veth.c revision 612c337306f00dc8d396830212de51c475844791 
1/*
2 *  drivers/net/veth.c
3 *
4 *  Copyright (C) 2007 OpenVZ http://openvz.org, SWsoft Inc
5 *
6 * Author: Pavel Emelianov <xemul@openvz.org>
7 * Ethtool interface from: Eric W. Biederman <ebiederm@xmission.com>
8 *
9 */
10
11#include <linux/netdevice.h>
12#include <linux/slab.h>
13#include <linux/ethtool.h>
14#include <linux/etherdevice.h>
15#include <linux/u64_stats_sync.h>
16
17#include <net/dst.h>
18#include <net/xfrm.h>
19#include <linux/veth.h>
20#include <linux/module.h>
21
22#define DRV_NAME	"veth"
23#define DRV_VERSION	"1.0"
24
25#define MIN_MTU 68		/* Min L3 MTU */
26#define MAX_MTU 65535		/* Max L3 MTU (arbitrary) */
27
28struct pcpu_vstats {
29	u64			packets;
30	u64			bytes;
31	struct u64_stats_sync	syncp;
32};
33
34struct veth_priv {
35	struct net_device __rcu	*peer;
36	atomic64_t		dropped;
37};
38
39/*
40 * ethtool interface
41 */
42
43static struct {
44	const char string[ETH_GSTRING_LEN];
45} ethtool_stats_keys[] = {
46	{ "peer_ifindex" },
47};
48
49static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
50{
51	cmd->supported		= 0;
52	cmd->advertising	= 0;
53	ethtool_cmd_speed_set(cmd, SPEED_10000);
54	cmd->duplex		= DUPLEX_FULL;
55	cmd->port		= PORT_TP;
56	cmd->phy_address	= 0;
57	cmd->transceiver	= XCVR_INTERNAL;
58	cmd->autoneg		= AUTONEG_DISABLE;
59	cmd->maxtxpkt		= 0;
60	cmd->maxrxpkt		= 0;
61	return 0;
62}
63
64static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
65{
66	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
67	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
68}
69
70static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
71{
72	switch(stringset) {
73	case ETH_SS_STATS:
74		memcpy(buf, &ethtool_stats_keys, sizeof(ethtool_stats_keys));
75		break;
76	}
77}
78
79static int veth_get_sset_count(struct net_device *dev, int sset)
80{
81	switch (sset) {
82	case ETH_SS_STATS:
83		return ARRAY_SIZE(ethtool_stats_keys);
84	default:
85		return -EOPNOTSUPP;
86	}
87}
88
89static void veth_get_ethtool_stats(struct net_device *dev,
90		struct ethtool_stats *stats, u64 *data)
91{
92	struct veth_priv *priv = netdev_priv(dev);
93	struct net_device *peer = rtnl_dereference(priv->peer);
94
95	data[0] = peer ? peer->ifindex : 0;
96}
97
98static const struct ethtool_ops veth_ethtool_ops = {
99	.get_settings		= veth_get_settings,
100	.get_drvinfo		= veth_get_drvinfo,
101	.get_link		= ethtool_op_get_link,
102	.get_strings		= veth_get_strings,
103	.get_sset_count		= veth_get_sset_count,
104	.get_ethtool_stats	= veth_get_ethtool_stats,
105};
106
107static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
108{
109	struct veth_priv *priv = netdev_priv(dev);
110	struct net_device *rcv;
111	int length = skb->len;
112
113	rcu_read_lock();
114	rcv = rcu_dereference(priv->peer);
115	if (unlikely(!rcv)) {
116		kfree_skb(skb);
117		goto drop;
118	}
119	/* don't change ip_summed == CHECKSUM_PARTIAL, as that
120	 * will cause bad checksum on forwarded packets
121	 */
122	if (skb->ip_summed == CHECKSUM_NONE &&
123	    rcv->features & NETIF_F_RXCSUM)
124		skb->ip_summed = CHECKSUM_UNNECESSARY;
125
126	if (likely(dev_forward_skb(rcv, skb) == NET_RX_SUCCESS)) {
127		struct pcpu_vstats *stats = this_cpu_ptr(dev->vstats);
128
129		u64_stats_update_begin(&stats->syncp);
130		stats->bytes += length;
131		stats->packets++;
132		u64_stats_update_end(&stats->syncp);
133	} else {
134drop:
135		atomic64_inc(&priv->dropped);
136	}
137	rcu_read_unlock();
138	return NETDEV_TX_OK;
139}
140
141/*
142 * general routines
143 */
144
145static u64 veth_stats_one(struct pcpu_vstats *result, struct net_device *dev)
146{
147	struct veth_priv *priv = netdev_priv(dev);
148	int cpu;
149
150	result->packets = 0;
151	result->bytes = 0;
152	for_each_possible_cpu(cpu) {
153		struct pcpu_vstats *stats = per_cpu_ptr(dev->vstats, cpu);
154		u64 packets, bytes;
155		unsigned int start;
156
157		do {
158			start = u64_stats_fetch_begin_bh(&stats->syncp);
159			packets = stats->packets;
160			bytes = stats->bytes;
161		} while (u64_stats_fetch_retry_bh(&stats->syncp, start));
162		result->packets += packets;
163		result->bytes += bytes;
164	}
165	return atomic64_read(&priv->dropped);
166}
167
168static struct rtnl_link_stats64 *veth_get_stats64(struct net_device *dev,
169						  struct rtnl_link_stats64 *tot)
170{
171	struct veth_priv *priv = netdev_priv(dev);
172	struct net_device *peer;
173	struct pcpu_vstats one;
174
175	tot->tx_dropped = veth_stats_one(&one, dev);
176	tot->tx_bytes = one.bytes;
177	tot->tx_packets = one.packets;
178
179	rcu_read_lock();
180	peer = rcu_dereference(priv->peer);
181	if (peer) {
182		tot->rx_dropped = veth_stats_one(&one, peer);
183		tot->rx_bytes = one.bytes;
184		tot->rx_packets = one.packets;
185	}
186	rcu_read_unlock();
187
188	return tot;
189}
190
191static int veth_open(struct net_device *dev)
192{
193	struct veth_priv *priv = netdev_priv(dev);
194	struct net_device *peer = rtnl_dereference(priv->peer);
195
196	if (!peer)
197		return -ENOTCONN;
198
199	if (peer->flags & IFF_UP) {
200		netif_carrier_on(dev);
201		netif_carrier_on(peer);
202	}
203	return 0;
204}
205
206static int veth_close(struct net_device *dev)
207{
208	struct veth_priv *priv = netdev_priv(dev);
209	struct net_device *peer = rtnl_dereference(priv->peer);
210
211	netif_carrier_off(dev);
212	if (peer)
213		netif_carrier_off(peer);
214
215	return 0;
216}
217
218static int is_valid_veth_mtu(int new_mtu)
219{
220	return new_mtu >= MIN_MTU && new_mtu <= MAX_MTU;
221}
222
223static int veth_change_mtu(struct net_device *dev, int new_mtu)
224{
225	if (!is_valid_veth_mtu(new_mtu))
226		return -EINVAL;
227	dev->mtu = new_mtu;
228	return 0;
229}
230
231static int veth_dev_init(struct net_device *dev)
232{
233	dev->vstats = alloc_percpu(struct pcpu_vstats);
234	if (!dev->vstats)
235		return -ENOMEM;
236
237	return 0;
238}
239
240static void veth_dev_free(struct net_device *dev)
241{
242	free_percpu(dev->vstats);
243	free_netdev(dev);
244}
245
246static const struct net_device_ops veth_netdev_ops = {
247	.ndo_init            = veth_dev_init,
248	.ndo_open            = veth_open,
249	.ndo_stop            = veth_close,
250	.ndo_start_xmit      = veth_xmit,
251	.ndo_change_mtu      = veth_change_mtu,
252	.ndo_get_stats64     = veth_get_stats64,
253	.ndo_set_mac_address = eth_mac_addr,
254};
255
256#define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_ALL_TSO |    \
257		       NETIF_F_HW_CSUM | NETIF_F_RXCSUM | NETIF_F_HIGHDMA | \
258		       NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
259		       NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
260
261static void veth_setup(struct net_device *dev)
262{
263	ether_setup(dev);
264
265	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
266	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
267
268	dev->netdev_ops = &veth_netdev_ops;
269	dev->ethtool_ops = &veth_ethtool_ops;
270	dev->features |= NETIF_F_LLTX;
271	dev->features |= VETH_FEATURES;
272	dev->vlan_features = dev->features;
273	dev->destructor = veth_dev_free;
274
275	dev->hw_features = VETH_FEATURES;
276}
277
278/*
279 * netlink interface
280 */
281
282static int veth_validate(struct nlattr *tb[], struct nlattr *data[])
283{
284	if (tb[IFLA_ADDRESS]) {
285		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
286			return -EINVAL;
287		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
288			return -EADDRNOTAVAIL;
289	}
290	if (tb[IFLA_MTU]) {
291		if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
292			return -EINVAL;
293	}
294	return 0;
295}
296
297static struct rtnl_link_ops veth_link_ops;
298
299static int veth_newlink(struct net *src_net, struct net_device *dev,
300			 struct nlattr *tb[], struct nlattr *data[])
301{
302	int err;
303	struct net_device *peer;
304	struct veth_priv *priv;
305	char ifname[IFNAMSIZ];
306	struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
307	struct ifinfomsg *ifmp;
308	struct net *net;
309
310	/*
311	 * create and register peer first
312	 */
313	if (data != NULL && data[VETH_INFO_PEER] != NULL) {
314		struct nlattr *nla_peer;
315
316		nla_peer = data[VETH_INFO_PEER];
317		ifmp = nla_data(nla_peer);
318		err = nla_parse(peer_tb, IFLA_MAX,
319				nla_data(nla_peer) + sizeof(struct ifinfomsg),
320				nla_len(nla_peer) - sizeof(struct ifinfomsg),
321				ifla_policy);
322		if (err < 0)
323			return err;
324
325		err = veth_validate(peer_tb, NULL);
326		if (err < 0)
327			return err;
328
329		tbp = peer_tb;
330	} else {
331		ifmp = NULL;
332		tbp = tb;
333	}
334
335	if (tbp[IFLA_IFNAME])
336		nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
337	else
338		snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
339
340	net = rtnl_link_get_net(src_net, tbp);
341	if (IS_ERR(net))
342		return PTR_ERR(net);
343
344	peer = rtnl_create_link(net, ifname, &veth_link_ops, tbp);
345	if (IS_ERR(peer)) {
346		put_net(net);
347		return PTR_ERR(peer);
348	}
349
350	if (tbp[IFLA_ADDRESS] == NULL)
351		eth_hw_addr_random(peer);
352
353	if (ifmp && (dev->ifindex != 0))
354		peer->ifindex = ifmp->ifi_index;
355
356	err = register_netdevice(peer);
357	put_net(net);
358	net = NULL;
359	if (err < 0)
360		goto err_register_peer;
361
362	netif_carrier_off(peer);
363
364	err = rtnl_configure_link(peer, ifmp);
365	if (err < 0)
366		goto err_configure_peer;
367
368	/*
369	 * register dev last
370	 *
371	 * note, that since we've registered new device the dev's name
372	 * should be re-allocated
373	 */
374
375	if (tb[IFLA_ADDRESS] == NULL)
376		eth_hw_addr_random(dev);
377
378	if (tb[IFLA_IFNAME])
379		nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
380	else
381		snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
382
383	err = register_netdevice(dev);
384	if (err < 0)
385		goto err_register_dev;
386
387	netif_carrier_off(dev);
388
389	/*
390	 * tie the deviced together
391	 */
392
393	priv = netdev_priv(dev);
394	rcu_assign_pointer(priv->peer, peer);
395
396	priv = netdev_priv(peer);
397	rcu_assign_pointer(priv->peer, dev);
398	return 0;
399
400err_register_dev:
401	/* nothing to do */
402err_configure_peer:
403	unregister_netdevice(peer);
404	return err;
405
406err_register_peer:
407	free_netdev(peer);
408	return err;
409}
410
411static void veth_dellink(struct net_device *dev, struct list_head *head)
412{
413	struct veth_priv *priv;
414	struct net_device *peer;
415
416	priv = netdev_priv(dev);
417	peer = rtnl_dereference(priv->peer);
418
419	/* Note : dellink() is called from default_device_exit_batch(),
420	 * before a rcu_synchronize() point. The devices are guaranteed
421	 * not being freed before one RCU grace period.
422	 */
423	RCU_INIT_POINTER(priv->peer, NULL);
424	unregister_netdevice_queue(dev, head);
425
426	if (peer) {
427		priv = netdev_priv(peer);
428		RCU_INIT_POINTER(priv->peer, NULL);
429		unregister_netdevice_queue(peer, head);
430	}
431}
432
433static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
434	[VETH_INFO_PEER]	= { .len = sizeof(struct ifinfomsg) },
435};
436
437static size_t veth_get_size(const struct net_device *dev)
438{
439	return nla_total_size(sizeof(u64)) + /* VETH_INFO_PEER */
440		0;
441}
442
443static int veth_fill_info(struct sk_buff *skb, const struct net_device *dev)
444{
445	struct veth_priv *priv = netdev_priv(dev);
446	struct net_device *peer = rtnl_dereference(priv->peer);
447	u64 peer_ifindex;
448
449	peer_ifindex = peer ? peer->ifindex : 0;
450	if (nla_put_u64(skb, VETH_INFO_PEER, peer_ifindex))
451		return -EMSGSIZE;
452
453	return 0;
454}
455
456static struct rtnl_link_ops veth_link_ops = {
457	.kind		= DRV_NAME,
458	.priv_size	= sizeof(struct veth_priv),
459	.setup		= veth_setup,
460	.validate	= veth_validate,
461	.newlink	= veth_newlink,
462	.dellink	= veth_dellink,
463	.policy		= veth_policy,
464	.maxtype	= VETH_INFO_MAX,
465	.get_size	= veth_get_size,
466	.fill_info	= veth_fill_info,
467};
468
469/*
470 * init/fini
471 */
472
473static __init int veth_init(void)
474{
475	return rtnl_link_register(&veth_link_ops);
476}
477
478static __exit void veth_exit(void)
479{
480	rtnl_link_unregister(&veth_link_ops);
481}
482
483module_init(veth_init);
484module_exit(veth_exit);
485
486MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
487MODULE_LICENSE("GPL v2");
488MODULE_ALIAS_RTNL_LINK(DRV_NAME);
489