xref: /openbmc/linux/drivers/net/veth.c (revision 9b9c2cd4)
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/rtnetlink.h>
18 #include <net/dst.h>
19 #include <net/xfrm.h>
20 #include <linux/veth.h>
21 #include <linux/module.h>
22 
23 #define DRV_NAME	"veth"
24 #define DRV_VERSION	"1.0"
25 
26 #define MIN_MTU 68		/* Min L3 MTU */
27 #define MAX_MTU 65535		/* Max L3 MTU (arbitrary) */
28 
29 struct pcpu_vstats {
30 	u64			packets;
31 	u64			bytes;
32 	struct u64_stats_sync	syncp;
33 };
34 
35 struct veth_priv {
36 	struct net_device __rcu	*peer;
37 	atomic64_t		dropped;
38 };
39 
40 /*
41  * ethtool interface
42  */
43 
44 static struct {
45 	const char string[ETH_GSTRING_LEN];
46 } ethtool_stats_keys[] = {
47 	{ "peer_ifindex" },
48 };
49 
50 static int veth_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
51 {
52 	cmd->supported		= 0;
53 	cmd->advertising	= 0;
54 	ethtool_cmd_speed_set(cmd, SPEED_10000);
55 	cmd->duplex		= DUPLEX_FULL;
56 	cmd->port		= PORT_TP;
57 	cmd->phy_address	= 0;
58 	cmd->transceiver	= XCVR_INTERNAL;
59 	cmd->autoneg		= AUTONEG_DISABLE;
60 	cmd->maxtxpkt		= 0;
61 	cmd->maxrxpkt		= 0;
62 	return 0;
63 }
64 
65 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
66 {
67 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
68 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
69 }
70 
71 static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
72 {
73 	switch(stringset) {
74 	case ETH_SS_STATS:
75 		memcpy(buf, &ethtool_stats_keys, sizeof(ethtool_stats_keys));
76 		break;
77 	}
78 }
79 
80 static int veth_get_sset_count(struct net_device *dev, int sset)
81 {
82 	switch (sset) {
83 	case ETH_SS_STATS:
84 		return ARRAY_SIZE(ethtool_stats_keys);
85 	default:
86 		return -EOPNOTSUPP;
87 	}
88 }
89 
90 static void veth_get_ethtool_stats(struct net_device *dev,
91 		struct ethtool_stats *stats, u64 *data)
92 {
93 	struct veth_priv *priv = netdev_priv(dev);
94 	struct net_device *peer = rtnl_dereference(priv->peer);
95 
96 	data[0] = peer ? peer->ifindex : 0;
97 }
98 
99 static const struct ethtool_ops veth_ethtool_ops = {
100 	.get_settings		= veth_get_settings,
101 	.get_drvinfo		= veth_get_drvinfo,
102 	.get_link		= ethtool_op_get_link,
103 	.get_strings		= veth_get_strings,
104 	.get_sset_count		= veth_get_sset_count,
105 	.get_ethtool_stats	= veth_get_ethtool_stats,
106 };
107 
108 static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
109 {
110 	struct veth_priv *priv = netdev_priv(dev);
111 	struct net_device *rcv;
112 	int length = skb->len;
113 
114 	rcu_read_lock();
115 	rcv = rcu_dereference(priv->peer);
116 	if (unlikely(!rcv)) {
117 		kfree_skb(skb);
118 		goto drop;
119 	}
120 
121 	if (likely(dev_forward_skb(rcv, skb) == NET_RX_SUCCESS)) {
122 		struct pcpu_vstats *stats = this_cpu_ptr(dev->vstats);
123 
124 		u64_stats_update_begin(&stats->syncp);
125 		stats->bytes += length;
126 		stats->packets++;
127 		u64_stats_update_end(&stats->syncp);
128 	} else {
129 drop:
130 		atomic64_inc(&priv->dropped);
131 	}
132 	rcu_read_unlock();
133 	return NETDEV_TX_OK;
134 }
135 
136 /*
137  * general routines
138  */
139 
140 static u64 veth_stats_one(struct pcpu_vstats *result, struct net_device *dev)
141 {
142 	struct veth_priv *priv = netdev_priv(dev);
143 	int cpu;
144 
145 	result->packets = 0;
146 	result->bytes = 0;
147 	for_each_possible_cpu(cpu) {
148 		struct pcpu_vstats *stats = per_cpu_ptr(dev->vstats, cpu);
149 		u64 packets, bytes;
150 		unsigned int start;
151 
152 		do {
153 			start = u64_stats_fetch_begin_irq(&stats->syncp);
154 			packets = stats->packets;
155 			bytes = stats->bytes;
156 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
157 		result->packets += packets;
158 		result->bytes += bytes;
159 	}
160 	return atomic64_read(&priv->dropped);
161 }
162 
163 static struct rtnl_link_stats64 *veth_get_stats64(struct net_device *dev,
164 						  struct rtnl_link_stats64 *tot)
165 {
166 	struct veth_priv *priv = netdev_priv(dev);
167 	struct net_device *peer;
168 	struct pcpu_vstats one;
169 
170 	tot->tx_dropped = veth_stats_one(&one, dev);
171 	tot->tx_bytes = one.bytes;
172 	tot->tx_packets = one.packets;
173 
174 	rcu_read_lock();
175 	peer = rcu_dereference(priv->peer);
176 	if (peer) {
177 		tot->rx_dropped = veth_stats_one(&one, peer);
178 		tot->rx_bytes = one.bytes;
179 		tot->rx_packets = one.packets;
180 	}
181 	rcu_read_unlock();
182 
183 	return tot;
184 }
185 
186 /* fake multicast ability */
187 static void veth_set_multicast_list(struct net_device *dev)
188 {
189 }
190 
191 static 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 
206 static 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 
218 static int is_valid_veth_mtu(int new_mtu)
219 {
220 	return new_mtu >= MIN_MTU && new_mtu <= MAX_MTU;
221 }
222 
223 static 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 
231 static int veth_dev_init(struct net_device *dev)
232 {
233 	dev->vstats = netdev_alloc_pcpu_stats(struct pcpu_vstats);
234 	if (!dev->vstats)
235 		return -ENOMEM;
236 	return 0;
237 }
238 
239 static void veth_dev_free(struct net_device *dev)
240 {
241 	free_percpu(dev->vstats);
242 	free_netdev(dev);
243 }
244 
245 #ifdef CONFIG_NET_POLL_CONTROLLER
246 static void veth_poll_controller(struct net_device *dev)
247 {
248 	/* veth only receives frames when its peer sends one
249 	 * Since it's a synchronous operation, we are guaranteed
250 	 * never to have pending data when we poll for it so
251 	 * there is nothing to do here.
252 	 *
253 	 * We need this though so netpoll recognizes us as an interface that
254 	 * supports polling, which enables bridge devices in virt setups to
255 	 * still use netconsole
256 	 */
257 }
258 #endif	/* CONFIG_NET_POLL_CONTROLLER */
259 
260 static int veth_get_iflink(const struct net_device *dev)
261 {
262 	struct veth_priv *priv = netdev_priv(dev);
263 	struct net_device *peer;
264 	int iflink;
265 
266 	rcu_read_lock();
267 	peer = rcu_dereference(priv->peer);
268 	iflink = peer ? peer->ifindex : 0;
269 	rcu_read_unlock();
270 
271 	return iflink;
272 }
273 
274 static const struct net_device_ops veth_netdev_ops = {
275 	.ndo_init            = veth_dev_init,
276 	.ndo_open            = veth_open,
277 	.ndo_stop            = veth_close,
278 	.ndo_start_xmit      = veth_xmit,
279 	.ndo_change_mtu      = veth_change_mtu,
280 	.ndo_get_stats64     = veth_get_stats64,
281 	.ndo_set_rx_mode     = veth_set_multicast_list,
282 	.ndo_set_mac_address = eth_mac_addr,
283 #ifdef CONFIG_NET_POLL_CONTROLLER
284 	.ndo_poll_controller	= veth_poll_controller,
285 #endif
286 	.ndo_get_iflink		= veth_get_iflink,
287 	.ndo_features_check	= passthru_features_check,
288 };
289 
290 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_ALL_TSO |    \
291 		       NETIF_F_HW_CSUM | NETIF_F_RXCSUM | NETIF_F_HIGHDMA | \
292 		       NETIF_F_GSO_GRE | NETIF_F_GSO_UDP_TUNNEL |	    \
293 		       NETIF_F_GSO_IPIP | NETIF_F_GSO_SIT | NETIF_F_UFO	|   \
294 		       NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
295 		       NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
296 
297 static void veth_setup(struct net_device *dev)
298 {
299 	ether_setup(dev);
300 
301 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
302 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
303 	dev->priv_flags |= IFF_NO_QUEUE;
304 
305 	dev->netdev_ops = &veth_netdev_ops;
306 	dev->ethtool_ops = &veth_ethtool_ops;
307 	dev->features |= NETIF_F_LLTX;
308 	dev->features |= VETH_FEATURES;
309 	dev->vlan_features = dev->features &
310 			     ~(NETIF_F_HW_VLAN_CTAG_TX |
311 			       NETIF_F_HW_VLAN_STAG_TX |
312 			       NETIF_F_HW_VLAN_CTAG_RX |
313 			       NETIF_F_HW_VLAN_STAG_RX);
314 	dev->destructor = veth_dev_free;
315 
316 	dev->hw_features = VETH_FEATURES;
317 	dev->hw_enc_features = VETH_FEATURES;
318 }
319 
320 /*
321  * netlink interface
322  */
323 
324 static int veth_validate(struct nlattr *tb[], struct nlattr *data[])
325 {
326 	if (tb[IFLA_ADDRESS]) {
327 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
328 			return -EINVAL;
329 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
330 			return -EADDRNOTAVAIL;
331 	}
332 	if (tb[IFLA_MTU]) {
333 		if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
334 			return -EINVAL;
335 	}
336 	return 0;
337 }
338 
339 static struct rtnl_link_ops veth_link_ops;
340 
341 static int veth_newlink(struct net *src_net, struct net_device *dev,
342 			 struct nlattr *tb[], struct nlattr *data[])
343 {
344 	int err;
345 	struct net_device *peer;
346 	struct veth_priv *priv;
347 	char ifname[IFNAMSIZ];
348 	struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
349 	unsigned char name_assign_type;
350 	struct ifinfomsg *ifmp;
351 	struct net *net;
352 
353 	/*
354 	 * create and register peer first
355 	 */
356 	if (data != NULL && data[VETH_INFO_PEER] != NULL) {
357 		struct nlattr *nla_peer;
358 
359 		nla_peer = data[VETH_INFO_PEER];
360 		ifmp = nla_data(nla_peer);
361 		err = rtnl_nla_parse_ifla(peer_tb,
362 					  nla_data(nla_peer) + sizeof(struct ifinfomsg),
363 					  nla_len(nla_peer) - sizeof(struct ifinfomsg));
364 		if (err < 0)
365 			return err;
366 
367 		err = veth_validate(peer_tb, NULL);
368 		if (err < 0)
369 			return err;
370 
371 		tbp = peer_tb;
372 	} else {
373 		ifmp = NULL;
374 		tbp = tb;
375 	}
376 
377 	if (tbp[IFLA_IFNAME]) {
378 		nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
379 		name_assign_type = NET_NAME_USER;
380 	} else {
381 		snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
382 		name_assign_type = NET_NAME_ENUM;
383 	}
384 
385 	net = rtnl_link_get_net(src_net, tbp);
386 	if (IS_ERR(net))
387 		return PTR_ERR(net);
388 
389 	peer = rtnl_create_link(net, ifname, name_assign_type,
390 				&veth_link_ops, tbp);
391 	if (IS_ERR(peer)) {
392 		put_net(net);
393 		return PTR_ERR(peer);
394 	}
395 
396 	if (tbp[IFLA_ADDRESS] == NULL)
397 		eth_hw_addr_random(peer);
398 
399 	if (ifmp && (dev->ifindex != 0))
400 		peer->ifindex = ifmp->ifi_index;
401 
402 	err = register_netdevice(peer);
403 	put_net(net);
404 	net = NULL;
405 	if (err < 0)
406 		goto err_register_peer;
407 
408 	netif_carrier_off(peer);
409 
410 	err = rtnl_configure_link(peer, ifmp);
411 	if (err < 0)
412 		goto err_configure_peer;
413 
414 	/*
415 	 * register dev last
416 	 *
417 	 * note, that since we've registered new device the dev's name
418 	 * should be re-allocated
419 	 */
420 
421 	if (tb[IFLA_ADDRESS] == NULL)
422 		eth_hw_addr_random(dev);
423 
424 	if (tb[IFLA_IFNAME])
425 		nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
426 	else
427 		snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
428 
429 	err = register_netdevice(dev);
430 	if (err < 0)
431 		goto err_register_dev;
432 
433 	netif_carrier_off(dev);
434 
435 	/*
436 	 * tie the deviced together
437 	 */
438 
439 	priv = netdev_priv(dev);
440 	rcu_assign_pointer(priv->peer, peer);
441 
442 	priv = netdev_priv(peer);
443 	rcu_assign_pointer(priv->peer, dev);
444 	return 0;
445 
446 err_register_dev:
447 	/* nothing to do */
448 err_configure_peer:
449 	unregister_netdevice(peer);
450 	return err;
451 
452 err_register_peer:
453 	free_netdev(peer);
454 	return err;
455 }
456 
457 static void veth_dellink(struct net_device *dev, struct list_head *head)
458 {
459 	struct veth_priv *priv;
460 	struct net_device *peer;
461 
462 	priv = netdev_priv(dev);
463 	peer = rtnl_dereference(priv->peer);
464 
465 	/* Note : dellink() is called from default_device_exit_batch(),
466 	 * before a rcu_synchronize() point. The devices are guaranteed
467 	 * not being freed before one RCU grace period.
468 	 */
469 	RCU_INIT_POINTER(priv->peer, NULL);
470 	unregister_netdevice_queue(dev, head);
471 
472 	if (peer) {
473 		priv = netdev_priv(peer);
474 		RCU_INIT_POINTER(priv->peer, NULL);
475 		unregister_netdevice_queue(peer, head);
476 	}
477 }
478 
479 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
480 	[VETH_INFO_PEER]	= { .len = sizeof(struct ifinfomsg) },
481 };
482 
483 static struct net *veth_get_link_net(const struct net_device *dev)
484 {
485 	struct veth_priv *priv = netdev_priv(dev);
486 	struct net_device *peer = rtnl_dereference(priv->peer);
487 
488 	return peer ? dev_net(peer) : dev_net(dev);
489 }
490 
491 static struct rtnl_link_ops veth_link_ops = {
492 	.kind		= DRV_NAME,
493 	.priv_size	= sizeof(struct veth_priv),
494 	.setup		= veth_setup,
495 	.validate	= veth_validate,
496 	.newlink	= veth_newlink,
497 	.dellink	= veth_dellink,
498 	.policy		= veth_policy,
499 	.maxtype	= VETH_INFO_MAX,
500 	.get_link_net	= veth_get_link_net,
501 };
502 
503 /*
504  * init/fini
505  */
506 
507 static __init int veth_init(void)
508 {
509 	return rtnl_link_register(&veth_link_ops);
510 }
511 
512 static __exit void veth_exit(void)
513 {
514 	rtnl_link_unregister(&veth_link_ops);
515 }
516 
517 module_init(veth_init);
518 module_exit(veth_exit);
519 
520 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
521 MODULE_LICENSE("GPL v2");
522 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
523