xref: /openbmc/linux/drivers/net/macvlan.c (revision 7490ca1e)
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_vlan.h>
30 #include <linux/if_link.h>
31 #include <linux/if_macvlan.h>
32 #include <net/rtnetlink.h>
33 #include <net/xfrm.h>
34 
35 #define MACVLAN_HASH_SIZE	(1 << BITS_PER_BYTE)
36 
37 struct macvlan_port {
38 	struct net_device	*dev;
39 	struct hlist_head	vlan_hash[MACVLAN_HASH_SIZE];
40 	struct list_head	vlans;
41 	struct rcu_head		rcu;
42 	bool 			passthru;
43 	int			count;
44 };
45 
46 static void macvlan_port_destroy(struct net_device *dev);
47 
48 #define macvlan_port_get_rcu(dev) \
49 	((struct macvlan_port *) rcu_dereference(dev->rx_handler_data))
50 #define macvlan_port_get(dev) ((struct macvlan_port *) dev->rx_handler_data)
51 #define macvlan_port_exists(dev) (dev->priv_flags & IFF_MACVLAN_PORT)
52 
53 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
54 					       const unsigned char *addr)
55 {
56 	struct macvlan_dev *vlan;
57 	struct hlist_node *n;
58 
59 	hlist_for_each_entry_rcu(vlan, n, &port->vlan_hash[addr[5]], hlist) {
60 		if (!compare_ether_addr_64bits(vlan->dev->dev_addr, addr))
61 			return vlan;
62 	}
63 	return NULL;
64 }
65 
66 static void macvlan_hash_add(struct macvlan_dev *vlan)
67 {
68 	struct macvlan_port *port = vlan->port;
69 	const unsigned char *addr = vlan->dev->dev_addr;
70 
71 	hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[addr[5]]);
72 }
73 
74 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync)
75 {
76 	hlist_del_rcu(&vlan->hlist);
77 	if (sync)
78 		synchronize_rcu();
79 }
80 
81 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
82 					const unsigned char *addr)
83 {
84 	macvlan_hash_del(vlan, true);
85 	/* Now that we are unhashed it is safe to change the device
86 	 * address without confusing packet delivery.
87 	 */
88 	memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
89 	macvlan_hash_add(vlan);
90 }
91 
92 static int macvlan_addr_busy(const struct macvlan_port *port,
93 				const unsigned char *addr)
94 {
95 	/* Test to see if the specified multicast address is
96 	 * currently in use by the underlying device or
97 	 * another macvlan.
98 	 */
99 	if (!compare_ether_addr_64bits(port->dev->dev_addr, addr))
100 		return 1;
101 
102 	if (macvlan_hash_lookup(port, addr))
103 		return 1;
104 
105 	return 0;
106 }
107 
108 
109 static int macvlan_broadcast_one(struct sk_buff *skb,
110 				 const struct macvlan_dev *vlan,
111 				 const struct ethhdr *eth, bool local)
112 {
113 	struct net_device *dev = vlan->dev;
114 	if (!skb)
115 		return NET_RX_DROP;
116 
117 	if (local)
118 		return vlan->forward(dev, skb);
119 
120 	skb->dev = dev;
121 	if (!compare_ether_addr_64bits(eth->h_dest,
122 				       dev->broadcast))
123 		skb->pkt_type = PACKET_BROADCAST;
124 	else
125 		skb->pkt_type = PACKET_MULTICAST;
126 
127 	return vlan->receive(skb);
128 }
129 
130 static void macvlan_broadcast(struct sk_buff *skb,
131 			      const struct macvlan_port *port,
132 			      struct net_device *src,
133 			      enum macvlan_mode mode)
134 {
135 	const struct ethhdr *eth = eth_hdr(skb);
136 	const struct macvlan_dev *vlan;
137 	struct hlist_node *n;
138 	struct sk_buff *nskb;
139 	unsigned int i;
140 	int err;
141 
142 	if (skb->protocol == htons(ETH_P_PAUSE))
143 		return;
144 
145 	for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
146 		hlist_for_each_entry_rcu(vlan, n, &port->vlan_hash[i], hlist) {
147 			if (vlan->dev == src || !(vlan->mode & mode))
148 				continue;
149 
150 			nskb = skb_clone(skb, GFP_ATOMIC);
151 			err = macvlan_broadcast_one(nskb, vlan, eth,
152 					 mode == MACVLAN_MODE_BRIDGE);
153 			macvlan_count_rx(vlan, skb->len + ETH_HLEN,
154 					 err == NET_RX_SUCCESS, 1);
155 		}
156 	}
157 }
158 
159 /* called under rcu_read_lock() from netif_receive_skb */
160 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
161 {
162 	struct macvlan_port *port;
163 	struct sk_buff *skb = *pskb;
164 	const struct ethhdr *eth = eth_hdr(skb);
165 	const struct macvlan_dev *vlan;
166 	const struct macvlan_dev *src;
167 	struct net_device *dev;
168 	unsigned int len = 0;
169 	int ret = NET_RX_DROP;
170 
171 	port = macvlan_port_get_rcu(skb->dev);
172 	if (is_multicast_ether_addr(eth->h_dest)) {
173 		skb = ip_check_defrag(skb, IP_DEFRAG_MACVLAN);
174 		if (!skb)
175 			return RX_HANDLER_CONSUMED;
176 		eth = eth_hdr(skb);
177 		src = macvlan_hash_lookup(port, eth->h_source);
178 		if (!src)
179 			/* frame comes from an external address */
180 			macvlan_broadcast(skb, port, NULL,
181 					  MACVLAN_MODE_PRIVATE |
182 					  MACVLAN_MODE_VEPA    |
183 					  MACVLAN_MODE_PASSTHRU|
184 					  MACVLAN_MODE_BRIDGE);
185 		else if (src->mode == MACVLAN_MODE_VEPA)
186 			/* flood to everyone except source */
187 			macvlan_broadcast(skb, port, src->dev,
188 					  MACVLAN_MODE_VEPA |
189 					  MACVLAN_MODE_BRIDGE);
190 		else if (src->mode == MACVLAN_MODE_BRIDGE)
191 			/*
192 			 * flood only to VEPA ports, bridge ports
193 			 * already saw the frame on the way out.
194 			 */
195 			macvlan_broadcast(skb, port, src->dev,
196 					  MACVLAN_MODE_VEPA);
197 		else {
198 			/* forward to original port. */
199 			vlan = src;
200 			ret = macvlan_broadcast_one(skb, vlan, eth, 0);
201 			goto out;
202 		}
203 
204 		return RX_HANDLER_PASS;
205 	}
206 
207 	if (port->passthru)
208 		vlan = list_first_entry(&port->vlans, struct macvlan_dev, list);
209 	else
210 		vlan = macvlan_hash_lookup(port, eth->h_dest);
211 	if (vlan == NULL)
212 		return RX_HANDLER_PASS;
213 
214 	dev = vlan->dev;
215 	if (unlikely(!(dev->flags & IFF_UP))) {
216 		kfree_skb(skb);
217 		return RX_HANDLER_CONSUMED;
218 	}
219 	len = skb->len + ETH_HLEN;
220 	skb = skb_share_check(skb, GFP_ATOMIC);
221 	if (!skb)
222 		goto out;
223 
224 	skb->dev = dev;
225 	skb->pkt_type = PACKET_HOST;
226 
227 	ret = vlan->receive(skb);
228 
229 out:
230 	macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, 0);
231 	return RX_HANDLER_CONSUMED;
232 }
233 
234 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
235 {
236 	const struct macvlan_dev *vlan = netdev_priv(dev);
237 	const struct macvlan_port *port = vlan->port;
238 	const struct macvlan_dev *dest;
239 	__u8 ip_summed = skb->ip_summed;
240 
241 	if (vlan->mode == MACVLAN_MODE_BRIDGE) {
242 		const struct ethhdr *eth = (void *)skb->data;
243 		skb->ip_summed = CHECKSUM_UNNECESSARY;
244 
245 		/* send to other bridge ports directly */
246 		if (is_multicast_ether_addr(eth->h_dest)) {
247 			macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
248 			goto xmit_world;
249 		}
250 
251 		dest = macvlan_hash_lookup(port, eth->h_dest);
252 		if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
253 			/* send to lowerdev first for its network taps */
254 			dev_forward_skb(vlan->lowerdev, skb);
255 
256 			return NET_XMIT_SUCCESS;
257 		}
258 	}
259 
260 xmit_world:
261 	skb->ip_summed = ip_summed;
262 	skb_set_dev(skb, vlan->lowerdev);
263 	return dev_queue_xmit(skb);
264 }
265 
266 netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
267 			       struct net_device *dev)
268 {
269 	unsigned int len = skb->len;
270 	int ret;
271 	const struct macvlan_dev *vlan = netdev_priv(dev);
272 
273 	ret = macvlan_queue_xmit(skb, dev);
274 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
275 		struct macvlan_pcpu_stats *pcpu_stats;
276 
277 		pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
278 		u64_stats_update_begin(&pcpu_stats->syncp);
279 		pcpu_stats->tx_packets++;
280 		pcpu_stats->tx_bytes += len;
281 		u64_stats_update_end(&pcpu_stats->syncp);
282 	} else {
283 		this_cpu_inc(vlan->pcpu_stats->tx_dropped);
284 	}
285 	return ret;
286 }
287 EXPORT_SYMBOL_GPL(macvlan_start_xmit);
288 
289 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
290 			       unsigned short type, const void *daddr,
291 			       const void *saddr, unsigned len)
292 {
293 	const struct macvlan_dev *vlan = netdev_priv(dev);
294 	struct net_device *lowerdev = vlan->lowerdev;
295 
296 	return dev_hard_header(skb, lowerdev, type, daddr,
297 			       saddr ? : dev->dev_addr, len);
298 }
299 
300 static const struct header_ops macvlan_hard_header_ops = {
301 	.create  	= macvlan_hard_header,
302 	.rebuild	= eth_rebuild_header,
303 	.parse		= eth_header_parse,
304 	.cache		= eth_header_cache,
305 	.cache_update	= eth_header_cache_update,
306 };
307 
308 static int macvlan_open(struct net_device *dev)
309 {
310 	struct macvlan_dev *vlan = netdev_priv(dev);
311 	struct net_device *lowerdev = vlan->lowerdev;
312 	int err;
313 
314 	if (vlan->port->passthru) {
315 		dev_set_promiscuity(lowerdev, 1);
316 		goto hash_add;
317 	}
318 
319 	err = -EBUSY;
320 	if (macvlan_addr_busy(vlan->port, dev->dev_addr))
321 		goto out;
322 
323 	err = dev_uc_add(lowerdev, dev->dev_addr);
324 	if (err < 0)
325 		goto out;
326 	if (dev->flags & IFF_ALLMULTI) {
327 		err = dev_set_allmulti(lowerdev, 1);
328 		if (err < 0)
329 			goto del_unicast;
330 	}
331 
332 hash_add:
333 	macvlan_hash_add(vlan);
334 	return 0;
335 
336 del_unicast:
337 	dev_uc_del(lowerdev, dev->dev_addr);
338 out:
339 	return err;
340 }
341 
342 static int macvlan_stop(struct net_device *dev)
343 {
344 	struct macvlan_dev *vlan = netdev_priv(dev);
345 	struct net_device *lowerdev = vlan->lowerdev;
346 
347 	if (vlan->port->passthru) {
348 		dev_set_promiscuity(lowerdev, -1);
349 		goto hash_del;
350 	}
351 
352 	dev_mc_unsync(lowerdev, dev);
353 	if (dev->flags & IFF_ALLMULTI)
354 		dev_set_allmulti(lowerdev, -1);
355 
356 	dev_uc_del(lowerdev, dev->dev_addr);
357 
358 hash_del:
359 	macvlan_hash_del(vlan, !dev->dismantle);
360 	return 0;
361 }
362 
363 static int macvlan_set_mac_address(struct net_device *dev, void *p)
364 {
365 	struct macvlan_dev *vlan = netdev_priv(dev);
366 	struct net_device *lowerdev = vlan->lowerdev;
367 	struct sockaddr *addr = p;
368 	int err;
369 
370 	if (!is_valid_ether_addr(addr->sa_data))
371 		return -EADDRNOTAVAIL;
372 
373 	if (!(dev->flags & IFF_UP)) {
374 		/* Just copy in the new address */
375 		memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
376 	} else {
377 		/* Rehash and update the device filters */
378 		if (macvlan_addr_busy(vlan->port, addr->sa_data))
379 			return -EBUSY;
380 
381 		err = dev_uc_add(lowerdev, addr->sa_data);
382 		if (err)
383 			return err;
384 
385 		dev_uc_del(lowerdev, dev->dev_addr);
386 
387 		macvlan_hash_change_addr(vlan, addr->sa_data);
388 	}
389 	return 0;
390 }
391 
392 static void macvlan_change_rx_flags(struct net_device *dev, int change)
393 {
394 	struct macvlan_dev *vlan = netdev_priv(dev);
395 	struct net_device *lowerdev = vlan->lowerdev;
396 
397 	if (change & IFF_ALLMULTI)
398 		dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
399 }
400 
401 static void macvlan_set_multicast_list(struct net_device *dev)
402 {
403 	struct macvlan_dev *vlan = netdev_priv(dev);
404 
405 	dev_mc_sync(vlan->lowerdev, dev);
406 }
407 
408 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
409 {
410 	struct macvlan_dev *vlan = netdev_priv(dev);
411 
412 	if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
413 		return -EINVAL;
414 	dev->mtu = new_mtu;
415 	return 0;
416 }
417 
418 /*
419  * macvlan network devices have devices nesting below it and are a special
420  * "super class" of normal network devices; split their locks off into a
421  * separate class since they always nest.
422  */
423 static struct lock_class_key macvlan_netdev_xmit_lock_key;
424 static struct lock_class_key macvlan_netdev_addr_lock_key;
425 
426 #define MACVLAN_FEATURES \
427 	(NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
428 	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
429 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
430 	 NETIF_F_HW_VLAN_FILTER)
431 
432 #define MACVLAN_STATE_MASK \
433 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
434 
435 static void macvlan_set_lockdep_class_one(struct net_device *dev,
436 					  struct netdev_queue *txq,
437 					  void *_unused)
438 {
439 	lockdep_set_class(&txq->_xmit_lock,
440 			  &macvlan_netdev_xmit_lock_key);
441 }
442 
443 static void macvlan_set_lockdep_class(struct net_device *dev)
444 {
445 	lockdep_set_class(&dev->addr_list_lock,
446 			  &macvlan_netdev_addr_lock_key);
447 	netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
448 }
449 
450 static int macvlan_init(struct net_device *dev)
451 {
452 	struct macvlan_dev *vlan = netdev_priv(dev);
453 	const struct net_device *lowerdev = vlan->lowerdev;
454 
455 	dev->state		= (dev->state & ~MACVLAN_STATE_MASK) |
456 				  (lowerdev->state & MACVLAN_STATE_MASK);
457 	dev->features 		= lowerdev->features & MACVLAN_FEATURES;
458 	dev->features		|= NETIF_F_LLTX;
459 	dev->gso_max_size	= lowerdev->gso_max_size;
460 	dev->iflink		= lowerdev->ifindex;
461 	dev->hard_header_len	= lowerdev->hard_header_len;
462 
463 	macvlan_set_lockdep_class(dev);
464 
465 	vlan->pcpu_stats = alloc_percpu(struct macvlan_pcpu_stats);
466 	if (!vlan->pcpu_stats)
467 		return -ENOMEM;
468 
469 	return 0;
470 }
471 
472 static void macvlan_uninit(struct net_device *dev)
473 {
474 	struct macvlan_dev *vlan = netdev_priv(dev);
475 	struct macvlan_port *port = vlan->port;
476 
477 	free_percpu(vlan->pcpu_stats);
478 
479 	port->count -= 1;
480 	if (!port->count)
481 		macvlan_port_destroy(port->dev);
482 }
483 
484 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
485 							 struct rtnl_link_stats64 *stats)
486 {
487 	struct macvlan_dev *vlan = netdev_priv(dev);
488 
489 	if (vlan->pcpu_stats) {
490 		struct macvlan_pcpu_stats *p;
491 		u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
492 		u32 rx_errors = 0, tx_dropped = 0;
493 		unsigned int start;
494 		int i;
495 
496 		for_each_possible_cpu(i) {
497 			p = per_cpu_ptr(vlan->pcpu_stats, i);
498 			do {
499 				start = u64_stats_fetch_begin_bh(&p->syncp);
500 				rx_packets	= p->rx_packets;
501 				rx_bytes	= p->rx_bytes;
502 				rx_multicast	= p->rx_multicast;
503 				tx_packets	= p->tx_packets;
504 				tx_bytes	= p->tx_bytes;
505 			} while (u64_stats_fetch_retry_bh(&p->syncp, start));
506 
507 			stats->rx_packets	+= rx_packets;
508 			stats->rx_bytes		+= rx_bytes;
509 			stats->multicast	+= rx_multicast;
510 			stats->tx_packets	+= tx_packets;
511 			stats->tx_bytes		+= tx_bytes;
512 			/* rx_errors & tx_dropped are u32, updated
513 			 * without syncp protection.
514 			 */
515 			rx_errors	+= p->rx_errors;
516 			tx_dropped	+= p->tx_dropped;
517 		}
518 		stats->rx_errors	= rx_errors;
519 		stats->rx_dropped	= rx_errors;
520 		stats->tx_dropped	= tx_dropped;
521 	}
522 	return stats;
523 }
524 
525 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
526 				    unsigned short vid)
527 {
528 	struct macvlan_dev *vlan = netdev_priv(dev);
529 	struct net_device *lowerdev = vlan->lowerdev;
530 
531 	return vlan_vid_add(lowerdev, vid);
532 }
533 
534 static int 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 
540 	vlan_vid_del(lowerdev, vid);
541 	return 0;
542 }
543 
544 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
545 					struct ethtool_drvinfo *drvinfo)
546 {
547 	snprintf(drvinfo->driver, 32, "macvlan");
548 	snprintf(drvinfo->version, 32, "0.1");
549 }
550 
551 static int macvlan_ethtool_get_settings(struct net_device *dev,
552 					struct ethtool_cmd *cmd)
553 {
554 	const struct macvlan_dev *vlan = netdev_priv(dev);
555 
556 	return __ethtool_get_settings(vlan->lowerdev, cmd);
557 }
558 
559 static const struct ethtool_ops macvlan_ethtool_ops = {
560 	.get_link		= ethtool_op_get_link,
561 	.get_settings		= macvlan_ethtool_get_settings,
562 	.get_drvinfo		= macvlan_ethtool_get_drvinfo,
563 };
564 
565 static const struct net_device_ops macvlan_netdev_ops = {
566 	.ndo_init		= macvlan_init,
567 	.ndo_uninit		= macvlan_uninit,
568 	.ndo_open		= macvlan_open,
569 	.ndo_stop		= macvlan_stop,
570 	.ndo_start_xmit		= macvlan_start_xmit,
571 	.ndo_change_mtu		= macvlan_change_mtu,
572 	.ndo_change_rx_flags	= macvlan_change_rx_flags,
573 	.ndo_set_mac_address	= macvlan_set_mac_address,
574 	.ndo_set_rx_mode	= macvlan_set_multicast_list,
575 	.ndo_get_stats64	= macvlan_dev_get_stats64,
576 	.ndo_validate_addr	= eth_validate_addr,
577 	.ndo_vlan_rx_add_vid	= macvlan_vlan_rx_add_vid,
578 	.ndo_vlan_rx_kill_vid	= macvlan_vlan_rx_kill_vid,
579 };
580 
581 void macvlan_common_setup(struct net_device *dev)
582 {
583 	ether_setup(dev);
584 
585 	dev->priv_flags	       &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
586 	dev->netdev_ops		= &macvlan_netdev_ops;
587 	dev->destructor		= free_netdev;
588 	dev->header_ops		= &macvlan_hard_header_ops,
589 	dev->ethtool_ops	= &macvlan_ethtool_ops;
590 }
591 EXPORT_SYMBOL_GPL(macvlan_common_setup);
592 
593 static void macvlan_setup(struct net_device *dev)
594 {
595 	macvlan_common_setup(dev);
596 	dev->tx_queue_len	= 0;
597 }
598 
599 static int macvlan_port_create(struct net_device *dev)
600 {
601 	struct macvlan_port *port;
602 	unsigned int i;
603 	int err;
604 
605 	if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
606 		return -EINVAL;
607 
608 	port = kzalloc(sizeof(*port), GFP_KERNEL);
609 	if (port == NULL)
610 		return -ENOMEM;
611 
612 	port->passthru = false;
613 	port->dev = dev;
614 	INIT_LIST_HEAD(&port->vlans);
615 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
616 		INIT_HLIST_HEAD(&port->vlan_hash[i]);
617 
618 	err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
619 	if (err)
620 		kfree(port);
621 	else
622 		dev->priv_flags |= IFF_MACVLAN_PORT;
623 	return err;
624 }
625 
626 static void macvlan_port_destroy(struct net_device *dev)
627 {
628 	struct macvlan_port *port = macvlan_port_get(dev);
629 
630 	dev->priv_flags &= ~IFF_MACVLAN_PORT;
631 	netdev_rx_handler_unregister(dev);
632 	kfree_rcu(port, rcu);
633 }
634 
635 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
636 {
637 	if (tb[IFLA_ADDRESS]) {
638 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
639 			return -EINVAL;
640 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
641 			return -EADDRNOTAVAIL;
642 	}
643 
644 	if (data && data[IFLA_MACVLAN_MODE]) {
645 		switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
646 		case MACVLAN_MODE_PRIVATE:
647 		case MACVLAN_MODE_VEPA:
648 		case MACVLAN_MODE_BRIDGE:
649 		case MACVLAN_MODE_PASSTHRU:
650 			break;
651 		default:
652 			return -EINVAL;
653 		}
654 	}
655 	return 0;
656 }
657 
658 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
659 			   struct nlattr *tb[], struct nlattr *data[],
660 			   int (*receive)(struct sk_buff *skb),
661 			   int (*forward)(struct net_device *dev,
662 					  struct sk_buff *skb))
663 {
664 	struct macvlan_dev *vlan = netdev_priv(dev);
665 	struct macvlan_port *port;
666 	struct net_device *lowerdev;
667 	int err;
668 
669 	if (!tb[IFLA_LINK])
670 		return -EINVAL;
671 
672 	lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
673 	if (lowerdev == NULL)
674 		return -ENODEV;
675 
676 	/* When creating macvlans on top of other macvlans - use
677 	 * the real device as the lowerdev.
678 	 */
679 	if (lowerdev->rtnl_link_ops == dev->rtnl_link_ops) {
680 		struct macvlan_dev *lowervlan = netdev_priv(lowerdev);
681 		lowerdev = lowervlan->lowerdev;
682 	}
683 
684 	if (!tb[IFLA_MTU])
685 		dev->mtu = lowerdev->mtu;
686 	else if (dev->mtu > lowerdev->mtu)
687 		return -EINVAL;
688 
689 	if (!tb[IFLA_ADDRESS])
690 		random_ether_addr(dev->dev_addr);
691 
692 	if (!macvlan_port_exists(lowerdev)) {
693 		err = macvlan_port_create(lowerdev);
694 		if (err < 0)
695 			return err;
696 	}
697 	port = macvlan_port_get(lowerdev);
698 
699 	/* Only 1 macvlan device can be created in passthru mode */
700 	if (port->passthru)
701 		return -EINVAL;
702 
703 	vlan->lowerdev = lowerdev;
704 	vlan->dev      = dev;
705 	vlan->port     = port;
706 	vlan->receive  = receive;
707 	vlan->forward  = forward;
708 
709 	vlan->mode     = MACVLAN_MODE_VEPA;
710 	if (data && data[IFLA_MACVLAN_MODE])
711 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
712 
713 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
714 		if (port->count)
715 			return -EINVAL;
716 		port->passthru = true;
717 		memcpy(dev->dev_addr, lowerdev->dev_addr, ETH_ALEN);
718 	}
719 
720 	port->count += 1;
721 	err = register_netdevice(dev);
722 	if (err < 0)
723 		goto destroy_port;
724 
725 	list_add_tail(&vlan->list, &port->vlans);
726 	netif_stacked_transfer_operstate(lowerdev, dev);
727 
728 	return 0;
729 
730 destroy_port:
731 	port->count -= 1;
732 	if (!port->count)
733 		macvlan_port_destroy(lowerdev);
734 
735 	return err;
736 }
737 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
738 
739 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
740 			   struct nlattr *tb[], struct nlattr *data[])
741 {
742 	return macvlan_common_newlink(src_net, dev, tb, data,
743 				      netif_rx,
744 				      dev_forward_skb);
745 }
746 
747 void macvlan_dellink(struct net_device *dev, struct list_head *head)
748 {
749 	struct macvlan_dev *vlan = netdev_priv(dev);
750 
751 	list_del(&vlan->list);
752 	unregister_netdevice_queue(dev, head);
753 }
754 EXPORT_SYMBOL_GPL(macvlan_dellink);
755 
756 static int macvlan_changelink(struct net_device *dev,
757 		struct nlattr *tb[], struct nlattr *data[])
758 {
759 	struct macvlan_dev *vlan = netdev_priv(dev);
760 	if (data && data[IFLA_MACVLAN_MODE])
761 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
762 	return 0;
763 }
764 
765 static size_t macvlan_get_size(const struct net_device *dev)
766 {
767 	return nla_total_size(4);
768 }
769 
770 static int macvlan_fill_info(struct sk_buff *skb,
771 				const struct net_device *dev)
772 {
773 	struct macvlan_dev *vlan = netdev_priv(dev);
774 
775 	NLA_PUT_U32(skb, IFLA_MACVLAN_MODE, vlan->mode);
776 	return 0;
777 
778 nla_put_failure:
779 	return -EMSGSIZE;
780 }
781 
782 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
783 	[IFLA_MACVLAN_MODE] = { .type = NLA_U32 },
784 };
785 
786 int macvlan_link_register(struct rtnl_link_ops *ops)
787 {
788 	/* common fields */
789 	ops->priv_size		= sizeof(struct macvlan_dev);
790 	ops->validate		= macvlan_validate;
791 	ops->maxtype		= IFLA_MACVLAN_MAX;
792 	ops->policy		= macvlan_policy;
793 	ops->changelink		= macvlan_changelink;
794 	ops->get_size		= macvlan_get_size;
795 	ops->fill_info		= macvlan_fill_info;
796 
797 	return rtnl_link_register(ops);
798 };
799 EXPORT_SYMBOL_GPL(macvlan_link_register);
800 
801 static struct rtnl_link_ops macvlan_link_ops = {
802 	.kind		= "macvlan",
803 	.setup		= macvlan_setup,
804 	.newlink	= macvlan_newlink,
805 	.dellink	= macvlan_dellink,
806 };
807 
808 static int macvlan_device_event(struct notifier_block *unused,
809 				unsigned long event, void *ptr)
810 {
811 	struct net_device *dev = ptr;
812 	struct macvlan_dev *vlan, *next;
813 	struct macvlan_port *port;
814 	LIST_HEAD(list_kill);
815 
816 	if (!macvlan_port_exists(dev))
817 		return NOTIFY_DONE;
818 
819 	port = macvlan_port_get(dev);
820 
821 	switch (event) {
822 	case NETDEV_CHANGE:
823 		list_for_each_entry(vlan, &port->vlans, list)
824 			netif_stacked_transfer_operstate(vlan->lowerdev,
825 							 vlan->dev);
826 		break;
827 	case NETDEV_FEAT_CHANGE:
828 		list_for_each_entry(vlan, &port->vlans, list) {
829 			vlan->dev->features = dev->features & MACVLAN_FEATURES;
830 			vlan->dev->gso_max_size = dev->gso_max_size;
831 			netdev_features_change(vlan->dev);
832 		}
833 		break;
834 	case NETDEV_UNREGISTER:
835 		/* twiddle thumbs on netns device moves */
836 		if (dev->reg_state != NETREG_UNREGISTERING)
837 			break;
838 
839 		list_for_each_entry_safe(vlan, next, &port->vlans, list)
840 			vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
841 		unregister_netdevice_many(&list_kill);
842 		list_del(&list_kill);
843 		break;
844 	case NETDEV_PRE_TYPE_CHANGE:
845 		/* Forbid underlaying device to change its type. */
846 		return NOTIFY_BAD;
847 	}
848 	return NOTIFY_DONE;
849 }
850 
851 static struct notifier_block macvlan_notifier_block __read_mostly = {
852 	.notifier_call	= macvlan_device_event,
853 };
854 
855 static int __init macvlan_init_module(void)
856 {
857 	int err;
858 
859 	register_netdevice_notifier(&macvlan_notifier_block);
860 
861 	err = macvlan_link_register(&macvlan_link_ops);
862 	if (err < 0)
863 		goto err1;
864 	return 0;
865 err1:
866 	unregister_netdevice_notifier(&macvlan_notifier_block);
867 	return err;
868 }
869 
870 static void __exit macvlan_cleanup_module(void)
871 {
872 	rtnl_link_unregister(&macvlan_link_ops);
873 	unregister_netdevice_notifier(&macvlan_notifier_block);
874 }
875 
876 module_init(macvlan_init_module);
877 module_exit(macvlan_cleanup_module);
878 
879 MODULE_LICENSE("GPL");
880 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
881 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
882 MODULE_ALIAS_RTNL_LINK("macvlan");
883