macvlan.c revision efbbced361f3ff4ff9e85310ccff894185c4d904
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
33#define MACVLAN_HASH_SIZE	(1 << BITS_PER_BYTE)
34
35struct macvlan_port {
36	struct net_device	*dev;
37	struct hlist_head	vlan_hash[MACVLAN_HASH_SIZE];
38	struct list_head	vlans;
39};
40
41struct macvlan_dev {
42	struct net_device	*dev;
43	struct list_head	list;
44	struct hlist_node	hlist;
45	struct macvlan_port	*port;
46	struct net_device	*lowerdev;
47};
48
49
50static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
51					       const unsigned char *addr)
52{
53	struct macvlan_dev *vlan;
54	struct hlist_node *n;
55
56	hlist_for_each_entry_rcu(vlan, n, &port->vlan_hash[addr[5]], hlist) {
57		if (!compare_ether_addr(vlan->dev->dev_addr, addr))
58			return vlan;
59	}
60	return NULL;
61}
62
63static void macvlan_broadcast(struct sk_buff *skb,
64			      const struct macvlan_port *port)
65{
66	const struct ethhdr *eth = eth_hdr(skb);
67	const struct macvlan_dev *vlan;
68	struct hlist_node *n;
69	struct net_device *dev;
70	struct sk_buff *nskb;
71	unsigned int i;
72
73	if (skb->protocol == htons(ETH_P_PAUSE))
74		return;
75
76	for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
77		hlist_for_each_entry_rcu(vlan, n, &port->vlan_hash[i], hlist) {
78			dev = vlan->dev;
79
80			nskb = skb_clone(skb, GFP_ATOMIC);
81			if (nskb == NULL) {
82				dev->stats.rx_errors++;
83				dev->stats.rx_dropped++;
84				continue;
85			}
86
87			dev->stats.rx_bytes += skb->len + ETH_HLEN;
88			dev->stats.rx_packets++;
89			dev->stats.multicast++;
90			dev->last_rx = jiffies;
91
92			nskb->dev = dev;
93			if (!compare_ether_addr(eth->h_dest, dev->broadcast))
94				nskb->pkt_type = PACKET_BROADCAST;
95			else
96				nskb->pkt_type = PACKET_MULTICAST;
97
98			netif_rx(nskb);
99		}
100	}
101}
102
103/* called under rcu_read_lock() from netif_receive_skb */
104static struct sk_buff *macvlan_handle_frame(struct sk_buff *skb)
105{
106	const struct ethhdr *eth = eth_hdr(skb);
107	const struct macvlan_port *port;
108	const struct macvlan_dev *vlan;
109	struct net_device *dev;
110
111	port = rcu_dereference(skb->dev->macvlan_port);
112	if (port == NULL)
113		return skb;
114
115	if (is_multicast_ether_addr(eth->h_dest)) {
116		macvlan_broadcast(skb, port);
117		return skb;
118	}
119
120	vlan = macvlan_hash_lookup(port, eth->h_dest);
121	if (vlan == NULL)
122		return skb;
123
124	dev = vlan->dev;
125	if (unlikely(!(dev->flags & IFF_UP))) {
126		kfree_skb(skb);
127		return NULL;
128	}
129
130	skb = skb_share_check(skb, GFP_ATOMIC);
131	if (skb == NULL) {
132		dev->stats.rx_errors++;
133		dev->stats.rx_dropped++;
134		return NULL;
135	}
136
137	dev->stats.rx_bytes += skb->len + ETH_HLEN;
138	dev->stats.rx_packets++;
139	dev->last_rx = jiffies;
140
141	skb->dev = dev;
142	skb->pkt_type = PACKET_HOST;
143
144	netif_rx(skb);
145	return NULL;
146}
147
148static int macvlan_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
149{
150	const struct macvlan_dev *vlan = netdev_priv(dev);
151	unsigned int len = skb->len;
152	int ret;
153
154	skb->dev = vlan->lowerdev;
155	ret = dev_queue_xmit(skb);
156
157	if (likely(ret == NET_XMIT_SUCCESS)) {
158		dev->stats.tx_packets++;
159		dev->stats.tx_bytes += len;
160	} else {
161		dev->stats.tx_errors++;
162		dev->stats.tx_aborted_errors++;
163	}
164	return NETDEV_TX_OK;
165}
166
167static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
168			       unsigned short type, const void *daddr,
169			       const void *saddr, unsigned len)
170{
171	const struct macvlan_dev *vlan = netdev_priv(dev);
172	struct net_device *lowerdev = vlan->lowerdev;
173
174	return dev_hard_header(skb, lowerdev, type, daddr,
175			       saddr ? : dev->dev_addr, len);
176}
177
178static const struct header_ops macvlan_hard_header_ops = {
179	.create  	= macvlan_hard_header,
180	.rebuild	= eth_rebuild_header,
181	.parse		= eth_header_parse,
182	.cache		= eth_header_cache,
183	.cache_update	= eth_header_cache_update,
184};
185
186static int macvlan_open(struct net_device *dev)
187{
188	struct macvlan_dev *vlan = netdev_priv(dev);
189	struct macvlan_port *port = vlan->port;
190	struct net_device *lowerdev = vlan->lowerdev;
191	int err;
192
193	err = dev_unicast_add(lowerdev, dev->dev_addr, ETH_ALEN);
194	if (err < 0)
195		goto out;
196	if (dev->flags & IFF_ALLMULTI) {
197		err = dev_set_allmulti(lowerdev, 1);
198		if (err < 0)
199			goto del_unicast;
200	}
201
202	hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[dev->dev_addr[5]]);
203	return 0;
204
205del_unicast:
206	dev_unicast_delete(lowerdev, dev->dev_addr, ETH_ALEN);
207out:
208	return err;
209}
210
211static int macvlan_stop(struct net_device *dev)
212{
213	struct macvlan_dev *vlan = netdev_priv(dev);
214	struct net_device *lowerdev = vlan->lowerdev;
215
216	dev_mc_unsync(lowerdev, dev);
217	if (dev->flags & IFF_ALLMULTI)
218		dev_set_allmulti(lowerdev, -1);
219
220	dev_unicast_delete(lowerdev, dev->dev_addr, ETH_ALEN);
221
222	hlist_del_rcu(&vlan->hlist);
223	synchronize_rcu();
224	return 0;
225}
226
227static int macvlan_set_mac_address(struct net_device *dev, void *p)
228{
229	struct macvlan_dev *vlan = netdev_priv(dev);
230	struct net_device *lowerdev = vlan->lowerdev;
231	struct sockaddr *addr = p;
232	int err;
233
234	if (!is_valid_ether_addr(addr->sa_data))
235		return -EADDRNOTAVAIL;
236
237	if (!(dev->flags & IFF_UP))
238		goto out;
239
240	err = dev_unicast_add(lowerdev, addr->sa_data, ETH_ALEN);
241	if (err < 0)
242		return err;
243	dev_unicast_delete(lowerdev, dev->dev_addr, ETH_ALEN);
244
245out:
246	memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
247	return 0;
248}
249
250static void macvlan_change_rx_flags(struct net_device *dev, int change)
251{
252	struct macvlan_dev *vlan = netdev_priv(dev);
253	struct net_device *lowerdev = vlan->lowerdev;
254
255	if (change & IFF_ALLMULTI)
256		dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
257}
258
259static void macvlan_set_multicast_list(struct net_device *dev)
260{
261	struct macvlan_dev *vlan = netdev_priv(dev);
262
263	dev_mc_sync(vlan->lowerdev, dev);
264}
265
266static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
267{
268	struct macvlan_dev *vlan = netdev_priv(dev);
269
270	if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
271		return -EINVAL;
272	dev->mtu = new_mtu;
273	return 0;
274}
275
276/*
277 * macvlan network devices have devices nesting below it and are a special
278 * "super class" of normal network devices; split their locks off into a
279 * separate class since they always nest.
280 */
281static struct lock_class_key macvlan_netdev_xmit_lock_key;
282static struct lock_class_key macvlan_netdev_addr_lock_key;
283
284#define MACVLAN_FEATURES \
285	(NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
286	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
287	 NETIF_F_TSO_ECN | NETIF_F_TSO6)
288
289#define MACVLAN_STATE_MASK \
290	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
291
292static void macvlan_set_lockdep_class_one(struct net_device *dev,
293					  struct netdev_queue *txq,
294					  void *_unused)
295{
296	lockdep_set_class(&txq->_xmit_lock,
297			  &macvlan_netdev_xmit_lock_key);
298}
299
300static void macvlan_set_lockdep_class(struct net_device *dev)
301{
302	lockdep_set_class(&dev->addr_list_lock,
303			  &macvlan_netdev_addr_lock_key);
304	netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
305}
306
307static int macvlan_init(struct net_device *dev)
308{
309	struct macvlan_dev *vlan = netdev_priv(dev);
310	const struct net_device *lowerdev = vlan->lowerdev;
311
312	dev->state		= (dev->state & ~MACVLAN_STATE_MASK) |
313				  (lowerdev->state & MACVLAN_STATE_MASK);
314	dev->features 		= lowerdev->features & MACVLAN_FEATURES;
315	dev->iflink		= lowerdev->ifindex;
316
317	macvlan_set_lockdep_class(dev);
318
319	return 0;
320}
321
322static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
323					struct ethtool_drvinfo *drvinfo)
324{
325	snprintf(drvinfo->driver, 32, "macvlan");
326	snprintf(drvinfo->version, 32, "0.1");
327}
328
329static u32 macvlan_ethtool_get_rx_csum(struct net_device *dev)
330{
331	const struct macvlan_dev *vlan = netdev_priv(dev);
332	struct net_device *lowerdev = vlan->lowerdev;
333
334	if (lowerdev->ethtool_ops->get_rx_csum == NULL)
335		return 0;
336	return lowerdev->ethtool_ops->get_rx_csum(lowerdev);
337}
338
339static const struct ethtool_ops macvlan_ethtool_ops = {
340	.get_link		= ethtool_op_get_link,
341	.get_rx_csum		= macvlan_ethtool_get_rx_csum,
342	.get_drvinfo		= macvlan_ethtool_get_drvinfo,
343};
344
345static void macvlan_setup(struct net_device *dev)
346{
347	ether_setup(dev);
348
349	dev->init		= macvlan_init;
350	dev->open		= macvlan_open;
351	dev->stop		= macvlan_stop;
352	dev->change_mtu		= macvlan_change_mtu;
353	dev->change_rx_flags	= macvlan_change_rx_flags;
354	dev->set_mac_address	= macvlan_set_mac_address;
355	dev->set_multicast_list	= macvlan_set_multicast_list;
356	dev->hard_start_xmit	= macvlan_hard_start_xmit;
357	dev->destructor		= free_netdev;
358	dev->header_ops		= &macvlan_hard_header_ops,
359	dev->ethtool_ops	= &macvlan_ethtool_ops;
360	dev->tx_queue_len	= 0;
361}
362
363static int macvlan_port_create(struct net_device *dev)
364{
365	struct macvlan_port *port;
366	unsigned int i;
367
368	if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
369		return -EINVAL;
370
371	port = kzalloc(sizeof(*port), GFP_KERNEL);
372	if (port == NULL)
373		return -ENOMEM;
374
375	port->dev = dev;
376	INIT_LIST_HEAD(&port->vlans);
377	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
378		INIT_HLIST_HEAD(&port->vlan_hash[i]);
379	rcu_assign_pointer(dev->macvlan_port, port);
380	return 0;
381}
382
383static void macvlan_port_destroy(struct net_device *dev)
384{
385	struct macvlan_port *port = dev->macvlan_port;
386
387	rcu_assign_pointer(dev->macvlan_port, NULL);
388	synchronize_rcu();
389	kfree(port);
390}
391
392static void macvlan_transfer_operstate(struct net_device *dev)
393{
394	struct macvlan_dev *vlan = netdev_priv(dev);
395	const struct net_device *lowerdev = vlan->lowerdev;
396
397	if (lowerdev->operstate == IF_OPER_DORMANT)
398		netif_dormant_on(dev);
399	else
400		netif_dormant_off(dev);
401
402	if (netif_carrier_ok(lowerdev)) {
403		if (!netif_carrier_ok(dev))
404			netif_carrier_on(dev);
405	} else {
406		if (netif_carrier_ok(dev))
407			netif_carrier_off(dev);
408	}
409}
410
411static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
412{
413	if (tb[IFLA_ADDRESS]) {
414		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
415			return -EINVAL;
416		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
417			return -EADDRNOTAVAIL;
418	}
419	return 0;
420}
421
422static int macvlan_newlink(struct net_device *dev,
423			   struct nlattr *tb[], struct nlattr *data[])
424{
425	struct macvlan_dev *vlan = netdev_priv(dev);
426	struct macvlan_port *port;
427	struct net_device *lowerdev;
428	int err;
429
430	if (!tb[IFLA_LINK])
431		return -EINVAL;
432
433	lowerdev = __dev_get_by_index(dev_net(dev), nla_get_u32(tb[IFLA_LINK]));
434	if (lowerdev == NULL)
435		return -ENODEV;
436
437	/* Don't allow macvlans on top of other macvlans - its not really
438	 * wrong, but lockdep can't handle it and its not useful for anything
439	 * you couldn't do directly on top of the real device.
440	 */
441	if (lowerdev->rtnl_link_ops == dev->rtnl_link_ops)
442		return -ENODEV;
443
444	if (!tb[IFLA_MTU])
445		dev->mtu = lowerdev->mtu;
446	else if (dev->mtu > lowerdev->mtu)
447		return -EINVAL;
448
449	if (!tb[IFLA_ADDRESS])
450		random_ether_addr(dev->dev_addr);
451
452	if (lowerdev->macvlan_port == NULL) {
453		err = macvlan_port_create(lowerdev);
454		if (err < 0)
455			return err;
456	}
457	port = lowerdev->macvlan_port;
458
459	vlan->lowerdev = lowerdev;
460	vlan->dev      = dev;
461	vlan->port     = port;
462
463	err = register_netdevice(dev);
464	if (err < 0)
465		return err;
466
467	list_add_tail(&vlan->list, &port->vlans);
468	macvlan_transfer_operstate(dev);
469	return 0;
470}
471
472static void macvlan_dellink(struct net_device *dev)
473{
474	struct macvlan_dev *vlan = netdev_priv(dev);
475	struct macvlan_port *port = vlan->port;
476
477	list_del(&vlan->list);
478	unregister_netdevice(dev);
479
480	if (list_empty(&port->vlans))
481		macvlan_port_destroy(port->dev);
482}
483
484static struct rtnl_link_ops macvlan_link_ops __read_mostly = {
485	.kind		= "macvlan",
486	.priv_size	= sizeof(struct macvlan_dev),
487	.setup		= macvlan_setup,
488	.validate	= macvlan_validate,
489	.newlink	= macvlan_newlink,
490	.dellink	= macvlan_dellink,
491};
492
493static int macvlan_device_event(struct notifier_block *unused,
494				unsigned long event, void *ptr)
495{
496	struct net_device *dev = ptr;
497	struct macvlan_dev *vlan, *next;
498	struct macvlan_port *port;
499
500	port = dev->macvlan_port;
501	if (port == NULL)
502		return NOTIFY_DONE;
503
504	switch (event) {
505	case NETDEV_CHANGE:
506		list_for_each_entry(vlan, &port->vlans, list)
507			macvlan_transfer_operstate(vlan->dev);
508		break;
509	case NETDEV_FEAT_CHANGE:
510		list_for_each_entry(vlan, &port->vlans, list) {
511			vlan->dev->features = dev->features & MACVLAN_FEATURES;
512			netdev_features_change(vlan->dev);
513		}
514		break;
515	case NETDEV_UNREGISTER:
516		list_for_each_entry_safe(vlan, next, &port->vlans, list)
517			macvlan_dellink(vlan->dev);
518		break;
519	}
520	return NOTIFY_DONE;
521}
522
523static struct notifier_block macvlan_notifier_block __read_mostly = {
524	.notifier_call	= macvlan_device_event,
525};
526
527static int __init macvlan_init_module(void)
528{
529	int err;
530
531	register_netdevice_notifier(&macvlan_notifier_block);
532	macvlan_handle_frame_hook = macvlan_handle_frame;
533
534	err = rtnl_link_register(&macvlan_link_ops);
535	if (err < 0)
536		goto err1;
537	return 0;
538err1:
539	macvlan_handle_frame_hook = NULL;
540	unregister_netdevice_notifier(&macvlan_notifier_block);
541	return err;
542}
543
544static void __exit macvlan_cleanup_module(void)
545{
546	rtnl_link_unregister(&macvlan_link_ops);
547	macvlan_handle_frame_hook = NULL;
548	unregister_netdevice_notifier(&macvlan_notifier_block);
549}
550
551module_init(macvlan_init_module);
552module_exit(macvlan_cleanup_module);
553
554MODULE_LICENSE("GPL");
555MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
556MODULE_DESCRIPTION("Driver for MAC address based VLANs");
557MODULE_ALIAS_RTNL_LINK("macvlan");
558