macvlan.c revision 87002b03baabd2b8f6281ab6411ed88d24958de1
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
37struct 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
46static 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
53static 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
66static 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
74static 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
81static 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
92static 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
109static 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
130static 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 */
160static 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		src = macvlan_hash_lookup(port, eth->h_source);
177		if (!src)
178			/* frame comes from an external address */
179			macvlan_broadcast(skb, port, NULL,
180					  MACVLAN_MODE_PRIVATE |
181					  MACVLAN_MODE_VEPA    |
182					  MACVLAN_MODE_PASSTHRU|
183					  MACVLAN_MODE_BRIDGE);
184		else if (src->mode == MACVLAN_MODE_VEPA)
185			/* flood to everyone except source */
186			macvlan_broadcast(skb, port, src->dev,
187					  MACVLAN_MODE_VEPA |
188					  MACVLAN_MODE_BRIDGE);
189		else if (src->mode == MACVLAN_MODE_BRIDGE)
190			/*
191			 * flood only to VEPA ports, bridge ports
192			 * already saw the frame on the way out.
193			 */
194			macvlan_broadcast(skb, port, src->dev,
195					  MACVLAN_MODE_VEPA);
196		else {
197			/* forward to original port. */
198			vlan = src;
199			ret = macvlan_broadcast_one(skb, vlan, eth, 0);
200			goto out;
201		}
202
203		return RX_HANDLER_PASS;
204	}
205
206	if (port->passthru)
207		vlan = list_first_entry(&port->vlans, struct macvlan_dev, list);
208	else
209		vlan = macvlan_hash_lookup(port, eth->h_dest);
210	if (vlan == NULL)
211		return RX_HANDLER_PASS;
212
213	dev = vlan->dev;
214	if (unlikely(!(dev->flags & IFF_UP))) {
215		kfree_skb(skb);
216		return RX_HANDLER_CONSUMED;
217	}
218	len = skb->len + ETH_HLEN;
219	skb = skb_share_check(skb, GFP_ATOMIC);
220	if (!skb)
221		goto out;
222
223	skb->dev = dev;
224	skb->pkt_type = PACKET_HOST;
225
226	ret = vlan->receive(skb);
227
228out:
229	macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, 0);
230	return RX_HANDLER_CONSUMED;
231}
232
233static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
234{
235	const struct macvlan_dev *vlan = netdev_priv(dev);
236	const struct macvlan_port *port = vlan->port;
237	const struct macvlan_dev *dest;
238	__u8 ip_summed = skb->ip_summed;
239
240	if (vlan->mode == MACVLAN_MODE_BRIDGE) {
241		const struct ethhdr *eth = (void *)skb->data;
242		skb->ip_summed = CHECKSUM_UNNECESSARY;
243
244		/* send to other bridge ports directly */
245		if (is_multicast_ether_addr(eth->h_dest)) {
246			macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
247			goto xmit_world;
248		}
249
250		dest = macvlan_hash_lookup(port, eth->h_dest);
251		if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
252			/* send to lowerdev first for its network taps */
253			dev_forward_skb(vlan->lowerdev, skb);
254
255			return NET_XMIT_SUCCESS;
256		}
257	}
258
259xmit_world:
260	skb->ip_summed = ip_summed;
261	skb_set_dev(skb, vlan->lowerdev);
262	return dev_queue_xmit(skb);
263}
264
265netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
266			       struct net_device *dev)
267{
268	unsigned int len = skb->len;
269	int ret;
270	const struct macvlan_dev *vlan = netdev_priv(dev);
271
272	ret = macvlan_queue_xmit(skb, dev);
273	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
274		struct macvlan_pcpu_stats *pcpu_stats;
275
276		pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
277		u64_stats_update_begin(&pcpu_stats->syncp);
278		pcpu_stats->tx_packets++;
279		pcpu_stats->tx_bytes += len;
280		u64_stats_update_end(&pcpu_stats->syncp);
281	} else {
282		this_cpu_inc(vlan->pcpu_stats->tx_dropped);
283	}
284	return ret;
285}
286EXPORT_SYMBOL_GPL(macvlan_start_xmit);
287
288static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
289			       unsigned short type, const void *daddr,
290			       const void *saddr, unsigned len)
291{
292	const struct macvlan_dev *vlan = netdev_priv(dev);
293	struct net_device *lowerdev = vlan->lowerdev;
294
295	return dev_hard_header(skb, lowerdev, type, daddr,
296			       saddr ? : dev->dev_addr, len);
297}
298
299static const struct header_ops macvlan_hard_header_ops = {
300	.create  	= macvlan_hard_header,
301	.rebuild	= eth_rebuild_header,
302	.parse		= eth_header_parse,
303	.cache		= eth_header_cache,
304	.cache_update	= eth_header_cache_update,
305};
306
307static int macvlan_open(struct net_device *dev)
308{
309	struct macvlan_dev *vlan = netdev_priv(dev);
310	struct net_device *lowerdev = vlan->lowerdev;
311	int err;
312
313	if (vlan->port->passthru) {
314		dev_set_promiscuity(lowerdev, 1);
315		goto hash_add;
316	}
317
318	err = -EBUSY;
319	if (macvlan_addr_busy(vlan->port, dev->dev_addr))
320		goto out;
321
322	err = dev_uc_add(lowerdev, dev->dev_addr);
323	if (err < 0)
324		goto out;
325	if (dev->flags & IFF_ALLMULTI) {
326		err = dev_set_allmulti(lowerdev, 1);
327		if (err < 0)
328			goto del_unicast;
329	}
330
331hash_add:
332	macvlan_hash_add(vlan);
333	return 0;
334
335del_unicast:
336	dev_uc_del(lowerdev, dev->dev_addr);
337out:
338	return err;
339}
340
341static int macvlan_stop(struct net_device *dev)
342{
343	struct macvlan_dev *vlan = netdev_priv(dev);
344	struct net_device *lowerdev = vlan->lowerdev;
345
346	if (vlan->port->passthru) {
347		dev_set_promiscuity(lowerdev, -1);
348		goto hash_del;
349	}
350
351	dev_mc_unsync(lowerdev, dev);
352	if (dev->flags & IFF_ALLMULTI)
353		dev_set_allmulti(lowerdev, -1);
354
355	dev_uc_del(lowerdev, dev->dev_addr);
356
357hash_del:
358	macvlan_hash_del(vlan, !dev->dismantle);
359	return 0;
360}
361
362static int macvlan_set_mac_address(struct net_device *dev, void *p)
363{
364	struct macvlan_dev *vlan = netdev_priv(dev);
365	struct net_device *lowerdev = vlan->lowerdev;
366	struct sockaddr *addr = p;
367	int err;
368
369	if (!is_valid_ether_addr(addr->sa_data))
370		return -EADDRNOTAVAIL;
371
372	if (!(dev->flags & IFF_UP)) {
373		/* Just copy in the new address */
374		memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
375	} else {
376		/* Rehash and update the device filters */
377		if (macvlan_addr_busy(vlan->port, addr->sa_data))
378			return -EBUSY;
379
380		err = dev_uc_add(lowerdev, addr->sa_data);
381		if (err)
382			return err;
383
384		dev_uc_del(lowerdev, dev->dev_addr);
385
386		macvlan_hash_change_addr(vlan, addr->sa_data);
387	}
388	return 0;
389}
390
391static void macvlan_change_rx_flags(struct net_device *dev, int change)
392{
393	struct macvlan_dev *vlan = netdev_priv(dev);
394	struct net_device *lowerdev = vlan->lowerdev;
395
396	if (change & IFF_ALLMULTI)
397		dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
398}
399
400static void macvlan_set_multicast_list(struct net_device *dev)
401{
402	struct macvlan_dev *vlan = netdev_priv(dev);
403
404	dev_mc_sync(vlan->lowerdev, dev);
405}
406
407static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
408{
409	struct macvlan_dev *vlan = netdev_priv(dev);
410
411	if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
412		return -EINVAL;
413	dev->mtu = new_mtu;
414	return 0;
415}
416
417/*
418 * macvlan network devices have devices nesting below it and are a special
419 * "super class" of normal network devices; split their locks off into a
420 * separate class since they always nest.
421 */
422static struct lock_class_key macvlan_netdev_xmit_lock_key;
423static struct lock_class_key macvlan_netdev_addr_lock_key;
424
425#define MACVLAN_FEATURES \
426	(NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
427	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
428	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
429	 NETIF_F_HW_VLAN_FILTER)
430
431#define MACVLAN_STATE_MASK \
432	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
433
434static void macvlan_set_lockdep_class_one(struct net_device *dev,
435					  struct netdev_queue *txq,
436					  void *_unused)
437{
438	lockdep_set_class(&txq->_xmit_lock,
439			  &macvlan_netdev_xmit_lock_key);
440}
441
442static void macvlan_set_lockdep_class(struct net_device *dev)
443{
444	lockdep_set_class(&dev->addr_list_lock,
445			  &macvlan_netdev_addr_lock_key);
446	netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
447}
448
449static int macvlan_init(struct net_device *dev)
450{
451	struct macvlan_dev *vlan = netdev_priv(dev);
452	const struct net_device *lowerdev = vlan->lowerdev;
453
454	dev->state		= (dev->state & ~MACVLAN_STATE_MASK) |
455				  (lowerdev->state & MACVLAN_STATE_MASK);
456	dev->features 		= lowerdev->features & MACVLAN_FEATURES;
457	dev->features		|= NETIF_F_LLTX;
458	dev->gso_max_size	= lowerdev->gso_max_size;
459	dev->iflink		= lowerdev->ifindex;
460	dev->hard_header_len	= lowerdev->hard_header_len;
461
462	macvlan_set_lockdep_class(dev);
463
464	vlan->pcpu_stats = alloc_percpu(struct macvlan_pcpu_stats);
465	if (!vlan->pcpu_stats)
466		return -ENOMEM;
467
468	return 0;
469}
470
471static void macvlan_uninit(struct net_device *dev)
472{
473	struct macvlan_dev *vlan = netdev_priv(dev);
474	struct macvlan_port *port = vlan->port;
475
476	free_percpu(vlan->pcpu_stats);
477
478	port->count -= 1;
479	if (!port->count)
480		macvlan_port_destroy(port->dev);
481}
482
483static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
484							 struct rtnl_link_stats64 *stats)
485{
486	struct macvlan_dev *vlan = netdev_priv(dev);
487
488	if (vlan->pcpu_stats) {
489		struct macvlan_pcpu_stats *p;
490		u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
491		u32 rx_errors = 0, tx_dropped = 0;
492		unsigned int start;
493		int i;
494
495		for_each_possible_cpu(i) {
496			p = per_cpu_ptr(vlan->pcpu_stats, i);
497			do {
498				start = u64_stats_fetch_begin_bh(&p->syncp);
499				rx_packets	= p->rx_packets;
500				rx_bytes	= p->rx_bytes;
501				rx_multicast	= p->rx_multicast;
502				tx_packets	= p->tx_packets;
503				tx_bytes	= p->tx_bytes;
504			} while (u64_stats_fetch_retry_bh(&p->syncp, start));
505
506			stats->rx_packets	+= rx_packets;
507			stats->rx_bytes		+= rx_bytes;
508			stats->multicast	+= rx_multicast;
509			stats->tx_packets	+= tx_packets;
510			stats->tx_bytes		+= tx_bytes;
511			/* rx_errors & tx_dropped are u32, updated
512			 * without syncp protection.
513			 */
514			rx_errors	+= p->rx_errors;
515			tx_dropped	+= p->tx_dropped;
516		}
517		stats->rx_errors	= rx_errors;
518		stats->rx_dropped	= rx_errors;
519		stats->tx_dropped	= tx_dropped;
520	}
521	return stats;
522}
523
524static int macvlan_vlan_rx_add_vid(struct net_device *dev,
525				    unsigned short vid)
526{
527	struct macvlan_dev *vlan = netdev_priv(dev);
528	struct net_device *lowerdev = vlan->lowerdev;
529
530	return vlan_vid_add(lowerdev, vid);
531}
532
533static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
534				     unsigned short vid)
535{
536	struct macvlan_dev *vlan = netdev_priv(dev);
537	struct net_device *lowerdev = vlan->lowerdev;
538
539	vlan_vid_del(lowerdev, vid);
540	return 0;
541}
542
543static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
544					struct ethtool_drvinfo *drvinfo)
545{
546	snprintf(drvinfo->driver, 32, "macvlan");
547	snprintf(drvinfo->version, 32, "0.1");
548}
549
550static int macvlan_ethtool_get_settings(struct net_device *dev,
551					struct ethtool_cmd *cmd)
552{
553	const struct macvlan_dev *vlan = netdev_priv(dev);
554
555	return __ethtool_get_settings(vlan->lowerdev, cmd);
556}
557
558static const struct ethtool_ops macvlan_ethtool_ops = {
559	.get_link		= ethtool_op_get_link,
560	.get_settings		= macvlan_ethtool_get_settings,
561	.get_drvinfo		= macvlan_ethtool_get_drvinfo,
562};
563
564static const struct net_device_ops macvlan_netdev_ops = {
565	.ndo_init		= macvlan_init,
566	.ndo_uninit		= macvlan_uninit,
567	.ndo_open		= macvlan_open,
568	.ndo_stop		= macvlan_stop,
569	.ndo_start_xmit		= macvlan_start_xmit,
570	.ndo_change_mtu		= macvlan_change_mtu,
571	.ndo_change_rx_flags	= macvlan_change_rx_flags,
572	.ndo_set_mac_address	= macvlan_set_mac_address,
573	.ndo_set_rx_mode	= macvlan_set_multicast_list,
574	.ndo_get_stats64	= macvlan_dev_get_stats64,
575	.ndo_validate_addr	= eth_validate_addr,
576	.ndo_vlan_rx_add_vid	= macvlan_vlan_rx_add_vid,
577	.ndo_vlan_rx_kill_vid	= macvlan_vlan_rx_kill_vid,
578};
579
580void macvlan_common_setup(struct net_device *dev)
581{
582	ether_setup(dev);
583
584	dev->priv_flags	       &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
585	dev->netdev_ops		= &macvlan_netdev_ops;
586	dev->destructor		= free_netdev;
587	dev->header_ops		= &macvlan_hard_header_ops,
588	dev->ethtool_ops	= &macvlan_ethtool_ops;
589}
590EXPORT_SYMBOL_GPL(macvlan_common_setup);
591
592static void macvlan_setup(struct net_device *dev)
593{
594	macvlan_common_setup(dev);
595	dev->tx_queue_len	= 0;
596}
597
598static int macvlan_port_create(struct net_device *dev)
599{
600	struct macvlan_port *port;
601	unsigned int i;
602	int err;
603
604	if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
605		return -EINVAL;
606
607	port = kzalloc(sizeof(*port), GFP_KERNEL);
608	if (port == NULL)
609		return -ENOMEM;
610
611	port->passthru = false;
612	port->dev = dev;
613	INIT_LIST_HEAD(&port->vlans);
614	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
615		INIT_HLIST_HEAD(&port->vlan_hash[i]);
616
617	err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
618	if (err)
619		kfree(port);
620	else
621		dev->priv_flags |= IFF_MACVLAN_PORT;
622	return err;
623}
624
625static void macvlan_port_destroy(struct net_device *dev)
626{
627	struct macvlan_port *port = macvlan_port_get(dev);
628
629	dev->priv_flags &= ~IFF_MACVLAN_PORT;
630	netdev_rx_handler_unregister(dev);
631	kfree_rcu(port, rcu);
632}
633
634static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
635{
636	if (tb[IFLA_ADDRESS]) {
637		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
638			return -EINVAL;
639		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
640			return -EADDRNOTAVAIL;
641	}
642
643	if (data && data[IFLA_MACVLAN_MODE]) {
644		switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
645		case MACVLAN_MODE_PRIVATE:
646		case MACVLAN_MODE_VEPA:
647		case MACVLAN_MODE_BRIDGE:
648		case MACVLAN_MODE_PASSTHRU:
649			break;
650		default:
651			return -EINVAL;
652		}
653	}
654	return 0;
655}
656
657int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
658			   struct nlattr *tb[], struct nlattr *data[],
659			   int (*receive)(struct sk_buff *skb),
660			   int (*forward)(struct net_device *dev,
661					  struct sk_buff *skb))
662{
663	struct macvlan_dev *vlan = netdev_priv(dev);
664	struct macvlan_port *port;
665	struct net_device *lowerdev;
666	int err;
667
668	if (!tb[IFLA_LINK])
669		return -EINVAL;
670
671	lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
672	if (lowerdev == NULL)
673		return -ENODEV;
674
675	/* When creating macvlans on top of other macvlans - use
676	 * the real device as the lowerdev.
677	 */
678	if (lowerdev->rtnl_link_ops == dev->rtnl_link_ops) {
679		struct macvlan_dev *lowervlan = netdev_priv(lowerdev);
680		lowerdev = lowervlan->lowerdev;
681	}
682
683	if (!tb[IFLA_MTU])
684		dev->mtu = lowerdev->mtu;
685	else if (dev->mtu > lowerdev->mtu)
686		return -EINVAL;
687
688	if (!tb[IFLA_ADDRESS])
689		random_ether_addr(dev->dev_addr);
690
691	if (!macvlan_port_exists(lowerdev)) {
692		err = macvlan_port_create(lowerdev);
693		if (err < 0)
694			return err;
695	}
696	port = macvlan_port_get(lowerdev);
697
698	/* Only 1 macvlan device can be created in passthru mode */
699	if (port->passthru)
700		return -EINVAL;
701
702	vlan->lowerdev = lowerdev;
703	vlan->dev      = dev;
704	vlan->port     = port;
705	vlan->receive  = receive;
706	vlan->forward  = forward;
707
708	vlan->mode     = MACVLAN_MODE_VEPA;
709	if (data && data[IFLA_MACVLAN_MODE])
710		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
711
712	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
713		if (port->count)
714			return -EINVAL;
715		port->passthru = true;
716		memcpy(dev->dev_addr, lowerdev->dev_addr, ETH_ALEN);
717	}
718
719	port->count += 1;
720	err = register_netdevice(dev);
721	if (err < 0)
722		goto destroy_port;
723
724	list_add_tail(&vlan->list, &port->vlans);
725	netif_stacked_transfer_operstate(lowerdev, dev);
726
727	return 0;
728
729destroy_port:
730	port->count -= 1;
731	if (!port->count)
732		macvlan_port_destroy(lowerdev);
733
734	return err;
735}
736EXPORT_SYMBOL_GPL(macvlan_common_newlink);
737
738static int macvlan_newlink(struct net *src_net, struct net_device *dev,
739			   struct nlattr *tb[], struct nlattr *data[])
740{
741	return macvlan_common_newlink(src_net, dev, tb, data,
742				      netif_rx,
743				      dev_forward_skb);
744}
745
746void macvlan_dellink(struct net_device *dev, struct list_head *head)
747{
748	struct macvlan_dev *vlan = netdev_priv(dev);
749
750	list_del(&vlan->list);
751	unregister_netdevice_queue(dev, head);
752}
753EXPORT_SYMBOL_GPL(macvlan_dellink);
754
755static int macvlan_changelink(struct net_device *dev,
756		struct nlattr *tb[], struct nlattr *data[])
757{
758	struct macvlan_dev *vlan = netdev_priv(dev);
759	if (data && data[IFLA_MACVLAN_MODE])
760		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
761	return 0;
762}
763
764static size_t macvlan_get_size(const struct net_device *dev)
765{
766	return nla_total_size(4);
767}
768
769static int macvlan_fill_info(struct sk_buff *skb,
770				const struct net_device *dev)
771{
772	struct macvlan_dev *vlan = netdev_priv(dev);
773
774	NLA_PUT_U32(skb, IFLA_MACVLAN_MODE, vlan->mode);
775	return 0;
776
777nla_put_failure:
778	return -EMSGSIZE;
779}
780
781static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
782	[IFLA_MACVLAN_MODE] = { .type = NLA_U32 },
783};
784
785int macvlan_link_register(struct rtnl_link_ops *ops)
786{
787	/* common fields */
788	ops->priv_size		= sizeof(struct macvlan_dev);
789	ops->validate		= macvlan_validate;
790	ops->maxtype		= IFLA_MACVLAN_MAX;
791	ops->policy		= macvlan_policy;
792	ops->changelink		= macvlan_changelink;
793	ops->get_size		= macvlan_get_size;
794	ops->fill_info		= macvlan_fill_info;
795
796	return rtnl_link_register(ops);
797};
798EXPORT_SYMBOL_GPL(macvlan_link_register);
799
800static struct rtnl_link_ops macvlan_link_ops = {
801	.kind		= "macvlan",
802	.setup		= macvlan_setup,
803	.newlink	= macvlan_newlink,
804	.dellink	= macvlan_dellink,
805};
806
807static int macvlan_device_event(struct notifier_block *unused,
808				unsigned long event, void *ptr)
809{
810	struct net_device *dev = ptr;
811	struct macvlan_dev *vlan, *next;
812	struct macvlan_port *port;
813	LIST_HEAD(list_kill);
814
815	if (!macvlan_port_exists(dev))
816		return NOTIFY_DONE;
817
818	port = macvlan_port_get(dev);
819
820	switch (event) {
821	case NETDEV_CHANGE:
822		list_for_each_entry(vlan, &port->vlans, list)
823			netif_stacked_transfer_operstate(vlan->lowerdev,
824							 vlan->dev);
825		break;
826	case NETDEV_FEAT_CHANGE:
827		list_for_each_entry(vlan, &port->vlans, list) {
828			vlan->dev->features = dev->features & MACVLAN_FEATURES;
829			vlan->dev->gso_max_size = dev->gso_max_size;
830			netdev_features_change(vlan->dev);
831		}
832		break;
833	case NETDEV_UNREGISTER:
834		/* twiddle thumbs on netns device moves */
835		if (dev->reg_state != NETREG_UNREGISTERING)
836			break;
837
838		list_for_each_entry_safe(vlan, next, &port->vlans, list)
839			vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
840		unregister_netdevice_many(&list_kill);
841		list_del(&list_kill);
842		break;
843	case NETDEV_PRE_TYPE_CHANGE:
844		/* Forbid underlaying device to change its type. */
845		return NOTIFY_BAD;
846	}
847	return NOTIFY_DONE;
848}
849
850static struct notifier_block macvlan_notifier_block __read_mostly = {
851	.notifier_call	= macvlan_device_event,
852};
853
854static int __init macvlan_init_module(void)
855{
856	int err;
857
858	register_netdevice_notifier(&macvlan_notifier_block);
859
860	err = macvlan_link_register(&macvlan_link_ops);
861	if (err < 0)
862		goto err1;
863	return 0;
864err1:
865	unregister_netdevice_notifier(&macvlan_notifier_block);
866	return err;
867}
868
869static void __exit macvlan_cleanup_module(void)
870{
871	rtnl_link_unregister(&macvlan_link_ops);
872	unregister_netdevice_notifier(&macvlan_notifier_block);
873}
874
875module_init(macvlan_init_module);
876module_exit(macvlan_cleanup_module);
877
878MODULE_LICENSE("GPL");
879MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
880MODULE_DESCRIPTION("Driver for MAC address based VLANs");
881MODULE_ALIAS_RTNL_LINK("macvlan");
882