xref: /linux/drivers/net/macvlan.c (revision a67ff6a54095e27093ea501fb143fefe51a536c2)
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 		skb = ip_check_defrag(skb, IP_DEFRAG_MACVLAN);
173 		if (!skb)
174 			return RX_HANDLER_CONSUMED;
175 		src = macvlan_hash_lookup(port, eth->h_source);
176 		if (!src)
177 			/* frame comes from an external address */
178 			macvlan_broadcast(skb, port, NULL,
179 					  MACVLAN_MODE_PRIVATE |
180 					  MACVLAN_MODE_VEPA    |
181 					  MACVLAN_MODE_PASSTHRU|
182 					  MACVLAN_MODE_BRIDGE);
183 		else if (src->mode == MACVLAN_MODE_VEPA)
184 			/* flood to everyone except source */
185 			macvlan_broadcast(skb, port, src->dev,
186 					  MACVLAN_MODE_VEPA |
187 					  MACVLAN_MODE_BRIDGE);
188 		else if (src->mode == MACVLAN_MODE_BRIDGE)
189 			/*
190 			 * flood only to VEPA ports, bridge ports
191 			 * already saw the frame on the way out.
192 			 */
193 			macvlan_broadcast(skb, port, src->dev,
194 					  MACVLAN_MODE_VEPA);
195 		else {
196 			/* forward to original port. */
197 			vlan = src;
198 			ret = macvlan_broadcast_one(skb, vlan, eth, 0);
199 			goto out;
200 		}
201 
202 		return RX_HANDLER_PASS;
203 	}
204 
205 	if (port->passthru)
206 		vlan = list_first_entry(&port->vlans, struct macvlan_dev, list);
207 	else
208 		vlan = macvlan_hash_lookup(port, eth->h_dest);
209 	if (vlan == NULL)
210 		return RX_HANDLER_PASS;
211 
212 	dev = vlan->dev;
213 	if (unlikely(!(dev->flags & IFF_UP))) {
214 		kfree_skb(skb);
215 		return RX_HANDLER_CONSUMED;
216 	}
217 	len = skb->len + ETH_HLEN;
218 	skb = skb_share_check(skb, GFP_ATOMIC);
219 	if (!skb)
220 		goto out;
221 
222 	skb->dev = dev;
223 	skb->pkt_type = PACKET_HOST;
224 
225 	ret = vlan->receive(skb);
226 
227 out:
228 	macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, 0);
229 	return RX_HANDLER_CONSUMED;
230 }
231 
232 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
233 {
234 	const struct macvlan_dev *vlan = netdev_priv(dev);
235 	const struct macvlan_port *port = vlan->port;
236 	const struct macvlan_dev *dest;
237 	__u8 ip_summed = skb->ip_summed;
238 
239 	if (vlan->mode == MACVLAN_MODE_BRIDGE) {
240 		const struct ethhdr *eth = (void *)skb->data;
241 		skb->ip_summed = CHECKSUM_UNNECESSARY;
242 
243 		/* send to other bridge ports directly */
244 		if (is_multicast_ether_addr(eth->h_dest)) {
245 			macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
246 			goto xmit_world;
247 		}
248 
249 		dest = macvlan_hash_lookup(port, eth->h_dest);
250 		if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
251 			/* send to lowerdev first for its network taps */
252 			dev_forward_skb(vlan->lowerdev, skb);
253 
254 			return NET_XMIT_SUCCESS;
255 		}
256 	}
257 
258 xmit_world:
259 	skb->ip_summed = ip_summed;
260 	skb_set_dev(skb, vlan->lowerdev);
261 	return dev_queue_xmit(skb);
262 }
263 
264 netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
265 			       struct net_device *dev)
266 {
267 	unsigned int len = skb->len;
268 	int ret;
269 	const struct macvlan_dev *vlan = netdev_priv(dev);
270 
271 	ret = macvlan_queue_xmit(skb, dev);
272 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
273 		struct macvlan_pcpu_stats *pcpu_stats;
274 
275 		pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
276 		u64_stats_update_begin(&pcpu_stats->syncp);
277 		pcpu_stats->tx_packets++;
278 		pcpu_stats->tx_bytes += len;
279 		u64_stats_update_end(&pcpu_stats->syncp);
280 	} else {
281 		this_cpu_inc(vlan->pcpu_stats->tx_dropped);
282 	}
283 	return ret;
284 }
285 EXPORT_SYMBOL_GPL(macvlan_start_xmit);
286 
287 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
288 			       unsigned short type, const void *daddr,
289 			       const void *saddr, unsigned len)
290 {
291 	const struct macvlan_dev *vlan = netdev_priv(dev);
292 	struct net_device *lowerdev = vlan->lowerdev;
293 
294 	return dev_hard_header(skb, lowerdev, type, daddr,
295 			       saddr ? : dev->dev_addr, len);
296 }
297 
298 static const struct header_ops macvlan_hard_header_ops = {
299 	.create  	= macvlan_hard_header,
300 	.rebuild	= eth_rebuild_header,
301 	.parse		= eth_header_parse,
302 	.cache		= eth_header_cache,
303 	.cache_update	= eth_header_cache_update,
304 };
305 
306 static int macvlan_open(struct net_device *dev)
307 {
308 	struct macvlan_dev *vlan = netdev_priv(dev);
309 	struct net_device *lowerdev = vlan->lowerdev;
310 	int err;
311 
312 	if (vlan->port->passthru) {
313 		dev_set_promiscuity(lowerdev, 1);
314 		goto hash_add;
315 	}
316 
317 	err = -EBUSY;
318 	if (macvlan_addr_busy(vlan->port, dev->dev_addr))
319 		goto out;
320 
321 	err = dev_uc_add(lowerdev, dev->dev_addr);
322 	if (err < 0)
323 		goto out;
324 	if (dev->flags & IFF_ALLMULTI) {
325 		err = dev_set_allmulti(lowerdev, 1);
326 		if (err < 0)
327 			goto del_unicast;
328 	}
329 
330 hash_add:
331 	macvlan_hash_add(vlan);
332 	return 0;
333 
334 del_unicast:
335 	dev_uc_del(lowerdev, dev->dev_addr);
336 out:
337 	return err;
338 }
339 
340 static int macvlan_stop(struct net_device *dev)
341 {
342 	struct macvlan_dev *vlan = netdev_priv(dev);
343 	struct net_device *lowerdev = vlan->lowerdev;
344 
345 	if (vlan->port->passthru) {
346 		dev_set_promiscuity(lowerdev, -1);
347 		goto hash_del;
348 	}
349 
350 	dev_mc_unsync(lowerdev, dev);
351 	if (dev->flags & IFF_ALLMULTI)
352 		dev_set_allmulti(lowerdev, -1);
353 
354 	dev_uc_del(lowerdev, dev->dev_addr);
355 
356 hash_del:
357 	macvlan_hash_del(vlan, !dev->dismantle);
358 	return 0;
359 }
360 
361 static int macvlan_set_mac_address(struct net_device *dev, void *p)
362 {
363 	struct macvlan_dev *vlan = netdev_priv(dev);
364 	struct net_device *lowerdev = vlan->lowerdev;
365 	struct sockaddr *addr = p;
366 	int err;
367 
368 	if (!is_valid_ether_addr(addr->sa_data))
369 		return -EADDRNOTAVAIL;
370 
371 	if (!(dev->flags & IFF_UP)) {
372 		/* Just copy in the new address */
373 		memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
374 	} else {
375 		/* Rehash and update the device filters */
376 		if (macvlan_addr_busy(vlan->port, addr->sa_data))
377 			return -EBUSY;
378 
379 		err = dev_uc_add(lowerdev, addr->sa_data);
380 		if (err)
381 			return err;
382 
383 		dev_uc_del(lowerdev, dev->dev_addr);
384 
385 		macvlan_hash_change_addr(vlan, addr->sa_data);
386 	}
387 	return 0;
388 }
389 
390 static void macvlan_change_rx_flags(struct net_device *dev, int change)
391 {
392 	struct macvlan_dev *vlan = netdev_priv(dev);
393 	struct net_device *lowerdev = vlan->lowerdev;
394 
395 	if (change & IFF_ALLMULTI)
396 		dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
397 }
398 
399 static void macvlan_set_multicast_list(struct net_device *dev)
400 {
401 	struct macvlan_dev *vlan = netdev_priv(dev);
402 
403 	dev_mc_sync(vlan->lowerdev, dev);
404 }
405 
406 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
407 {
408 	struct macvlan_dev *vlan = netdev_priv(dev);
409 
410 	if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
411 		return -EINVAL;
412 	dev->mtu = new_mtu;
413 	return 0;
414 }
415 
416 /*
417  * macvlan network devices have devices nesting below it and are a special
418  * "super class" of normal network devices; split their locks off into a
419  * separate class since they always nest.
420  */
421 static struct lock_class_key macvlan_netdev_xmit_lock_key;
422 static struct lock_class_key macvlan_netdev_addr_lock_key;
423 
424 #define MACVLAN_FEATURES \
425 	(NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
426 	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
427 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
428 	 NETIF_F_HW_VLAN_FILTER)
429 
430 #define MACVLAN_STATE_MASK \
431 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
432 
433 static void macvlan_set_lockdep_class_one(struct net_device *dev,
434 					  struct netdev_queue *txq,
435 					  void *_unused)
436 {
437 	lockdep_set_class(&txq->_xmit_lock,
438 			  &macvlan_netdev_xmit_lock_key);
439 }
440 
441 static void macvlan_set_lockdep_class(struct net_device *dev)
442 {
443 	lockdep_set_class(&dev->addr_list_lock,
444 			  &macvlan_netdev_addr_lock_key);
445 	netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
446 }
447 
448 static int macvlan_init(struct net_device *dev)
449 {
450 	struct macvlan_dev *vlan = netdev_priv(dev);
451 	const struct net_device *lowerdev = vlan->lowerdev;
452 
453 	dev->state		= (dev->state & ~MACVLAN_STATE_MASK) |
454 				  (lowerdev->state & MACVLAN_STATE_MASK);
455 	dev->features 		= lowerdev->features & MACVLAN_FEATURES;
456 	dev->features		|= NETIF_F_LLTX;
457 	dev->gso_max_size	= lowerdev->gso_max_size;
458 	dev->iflink		= lowerdev->ifindex;
459 	dev->hard_header_len	= lowerdev->hard_header_len;
460 
461 	macvlan_set_lockdep_class(dev);
462 
463 	vlan->pcpu_stats = alloc_percpu(struct macvlan_pcpu_stats);
464 	if (!vlan->pcpu_stats)
465 		return -ENOMEM;
466 
467 	return 0;
468 }
469 
470 static void macvlan_uninit(struct net_device *dev)
471 {
472 	struct macvlan_dev *vlan = netdev_priv(dev);
473 	struct macvlan_port *port = vlan->port;
474 
475 	free_percpu(vlan->pcpu_stats);
476 
477 	port->count -= 1;
478 	if (!port->count)
479 		macvlan_port_destroy(port->dev);
480 }
481 
482 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
483 							 struct rtnl_link_stats64 *stats)
484 {
485 	struct macvlan_dev *vlan = netdev_priv(dev);
486 
487 	if (vlan->pcpu_stats) {
488 		struct macvlan_pcpu_stats *p;
489 		u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
490 		u32 rx_errors = 0, tx_dropped = 0;
491 		unsigned int start;
492 		int i;
493 
494 		for_each_possible_cpu(i) {
495 			p = per_cpu_ptr(vlan->pcpu_stats, i);
496 			do {
497 				start = u64_stats_fetch_begin_bh(&p->syncp);
498 				rx_packets	= p->rx_packets;
499 				rx_bytes	= p->rx_bytes;
500 				rx_multicast	= p->rx_multicast;
501 				tx_packets	= p->tx_packets;
502 				tx_bytes	= p->tx_bytes;
503 			} while (u64_stats_fetch_retry_bh(&p->syncp, start));
504 
505 			stats->rx_packets	+= rx_packets;
506 			stats->rx_bytes		+= rx_bytes;
507 			stats->multicast	+= rx_multicast;
508 			stats->tx_packets	+= tx_packets;
509 			stats->tx_bytes		+= tx_bytes;
510 			/* rx_errors & tx_dropped are u32, updated
511 			 * without syncp protection.
512 			 */
513 			rx_errors	+= p->rx_errors;
514 			tx_dropped	+= p->tx_dropped;
515 		}
516 		stats->rx_errors	= rx_errors;
517 		stats->rx_dropped	= rx_errors;
518 		stats->tx_dropped	= tx_dropped;
519 	}
520 	return stats;
521 }
522 
523 static void macvlan_vlan_rx_add_vid(struct net_device *dev,
524 				    unsigned short vid)
525 {
526 	struct macvlan_dev *vlan = netdev_priv(dev);
527 	struct net_device *lowerdev = vlan->lowerdev;
528 	const struct net_device_ops *ops = lowerdev->netdev_ops;
529 
530 	if (ops->ndo_vlan_rx_add_vid)
531 		ops->ndo_vlan_rx_add_vid(lowerdev, vid);
532 }
533 
534 static void macvlan_vlan_rx_kill_vid(struct net_device *dev,
535 				     unsigned short vid)
536 {
537 	struct macvlan_dev *vlan = netdev_priv(dev);
538 	struct net_device *lowerdev = vlan->lowerdev;
539 	const struct net_device_ops *ops = lowerdev->netdev_ops;
540 
541 	if (ops->ndo_vlan_rx_kill_vid)
542 		ops->ndo_vlan_rx_kill_vid(lowerdev, vid);
543 }
544 
545 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
546 					struct ethtool_drvinfo *drvinfo)
547 {
548 	snprintf(drvinfo->driver, 32, "macvlan");
549 	snprintf(drvinfo->version, 32, "0.1");
550 }
551 
552 static int macvlan_ethtool_get_settings(struct net_device *dev,
553 					struct ethtool_cmd *cmd)
554 {
555 	const struct macvlan_dev *vlan = netdev_priv(dev);
556 
557 	return __ethtool_get_settings(vlan->lowerdev, cmd);
558 }
559 
560 static const struct ethtool_ops macvlan_ethtool_ops = {
561 	.get_link		= ethtool_op_get_link,
562 	.get_settings		= macvlan_ethtool_get_settings,
563 	.get_drvinfo		= macvlan_ethtool_get_drvinfo,
564 };
565 
566 static const struct net_device_ops macvlan_netdev_ops = {
567 	.ndo_init		= macvlan_init,
568 	.ndo_uninit		= macvlan_uninit,
569 	.ndo_open		= macvlan_open,
570 	.ndo_stop		= macvlan_stop,
571 	.ndo_start_xmit		= macvlan_start_xmit,
572 	.ndo_change_mtu		= macvlan_change_mtu,
573 	.ndo_change_rx_flags	= macvlan_change_rx_flags,
574 	.ndo_set_mac_address	= macvlan_set_mac_address,
575 	.ndo_set_rx_mode	= macvlan_set_multicast_list,
576 	.ndo_get_stats64	= macvlan_dev_get_stats64,
577 	.ndo_validate_addr	= eth_validate_addr,
578 	.ndo_vlan_rx_add_vid	= macvlan_vlan_rx_add_vid,
579 	.ndo_vlan_rx_kill_vid	= macvlan_vlan_rx_kill_vid,
580 };
581 
582 void macvlan_common_setup(struct net_device *dev)
583 {
584 	ether_setup(dev);
585 
586 	dev->priv_flags	       &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
587 	dev->netdev_ops		= &macvlan_netdev_ops;
588 	dev->destructor		= free_netdev;
589 	dev->header_ops		= &macvlan_hard_header_ops,
590 	dev->ethtool_ops	= &macvlan_ethtool_ops;
591 }
592 EXPORT_SYMBOL_GPL(macvlan_common_setup);
593 
594 static void macvlan_setup(struct net_device *dev)
595 {
596 	macvlan_common_setup(dev);
597 	dev->tx_queue_len	= 0;
598 }
599 
600 static int macvlan_port_create(struct net_device *dev)
601 {
602 	struct macvlan_port *port;
603 	unsigned int i;
604 	int err;
605 
606 	if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
607 		return -EINVAL;
608 
609 	port = kzalloc(sizeof(*port), GFP_KERNEL);
610 	if (port == NULL)
611 		return -ENOMEM;
612 
613 	port->passthru = false;
614 	port->dev = dev;
615 	INIT_LIST_HEAD(&port->vlans);
616 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
617 		INIT_HLIST_HEAD(&port->vlan_hash[i]);
618 
619 	err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
620 	if (err)
621 		kfree(port);
622 	else
623 		dev->priv_flags |= IFF_MACVLAN_PORT;
624 	return err;
625 }
626 
627 static void macvlan_port_destroy(struct net_device *dev)
628 {
629 	struct macvlan_port *port = macvlan_port_get(dev);
630 
631 	dev->priv_flags &= ~IFF_MACVLAN_PORT;
632 	netdev_rx_handler_unregister(dev);
633 	kfree_rcu(port, rcu);
634 }
635 
636 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
637 {
638 	if (tb[IFLA_ADDRESS]) {
639 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
640 			return -EINVAL;
641 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
642 			return -EADDRNOTAVAIL;
643 	}
644 
645 	if (data && data[IFLA_MACVLAN_MODE]) {
646 		switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
647 		case MACVLAN_MODE_PRIVATE:
648 		case MACVLAN_MODE_VEPA:
649 		case MACVLAN_MODE_BRIDGE:
650 		case MACVLAN_MODE_PASSTHRU:
651 			break;
652 		default:
653 			return -EINVAL;
654 		}
655 	}
656 	return 0;
657 }
658 
659 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
660 			   struct nlattr *tb[], struct nlattr *data[],
661 			   int (*receive)(struct sk_buff *skb),
662 			   int (*forward)(struct net_device *dev,
663 					  struct sk_buff *skb))
664 {
665 	struct macvlan_dev *vlan = netdev_priv(dev);
666 	struct macvlan_port *port;
667 	struct net_device *lowerdev;
668 	int err;
669 
670 	if (!tb[IFLA_LINK])
671 		return -EINVAL;
672 
673 	lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
674 	if (lowerdev == NULL)
675 		return -ENODEV;
676 
677 	/* When creating macvlans on top of other macvlans - use
678 	 * the real device as the lowerdev.
679 	 */
680 	if (lowerdev->rtnl_link_ops == dev->rtnl_link_ops) {
681 		struct macvlan_dev *lowervlan = netdev_priv(lowerdev);
682 		lowerdev = lowervlan->lowerdev;
683 	}
684 
685 	if (!tb[IFLA_MTU])
686 		dev->mtu = lowerdev->mtu;
687 	else if (dev->mtu > lowerdev->mtu)
688 		return -EINVAL;
689 
690 	if (!tb[IFLA_ADDRESS])
691 		random_ether_addr(dev->dev_addr);
692 
693 	if (!macvlan_port_exists(lowerdev)) {
694 		err = macvlan_port_create(lowerdev);
695 		if (err < 0)
696 			return err;
697 	}
698 	port = macvlan_port_get(lowerdev);
699 
700 	/* Only 1 macvlan device can be created in passthru mode */
701 	if (port->passthru)
702 		return -EINVAL;
703 
704 	vlan->lowerdev = lowerdev;
705 	vlan->dev      = dev;
706 	vlan->port     = port;
707 	vlan->receive  = receive;
708 	vlan->forward  = forward;
709 
710 	vlan->mode     = MACVLAN_MODE_VEPA;
711 	if (data && data[IFLA_MACVLAN_MODE])
712 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
713 
714 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
715 		if (port->count)
716 			return -EINVAL;
717 		port->passthru = true;
718 		memcpy(dev->dev_addr, lowerdev->dev_addr, ETH_ALEN);
719 	}
720 
721 	port->count += 1;
722 	err = register_netdevice(dev);
723 	if (err < 0)
724 		goto destroy_port;
725 
726 	list_add_tail(&vlan->list, &port->vlans);
727 	netif_stacked_transfer_operstate(lowerdev, dev);
728 
729 	return 0;
730 
731 destroy_port:
732 	port->count -= 1;
733 	if (!port->count)
734 		macvlan_port_destroy(lowerdev);
735 
736 	return err;
737 }
738 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
739 
740 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
741 			   struct nlattr *tb[], struct nlattr *data[])
742 {
743 	return macvlan_common_newlink(src_net, dev, tb, data,
744 				      netif_rx,
745 				      dev_forward_skb);
746 }
747 
748 void macvlan_dellink(struct net_device *dev, struct list_head *head)
749 {
750 	struct macvlan_dev *vlan = netdev_priv(dev);
751 
752 	list_del(&vlan->list);
753 	unregister_netdevice_queue(dev, head);
754 }
755 EXPORT_SYMBOL_GPL(macvlan_dellink);
756 
757 static int macvlan_changelink(struct net_device *dev,
758 		struct nlattr *tb[], struct nlattr *data[])
759 {
760 	struct macvlan_dev *vlan = netdev_priv(dev);
761 	if (data && data[IFLA_MACVLAN_MODE])
762 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
763 	return 0;
764 }
765 
766 static size_t macvlan_get_size(const struct net_device *dev)
767 {
768 	return nla_total_size(4);
769 }
770 
771 static int macvlan_fill_info(struct sk_buff *skb,
772 				const struct net_device *dev)
773 {
774 	struct macvlan_dev *vlan = netdev_priv(dev);
775 
776 	NLA_PUT_U32(skb, IFLA_MACVLAN_MODE, vlan->mode);
777 	return 0;
778 
779 nla_put_failure:
780 	return -EMSGSIZE;
781 }
782 
783 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
784 	[IFLA_MACVLAN_MODE] = { .type = NLA_U32 },
785 };
786 
787 int macvlan_link_register(struct rtnl_link_ops *ops)
788 {
789 	/* common fields */
790 	ops->priv_size		= sizeof(struct macvlan_dev);
791 	ops->validate		= macvlan_validate;
792 	ops->maxtype		= IFLA_MACVLAN_MAX;
793 	ops->policy		= macvlan_policy;
794 	ops->changelink		= macvlan_changelink;
795 	ops->get_size		= macvlan_get_size;
796 	ops->fill_info		= macvlan_fill_info;
797 
798 	return rtnl_link_register(ops);
799 };
800 EXPORT_SYMBOL_GPL(macvlan_link_register);
801 
802 static struct rtnl_link_ops macvlan_link_ops = {
803 	.kind		= "macvlan",
804 	.setup		= macvlan_setup,
805 	.newlink	= macvlan_newlink,
806 	.dellink	= macvlan_dellink,
807 };
808 
809 static int macvlan_device_event(struct notifier_block *unused,
810 				unsigned long event, void *ptr)
811 {
812 	struct net_device *dev = ptr;
813 	struct macvlan_dev *vlan, *next;
814 	struct macvlan_port *port;
815 	LIST_HEAD(list_kill);
816 
817 	if (!macvlan_port_exists(dev))
818 		return NOTIFY_DONE;
819 
820 	port = macvlan_port_get(dev);
821 
822 	switch (event) {
823 	case NETDEV_CHANGE:
824 		list_for_each_entry(vlan, &port->vlans, list)
825 			netif_stacked_transfer_operstate(vlan->lowerdev,
826 							 vlan->dev);
827 		break;
828 	case NETDEV_FEAT_CHANGE:
829 		list_for_each_entry(vlan, &port->vlans, list) {
830 			vlan->dev->features = dev->features & MACVLAN_FEATURES;
831 			vlan->dev->gso_max_size = dev->gso_max_size;
832 			netdev_features_change(vlan->dev);
833 		}
834 		break;
835 	case NETDEV_UNREGISTER:
836 		/* twiddle thumbs on netns device moves */
837 		if (dev->reg_state != NETREG_UNREGISTERING)
838 			break;
839 
840 		list_for_each_entry_safe(vlan, next, &port->vlans, list)
841 			vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
842 		unregister_netdevice_many(&list_kill);
843 		list_del(&list_kill);
844 		break;
845 	case NETDEV_PRE_TYPE_CHANGE:
846 		/* Forbid underlaying device to change its type. */
847 		return NOTIFY_BAD;
848 	}
849 	return NOTIFY_DONE;
850 }
851 
852 static struct notifier_block macvlan_notifier_block __read_mostly = {
853 	.notifier_call	= macvlan_device_event,
854 };
855 
856 static int __init macvlan_init_module(void)
857 {
858 	int err;
859 
860 	register_netdevice_notifier(&macvlan_notifier_block);
861 
862 	err = macvlan_link_register(&macvlan_link_ops);
863 	if (err < 0)
864 		goto err1;
865 	return 0;
866 err1:
867 	unregister_netdevice_notifier(&macvlan_notifier_block);
868 	return err;
869 }
870 
871 static void __exit macvlan_cleanup_module(void)
872 {
873 	rtnl_link_unregister(&macvlan_link_ops);
874 	unregister_netdevice_notifier(&macvlan_notifier_block);
875 }
876 
877 module_init(macvlan_init_module);
878 module_exit(macvlan_cleanup_module);
879 
880 MODULE_LICENSE("GPL");
881 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
882 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
883 MODULE_ALIAS_RTNL_LINK("macvlan");
884