xref: /openbmc/linux/drivers/net/ipvlan/ipvlan_core.c (revision 7effbd18)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Copyright (c) 2014 Mahesh Bandewar <maheshb@google.com>
3  */
4 
5 #include "ipvlan.h"
6 
7 static u32 ipvlan_jhash_secret __read_mostly;
8 
9 void ipvlan_init_secret(void)
10 {
11 	net_get_random_once(&ipvlan_jhash_secret, sizeof(ipvlan_jhash_secret));
12 }
13 
14 void ipvlan_count_rx(const struct ipvl_dev *ipvlan,
15 			    unsigned int len, bool success, bool mcast)
16 {
17 	if (likely(success)) {
18 		struct ipvl_pcpu_stats *pcptr;
19 
20 		pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
21 		u64_stats_update_begin(&pcptr->syncp);
22 		u64_stats_inc(&pcptr->rx_pkts);
23 		u64_stats_add(&pcptr->rx_bytes, len);
24 		if (mcast)
25 			u64_stats_inc(&pcptr->rx_mcast);
26 		u64_stats_update_end(&pcptr->syncp);
27 	} else {
28 		this_cpu_inc(ipvlan->pcpu_stats->rx_errs);
29 	}
30 }
31 EXPORT_SYMBOL_GPL(ipvlan_count_rx);
32 
33 #if IS_ENABLED(CONFIG_IPV6)
34 static u8 ipvlan_get_v6_hash(const void *iaddr)
35 {
36 	const struct in6_addr *ip6_addr = iaddr;
37 
38 	return __ipv6_addr_jhash(ip6_addr, ipvlan_jhash_secret) &
39 	       IPVLAN_HASH_MASK;
40 }
41 #else
42 static u8 ipvlan_get_v6_hash(const void *iaddr)
43 {
44 	return 0;
45 }
46 #endif
47 
48 static u8 ipvlan_get_v4_hash(const void *iaddr)
49 {
50 	const struct in_addr *ip4_addr = iaddr;
51 
52 	return jhash_1word(ip4_addr->s_addr, ipvlan_jhash_secret) &
53 	       IPVLAN_HASH_MASK;
54 }
55 
56 static bool addr_equal(bool is_v6, struct ipvl_addr *addr, const void *iaddr)
57 {
58 	if (!is_v6 && addr->atype == IPVL_IPV4) {
59 		struct in_addr *i4addr = (struct in_addr *)iaddr;
60 
61 		return addr->ip4addr.s_addr == i4addr->s_addr;
62 #if IS_ENABLED(CONFIG_IPV6)
63 	} else if (is_v6 && addr->atype == IPVL_IPV6) {
64 		struct in6_addr *i6addr = (struct in6_addr *)iaddr;
65 
66 		return ipv6_addr_equal(&addr->ip6addr, i6addr);
67 #endif
68 	}
69 
70 	return false;
71 }
72 
73 static struct ipvl_addr *ipvlan_ht_addr_lookup(const struct ipvl_port *port,
74 					       const void *iaddr, bool is_v6)
75 {
76 	struct ipvl_addr *addr;
77 	u8 hash;
78 
79 	hash = is_v6 ? ipvlan_get_v6_hash(iaddr) :
80 	       ipvlan_get_v4_hash(iaddr);
81 	hlist_for_each_entry_rcu(addr, &port->hlhead[hash], hlnode)
82 		if (addr_equal(is_v6, addr, iaddr))
83 			return addr;
84 	return NULL;
85 }
86 
87 void ipvlan_ht_addr_add(struct ipvl_dev *ipvlan, struct ipvl_addr *addr)
88 {
89 	struct ipvl_port *port = ipvlan->port;
90 	u8 hash;
91 
92 	hash = (addr->atype == IPVL_IPV6) ?
93 	       ipvlan_get_v6_hash(&addr->ip6addr) :
94 	       ipvlan_get_v4_hash(&addr->ip4addr);
95 	if (hlist_unhashed(&addr->hlnode))
96 		hlist_add_head_rcu(&addr->hlnode, &port->hlhead[hash]);
97 }
98 
99 void ipvlan_ht_addr_del(struct ipvl_addr *addr)
100 {
101 	hlist_del_init_rcu(&addr->hlnode);
102 }
103 
104 struct ipvl_addr *ipvlan_find_addr(const struct ipvl_dev *ipvlan,
105 				   const void *iaddr, bool is_v6)
106 {
107 	struct ipvl_addr *addr, *ret = NULL;
108 
109 	rcu_read_lock();
110 	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode) {
111 		if (addr_equal(is_v6, addr, iaddr)) {
112 			ret = addr;
113 			break;
114 		}
115 	}
116 	rcu_read_unlock();
117 	return ret;
118 }
119 
120 bool ipvlan_addr_busy(struct ipvl_port *port, void *iaddr, bool is_v6)
121 {
122 	struct ipvl_dev *ipvlan;
123 	bool ret = false;
124 
125 	rcu_read_lock();
126 	list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
127 		if (ipvlan_find_addr(ipvlan, iaddr, is_v6)) {
128 			ret = true;
129 			break;
130 		}
131 	}
132 	rcu_read_unlock();
133 	return ret;
134 }
135 
136 void *ipvlan_get_L3_hdr(struct ipvl_port *port, struct sk_buff *skb, int *type)
137 {
138 	void *lyr3h = NULL;
139 
140 	switch (skb->protocol) {
141 	case htons(ETH_P_ARP): {
142 		struct arphdr *arph;
143 
144 		if (unlikely(!pskb_may_pull(skb, arp_hdr_len(port->dev))))
145 			return NULL;
146 
147 		arph = arp_hdr(skb);
148 		*type = IPVL_ARP;
149 		lyr3h = arph;
150 		break;
151 	}
152 	case htons(ETH_P_IP): {
153 		u32 pktlen;
154 		struct iphdr *ip4h;
155 
156 		if (unlikely(!pskb_may_pull(skb, sizeof(*ip4h))))
157 			return NULL;
158 
159 		ip4h = ip_hdr(skb);
160 		pktlen = skb_ip_totlen(skb);
161 		if (ip4h->ihl < 5 || ip4h->version != 4)
162 			return NULL;
163 		if (skb->len < pktlen || pktlen < (ip4h->ihl * 4))
164 			return NULL;
165 
166 		*type = IPVL_IPV4;
167 		lyr3h = ip4h;
168 		break;
169 	}
170 #if IS_ENABLED(CONFIG_IPV6)
171 	case htons(ETH_P_IPV6): {
172 		struct ipv6hdr *ip6h;
173 
174 		if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h))))
175 			return NULL;
176 
177 		ip6h = ipv6_hdr(skb);
178 		if (ip6h->version != 6)
179 			return NULL;
180 
181 		*type = IPVL_IPV6;
182 		lyr3h = ip6h;
183 		/* Only Neighbour Solicitation pkts need different treatment */
184 		if (ipv6_addr_any(&ip6h->saddr) &&
185 		    ip6h->nexthdr == NEXTHDR_ICMP) {
186 			struct icmp6hdr	*icmph;
187 
188 			if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h) + sizeof(*icmph))))
189 				return NULL;
190 
191 			ip6h = ipv6_hdr(skb);
192 			icmph = (struct icmp6hdr *)(ip6h + 1);
193 
194 			if (icmph->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
195 				/* Need to access the ipv6 address in body */
196 				if (unlikely(!pskb_may_pull(skb, sizeof(*ip6h) + sizeof(*icmph)
197 						+ sizeof(struct in6_addr))))
198 					return NULL;
199 
200 				ip6h = ipv6_hdr(skb);
201 				icmph = (struct icmp6hdr *)(ip6h + 1);
202 			}
203 
204 			*type = IPVL_ICMPV6;
205 			lyr3h = icmph;
206 		}
207 		break;
208 	}
209 #endif
210 	default:
211 		return NULL;
212 	}
213 
214 	return lyr3h;
215 }
216 
217 unsigned int ipvlan_mac_hash(const unsigned char *addr)
218 {
219 	u32 hash = jhash_1word(__get_unaligned_cpu32(addr+2),
220 			       ipvlan_jhash_secret);
221 
222 	return hash & IPVLAN_MAC_FILTER_MASK;
223 }
224 
225 void ipvlan_process_multicast(struct work_struct *work)
226 {
227 	struct ipvl_port *port = container_of(work, struct ipvl_port, wq);
228 	struct ethhdr *ethh;
229 	struct ipvl_dev *ipvlan;
230 	struct sk_buff *skb, *nskb;
231 	struct sk_buff_head list;
232 	unsigned int len;
233 	unsigned int mac_hash;
234 	int ret;
235 	u8 pkt_type;
236 	bool tx_pkt;
237 
238 	__skb_queue_head_init(&list);
239 
240 	spin_lock_bh(&port->backlog.lock);
241 	skb_queue_splice_tail_init(&port->backlog, &list);
242 	spin_unlock_bh(&port->backlog.lock);
243 
244 	while ((skb = __skb_dequeue(&list)) != NULL) {
245 		struct net_device *dev = skb->dev;
246 		bool consumed = false;
247 
248 		ethh = eth_hdr(skb);
249 		tx_pkt = IPVL_SKB_CB(skb)->tx_pkt;
250 		mac_hash = ipvlan_mac_hash(ethh->h_dest);
251 
252 		if (ether_addr_equal(ethh->h_dest, port->dev->broadcast))
253 			pkt_type = PACKET_BROADCAST;
254 		else
255 			pkt_type = PACKET_MULTICAST;
256 
257 		rcu_read_lock();
258 		list_for_each_entry_rcu(ipvlan, &port->ipvlans, pnode) {
259 			if (tx_pkt && (ipvlan->dev == skb->dev))
260 				continue;
261 			if (!test_bit(mac_hash, ipvlan->mac_filters))
262 				continue;
263 			if (!(ipvlan->dev->flags & IFF_UP))
264 				continue;
265 			ret = NET_RX_DROP;
266 			len = skb->len + ETH_HLEN;
267 			nskb = skb_clone(skb, GFP_ATOMIC);
268 			local_bh_disable();
269 			if (nskb) {
270 				consumed = true;
271 				nskb->pkt_type = pkt_type;
272 				nskb->dev = ipvlan->dev;
273 				if (tx_pkt)
274 					ret = dev_forward_skb(ipvlan->dev, nskb);
275 				else
276 					ret = netif_rx(nskb);
277 			}
278 			ipvlan_count_rx(ipvlan, len, ret == NET_RX_SUCCESS, true);
279 			local_bh_enable();
280 		}
281 		rcu_read_unlock();
282 
283 		if (tx_pkt) {
284 			/* If the packet originated here, send it out. */
285 			skb->dev = port->dev;
286 			skb->pkt_type = pkt_type;
287 			dev_queue_xmit(skb);
288 		} else {
289 			if (consumed)
290 				consume_skb(skb);
291 			else
292 				kfree_skb(skb);
293 		}
294 		dev_put(dev);
295 		cond_resched();
296 	}
297 }
298 
299 static void ipvlan_skb_crossing_ns(struct sk_buff *skb, struct net_device *dev)
300 {
301 	bool xnet = true;
302 
303 	if (dev)
304 		xnet = !net_eq(dev_net(skb->dev), dev_net(dev));
305 
306 	skb_scrub_packet(skb, xnet);
307 	if (dev)
308 		skb->dev = dev;
309 }
310 
311 static int ipvlan_rcv_frame(struct ipvl_addr *addr, struct sk_buff **pskb,
312 			    bool local)
313 {
314 	struct ipvl_dev *ipvlan = addr->master;
315 	struct net_device *dev = ipvlan->dev;
316 	unsigned int len;
317 	rx_handler_result_t ret = RX_HANDLER_CONSUMED;
318 	bool success = false;
319 	struct sk_buff *skb = *pskb;
320 
321 	len = skb->len + ETH_HLEN;
322 	/* Only packets exchanged between two local slaves need to have
323 	 * device-up check as well as skb-share check.
324 	 */
325 	if (local) {
326 		if (unlikely(!(dev->flags & IFF_UP))) {
327 			kfree_skb(skb);
328 			goto out;
329 		}
330 
331 		skb = skb_share_check(skb, GFP_ATOMIC);
332 		if (!skb)
333 			goto out;
334 
335 		*pskb = skb;
336 	}
337 
338 	if (local) {
339 		skb->pkt_type = PACKET_HOST;
340 		if (dev_forward_skb(ipvlan->dev, skb) == NET_RX_SUCCESS)
341 			success = true;
342 	} else {
343 		skb->dev = dev;
344 		ret = RX_HANDLER_ANOTHER;
345 		success = true;
346 	}
347 
348 out:
349 	ipvlan_count_rx(ipvlan, len, success, false);
350 	return ret;
351 }
352 
353 struct ipvl_addr *ipvlan_addr_lookup(struct ipvl_port *port, void *lyr3h,
354 				     int addr_type, bool use_dest)
355 {
356 	struct ipvl_addr *addr = NULL;
357 
358 	switch (addr_type) {
359 #if IS_ENABLED(CONFIG_IPV6)
360 	case IPVL_IPV6: {
361 		struct ipv6hdr *ip6h;
362 		struct in6_addr *i6addr;
363 
364 		ip6h = (struct ipv6hdr *)lyr3h;
365 		i6addr = use_dest ? &ip6h->daddr : &ip6h->saddr;
366 		addr = ipvlan_ht_addr_lookup(port, i6addr, true);
367 		break;
368 	}
369 	case IPVL_ICMPV6: {
370 		struct nd_msg *ndmh;
371 		struct in6_addr *i6addr;
372 
373 		/* Make sure that the NeighborSolicitation ICMPv6 packets
374 		 * are handled to avoid DAD issue.
375 		 */
376 		ndmh = (struct nd_msg *)lyr3h;
377 		if (ndmh->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) {
378 			i6addr = &ndmh->target;
379 			addr = ipvlan_ht_addr_lookup(port, i6addr, true);
380 		}
381 		break;
382 	}
383 #endif
384 	case IPVL_IPV4: {
385 		struct iphdr *ip4h;
386 		__be32 *i4addr;
387 
388 		ip4h = (struct iphdr *)lyr3h;
389 		i4addr = use_dest ? &ip4h->daddr : &ip4h->saddr;
390 		addr = ipvlan_ht_addr_lookup(port, i4addr, false);
391 		break;
392 	}
393 	case IPVL_ARP: {
394 		struct arphdr *arph;
395 		unsigned char *arp_ptr;
396 		__be32 dip;
397 
398 		arph = (struct arphdr *)lyr3h;
399 		arp_ptr = (unsigned char *)(arph + 1);
400 		if (use_dest)
401 			arp_ptr += (2 * port->dev->addr_len) + 4;
402 		else
403 			arp_ptr += port->dev->addr_len;
404 
405 		memcpy(&dip, arp_ptr, 4);
406 		addr = ipvlan_ht_addr_lookup(port, &dip, false);
407 		break;
408 	}
409 	}
410 
411 	return addr;
412 }
413 
414 static int ipvlan_process_v4_outbound(struct sk_buff *skb)
415 {
416 	const struct iphdr *ip4h = ip_hdr(skb);
417 	struct net_device *dev = skb->dev;
418 	struct net *net = dev_net(dev);
419 	struct rtable *rt;
420 	int err, ret = NET_XMIT_DROP;
421 	struct flowi4 fl4 = {
422 		.flowi4_oif = dev->ifindex,
423 		.flowi4_tos = RT_TOS(ip4h->tos),
424 		.flowi4_flags = FLOWI_FLAG_ANYSRC,
425 		.flowi4_mark = skb->mark,
426 		.daddr = ip4h->daddr,
427 		.saddr = ip4h->saddr,
428 	};
429 
430 	rt = ip_route_output_flow(net, &fl4, NULL);
431 	if (IS_ERR(rt))
432 		goto err;
433 
434 	if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
435 		ip_rt_put(rt);
436 		goto err;
437 	}
438 	skb_dst_set(skb, &rt->dst);
439 	err = ip_local_out(net, skb->sk, skb);
440 	if (unlikely(net_xmit_eval(err)))
441 		dev->stats.tx_errors++;
442 	else
443 		ret = NET_XMIT_SUCCESS;
444 	goto out;
445 err:
446 	dev->stats.tx_errors++;
447 	kfree_skb(skb);
448 out:
449 	return ret;
450 }
451 
452 #if IS_ENABLED(CONFIG_IPV6)
453 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
454 {
455 	const struct ipv6hdr *ip6h = ipv6_hdr(skb);
456 	struct net_device *dev = skb->dev;
457 	struct net *net = dev_net(dev);
458 	struct dst_entry *dst;
459 	int err, ret = NET_XMIT_DROP;
460 	struct flowi6 fl6 = {
461 		.flowi6_oif = dev->ifindex,
462 		.daddr = ip6h->daddr,
463 		.saddr = ip6h->saddr,
464 		.flowi6_flags = FLOWI_FLAG_ANYSRC,
465 		.flowlabel = ip6_flowinfo(ip6h),
466 		.flowi6_mark = skb->mark,
467 		.flowi6_proto = ip6h->nexthdr,
468 	};
469 
470 	dst = ip6_route_output(net, NULL, &fl6);
471 	if (dst->error) {
472 		ret = dst->error;
473 		dst_release(dst);
474 		goto err;
475 	}
476 	skb_dst_set(skb, dst);
477 	err = ip6_local_out(net, skb->sk, skb);
478 	if (unlikely(net_xmit_eval(err)))
479 		dev->stats.tx_errors++;
480 	else
481 		ret = NET_XMIT_SUCCESS;
482 	goto out;
483 err:
484 	dev->stats.tx_errors++;
485 	kfree_skb(skb);
486 out:
487 	return ret;
488 }
489 #else
490 static int ipvlan_process_v6_outbound(struct sk_buff *skb)
491 {
492 	return NET_XMIT_DROP;
493 }
494 #endif
495 
496 static int ipvlan_process_outbound(struct sk_buff *skb)
497 {
498 	int ret = NET_XMIT_DROP;
499 
500 	/* The ipvlan is a pseudo-L2 device, so the packets that we receive
501 	 * will have L2; which need to discarded and processed further
502 	 * in the net-ns of the main-device.
503 	 */
504 	if (skb_mac_header_was_set(skb)) {
505 		/* In this mode we dont care about
506 		 * multicast and broadcast traffic */
507 		struct ethhdr *ethh = eth_hdr(skb);
508 
509 		if (is_multicast_ether_addr(ethh->h_dest)) {
510 			pr_debug_ratelimited(
511 				"Dropped {multi|broad}cast of type=[%x]\n",
512 				ntohs(skb->protocol));
513 			kfree_skb(skb);
514 			goto out;
515 		}
516 
517 		skb_pull(skb, sizeof(*ethh));
518 		skb->mac_header = (typeof(skb->mac_header))~0U;
519 		skb_reset_network_header(skb);
520 	}
521 
522 	if (skb->protocol == htons(ETH_P_IPV6))
523 		ret = ipvlan_process_v6_outbound(skb);
524 	else if (skb->protocol == htons(ETH_P_IP))
525 		ret = ipvlan_process_v4_outbound(skb);
526 	else {
527 		pr_warn_ratelimited("Dropped outbound packet type=%x\n",
528 				    ntohs(skb->protocol));
529 		kfree_skb(skb);
530 	}
531 out:
532 	return ret;
533 }
534 
535 static void ipvlan_multicast_enqueue(struct ipvl_port *port,
536 				     struct sk_buff *skb, bool tx_pkt)
537 {
538 	if (skb->protocol == htons(ETH_P_PAUSE)) {
539 		kfree_skb(skb);
540 		return;
541 	}
542 
543 	/* Record that the deferred packet is from TX or RX path. By
544 	 * looking at mac-addresses on packet will lead to erronus decisions.
545 	 * (This would be true for a loopback-mode on master device or a
546 	 * hair-pin mode of the switch.)
547 	 */
548 	IPVL_SKB_CB(skb)->tx_pkt = tx_pkt;
549 
550 	spin_lock(&port->backlog.lock);
551 	if (skb_queue_len(&port->backlog) < IPVLAN_QBACKLOG_LIMIT) {
552 		if (skb->dev)
553 			dev_hold(skb->dev);
554 		__skb_queue_tail(&port->backlog, skb);
555 		spin_unlock(&port->backlog.lock);
556 		schedule_work(&port->wq);
557 	} else {
558 		spin_unlock(&port->backlog.lock);
559 		dev_core_stats_rx_dropped_inc(skb->dev);
560 		kfree_skb(skb);
561 	}
562 }
563 
564 static int ipvlan_xmit_mode_l3(struct sk_buff *skb, struct net_device *dev)
565 {
566 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
567 	void *lyr3h;
568 	struct ipvl_addr *addr;
569 	int addr_type;
570 
571 	lyr3h = ipvlan_get_L3_hdr(ipvlan->port, skb, &addr_type);
572 	if (!lyr3h)
573 		goto out;
574 
575 	if (!ipvlan_is_vepa(ipvlan->port)) {
576 		addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
577 		if (addr) {
578 			if (ipvlan_is_private(ipvlan->port)) {
579 				consume_skb(skb);
580 				return NET_XMIT_DROP;
581 			}
582 			return ipvlan_rcv_frame(addr, &skb, true);
583 		}
584 	}
585 out:
586 	ipvlan_skb_crossing_ns(skb, ipvlan->phy_dev);
587 	return ipvlan_process_outbound(skb);
588 }
589 
590 static int ipvlan_xmit_mode_l2(struct sk_buff *skb, struct net_device *dev)
591 {
592 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
593 	struct ethhdr *eth = skb_eth_hdr(skb);
594 	struct ipvl_addr *addr;
595 	void *lyr3h;
596 	int addr_type;
597 
598 	if (!ipvlan_is_vepa(ipvlan->port) &&
599 	    ether_addr_equal(eth->h_dest, eth->h_source)) {
600 		lyr3h = ipvlan_get_L3_hdr(ipvlan->port, skb, &addr_type);
601 		if (lyr3h) {
602 			addr = ipvlan_addr_lookup(ipvlan->port, lyr3h, addr_type, true);
603 			if (addr) {
604 				if (ipvlan_is_private(ipvlan->port)) {
605 					consume_skb(skb);
606 					return NET_XMIT_DROP;
607 				}
608 				return ipvlan_rcv_frame(addr, &skb, true);
609 			}
610 		}
611 		skb = skb_share_check(skb, GFP_ATOMIC);
612 		if (!skb)
613 			return NET_XMIT_DROP;
614 
615 		/* Packet definitely does not belong to any of the
616 		 * virtual devices, but the dest is local. So forward
617 		 * the skb for the main-dev. At the RX side we just return
618 		 * RX_PASS for it to be processed further on the stack.
619 		 */
620 		return dev_forward_skb(ipvlan->phy_dev, skb);
621 
622 	} else if (is_multicast_ether_addr(eth->h_dest)) {
623 		skb_reset_mac_header(skb);
624 		ipvlan_skb_crossing_ns(skb, NULL);
625 		ipvlan_multicast_enqueue(ipvlan->port, skb, true);
626 		return NET_XMIT_SUCCESS;
627 	}
628 
629 	skb->dev = ipvlan->phy_dev;
630 	return dev_queue_xmit(skb);
631 }
632 
633 int ipvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
634 {
635 	struct ipvl_dev *ipvlan = netdev_priv(dev);
636 	struct ipvl_port *port = ipvlan_port_get_rcu_bh(ipvlan->phy_dev);
637 
638 	if (!port)
639 		goto out;
640 
641 	if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
642 		goto out;
643 
644 	switch(port->mode) {
645 	case IPVLAN_MODE_L2:
646 		return ipvlan_xmit_mode_l2(skb, dev);
647 	case IPVLAN_MODE_L3:
648 #ifdef CONFIG_IPVLAN_L3S
649 	case IPVLAN_MODE_L3S:
650 #endif
651 		return ipvlan_xmit_mode_l3(skb, dev);
652 	}
653 
654 	/* Should not reach here */
655 	WARN_ONCE(true, "%s called for mode = [%x]\n", __func__, port->mode);
656 out:
657 	kfree_skb(skb);
658 	return NET_XMIT_DROP;
659 }
660 
661 static bool ipvlan_external_frame(struct sk_buff *skb, struct ipvl_port *port)
662 {
663 	struct ethhdr *eth = eth_hdr(skb);
664 	struct ipvl_addr *addr;
665 	void *lyr3h;
666 	int addr_type;
667 
668 	if (ether_addr_equal(eth->h_source, skb->dev->dev_addr)) {
669 		lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
670 		if (!lyr3h)
671 			return true;
672 
673 		addr = ipvlan_addr_lookup(port, lyr3h, addr_type, false);
674 		if (addr)
675 			return false;
676 	}
677 
678 	return true;
679 }
680 
681 static rx_handler_result_t ipvlan_handle_mode_l3(struct sk_buff **pskb,
682 						 struct ipvl_port *port)
683 {
684 	void *lyr3h;
685 	int addr_type;
686 	struct ipvl_addr *addr;
687 	struct sk_buff *skb = *pskb;
688 	rx_handler_result_t ret = RX_HANDLER_PASS;
689 
690 	lyr3h = ipvlan_get_L3_hdr(port, skb, &addr_type);
691 	if (!lyr3h)
692 		goto out;
693 
694 	addr = ipvlan_addr_lookup(port, lyr3h, addr_type, true);
695 	if (addr)
696 		ret = ipvlan_rcv_frame(addr, pskb, false);
697 
698 out:
699 	return ret;
700 }
701 
702 static rx_handler_result_t ipvlan_handle_mode_l2(struct sk_buff **pskb,
703 						 struct ipvl_port *port)
704 {
705 	struct sk_buff *skb = *pskb;
706 	struct ethhdr *eth = eth_hdr(skb);
707 	rx_handler_result_t ret = RX_HANDLER_PASS;
708 
709 	if (is_multicast_ether_addr(eth->h_dest)) {
710 		if (ipvlan_external_frame(skb, port)) {
711 			struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
712 
713 			/* External frames are queued for device local
714 			 * distribution, but a copy is given to master
715 			 * straight away to avoid sending duplicates later
716 			 * when work-queue processes this frame. This is
717 			 * achieved by returning RX_HANDLER_PASS.
718 			 */
719 			if (nskb) {
720 				ipvlan_skb_crossing_ns(nskb, NULL);
721 				ipvlan_multicast_enqueue(port, nskb, false);
722 			}
723 		}
724 	} else {
725 		/* Perform like l3 mode for non-multicast packet */
726 		ret = ipvlan_handle_mode_l3(pskb, port);
727 	}
728 
729 	return ret;
730 }
731 
732 rx_handler_result_t ipvlan_handle_frame(struct sk_buff **pskb)
733 {
734 	struct sk_buff *skb = *pskb;
735 	struct ipvl_port *port = ipvlan_port_get_rcu(skb->dev);
736 
737 	if (!port)
738 		return RX_HANDLER_PASS;
739 
740 	switch (port->mode) {
741 	case IPVLAN_MODE_L2:
742 		return ipvlan_handle_mode_l2(pskb, port);
743 	case IPVLAN_MODE_L3:
744 		return ipvlan_handle_mode_l3(pskb, port);
745 #ifdef CONFIG_IPVLAN_L3S
746 	case IPVLAN_MODE_L3S:
747 		return RX_HANDLER_PASS;
748 #endif
749 	}
750 
751 	/* Should not reach here */
752 	WARN_ONCE(true, "%s called for mode = [%x]\n", __func__, port->mode);
753 	kfree_skb(skb);
754 	return RX_HANDLER_CONSUMED;
755 }
756