if (ether_addr_equal(eth->h_source, skb->dev->dev_addr)) {
627 		lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
628 		if (!lyr3h)
629 			return true;
630 
631 		addr = ipvlan_addr_lookup(port, lyr3h, addr_type, false);
632 		if (addr)
633 			return false;
634 	}
635 
636 	return true;
637 }
638 
639 static rx_handler_result_t ipvlan_handle_mode_l3(struct sk_buff **pskb,
640 						 struct ipvl_port *port)
641 {
642 	void *lyr3h;
643 	int addr_type;
644 	struct ipvl_addr *addr;
645 	struct sk_buff *skb = *pskb;
646 	rx_handler_result_t ret = RX_HANDLER_PASS;
647 
648 	lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
649 	if (!lyr3h)
650 		goto out;
651 
652 	addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
653 	if (addr)
654 		ret = ipvlan_rcv_frame(addr, pskb, false);
655 
656 out:
657 	return ret;
658 }
659 
660 static rx_handler_result_t ipvlan_handle_mode_l2(struct sk_buff **pskb,
661 						 struct ipvl_port *port)
662 {
663 	struct sk_buff *skb = *pskb;
664 	struct ethhdr *eth = eth_hdr(skb);
665 	rx_handler_result_t ret = RX_HANDLER_PASS;
666 	void *lyr3h;
667 	int addr_type;
668 
669 	if (is_multicast_ether_addr(eth->h_dest)) {
670 		if (ipvlan_external_frame(skb, port)) {
671 			struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
672 
673 			/* External frames are queued for device local
674 			 * distribution, but a copy is given to master
675 			 * straight away to avoid sending duplicates later
676 			 * when work-queue processes this frame. This is
677 			 * achieved by returning RX_HANDLER_PASS.
678 			 */
679 			if (nskb) {
680 				ipvlan_skb_crossing_ns(nskb, NULL);
681 				ipvlan_multicast_enqueue(port, nskb, false);
682 			}
683 		}
684 	} else {
685 		struct ipvl_addr *addr;
686 
687 		lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
688 		if (!lyr3h)
689 			return ret;
690 
691 		addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
692 		if (addr)
693 			ret = ipvlan_rcv_frame(addr, pskb, false);
694 	}
695 
696 	return ret;
697 }
698 
699 rx_handler_result_t ipvlan_handle_frame(struct sk_buff **pskb)
700 {
701 	struct sk_buff *skb = *pskb;
702 	struct ipvl_port *port = ipvlan_port_get_rcu(skb->dev);
703 
704 	if (!port)
705 		return RX_HANDLER_PASS;
706 
707 	switch (port->mode) {
708 	case IPVLAN_MODE_L2:
709 		return ipvlan_handle_mode_l2(pskb, port);
710 	case IPVLAN_MODE_L3:
711 		return ipvlan_handle_mode_l3(pskb, port);
712 	case IPVLAN_MODE_L3S:
713 		return RX_HANDLER_PASS;
714 	}
715 
716 	/* Should not reach here */
717 	WARN_ONCE(true, "ipvlan_handle_frame() called for mode = [%hx]\n",
718 			  port->mode);
719 	kfree_skb(skb);
720 	return RX_HANDLER_CONSUMED;
721 }
722 
723 static struct ipvl_addr *ipvlan_skb_to_addr(struct sk_buff *skb,
724 					    struct net_device *dev)
725 {
726 	struct ipvl_addr *addr = NULL;
727 	struct ipvl_port *port;
728 	void *lyr3h;
729 	int addr_type;
730 
731 	if (!dev || !netif_is_ipvlan_port(dev))
732 		goto out;
733 
734 	port = ipvlan_port_get_rcu(dev);
735 	if (!port || port->mode != IPVLAN_MODE_L3S)
736 		goto out;
737 
738 	lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
739 	if (!lyr3h)
740 		goto out;
741 
742 	addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
743 out:
744 	return addr;
745 }
746 
747 struct sk_buff *ipvlan_l3_rcv(struct net_device *dev, struct sk_buff *skb,
748 			      u16 proto)
749 {
750 	struct ipvl_addr *addr;
751 	struct net_device *sdev;
752 
753 	addr = ipvlan_skb_to_addr(skb, dev);
754 	if (!addr)
755 		goto out;
756 
757 	sdev = addr->master->dev;
758 	switch (proto) {
759 	case AF_INET:
760 	{
761 		int err;
762 		struct iphdr *ip4h = ip_hdr(skb);
763 
764 		err = ip_route_input_noref(skb, ip4h->daddr, ip4h->saddr,
765 					   ip4h->tos, sdev);
766 		if (unlikely(err))
767 			goto out;
768 		break;
769 	}
770 	case AF_INET6:
771 	{
772 		struct dst_entry *dst;
773 		struct ipv6hdr *ip6h = ipv6_hdr(skb);
774 		int flags = RT6_LOOKUP_F_HAS_SADDR;
775 		struct flowi6 fl6 = {
776 			.flowi6_iif   = sdev->ifindex,
777 			.daddr        = ip6h->daddr,
778 			.saddr        = ip6h->saddr,
779 			.flowlabel    = ip6_flowinfo(ip6h),
780 			.flowi6_mark  = skb->mark,
781 			.flowi6_proto = ip6h->nexthdr,
782 		};
783 
784 		skb_dst_drop(skb);
785 		dst = ip6_route_input_lookup(dev_net(sdev), sdev, &fl6, flags);
786 		skb_dst_set(skb, dst);
787 		break;
788 	}
789 	default:
790 		break;
791 	}
792 
793 out:
794 	return skb;
795 }
796 
797 unsigned int ipvlan_nf_input(void *priv, struct sk_buff *skb,
798 			     const struct nf_hook_state *state)
799 {
800 	struct ipvl_addr *addr;
801 	unsigned int len;
802 
803 	addr = ipvlan_skb_to_addr(skb, skb->dev);
804 	if (!addr)
805 		goto out;
806 
807 	skb->dev = addr->master->dev;
808 	len = skb->len + ETH_HLEN;
809 	ipvlan_count_rx(addr->master, len, true, false);
810 out:
811 	return NF_ACCEPT;
812 }
813