xref: /linux/drivers/net/macvlan.c (revision db4e83957f961f9053282409c5062c6baef857a4)
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 		return RX_HANDLER_PASS;
196 	}
197 
198 	if (port->passthru)
199 		vlan = list_first_entry(&port->vlans, struct macvlan_dev, list);
200 	else
201 		vlan = macvlan_hash_lookup(port, eth->h_dest);
202 	if (vlan == NULL)
203 		return RX_HANDLER_PASS;
204 
205 	dev = vlan->dev;
206 	if (unlikely(!(dev->flags & IFF_UP))) {
207 		kfree_skb(skb);
208 		return RX_HANDLER_CONSUMED;
209 	}
210 	len = skb->len + ETH_HLEN;
211 	skb = skb_share_check(skb, GFP_ATOMIC);
212 	if (!skb)
213 		goto out;
214 
215 	skb->dev = dev;
216 	skb->pkt_type = PACKET_HOST;
217 
218 	ret = vlan->receive(skb);
219 
220 out:
221 	macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, 0);
222 	return RX_HANDLER_CONSUMED;
223 }
224 
225 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
226 {
227 	const struct macvlan_dev *vlan = netdev_priv(dev);
228 	const struct macvlan_port *port = vlan->port;
229 	const struct macvlan_dev *dest;
230 	__u8 ip_summed = skb->ip_summed;
231 
232 	if (vlan->mode == MACVLAN_MODE_BRIDGE) {
233 		const struct ethhdr *eth = (void *)skb->data;
234 		skb->ip_summed = CHECKSUM_UNNECESSARY;
235 
236 		/* send to other bridge ports directly */
237 		if (is_multicast_ether_addr(eth->h_dest)) {
238 			macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
239 			goto xmit_world;
240 		}
241 
242 		dest = macvlan_hash_lookup(port, eth->h_dest);
243 		if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
244 			/* send to lowerdev first for its network taps */
245 			dev_forward_skb(vlan->lowerdev, skb);
246 
247 			return NET_XMIT_SUCCESS;
248 		}
249 	}
250 
251 xmit_world:
252 	skb->ip_summed = ip_summed;
253 	skb_set_dev(skb, vlan->lowerdev);
254 	return dev_queue_xmit(skb);
255 }
256 
257 netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
258 			       struct net_device *dev)
259 {
260 	unsigned int len = skb->len;
261 	int ret;
262 	const struct macvlan_dev *vlan = netdev_priv(dev);
263 
264 	ret = macvlan_queue_xmit(skb, dev);
265 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
266 		struct macvlan_pcpu_stats *pcpu_stats;
267 
268 		pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
269 		u64_stats_update_begin(&pcpu_stats->syncp);
270 		pcpu_stats->tx_packets++;
271 		pcpu_stats->tx_bytes += len;
272 		u64_stats_update_end(&pcpu_stats->syncp);
273 	} else {
274 		this_cpu_inc(vlan->pcpu_stats->tx_dropped);
275 	}
276 	return ret;
277 }
278 EXPORT_SYMBOL_GPL(macvlan_start_xmit);
279 
280 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
281 			       unsigned short type, const void *daddr,
282 			       const void *saddr, unsigned len)
283 {
284 	const struct macvlan_dev *vlan = netdev_priv(dev);
285 	struct net_device *lowerdev = vlan->lowerdev;
286 
287 	return dev_hard_header(skb, lowerdev, type, daddr,
288 			       saddr ? : dev->dev_addr, len);
289 }
290 
291 static const struct header_ops macvlan_hard_header_ops = {
292 	.create  	= macvlan_hard_header,
293 	.rebuild	= eth_rebuild_header,
294 	.parse		= eth_header_parse,
295 	.cache		= eth_header_cache,
296 	.cache_update	= eth_header_cache_update,
297 };
298 
299 static int macvlan_open(struct net_device *dev)
300 {
301 	struct macvlan_dev *vlan = netdev_priv(dev);
302 	struct net_device *lowerdev = vlan->lowerdev;
303 	int err;
304 
305 	if (vlan->port->passthru) {
306 		dev_set_promiscuity(lowerdev, 1);
307 		goto hash_add;
308 	}
309 
310 	err = -EBUSY;
311 	if (macvlan_addr_busy(vlan->port, dev->dev_addr))
312 		goto out;
313 
314 	err = dev_uc_add(lowerdev, dev->dev_addr);
315 	if (err < 0)
316 		goto out;
317 	if (dev->flags & IFF_ALLMULTI) {
318 		err = dev_set_allmulti(lowerdev, 1);
319 		if (err < 0)
320 			goto del_unicast;
321 	}
322 
323 hash_add:
324 	macvlan_hash_add(vlan);
325 	return 0;
326 
327 del_unicast:
328 	dev_uc_del(lowerdev, dev->dev_addr);
329 out:
330 	return err;
331 }
332 
333 static int macvlan_stop(struct net_device *dev)
334 {
335 	struct macvlan_dev *vlan = netdev_priv(dev);
336 	struct net_device *lowerdev = vlan->lowerdev;
337 
338 	if (vlan->port->passthru) {
339 		dev_set_promiscuity(lowerdev, -1);
340 		goto hash_del;
341 	}
342 
343 	dev_mc_unsync(lowerdev, dev);
344 	if (dev->flags & IFF_ALLMULTI)
345 		dev_set_allmulti(lowerdev, -1);
346 
347 	dev_uc_del(lowerdev, dev->dev_addr);
348 
349 hash_del:
350 	macvlan_hash_del(vlan, !dev->dismantle);
351 	return 0;
352 }
353 
354 static int macvlan_set_mac_address(struct net_device *dev, void *p)
355 {
356 	struct macvlan_dev *vlan = netdev_priv(dev);
357 	struct net_device *lowerdev = vlan->lowerdev;
358 	struct sockaddr *addr = p;
359 	int err;
360 
361 	if (!is_valid_ether_addr(addr->sa_data))
362 		return -EADDRNOTAVAIL;
363 
364 	if (!(dev->flags & IFF_UP)) {
365 		/* Just copy in the new address */
366 		memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
367 	} else {
368 		/* Rehash and update the device filters */
369 		if (macvlan_addr_busy(vlan->port, addr->sa_data))
370 			return -EBUSY;
371 
372 		err = dev_uc_add(lowerdev, addr->sa_data);
373 		if (err)
374 			return err;
375 
376 		dev_uc_del(lowerdev, dev->dev_addr);
377 
378 		macvlan_hash_change_addr(vlan, addr->sa_data);
379 	}
380 	return 0;
381 }
382 
383 static void macvlan_change_rx_flags(struct net_device *dev, int change)
384 {
385 	struct macvlan_dev *vlan = netdev_priv(dev);
386 	struct net_device *lowerdev = vlan->lowerdev;
387 
388 	if (change & IFF_ALLMULTI)
389 		dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
390 }
391 
392 static void macvlan_set_multicast_list(struct net_device *dev)
393 {
394 	struct macvlan_dev *vlan = netdev_priv(dev);
395 
396 	dev_mc_sync(vlan->lowerdev, dev);
397 }
398 
399 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
400 {
401 	struct macvlan_dev *vlan = netdev_priv(dev);
402 
403 	if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
404 		return -EINVAL;
405 	dev->mtu = new_mtu;
406 	return 0;
407 }
408 
409 /*
410  * macvlan network devices have devices nesting below it and are a special
411  * "super class" of normal network devices; split their locks off into a
412  * separate class since they always nest.
413  */
414 static struct lock_class_key macvlan_netdev_xmit_lock_key;
415 static struct lock_class_key macvlan_netdev_addr_lock_key;
416 
417 #define MACVLAN_FEATURES \
418 	(NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
419 	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
420 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
421 	 NETIF_F_HW_VLAN_FILTER)
422 
423 #define MACVLAN_STATE_MASK \
424 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
425 
426 static void macvlan_set_lockdep_class_one(struct net_device *dev,
427 					  struct netdev_queue *txq,
428 					  void *_unused)
429 {
430 	lockdep_set_class(&txq->_xmit_lock,
431 			  &macvlan_netdev_xmit_lock_key);
432 }
433 
434 static void macvlan_set_lockdep_class(struct net_device *dev)
435 {
436 	lockdep_set_class(&dev->addr_list_lock,
437 			  &macvlan_netdev_addr_lock_key);
438 	netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
439 }
440 
441 static int macvlan_init(struct net_device *dev)
442 {
443 	struct macvlan_dev *vlan = netdev_priv(dev);
444 	const struct net_device *lowerdev = vlan->lowerdev;
445 
446 	dev->state		= (dev->state & ~MACVLAN_STATE_MASK) |
447 				  (lowerdev->state & MACVLAN_STATE_MASK);
448 	dev->features 		= lowerdev->features & MACVLAN_FEATURES;
449 	dev->features		|= NETIF_F_LLTX;
450 	dev->gso_max_size	= lowerdev->gso_max_size;
451 	dev->iflink		= lowerdev->ifindex;
452 	dev->hard_header_len	= lowerdev->hard_header_len;
453 
454 	macvlan_set_lockdep_class(dev);
455 
456 	vlan->pcpu_stats = alloc_percpu(struct macvlan_pcpu_stats);
457 	if (!vlan->pcpu_stats)
458 		return -ENOMEM;
459 
460 	return 0;
461 }
462 
463 static void macvlan_uninit(struct net_device *dev)
464 {
465 	struct macvlan_dev *vlan = netdev_priv(dev);
466 	struct macvlan_port *port = vlan->port;
467 
468 	free_percpu(vlan->pcpu_stats);
469 
470 	port->count -= 1;
471 	if (!port->count)
472 		macvlan_port_destroy(port->dev);
473 }
474 
475 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
476 							 struct rtnl_link_stats64 *stats)
477 {
478 	struct macvlan_dev *vlan = netdev_priv(dev);
479 
480 	if (vlan->pcpu_stats) {
481 		struct macvlan_pcpu_stats *p;
482 		u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
483 		u32 rx_errors = 0, tx_dropped = 0;
484 		unsigned int start;
485 		int i;
486 
487 		for_each_possible_cpu(i) {
488 			p = per_cpu_ptr(vlan->pcpu_stats, i);
489 			do {
490 				start = u64_stats_fetch_begin_bh(&p->syncp);
491 				rx_packets	= p->rx_packets;
492 				rx_bytes	= p->rx_bytes;
493 				rx_multicast	= p->rx_multicast;
494 				tx_packets	= p->tx_packets;
495 				tx_bytes	= p->tx_bytes;
496 			} while (u64_stats_fetch_retry_bh(&p->syncp, start));
497 
498 			stats->rx_packets	+= rx_packets;
499 			stats->rx_bytes		+= rx_bytes;
500 			stats->multicast	+= rx_multicast;
501 			stats->tx_packets	+= tx_packets;
502 			stats->tx_bytes		+= tx_bytes;
503 			/* rx_errors & tx_dropped are u32, updated
504 			 * without syncp protection.
505 			 */
506 			rx_errors	+= p->rx_errors;
507 			tx_dropped	+= p->tx_dropped;
508 		}
509 		stats->rx_errors	= rx_errors;
510 		stats->rx_dropped	= rx_errors;
511 		stats->tx_dropped	= tx_dropped;
512 	}
513 	return stats;
514 }
515 
516 static void macvlan_vlan_rx_add_vid(struct net_device *dev,
517 				    unsigned short vid)
518 {
519 	struct macvlan_dev *vlan = netdev_priv(dev);
520 	struct net_device *lowerdev = vlan->lowerdev;
521 	const struct net_device_ops *ops = lowerdev->netdev_ops;
522 
523 	if (ops->ndo_vlan_rx_add_vid)
524 		ops->ndo_vlan_rx_add_vid(lowerdev, vid);
525 }
526 
527 static void macvlan_vlan_rx_kill_vid(struct net_device *dev,
528 				     unsigned short vid)
529 {
530 	struct macvlan_dev *vlan = netdev_priv(dev);
531 	struct net_device *lowerdev = vlan->lowerdev;
532 	const struct net_device_ops *ops = lowerdev->netdev_ops;
533 
534 	if (ops->ndo_vlan_rx_kill_vid)
535 		ops->ndo_vlan_rx_kill_vid(lowerdev, vid);
536 }
537 
538 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
539 					struct ethtool_drvinfo *drvinfo)
540 {
541 	snprintf(drvinfo->driver, 32, "macvlan");
542 	snprintf(drvinfo->version, 32, "0.1");
543 }
544 
545 static int macvlan_ethtool_get_settings(struct net_device *dev,
546 					struct ethtool_cmd *cmd)
547 {
548 	const struct macvlan_dev *vlan = netdev_priv(dev);
549 
550 	return __ethtool_get_settings(vlan->lowerdev, cmd);
551 }
552 
553 static const struct ethtool_ops macvlan_ethtool_ops = {
554 	.get_link		= ethtool_op_get_link,
555 	.get_settings		= macvlan_ethtool_get_settings,
556 	.get_drvinfo		= macvlan_ethtool_get_drvinfo,
557 };
558 
559 static const struct net_device_ops macvlan_netdev_ops = {
560 	.ndo_init		= macvlan_init,
561 	.ndo_uninit		= macvlan_uninit,
562 	.ndo_open		= macvlan_open,
563 	.ndo_stop		= macvlan_stop,
564 	.ndo_start_xmit		= macvlan_start_xmit,
565 	.ndo_change_mtu		= macvlan_change_mtu,
566 	.ndo_change_rx_flags	= macvlan_change_rx_flags,
567 	.ndo_set_mac_address	= macvlan_set_mac_address,
568 	.ndo_set_rx_mode	= macvlan_set_multicast_list,
569 	.ndo_get_stats64	= macvlan_dev_get_stats64,
570 	.ndo_validate_addr	= eth_validate_addr,
571 	.ndo_vlan_rx_add_vid	= macvlan_vlan_rx_add_vid,
572 	.ndo_vlan_rx_kill_vid	= macvlan_vlan_rx_kill_vid,
573 };
574 
575 void macvlan_common_setup(struct net_device *dev)
576 {
577 	ether_setup(dev);
578 
579 	dev->priv_flags	       &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
580 	dev->netdev_ops		= &macvlan_netdev_ops;
581 	dev->destructor		= free_netdev;
582 	dev->header_ops		= &macvlan_hard_header_ops,
583 	dev->ethtool_ops	= &macvlan_ethtool_ops;
584 }
585 EXPORT_SYMBOL_GPL(macvlan_common_setup);
586 
587 static void macvlan_setup(struct net_device *dev)
588 {
589 	macvlan_common_setup(dev);
590 	dev->tx_queue_len	= 0;
591 }
592 
593 static int macvlan_port_create(struct net_device *dev)
594 {
595 	struct macvlan_port *port;
596 	unsigned int i;
597 	int err;
598 
599 	if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
600 		return -EINVAL;
601 
602 	port = kzalloc(sizeof(*port), GFP_KERNEL);
603 	if (port == NULL)
604 		return -ENOMEM;
605 
606 	port->passthru = false;
607 	port->dev = dev;
608 	INIT_LIST_HEAD(&port->vlans);
609 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
610 		INIT_HLIST_HEAD(&port->vlan_hash[i]);
611 
612 	err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
613 	if (err)
614 		kfree(port);
615 	else
616 		dev->priv_flags |= IFF_MACVLAN_PORT;
617 	return err;
618 }
619 
620 static void macvlan_port_destroy(struct net_device *dev)
621 {
622 	struct macvlan_port *port = macvlan_port_get(dev);
623 
624 	dev->priv_flags &= ~IFF_MACVLAN_PORT;
625 	netdev_rx_handler_unregister(dev);
626 	kfree_rcu(port, rcu);
627 }
628 
629 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
630 {
631 	if (tb[IFLA_ADDRESS]) {
632 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
633 			return -EINVAL;
634 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
635 			return -EADDRNOTAVAIL;
636 	}
637 
638 	if (data && data[IFLA_MACVLAN_MODE]) {
639 		switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
640 		case MACVLAN_MODE_PRIVATE:
641 		case MACVLAN_MODE_VEPA:
642 		case MACVLAN_MODE_BRIDGE:
643 		case MACVLAN_MODE_PASSTHRU:
644 			break;
645 		default:
646 			return -EINVAL;
647 		}
648 	}
649 	return 0;
650 }
651 
652 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
653 			   struct nlattr *tb[], struct nlattr *data[],
654 			   int (*receive)(struct sk_buff *skb),
655 			   int (*forward)(struct net_device *dev,
656 					  struct sk_buff *skb))
657 {
658 	struct macvlan_dev *vlan = netdev_priv(dev);
659 	struct macvlan_port *port;
660 	struct net_device *lowerdev;
661 	int err;
662 
663 	if (!tb[IFLA_LINK])
664 		return -EINVAL;
665 
666 	lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
667 	if (lowerdev == NULL)
668 		return -ENODEV;
669 
670 	/* When creating macvlans on top of other macvlans - use
671 	 * the real device as the lowerdev.
672 	 */
673 	if (lowerdev->rtnl_link_ops == dev->rtnl_link_ops) {
674 		struct macvlan_dev *lowervlan = netdev_priv(lowerdev);
675 		lowerdev = lowervlan->lowerdev;
676 	}
677 
678 	if (!tb[IFLA_MTU])
679 		dev->mtu = lowerdev->mtu;
680 	else if (dev->mtu > lowerdev->mtu)
681 		return -EINVAL;
682 
683 	if (!tb[IFLA_ADDRESS])
684 		random_ether_addr(dev->dev_addr);
685 
686 	if (!macvlan_port_exists(lowerdev)) {
687 		err = macvlan_port_create(lowerdev);
688 		if (err < 0)
689 			return err;
690 	}
691 	port = macvlan_port_get(lowerdev);
692 
693 	/* Only 1 macvlan device can be created in passthru mode */
694 	if (port->passthru)
695 		return -EINVAL;
696 
697 	vlan->lowerdev = lowerdev;
698 	vlan->dev      = dev;
699 	vlan->port     = port;
700 	vlan->receive  = receive;
701 	vlan->forward  = forward;
702 
703 	vlan->mode     = MACVLAN_MODE_VEPA;
704 	if (data && data[IFLA_MACVLAN_MODE])
705 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
706 
707 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
708 		if (port->count)
709 			return -EINVAL;
710 		port->passthru = true;
711 		memcpy(dev->dev_addr, lowerdev->dev_addr, ETH_ALEN);
712 	}
713 
714 	port->count += 1;
715 	err = register_netdevice(dev);
716 	if (err < 0)
717 		goto destroy_port;
718 
719 	list_add_tail(&vlan->list, &port->vlans);
720 	netif_stacked_transfer_operstate(lowerdev, dev);
721 
722 	return 0;
723 
724 destroy_port:
725 	port->count -= 1;
726 	if (!port->count)
727 		macvlan_port_destroy(lowerdev);
728 
729 	return err;
730 }
731 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
732 
733 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
734 			   struct nlattr *tb[], struct nlattr *data[])
735 {
736 	return macvlan_common_newlink(src_net, dev, tb, data,
737 				      netif_rx,
738 				      dev_forward_skb);
739 }
740 
741 void macvlan_dellink(struct net_device *dev, struct list_head *head)
742 {
743 	struct macvlan_dev *vlan = netdev_priv(dev);
744 
745 	list_del(&vlan->list);
746 	unregister_netdevice_queue(dev, head);
747 }
748 EXPORT_SYMBOL_GPL(macvlan_dellink);
749 
750 static int macvlan_changelink(struct net_device *dev,
751 		struct nlattr *tb[], struct nlattr *data[])
752 {
753 	struct macvlan_dev *vlan = netdev_priv(dev);
754 	if (data && data[IFLA_MACVLAN_MODE])
755 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
756 	return 0;
757 }
758 
759 static size_t macvlan_get_size(const struct net_device *dev)
760 {
761 	return nla_total_size(4);
762 }
763 
764 static int macvlan_fill_info(struct sk_buff *skb,
765 				const struct net_device *dev)
766 {
767 	struct macvlan_dev *vlan = netdev_priv(dev);
768 
769 	NLA_PUT_U32(skb, IFLA_MACVLAN_MODE, vlan->mode);
770 	return 0;
771 
772 nla_put_failure:
773 	return -EMSGSIZE;
774 }
775 
776 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
777 	[IFLA_MACVLAN_MODE] = { .type = NLA_U32 },
778 };
779 
780 int macvlan_link_register(struct rtnl_link_ops *ops)
781 {
782 	/* common fields */
783 	ops->priv_size		= sizeof(struct macvlan_dev);
784 	ops->validate		= macvlan_validate;
785 	ops->maxtype		= IFLA_MACVLAN_MAX;
786 	ops->policy		= macvlan_policy;
787 	ops->changelink		= macvlan_changelink;
788 	ops->get_size		= macvlan_get_size;
789 	ops->fill_info		= macvlan_fill_info;
790 
791 	return rtnl_link_register(ops);
792 };
793 EXPORT_SYMBOL_GPL(macvlan_link_register);
794 
795 static struct rtnl_link_ops macvlan_link_ops = {
796 	.kind		= "macvlan",
797 	.setup		= macvlan_setup,
798 	.newlink	= macvlan_newlink,
799 	.dellink	= macvlan_dellink,
800 };
801 
802 static int macvlan_device_event(struct notifier_block *unused,
803 				unsigned long event, void *ptr)
804 {
805 	struct net_device *dev = ptr;
806 	struct macvlan_dev *vlan, *next;
807 	struct macvlan_port *port;
808 	LIST_HEAD(list_kill);
809 
810 	if (!macvlan_port_exists(dev))
811 		return NOTIFY_DONE;
812 
813 	port = macvlan_port_get(dev);
814 
815 	switch (event) {
816 	case NETDEV_CHANGE:
817 		list_for_each_entry(vlan, &port->vlans, list)
818 			netif_stacked_transfer_operstate(vlan->lowerdev,
819 							 vlan->dev);
820 		break;
821 	case NETDEV_FEAT_CHANGE:
822 		list_for_each_entry(vlan, &port->vlans, list) {
823 			vlan->dev->features = dev->features & MACVLAN_FEATURES;
824 			vlan->dev->gso_max_size = dev->gso_max_size;
825 			netdev_features_change(vlan->dev);
826 		}
827 		break;
828 	case NETDEV_UNREGISTER:
829 		/* twiddle thumbs on netns device moves */
830 		if (dev->reg_state != NETREG_UNREGISTERING)
831 			break;
832 
833 		list_for_each_entry_safe(vlan, next, &port->vlans, list)
834 			vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
835 		unregister_netdevice_many(&list_kill);
836 		list_del(&list_kill);
837 		break;
838 	case NETDEV_PRE_TYPE_CHANGE:
839 		/* Forbid underlaying device to change its type. */
840 		return NOTIFY_BAD;
841 	}
842 	return NOTIFY_DONE;
843 }
844 
845 static struct notifier_block macvlan_notifier_block __read_mostly = {
846 	.notifier_call	= macvlan_device_event,
847 };
848 
849 static int __init macvlan_init_module(void)
850 {
851 	int err;
852 
853 	register_netdevice_notifier(&macvlan_notifier_block);
854 
855 	err = macvlan_link_register(&macvlan_link_ops);
856 	if (err < 0)
857 		goto err1;
858 	return 0;
859 err1:
860 	unregister_netdevice_notifier(&macvlan_notifier_block);
861 	return err;
862 }
863 
864 static void __exit macvlan_cleanup_module(void)
865 {
866 	rtnl_link_unregister(&macvlan_link_ops);
867 	unregister_netdevice_notifier(&macvlan_notifier_block);
868 }
869 
870 module_init(macvlan_init_module);
871 module_exit(macvlan_cleanup_module);
872 
873 MODULE_LICENSE("GPL");
874 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
875 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
876 MODULE_ALIAS_RTNL_LINK("macvlan");
877