xref: /openbmc/linux/drivers/net/macvlan.c (revision 06f4e926d256d902dd9a53dcb400fd74974ce087)
1 /*
2  * Copyright (c) 2007 Patrick McHardy <kaber@trash.net>
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public License as
6  * published by the Free Software Foundation; either version 2 of
7  * the License, or (at your option) any later version.
8  *
9  * The code this is based on carried the following copyright notice:
10  * ---
11  * (C) Copyright 2001-2006
12  * Alex Zeffertt, Cambridge Broadband Ltd, ajz@cambridgebroadband.com
13  * Re-worked by Ben Greear <greearb@candelatech.com>
14  * ---
15  */
16 #include <linux/kernel.h>
17 #include <linux/types.h>
18 #include <linux/module.h>
19 #include <linux/init.h>
20 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/string.h>
23 #include <linux/rculist.h>
24 #include <linux/notifier.h>
25 #include <linux/netdevice.h>
26 #include <linux/etherdevice.h>
27 #include <linux/ethtool.h>
28 #include <linux/if_arp.h>
29 #include <linux/if_link.h>
30 #include <linux/if_macvlan.h>
31 #include <net/rtnetlink.h>
32 #include <net/xfrm.h>
33 
34 #define MACVLAN_HASH_SIZE	(1 << BITS_PER_BYTE)
35 
36 struct macvlan_port {
37 	struct net_device	*dev;
38 	struct hlist_head	vlan_hash[MACVLAN_HASH_SIZE];
39 	struct list_head	vlans;
40 	struct rcu_head		rcu;
41 	bool 			passthru;
42 	int			count;
43 };
44 
45 static void macvlan_port_destroy(struct net_device *dev);
46 
47 #define macvlan_port_get_rcu(dev) \
48 	((struct macvlan_port *) rcu_dereference(dev->rx_handler_data))
49 #define macvlan_port_get(dev) ((struct macvlan_port *) dev->rx_handler_data)
50 #define macvlan_port_exists(dev) (dev->priv_flags & IFF_MACVLAN_PORT)
51 
52 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
53 					       const unsigned char *addr)
54 {
55 	struct macvlan_dev *vlan;
56 	struct hlist_node *n;
57 
58 	hlist_for_each_entry_rcu(vlan, n, &port->vlan_hash[addr[5]], hlist) {
59 		if (!compare_ether_addr_64bits(vlan->dev->dev_addr, addr))
60 			return vlan;
61 	}
62 	return NULL;
63 }
64 
65 static void macvlan_hash_add(struct macvlan_dev *vlan)
66 {
67 	struct macvlan_port *port = vlan->port;
68 	const unsigned char *addr = vlan->dev->dev_addr;
69 
70 	hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[addr[5]]);
71 }
72 
73 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync)
74 {
75 	hlist_del_rcu(&vlan->hlist);
76 	if (sync)
77 		synchronize_rcu();
78 }
79 
80 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
81 					const unsigned char *addr)
82 {
83 	macvlan_hash_del(vlan, true);
84 	/* Now that we are unhashed it is safe to change the device
85 	 * address without confusing packet delivery.
86 	 */
87 	memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
88 	macvlan_hash_add(vlan);
89 }
90 
91 static int macvlan_addr_busy(const struct macvlan_port *port,
92 				const unsigned char *addr)
93 {
94 	/* Test to see if the specified multicast address is
95 	 * currently in use by the underlying device or
96 	 * another macvlan.
97 	 */
98 	if (!compare_ether_addr_64bits(port->dev->dev_addr, addr))
99 		return 1;
100 
101 	if (macvlan_hash_lookup(port, addr))
102 		return 1;
103 
104 	return 0;
105 }
106 
107 
108 static int macvlan_broadcast_one(struct sk_buff *skb,
109 				 const struct macvlan_dev *vlan,
110 				 const struct ethhdr *eth, bool local)
111 {
112 	struct net_device *dev = vlan->dev;
113 	if (!skb)
114 		return NET_RX_DROP;
115 
116 	if (local)
117 		return vlan->forward(dev, skb);
118 
119 	skb->dev = dev;
120 	if (!compare_ether_addr_64bits(eth->h_dest,
121 				       dev->broadcast))
122 		skb->pkt_type = PACKET_BROADCAST;
123 	else
124 		skb->pkt_type = PACKET_MULTICAST;
125 
126 	return vlan->receive(skb);
127 }
128 
129 static void macvlan_broadcast(struct sk_buff *skb,
130 			      const struct macvlan_port *port,
131 			      struct net_device *src,
132 			      enum macvlan_mode mode)
133 {
134 	const struct ethhdr *eth = eth_hdr(skb);
135 	const struct macvlan_dev *vlan;
136 	struct hlist_node *n;
137 	struct sk_buff *nskb;
138 	unsigned int i;
139 	int err;
140 
141 	if (skb->protocol == htons(ETH_P_PAUSE))
142 		return;
143 
144 	for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
145 		hlist_for_each_entry_rcu(vlan, n, &port->vlan_hash[i], hlist) {
146 			if (vlan->dev == src || !(vlan->mode & mode))
147 				continue;
148 
149 			nskb = skb_clone(skb, GFP_ATOMIC);
150 			err = macvlan_broadcast_one(nskb, vlan, eth,
151 					 mode == MACVLAN_MODE_BRIDGE);
152 			macvlan_count_rx(vlan, skb->len + ETH_HLEN,
153 					 err == NET_RX_SUCCESS, 1);
154 		}
155 	}
156 }
157 
158 /* called under rcu_read_lock() from netif_receive_skb */
159 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
160 {
161 	struct macvlan_port *port;
162 	struct sk_buff *skb = *pskb;
163 	const struct ethhdr *eth = eth_hdr(skb);
164 	const struct macvlan_dev *vlan;
165 	const struct macvlan_dev *src;
166 	struct net_device *dev;
167 	unsigned int len = 0;
168 	int ret = NET_RX_DROP;
169 
170 	port = macvlan_port_get_rcu(skb->dev);
171 	if (is_multicast_ether_addr(eth->h_dest)) {
172 		src = macvlan_hash_lookup(port, eth->h_source);
173 		if (!src)
174 			/* frame comes from an external address */
175 			macvlan_broadcast(skb, port, NULL,
176 					  MACVLAN_MODE_PRIVATE |
177 					  MACVLAN_MODE_VEPA    |
178 					  MACVLAN_MODE_PASSTHRU|
179 					  MACVLAN_MODE_BRIDGE);
180 		else if (src->mode == MACVLAN_MODE_VEPA)
181 			/* flood to everyone except source */
182 			macvlan_broadcast(skb, port, src->dev,
183 					  MACVLAN_MODE_VEPA |
184 					  MACVLAN_MODE_BRIDGE);
185 		else if (src->mode == MACVLAN_MODE_BRIDGE)
186 			/*
187 			 * flood only to VEPA ports, bridge ports
188 			 * already saw the frame on the way out.
189 			 */
190 			macvlan_broadcast(skb, port, src->dev,
191 					  MACVLAN_MODE_VEPA);
192 		return RX_HANDLER_PASS;
193 	}
194 
195 	if (port->passthru)
196 		vlan = list_first_entry(&port->vlans, struct macvlan_dev, list);
197 	else
198 		vlan = macvlan_hash_lookup(port, eth->h_dest);
199 	if (vlan == NULL)
200 		return RX_HANDLER_PASS;
201 
202 	dev = vlan->dev;
203 	if (unlikely(!(dev->flags & IFF_UP))) {
204 		kfree_skb(skb);
205 		return RX_HANDLER_CONSUMED;
206 	}
207 	len = skb->len + ETH_HLEN;
208 	skb = skb_share_check(skb, GFP_ATOMIC);
209 	if (!skb)
210 		goto out;
211 
212 	skb->dev = dev;
213 	skb->pkt_type = PACKET_HOST;
214 
215 	ret = vlan->receive(skb);
216 
217 out:
218 	macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, 0);
219 	return RX_HANDLER_CONSUMED;
220 }
221 
222 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
223 {
224 	const struct macvlan_dev *vlan = netdev_priv(dev);
225 	const struct macvlan_port *port = vlan->port;
226 	const struct macvlan_dev *dest;
227 	__u8 ip_summed = skb->ip_summed;
228 
229 	if (vlan->mode == MACVLAN_MODE_BRIDGE) {
230 		const struct ethhdr *eth = (void *)skb->data;
231 		skb->ip_summed = CHECKSUM_UNNECESSARY;
232 
233 		/* send to other bridge ports directly */
234 		if (is_multicast_ether_addr(eth->h_dest)) {
235 			macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
236 			goto xmit_world;
237 		}
238 
239 		dest = macvlan_hash_lookup(port, eth->h_dest);
240 		if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
241 			unsigned int length = skb->len + ETH_HLEN;
242 			int ret = dest->forward(dest->dev, skb);
243 			macvlan_count_rx(dest, length,
244 					 ret == NET_RX_SUCCESS, 0);
245 
246 			return NET_XMIT_SUCCESS;
247 		}
248 	}
249 
250 xmit_world:
251 	skb->ip_summed = ip_summed;
252 	skb_set_dev(skb, vlan->lowerdev);
253 	return dev_queue_xmit(skb);
254 }
255 
256 netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
257 			       struct net_device *dev)
258 {
259 	unsigned int len = skb->len;
260 	int ret;
261 	const struct macvlan_dev *vlan = netdev_priv(dev);
262 
263 	ret = macvlan_queue_xmit(skb, dev);
264 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
265 		struct macvlan_pcpu_stats *pcpu_stats;
266 
267 		pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
268 		u64_stats_update_begin(&pcpu_stats->syncp);
269 		pcpu_stats->tx_packets++;
270 		pcpu_stats->tx_bytes += len;
271 		u64_stats_update_end(&pcpu_stats->syncp);
272 	} else {
273 		this_cpu_inc(vlan->pcpu_stats->tx_dropped);
274 	}
275 	return ret;
276 }
277 EXPORT_SYMBOL_GPL(macvlan_start_xmit);
278 
279 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
280 			       unsigned short type, const void *daddr,
281 			       const void *saddr, unsigned len)
282 {
283 	const struct macvlan_dev *vlan = netdev_priv(dev);
284 	struct net_device *lowerdev = vlan->lowerdev;
285 
286 	return dev_hard_header(skb, lowerdev, type, daddr,
287 			       saddr ? : dev->dev_addr, len);
288 }
289 
290 static const struct header_ops macvlan_hard_header_ops = {
291 	.create  	= macvlan_hard_header,
292 	.rebuild	= eth_rebuild_header,
293 	.parse		= eth_header_parse,
294 	.cache		= eth_header_cache,
295 	.cache_update	= eth_header_cache_update,
296 };
297 
298 static int macvlan_open(struct net_device *dev)
299 {
300 	struct macvlan_dev *vlan = netdev_priv(dev);
301 	struct net_device *lowerdev = vlan->lowerdev;
302 	int err;
303 
304 	if (vlan->port->passthru) {
305 		dev_set_promiscuity(lowerdev, 1);
306 		goto hash_add;
307 	}
308 
309 	err = -EBUSY;
310 	if (macvlan_addr_busy(vlan->port, dev->dev_addr))
311 		goto out;
312 
313 	err = dev_uc_add(lowerdev, dev->dev_addr);
314 	if (err < 0)
315 		goto out;
316 	if (dev->flags & IFF_ALLMULTI) {
317 		err = dev_set_allmulti(lowerdev, 1);
318 		if (err < 0)
319 			goto del_unicast;
320 	}
321 
322 hash_add:
323 	macvlan_hash_add(vlan);
324 	return 0;
325 
326 del_unicast:
327 	dev_uc_del(lowerdev, dev->dev_addr);
328 out:
329 	return err;
330 }
331 
332 static int macvlan_stop(struct net_device *dev)
333 {
334 	struct macvlan_dev *vlan = netdev_priv(dev);
335 	struct net_device *lowerdev = vlan->lowerdev;
336 
337 	if (vlan->port->passthru) {
338 		dev_set_promiscuity(lowerdev, -1);
339 		goto hash_del;
340 	}
341 
342 	dev_mc_unsync(lowerdev, dev);
343 	if (dev->flags & IFF_ALLMULTI)
344 		dev_set_allmulti(lowerdev, -1);
345 
346 	dev_uc_del(lowerdev, dev->dev_addr);
347 
348 hash_del:
349 	macvlan_hash_del(vlan, !dev->dismantle);
350 	return 0;
351 }
352 
353 static int macvlan_set_mac_address(struct net_device *dev, void *p)
354 {
355 	struct macvlan_dev *vlan = netdev_priv(dev);
356 	struct net_device *lowerdev = vlan->lowerdev;
357 	struct sockaddr *addr = p;
358 	int err;
359 
360 	if (!is_valid_ether_addr(addr->sa_data))
361 		return -EADDRNOTAVAIL;
362 
363 	if (!(dev->flags & IFF_UP)) {
364 		/* Just copy in the new address */
365 		memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
366 	} else {
367 		/* Rehash and update the device filters */
368 		if (macvlan_addr_busy(vlan->port, addr->sa_data))
369 			return -EBUSY;
370 
371 		err = dev_uc_add(lowerdev, addr->sa_data);
372 		if (err)
373 			return err;
374 
375 		dev_uc_del(lowerdev, dev->dev_addr);
376 
377 		macvlan_hash_change_addr(vlan, addr->sa_data);
378 	}
379 	return 0;
380 }
381 
382 static void macvlan_change_rx_flags(struct net_device *dev, int change)
383 {
384 	struct macvlan_dev *vlan = netdev_priv(dev);
385 	struct net_device *lowerdev = vlan->lowerdev;
386 
387 	if (change & IFF_ALLMULTI)
388 		dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
389 }
390 
391 static void macvlan_set_multicast_list(struct net_device *dev)
392 {
393 	struct macvlan_dev *vlan = netdev_priv(dev);
394 
395 	dev_mc_sync(vlan->lowerdev, dev);
396 }
397 
398 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
399 {
400 	struct macvlan_dev *vlan = netdev_priv(dev);
401 
402 	if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
403 		return -EINVAL;
404 	dev->mtu = new_mtu;
405 	return 0;
406 }
407 
408 /*
409  * macvlan network devices have devices nesting below it and are a special
410  * "super class" of normal network devices; split their locks off into a
411  * separate class since they always nest.
412  */
413 static struct lock_class_key macvlan_netdev_xmit_lock_key;
414 static struct lock_class_key macvlan_netdev_addr_lock_key;
415 
416 #define MACVLAN_FEATURES \
417 	(NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
418 	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
419 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM)
420 
421 #define MACVLAN_STATE_MASK \
422 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
423 
424 static void macvlan_set_lockdep_class_one(struct net_device *dev,
425 					  struct netdev_queue *txq,
426 					  void *_unused)
427 {
428 	lockdep_set_class(&txq->_xmit_lock,
429 			  &macvlan_netdev_xmit_lock_key);
430 }
431 
432 static void macvlan_set_lockdep_class(struct net_device *dev)
433 {
434 	lockdep_set_class(&dev->addr_list_lock,
435 			  &macvlan_netdev_addr_lock_key);
436 	netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
437 }
438 
439 static int macvlan_init(struct net_device *dev)
440 {
441 	struct macvlan_dev *vlan = netdev_priv(dev);
442 	const struct net_device *lowerdev = vlan->lowerdev;
443 
444 	dev->state		= (dev->state & ~MACVLAN_STATE_MASK) |
445 				  (lowerdev->state & MACVLAN_STATE_MASK);
446 	dev->features 		= lowerdev->features & MACVLAN_FEATURES;
447 	dev->features		|= NETIF_F_LLTX;
448 	dev->gso_max_size	= lowerdev->gso_max_size;
449 	dev->iflink		= lowerdev->ifindex;
450 	dev->hard_header_len	= lowerdev->hard_header_len;
451 
452 	macvlan_set_lockdep_class(dev);
453 
454 	vlan->pcpu_stats = alloc_percpu(struct macvlan_pcpu_stats);
455 	if (!vlan->pcpu_stats)
456 		return -ENOMEM;
457 
458 	return 0;
459 }
460 
461 static void macvlan_uninit(struct net_device *dev)
462 {
463 	struct macvlan_dev *vlan = netdev_priv(dev);
464 	struct macvlan_port *port = vlan->port;
465 
466 	free_percpu(vlan->pcpu_stats);
467 
468 	port->count -= 1;
469 	if (!port->count)
470 		macvlan_port_destroy(port->dev);
471 }
472 
473 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
474 							 struct rtnl_link_stats64 *stats)
475 {
476 	struct macvlan_dev *vlan = netdev_priv(dev);
477 
478 	if (vlan->pcpu_stats) {
479 		struct macvlan_pcpu_stats *p;
480 		u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
481 		u32 rx_errors = 0, tx_dropped = 0;
482 		unsigned int start;
483 		int i;
484 
485 		for_each_possible_cpu(i) {
486 			p = per_cpu_ptr(vlan->pcpu_stats, i);
487 			do {
488 				start = u64_stats_fetch_begin_bh(&p->syncp);
489 				rx_packets	= p->rx_packets;
490 				rx_bytes	= p->rx_bytes;
491 				rx_multicast	= p->rx_multicast;
492 				tx_packets	= p->tx_packets;
493 				tx_bytes	= p->tx_bytes;
494 			} while (u64_stats_fetch_retry_bh(&p->syncp, start));
495 
496 			stats->rx_packets	+= rx_packets;
497 			stats->rx_bytes		+= rx_bytes;
498 			stats->multicast	+= rx_multicast;
499 			stats->tx_packets	+= tx_packets;
500 			stats->tx_bytes		+= tx_bytes;
501 			/* rx_errors & tx_dropped are u32, updated
502 			 * without syncp protection.
503 			 */
504 			rx_errors	+= p->rx_errors;
505 			tx_dropped	+= p->tx_dropped;
506 		}
507 		stats->rx_errors	= rx_errors;
508 		stats->rx_dropped	= rx_errors;
509 		stats->tx_dropped	= tx_dropped;
510 	}
511 	return stats;
512 }
513 
514 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
515 					struct ethtool_drvinfo *drvinfo)
516 {
517 	snprintf(drvinfo->driver, 32, "macvlan");
518 	snprintf(drvinfo->version, 32, "0.1");
519 }
520 
521 static int macvlan_ethtool_get_settings(struct net_device *dev,
522 					struct ethtool_cmd *cmd)
523 {
524 	const struct macvlan_dev *vlan = netdev_priv(dev);
525 	return dev_ethtool_get_settings(vlan->lowerdev, cmd);
526 }
527 
528 static const struct ethtool_ops macvlan_ethtool_ops = {
529 	.get_link		= ethtool_op_get_link,
530 	.get_settings		= macvlan_ethtool_get_settings,
531 	.get_drvinfo		= macvlan_ethtool_get_drvinfo,
532 };
533 
534 static const struct net_device_ops macvlan_netdev_ops = {
535 	.ndo_init		= macvlan_init,
536 	.ndo_uninit		= macvlan_uninit,
537 	.ndo_open		= macvlan_open,
538 	.ndo_stop		= macvlan_stop,
539 	.ndo_start_xmit		= macvlan_start_xmit,
540 	.ndo_change_mtu		= macvlan_change_mtu,
541 	.ndo_change_rx_flags	= macvlan_change_rx_flags,
542 	.ndo_set_mac_address	= macvlan_set_mac_address,
543 	.ndo_set_multicast_list	= macvlan_set_multicast_list,
544 	.ndo_get_stats64	= macvlan_dev_get_stats64,
545 	.ndo_validate_addr	= eth_validate_addr,
546 };
547 
548 void macvlan_common_setup(struct net_device *dev)
549 {
550 	ether_setup(dev);
551 
552 	dev->priv_flags	       &= ~IFF_XMIT_DST_RELEASE;
553 	dev->netdev_ops		= &macvlan_netdev_ops;
554 	dev->destructor		= free_netdev;
555 	dev->header_ops		= &macvlan_hard_header_ops,
556 	dev->ethtool_ops	= &macvlan_ethtool_ops;
557 }
558 EXPORT_SYMBOL_GPL(macvlan_common_setup);
559 
560 static void macvlan_setup(struct net_device *dev)
561 {
562 	macvlan_common_setup(dev);
563 	dev->tx_queue_len	= 0;
564 }
565 
566 static int macvlan_port_create(struct net_device *dev)
567 {
568 	struct macvlan_port *port;
569 	unsigned int i;
570 	int err;
571 
572 	if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
573 		return -EINVAL;
574 
575 	port = kzalloc(sizeof(*port), GFP_KERNEL);
576 	if (port == NULL)
577 		return -ENOMEM;
578 
579 	port->passthru = false;
580 	port->dev = dev;
581 	INIT_LIST_HEAD(&port->vlans);
582 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
583 		INIT_HLIST_HEAD(&port->vlan_hash[i]);
584 
585 	err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
586 	if (err)
587 		kfree(port);
588 	else
589 		dev->priv_flags |= IFF_MACVLAN_PORT;
590 	return err;
591 }
592 
593 static void macvlan_port_destroy(struct net_device *dev)
594 {
595 	struct macvlan_port *port = macvlan_port_get(dev);
596 
597 	dev->priv_flags &= ~IFF_MACVLAN_PORT;
598 	netdev_rx_handler_unregister(dev);
599 	kfree_rcu(port, rcu);
600 }
601 
602 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
603 {
604 	if (tb[IFLA_ADDRESS]) {
605 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
606 			return -EINVAL;
607 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
608 			return -EADDRNOTAVAIL;
609 	}
610 
611 	if (data && data[IFLA_MACVLAN_MODE]) {
612 		switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
613 		case MACVLAN_MODE_PRIVATE:
614 		case MACVLAN_MODE_VEPA:
615 		case MACVLAN_MODE_BRIDGE:
616 		case MACVLAN_MODE_PASSTHRU:
617 			break;
618 		default:
619 			return -EINVAL;
620 		}
621 	}
622 	return 0;
623 }
624 
625 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
626 			   struct nlattr *tb[], struct nlattr *data[],
627 			   int (*receive)(struct sk_buff *skb),
628 			   int (*forward)(struct net_device *dev,
629 					  struct sk_buff *skb))
630 {
631 	struct macvlan_dev *vlan = netdev_priv(dev);
632 	struct macvlan_port *port;
633 	struct net_device *lowerdev;
634 	int err;
635 
636 	if (!tb[IFLA_LINK])
637 		return -EINVAL;
638 
639 	lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
640 	if (lowerdev == NULL)
641 		return -ENODEV;
642 
643 	/* When creating macvlans on top of other macvlans - use
644 	 * the real device as the lowerdev.
645 	 */
646 	if (lowerdev->rtnl_link_ops == dev->rtnl_link_ops) {
647 		struct macvlan_dev *lowervlan = netdev_priv(lowerdev);
648 		lowerdev = lowervlan->lowerdev;
649 	}
650 
651 	if (!tb[IFLA_MTU])
652 		dev->mtu = lowerdev->mtu;
653 	else if (dev->mtu > lowerdev->mtu)
654 		return -EINVAL;
655 
656 	if (!tb[IFLA_ADDRESS])
657 		random_ether_addr(dev->dev_addr);
658 
659 	if (!macvlan_port_exists(lowerdev)) {
660 		err = macvlan_port_create(lowerdev);
661 		if (err < 0)
662 			return err;
663 	}
664 	port = macvlan_port_get(lowerdev);
665 
666 	/* Only 1 macvlan device can be created in passthru mode */
667 	if (port->passthru)
668 		return -EINVAL;
669 
670 	vlan->lowerdev = lowerdev;
671 	vlan->dev      = dev;
672 	vlan->port     = port;
673 	vlan->receive  = receive;
674 	vlan->forward  = forward;
675 
676 	vlan->mode     = MACVLAN_MODE_VEPA;
677 	if (data && data[IFLA_MACVLAN_MODE])
678 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
679 
680 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
681 		if (port->count)
682 			return -EINVAL;
683 		port->passthru = true;
684 		memcpy(dev->dev_addr, lowerdev->dev_addr, ETH_ALEN);
685 	}
686 
687 	port->count += 1;
688 	err = register_netdevice(dev);
689 	if (err < 0)
690 		goto destroy_port;
691 
692 	list_add_tail(&vlan->list, &port->vlans);
693 	netif_stacked_transfer_operstate(lowerdev, dev);
694 
695 	return 0;
696 
697 destroy_port:
698 	port->count -= 1;
699 	if (!port->count)
700 		macvlan_port_destroy(lowerdev);
701 
702 	return err;
703 }
704 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
705 
706 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
707 			   struct nlattr *tb[], struct nlattr *data[])
708 {
709 	return macvlan_common_newlink(src_net, dev, tb, data,
710 				      netif_rx,
711 				      dev_forward_skb);
712 }
713 
714 void macvlan_dellink(struct net_device *dev, struct list_head *head)
715 {
716 	struct macvlan_dev *vlan = netdev_priv(dev);
717 
718 	list_del(&vlan->list);
719 	unregister_netdevice_queue(dev, head);
720 }
721 EXPORT_SYMBOL_GPL(macvlan_dellink);
722 
723 static int macvlan_changelink(struct net_device *dev,
724 		struct nlattr *tb[], struct nlattr *data[])
725 {
726 	struct macvlan_dev *vlan = netdev_priv(dev);
727 	if (data && data[IFLA_MACVLAN_MODE])
728 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
729 	return 0;
730 }
731 
732 static size_t macvlan_get_size(const struct net_device *dev)
733 {
734 	return nla_total_size(4);
735 }
736 
737 static int macvlan_fill_info(struct sk_buff *skb,
738 				const struct net_device *dev)
739 {
740 	struct macvlan_dev *vlan = netdev_priv(dev);
741 
742 	NLA_PUT_U32(skb, IFLA_MACVLAN_MODE, vlan->mode);
743 	return 0;
744 
745 nla_put_failure:
746 	return -EMSGSIZE;
747 }
748 
749 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
750 	[IFLA_MACVLAN_MODE] = { .type = NLA_U32 },
751 };
752 
753 int macvlan_link_register(struct rtnl_link_ops *ops)
754 {
755 	/* common fields */
756 	ops->priv_size		= sizeof(struct macvlan_dev);
757 	ops->validate		= macvlan_validate;
758 	ops->maxtype		= IFLA_MACVLAN_MAX;
759 	ops->policy		= macvlan_policy;
760 	ops->changelink		= macvlan_changelink;
761 	ops->get_size		= macvlan_get_size;
762 	ops->fill_info		= macvlan_fill_info;
763 
764 	return rtnl_link_register(ops);
765 };
766 EXPORT_SYMBOL_GPL(macvlan_link_register);
767 
768 static struct rtnl_link_ops macvlan_link_ops = {
769 	.kind		= "macvlan",
770 	.setup		= macvlan_setup,
771 	.newlink	= macvlan_newlink,
772 	.dellink	= macvlan_dellink,
773 };
774 
775 static int macvlan_device_event(struct notifier_block *unused,
776 				unsigned long event, void *ptr)
777 {
778 	struct net_device *dev = ptr;
779 	struct macvlan_dev *vlan, *next;
780 	struct macvlan_port *port;
781 	LIST_HEAD(list_kill);
782 
783 	if (!macvlan_port_exists(dev))
784 		return NOTIFY_DONE;
785 
786 	port = macvlan_port_get(dev);
787 
788 	switch (event) {
789 	case NETDEV_CHANGE:
790 		list_for_each_entry(vlan, &port->vlans, list)
791 			netif_stacked_transfer_operstate(vlan->lowerdev,
792 							 vlan->dev);
793 		break;
794 	case NETDEV_FEAT_CHANGE:
795 		list_for_each_entry(vlan, &port->vlans, list) {
796 			vlan->dev->features = dev->features & MACVLAN_FEATURES;
797 			vlan->dev->gso_max_size = dev->gso_max_size;
798 			netdev_features_change(vlan->dev);
799 		}
800 		break;
801 	case NETDEV_UNREGISTER:
802 		/* twiddle thumbs on netns device moves */
803 		if (dev->reg_state != NETREG_UNREGISTERING)
804 			break;
805 
806 		list_for_each_entry_safe(vlan, next, &port->vlans, list)
807 			vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
808 		unregister_netdevice_many(&list_kill);
809 		list_del(&list_kill);
810 		break;
811 	case NETDEV_PRE_TYPE_CHANGE:
812 		/* Forbid underlaying device to change its type. */
813 		return NOTIFY_BAD;
814 	}
815 	return NOTIFY_DONE;
816 }
817 
818 static struct notifier_block macvlan_notifier_block __read_mostly = {
819 	.notifier_call	= macvlan_device_event,
820 };
821 
822 static int __init macvlan_init_module(void)
823 {
824 	int err;
825 
826 	register_netdevice_notifier(&macvlan_notifier_block);
827 
828 	err = macvlan_link_register(&macvlan_link_ops);
829 	if (err < 0)
830 		goto err1;
831 	return 0;
832 err1:
833 	unregister_netdevice_notifier(&macvlan_notifier_block);
834 	return err;
835 }
836 
837 static void __exit macvlan_cleanup_module(void)
838 {
839 	rtnl_link_unregister(&macvlan_link_ops);
840 	unregister_netdevice_notifier(&macvlan_notifier_block);
841 }
842 
843 module_init(macvlan_init_module);
844 module_exit(macvlan_cleanup_module);
845 
846 MODULE_LICENSE("GPL");
847 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
848 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
849 MODULE_ALIAS_RTNL_LINK("macvlan");
850