xref: /linux/drivers/net/macvlan.c (revision c54ea4918c2b7722d7242ea53271356501988a9b)
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 };
43 
44 #define macvlan_port_get_rcu(dev) \
45 	((struct macvlan_port *) rcu_dereference(dev->rx_handler_data))
46 #define macvlan_port_get(dev) ((struct macvlan_port *) dev->rx_handler_data)
47 #define macvlan_port_exists(dev) (dev->priv_flags & IFF_MACVLAN_PORT)
48 
49 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
50 					       const unsigned char *addr)
51 {
52 	struct macvlan_dev *vlan;
53 	struct hlist_node *n;
54 
55 	hlist_for_each_entry_rcu(vlan, n, &port->vlan_hash[addr[5]], hlist) {
56 		if (!compare_ether_addr_64bits(vlan->dev->dev_addr, addr))
57 			return vlan;
58 	}
59 	return NULL;
60 }
61 
62 static void macvlan_hash_add(struct macvlan_dev *vlan)
63 {
64 	struct macvlan_port *port = vlan->port;
65 	const unsigned char *addr = vlan->dev->dev_addr;
66 
67 	hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[addr[5]]);
68 }
69 
70 static void macvlan_hash_del(struct macvlan_dev *vlan)
71 {
72 	hlist_del_rcu(&vlan->hlist);
73 	synchronize_rcu();
74 }
75 
76 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
77 					const unsigned char *addr)
78 {
79 	macvlan_hash_del(vlan);
80 	/* Now that we are unhashed it is safe to change the device
81 	 * address without confusing packet delivery.
82 	 */
83 	memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
84 	macvlan_hash_add(vlan);
85 }
86 
87 static int macvlan_addr_busy(const struct macvlan_port *port,
88 				const unsigned char *addr)
89 {
90 	/* Test to see if the specified multicast address is
91 	 * currently in use by the underlying device or
92 	 * another macvlan.
93 	 */
94 	if (!compare_ether_addr_64bits(port->dev->dev_addr, addr))
95 		return 1;
96 
97 	if (macvlan_hash_lookup(port, addr))
98 		return 1;
99 
100 	return 0;
101 }
102 
103 
104 static int macvlan_broadcast_one(struct sk_buff *skb,
105 				 const struct macvlan_dev *vlan,
106 				 const struct ethhdr *eth, bool local)
107 {
108 	struct net_device *dev = vlan->dev;
109 	if (!skb)
110 		return NET_RX_DROP;
111 
112 	if (local)
113 		return vlan->forward(dev, skb);
114 
115 	skb->dev = dev;
116 	if (!compare_ether_addr_64bits(eth->h_dest,
117 				       dev->broadcast))
118 		skb->pkt_type = PACKET_BROADCAST;
119 	else
120 		skb->pkt_type = PACKET_MULTICAST;
121 
122 	return vlan->receive(skb);
123 }
124 
125 static void macvlan_broadcast(struct sk_buff *skb,
126 			      const struct macvlan_port *port,
127 			      struct net_device *src,
128 			      enum macvlan_mode mode)
129 {
130 	const struct ethhdr *eth = eth_hdr(skb);
131 	const struct macvlan_dev *vlan;
132 	struct hlist_node *n;
133 	struct sk_buff *nskb;
134 	unsigned int i;
135 	int err;
136 
137 	if (skb->protocol == htons(ETH_P_PAUSE))
138 		return;
139 
140 	for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
141 		hlist_for_each_entry_rcu(vlan, n, &port->vlan_hash[i], hlist) {
142 			if (vlan->dev == src || !(vlan->mode & mode))
143 				continue;
144 
145 			nskb = skb_clone(skb, GFP_ATOMIC);
146 			err = macvlan_broadcast_one(nskb, vlan, eth,
147 					 mode == MACVLAN_MODE_BRIDGE);
148 			macvlan_count_rx(vlan, skb->len + ETH_HLEN,
149 					 err == NET_RX_SUCCESS, 1);
150 		}
151 	}
152 }
153 
154 /* called under rcu_read_lock() from netif_receive_skb */
155 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
156 {
157 	struct macvlan_port *port;
158 	struct sk_buff *skb = *pskb;
159 	const struct ethhdr *eth = eth_hdr(skb);
160 	const struct macvlan_dev *vlan;
161 	const struct macvlan_dev *src;
162 	struct net_device *dev;
163 	unsigned int len = 0;
164 	int ret = NET_RX_DROP;
165 
166 	port = macvlan_port_get_rcu(skb->dev);
167 	if (is_multicast_ether_addr(eth->h_dest)) {
168 		src = macvlan_hash_lookup(port, eth->h_source);
169 		if (!src)
170 			/* frame comes from an external address */
171 			macvlan_broadcast(skb, port, NULL,
172 					  MACVLAN_MODE_PRIVATE |
173 					  MACVLAN_MODE_VEPA    |
174 					  MACVLAN_MODE_PASSTHRU|
175 					  MACVLAN_MODE_BRIDGE);
176 		else if (src->mode == MACVLAN_MODE_VEPA)
177 			/* flood to everyone except source */
178 			macvlan_broadcast(skb, port, src->dev,
179 					  MACVLAN_MODE_VEPA |
180 					  MACVLAN_MODE_BRIDGE);
181 		else if (src->mode == MACVLAN_MODE_BRIDGE)
182 			/*
183 			 * flood only to VEPA ports, bridge ports
184 			 * already saw the frame on the way out.
185 			 */
186 			macvlan_broadcast(skb, port, src->dev,
187 					  MACVLAN_MODE_VEPA);
188 		return RX_HANDLER_PASS;
189 	}
190 
191 	if (port->passthru)
192 		vlan = list_first_entry(&port->vlans, struct macvlan_dev, list);
193 	else
194 		vlan = macvlan_hash_lookup(port, eth->h_dest);
195 	if (vlan == NULL)
196 		return RX_HANDLER_PASS;
197 
198 	dev = vlan->dev;
199 	if (unlikely(!(dev->flags & IFF_UP))) {
200 		kfree_skb(skb);
201 		return RX_HANDLER_CONSUMED;
202 	}
203 	len = skb->len + ETH_HLEN;
204 	skb = skb_share_check(skb, GFP_ATOMIC);
205 	if (!skb)
206 		goto out;
207 
208 	skb->dev = dev;
209 	skb->pkt_type = PACKET_HOST;
210 
211 	ret = vlan->receive(skb);
212 
213 out:
214 	macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, 0);
215 	return RX_HANDLER_CONSUMED;
216 }
217 
218 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
219 {
220 	const struct macvlan_dev *vlan = netdev_priv(dev);
221 	const struct macvlan_port *port = vlan->port;
222 	const struct macvlan_dev *dest;
223 	__u8 ip_summed = skb->ip_summed;
224 
225 	if (vlan->mode == MACVLAN_MODE_BRIDGE) {
226 		const struct ethhdr *eth = (void *)skb->data;
227 		skb->ip_summed = CHECKSUM_UNNECESSARY;
228 
229 		/* send to other bridge ports directly */
230 		if (is_multicast_ether_addr(eth->h_dest)) {
231 			macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
232 			goto xmit_world;
233 		}
234 
235 		dest = macvlan_hash_lookup(port, eth->h_dest);
236 		if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
237 			unsigned int length = skb->len + ETH_HLEN;
238 			int ret = dest->forward(dest->dev, skb);
239 			macvlan_count_rx(dest, length,
240 					 ret == NET_RX_SUCCESS, 0);
241 
242 			return NET_XMIT_SUCCESS;
243 		}
244 	}
245 
246 xmit_world:
247 	skb->ip_summed = ip_summed;
248 	skb_set_dev(skb, vlan->lowerdev);
249 	return dev_queue_xmit(skb);
250 }
251 
252 netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
253 			       struct net_device *dev)
254 {
255 	unsigned int len = skb->len;
256 	int ret;
257 	const struct macvlan_dev *vlan = netdev_priv(dev);
258 
259 	ret = macvlan_queue_xmit(skb, dev);
260 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
261 		struct macvlan_pcpu_stats *pcpu_stats;
262 
263 		pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
264 		u64_stats_update_begin(&pcpu_stats->syncp);
265 		pcpu_stats->tx_packets++;
266 		pcpu_stats->tx_bytes += len;
267 		u64_stats_update_end(&pcpu_stats->syncp);
268 	} else {
269 		this_cpu_inc(vlan->pcpu_stats->tx_dropped);
270 	}
271 	return ret;
272 }
273 EXPORT_SYMBOL_GPL(macvlan_start_xmit);
274 
275 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
276 			       unsigned short type, const void *daddr,
277 			       const void *saddr, unsigned len)
278 {
279 	const struct macvlan_dev *vlan = netdev_priv(dev);
280 	struct net_device *lowerdev = vlan->lowerdev;
281 
282 	return dev_hard_header(skb, lowerdev, type, daddr,
283 			       saddr ? : dev->dev_addr, len);
284 }
285 
286 static const struct header_ops macvlan_hard_header_ops = {
287 	.create  	= macvlan_hard_header,
288 	.rebuild	= eth_rebuild_header,
289 	.parse		= eth_header_parse,
290 	.cache		= eth_header_cache,
291 	.cache_update	= eth_header_cache_update,
292 };
293 
294 static int macvlan_open(struct net_device *dev)
295 {
296 	struct macvlan_dev *vlan = netdev_priv(dev);
297 	struct net_device *lowerdev = vlan->lowerdev;
298 	int err;
299 
300 	if (vlan->port->passthru) {
301 		dev_set_promiscuity(lowerdev, 1);
302 		goto hash_add;
303 	}
304 
305 	err = -EBUSY;
306 	if (macvlan_addr_busy(vlan->port, dev->dev_addr))
307 		goto out;
308 
309 	err = dev_uc_add(lowerdev, dev->dev_addr);
310 	if (err < 0)
311 		goto out;
312 	if (dev->flags & IFF_ALLMULTI) {
313 		err = dev_set_allmulti(lowerdev, 1);
314 		if (err < 0)
315 			goto del_unicast;
316 	}
317 
318 hash_add:
319 	macvlan_hash_add(vlan);
320 	return 0;
321 
322 del_unicast:
323 	dev_uc_del(lowerdev, dev->dev_addr);
324 out:
325 	return err;
326 }
327 
328 static int macvlan_stop(struct net_device *dev)
329 {
330 	struct macvlan_dev *vlan = netdev_priv(dev);
331 	struct net_device *lowerdev = vlan->lowerdev;
332 
333 	if (vlan->port->passthru) {
334 		dev_set_promiscuity(lowerdev, -1);
335 		goto hash_del;
336 	}
337 
338 	dev_mc_unsync(lowerdev, dev);
339 	if (dev->flags & IFF_ALLMULTI)
340 		dev_set_allmulti(lowerdev, -1);
341 
342 	dev_uc_del(lowerdev, dev->dev_addr);
343 
344 hash_del:
345 	macvlan_hash_del(vlan);
346 	return 0;
347 }
348 
349 static int macvlan_set_mac_address(struct net_device *dev, void *p)
350 {
351 	struct macvlan_dev *vlan = netdev_priv(dev);
352 	struct net_device *lowerdev = vlan->lowerdev;
353 	struct sockaddr *addr = p;
354 	int err;
355 
356 	if (!is_valid_ether_addr(addr->sa_data))
357 		return -EADDRNOTAVAIL;
358 
359 	if (!(dev->flags & IFF_UP)) {
360 		/* Just copy in the new address */
361 		memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
362 	} else {
363 		/* Rehash and update the device filters */
364 		if (macvlan_addr_busy(vlan->port, addr->sa_data))
365 			return -EBUSY;
366 
367 		err = dev_uc_add(lowerdev, addr->sa_data);
368 		if (err)
369 			return err;
370 
371 		dev_uc_del(lowerdev, dev->dev_addr);
372 
373 		macvlan_hash_change_addr(vlan, addr->sa_data);
374 	}
375 	return 0;
376 }
377 
378 static void macvlan_change_rx_flags(struct net_device *dev, int change)
379 {
380 	struct macvlan_dev *vlan = netdev_priv(dev);
381 	struct net_device *lowerdev = vlan->lowerdev;
382 
383 	if (change & IFF_ALLMULTI)
384 		dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
385 }
386 
387 static void macvlan_set_multicast_list(struct net_device *dev)
388 {
389 	struct macvlan_dev *vlan = netdev_priv(dev);
390 
391 	dev_mc_sync(vlan->lowerdev, dev);
392 }
393 
394 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
395 {
396 	struct macvlan_dev *vlan = netdev_priv(dev);
397 
398 	if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
399 		return -EINVAL;
400 	dev->mtu = new_mtu;
401 	return 0;
402 }
403 
404 /*
405  * macvlan network devices have devices nesting below it and are a special
406  * "super class" of normal network devices; split their locks off into a
407  * separate class since they always nest.
408  */
409 static struct lock_class_key macvlan_netdev_xmit_lock_key;
410 static struct lock_class_key macvlan_netdev_addr_lock_key;
411 
412 #define MACVLAN_FEATURES \
413 	(NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
414 	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
415 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO)
416 
417 #define MACVLAN_STATE_MASK \
418 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
419 
420 static void macvlan_set_lockdep_class_one(struct net_device *dev,
421 					  struct netdev_queue *txq,
422 					  void *_unused)
423 {
424 	lockdep_set_class(&txq->_xmit_lock,
425 			  &macvlan_netdev_xmit_lock_key);
426 }
427 
428 static void macvlan_set_lockdep_class(struct net_device *dev)
429 {
430 	lockdep_set_class(&dev->addr_list_lock,
431 			  &macvlan_netdev_addr_lock_key);
432 	netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
433 }
434 
435 static int macvlan_init(struct net_device *dev)
436 {
437 	struct macvlan_dev *vlan = netdev_priv(dev);
438 	const struct net_device *lowerdev = vlan->lowerdev;
439 
440 	dev->state		= (dev->state & ~MACVLAN_STATE_MASK) |
441 				  (lowerdev->state & MACVLAN_STATE_MASK);
442 	dev->features 		= lowerdev->features & MACVLAN_FEATURES;
443 	dev->features		|= NETIF_F_LLTX;
444 	dev->gso_max_size	= lowerdev->gso_max_size;
445 	dev->iflink		= lowerdev->ifindex;
446 	dev->hard_header_len	= lowerdev->hard_header_len;
447 
448 	macvlan_set_lockdep_class(dev);
449 
450 	vlan->pcpu_stats = alloc_percpu(struct macvlan_pcpu_stats);
451 	if (!vlan->pcpu_stats)
452 		return -ENOMEM;
453 
454 	return 0;
455 }
456 
457 static void macvlan_uninit(struct net_device *dev)
458 {
459 	struct macvlan_dev *vlan = netdev_priv(dev);
460 
461 	free_percpu(vlan->pcpu_stats);
462 }
463 
464 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
465 							 struct rtnl_link_stats64 *stats)
466 {
467 	struct macvlan_dev *vlan = netdev_priv(dev);
468 
469 	if (vlan->pcpu_stats) {
470 		struct macvlan_pcpu_stats *p;
471 		u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
472 		u32 rx_errors = 0, tx_dropped = 0;
473 		unsigned int start;
474 		int i;
475 
476 		for_each_possible_cpu(i) {
477 			p = per_cpu_ptr(vlan->pcpu_stats, i);
478 			do {
479 				start = u64_stats_fetch_begin_bh(&p->syncp);
480 				rx_packets	= p->rx_packets;
481 				rx_bytes	= p->rx_bytes;
482 				rx_multicast	= p->rx_multicast;
483 				tx_packets	= p->tx_packets;
484 				tx_bytes	= p->tx_bytes;
485 			} while (u64_stats_fetch_retry_bh(&p->syncp, start));
486 
487 			stats->rx_packets	+= rx_packets;
488 			stats->rx_bytes		+= rx_bytes;
489 			stats->multicast	+= rx_multicast;
490 			stats->tx_packets	+= tx_packets;
491 			stats->tx_bytes		+= tx_bytes;
492 			/* rx_errors & tx_dropped are u32, updated
493 			 * without syncp protection.
494 			 */
495 			rx_errors	+= p->rx_errors;
496 			tx_dropped	+= p->tx_dropped;
497 		}
498 		stats->rx_errors	= rx_errors;
499 		stats->rx_dropped	= rx_errors;
500 		stats->tx_dropped	= tx_dropped;
501 	}
502 	return stats;
503 }
504 
505 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
506 					struct ethtool_drvinfo *drvinfo)
507 {
508 	snprintf(drvinfo->driver, 32, "macvlan");
509 	snprintf(drvinfo->version, 32, "0.1");
510 }
511 
512 static u32 macvlan_ethtool_get_rx_csum(struct net_device *dev)
513 {
514 	const struct macvlan_dev *vlan = netdev_priv(dev);
515 	return dev_ethtool_get_rx_csum(vlan->lowerdev);
516 }
517 
518 static int macvlan_ethtool_get_settings(struct net_device *dev,
519 					struct ethtool_cmd *cmd)
520 {
521 	const struct macvlan_dev *vlan = netdev_priv(dev);
522 	return dev_ethtool_get_settings(vlan->lowerdev, cmd);
523 }
524 
525 static u32 macvlan_ethtool_get_flags(struct net_device *dev)
526 {
527 	const struct macvlan_dev *vlan = netdev_priv(dev);
528 	return dev_ethtool_get_flags(vlan->lowerdev);
529 }
530 
531 static const struct ethtool_ops macvlan_ethtool_ops = {
532 	.get_link		= ethtool_op_get_link,
533 	.get_settings		= macvlan_ethtool_get_settings,
534 	.get_rx_csum		= macvlan_ethtool_get_rx_csum,
535 	.get_drvinfo		= macvlan_ethtool_get_drvinfo,
536 	.get_flags		= macvlan_ethtool_get_flags,
537 };
538 
539 static const struct net_device_ops macvlan_netdev_ops = {
540 	.ndo_init		= macvlan_init,
541 	.ndo_uninit		= macvlan_uninit,
542 	.ndo_open		= macvlan_open,
543 	.ndo_stop		= macvlan_stop,
544 	.ndo_start_xmit		= macvlan_start_xmit,
545 	.ndo_change_mtu		= macvlan_change_mtu,
546 	.ndo_change_rx_flags	= macvlan_change_rx_flags,
547 	.ndo_set_mac_address	= macvlan_set_mac_address,
548 	.ndo_set_multicast_list	= macvlan_set_multicast_list,
549 	.ndo_get_stats64	= macvlan_dev_get_stats64,
550 	.ndo_validate_addr	= eth_validate_addr,
551 };
552 
553 void macvlan_common_setup(struct net_device *dev)
554 {
555 	ether_setup(dev);
556 
557 	dev->priv_flags	       &= ~IFF_XMIT_DST_RELEASE;
558 	dev->netdev_ops		= &macvlan_netdev_ops;
559 	dev->destructor		= free_netdev;
560 	dev->header_ops		= &macvlan_hard_header_ops,
561 	dev->ethtool_ops	= &macvlan_ethtool_ops;
562 }
563 EXPORT_SYMBOL_GPL(macvlan_common_setup);
564 
565 static void macvlan_setup(struct net_device *dev)
566 {
567 	macvlan_common_setup(dev);
568 	dev->tx_queue_len	= 0;
569 }
570 
571 static int macvlan_port_create(struct net_device *dev)
572 {
573 	struct macvlan_port *port;
574 	unsigned int i;
575 	int err;
576 
577 	if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
578 		return -EINVAL;
579 
580 	port = kzalloc(sizeof(*port), GFP_KERNEL);
581 	if (port == NULL)
582 		return -ENOMEM;
583 
584 	port->passthru = false;
585 	port->dev = dev;
586 	INIT_LIST_HEAD(&port->vlans);
587 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
588 		INIT_HLIST_HEAD(&port->vlan_hash[i]);
589 
590 	err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
591 	if (err)
592 		kfree(port);
593 
594 	dev->priv_flags |= IFF_MACVLAN_PORT;
595 	return err;
596 }
597 
598 static void macvlan_port_rcu_free(struct rcu_head *head)
599 {
600 	struct macvlan_port *port;
601 
602 	port = container_of(head, struct macvlan_port, rcu);
603 	kfree(port);
604 }
605 
606 static void macvlan_port_destroy(struct net_device *dev)
607 {
608 	struct macvlan_port *port = macvlan_port_get(dev);
609 
610 	dev->priv_flags &= ~IFF_MACVLAN_PORT;
611 	netdev_rx_handler_unregister(dev);
612 	call_rcu(&port->rcu, macvlan_port_rcu_free);
613 }
614 
615 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
616 {
617 	if (tb[IFLA_ADDRESS]) {
618 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
619 			return -EINVAL;
620 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
621 			return -EADDRNOTAVAIL;
622 	}
623 
624 	if (data && data[IFLA_MACVLAN_MODE]) {
625 		switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
626 		case MACVLAN_MODE_PRIVATE:
627 		case MACVLAN_MODE_VEPA:
628 		case MACVLAN_MODE_BRIDGE:
629 		case MACVLAN_MODE_PASSTHRU:
630 			break;
631 		default:
632 			return -EINVAL;
633 		}
634 	}
635 	return 0;
636 }
637 
638 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
639 			   struct nlattr *tb[], struct nlattr *data[],
640 			   int (*receive)(struct sk_buff *skb),
641 			   int (*forward)(struct net_device *dev,
642 					  struct sk_buff *skb))
643 {
644 	struct macvlan_dev *vlan = netdev_priv(dev);
645 	struct macvlan_port *port;
646 	struct net_device *lowerdev;
647 	int err;
648 
649 	if (!tb[IFLA_LINK])
650 		return -EINVAL;
651 
652 	lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
653 	if (lowerdev == NULL)
654 		return -ENODEV;
655 
656 	/* When creating macvlans on top of other macvlans - use
657 	 * the real device as the lowerdev.
658 	 */
659 	if (lowerdev->rtnl_link_ops == dev->rtnl_link_ops) {
660 		struct macvlan_dev *lowervlan = netdev_priv(lowerdev);
661 		lowerdev = lowervlan->lowerdev;
662 	}
663 
664 	if (!tb[IFLA_MTU])
665 		dev->mtu = lowerdev->mtu;
666 	else if (dev->mtu > lowerdev->mtu)
667 		return -EINVAL;
668 
669 	if (!tb[IFLA_ADDRESS])
670 		random_ether_addr(dev->dev_addr);
671 
672 	if (!macvlan_port_exists(lowerdev)) {
673 		err = macvlan_port_create(lowerdev);
674 		if (err < 0)
675 			return err;
676 	}
677 	port = macvlan_port_get(lowerdev);
678 
679 	/* Only 1 macvlan device can be created in passthru mode */
680 	if (port->passthru)
681 		return -EINVAL;
682 
683 	vlan->lowerdev = lowerdev;
684 	vlan->dev      = dev;
685 	vlan->port     = port;
686 	vlan->receive  = receive;
687 	vlan->forward  = forward;
688 
689 	vlan->mode     = MACVLAN_MODE_VEPA;
690 	if (data && data[IFLA_MACVLAN_MODE])
691 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
692 
693 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
694 		if (!list_empty(&port->vlans))
695 			return -EINVAL;
696 		port->passthru = true;
697 		memcpy(dev->dev_addr, lowerdev->dev_addr, ETH_ALEN);
698 	}
699 
700 	err = register_netdevice(dev);
701 	if (err < 0)
702 		goto destroy_port;
703 
704 	list_add_tail(&vlan->list, &port->vlans);
705 	netif_stacked_transfer_operstate(lowerdev, dev);
706 
707 	return 0;
708 
709 destroy_port:
710 	if (list_empty(&port->vlans))
711 		macvlan_port_destroy(lowerdev);
712 
713 	return err;
714 }
715 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
716 
717 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
718 			   struct nlattr *tb[], struct nlattr *data[])
719 {
720 	return macvlan_common_newlink(src_net, dev, tb, data,
721 				      netif_rx,
722 				      dev_forward_skb);
723 }
724 
725 void macvlan_dellink(struct net_device *dev, struct list_head *head)
726 {
727 	struct macvlan_dev *vlan = netdev_priv(dev);
728 	struct macvlan_port *port = vlan->port;
729 
730 	list_del(&vlan->list);
731 	unregister_netdevice_queue(dev, head);
732 
733 	if (list_empty(&port->vlans))
734 		macvlan_port_destroy(port->dev);
735 }
736 EXPORT_SYMBOL_GPL(macvlan_dellink);
737 
738 static int macvlan_changelink(struct net_device *dev,
739 		struct nlattr *tb[], struct nlattr *data[])
740 {
741 	struct macvlan_dev *vlan = netdev_priv(dev);
742 	if (data && data[IFLA_MACVLAN_MODE])
743 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
744 	return 0;
745 }
746 
747 static size_t macvlan_get_size(const struct net_device *dev)
748 {
749 	return nla_total_size(4);
750 }
751 
752 static int macvlan_fill_info(struct sk_buff *skb,
753 				const struct net_device *dev)
754 {
755 	struct macvlan_dev *vlan = netdev_priv(dev);
756 
757 	NLA_PUT_U32(skb, IFLA_MACVLAN_MODE, vlan->mode);
758 	return 0;
759 
760 nla_put_failure:
761 	return -EMSGSIZE;
762 }
763 
764 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
765 	[IFLA_MACVLAN_MODE] = { .type = NLA_U32 },
766 };
767 
768 int macvlan_link_register(struct rtnl_link_ops *ops)
769 {
770 	/* common fields */
771 	ops->priv_size		= sizeof(struct macvlan_dev);
772 	ops->validate		= macvlan_validate;
773 	ops->maxtype		= IFLA_MACVLAN_MAX;
774 	ops->policy		= macvlan_policy;
775 	ops->changelink		= macvlan_changelink;
776 	ops->get_size		= macvlan_get_size;
777 	ops->fill_info		= macvlan_fill_info;
778 
779 	return rtnl_link_register(ops);
780 };
781 EXPORT_SYMBOL_GPL(macvlan_link_register);
782 
783 static struct rtnl_link_ops macvlan_link_ops = {
784 	.kind		= "macvlan",
785 	.setup		= macvlan_setup,
786 	.newlink	= macvlan_newlink,
787 	.dellink	= macvlan_dellink,
788 };
789 
790 static int macvlan_device_event(struct notifier_block *unused,
791 				unsigned long event, void *ptr)
792 {
793 	struct net_device *dev = ptr;
794 	struct macvlan_dev *vlan, *next;
795 	struct macvlan_port *port;
796 
797 	if (!macvlan_port_exists(dev))
798 		return NOTIFY_DONE;
799 
800 	port = macvlan_port_get(dev);
801 
802 	switch (event) {
803 	case NETDEV_CHANGE:
804 		list_for_each_entry(vlan, &port->vlans, list)
805 			netif_stacked_transfer_operstate(vlan->lowerdev,
806 							 vlan->dev);
807 		break;
808 	case NETDEV_FEAT_CHANGE:
809 		list_for_each_entry(vlan, &port->vlans, list) {
810 			vlan->dev->features = dev->features & MACVLAN_FEATURES;
811 			vlan->dev->gso_max_size = dev->gso_max_size;
812 			netdev_features_change(vlan->dev);
813 		}
814 		break;
815 	case NETDEV_UNREGISTER:
816 		/* twiddle thumbs on netns device moves */
817 		if (dev->reg_state != NETREG_UNREGISTERING)
818 			break;
819 
820 		list_for_each_entry_safe(vlan, next, &port->vlans, list)
821 			vlan->dev->rtnl_link_ops->dellink(vlan->dev, NULL);
822 		break;
823 	case NETDEV_PRE_TYPE_CHANGE:
824 		/* Forbid underlaying device to change its type. */
825 		return NOTIFY_BAD;
826 	}
827 	return NOTIFY_DONE;
828 }
829 
830 static struct notifier_block macvlan_notifier_block __read_mostly = {
831 	.notifier_call	= macvlan_device_event,
832 };
833 
834 static int __init macvlan_init_module(void)
835 {
836 	int err;
837 
838 	register_netdevice_notifier(&macvlan_notifier_block);
839 
840 	err = macvlan_link_register(&macvlan_link_ops);
841 	if (err < 0)
842 		goto err1;
843 	return 0;
844 err1:
845 	unregister_netdevice_notifier(&macvlan_notifier_block);
846 	return err;
847 }
848 
849 static void __exit macvlan_cleanup_module(void)
850 {
851 	rtnl_link_unregister(&macvlan_link_ops);
852 	unregister_netdevice_notifier(&macvlan_notifier_block);
853 }
854 
855 module_init(macvlan_init_module);
856 module_exit(macvlan_cleanup_module);
857 
858 MODULE_LICENSE("GPL");
859 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
860 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
861 MODULE_ALIAS_RTNL_LINK("macvlan");
862