xref: /openbmc/linux/drivers/net/veth.c (revision bb0eb050)
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 struct pcpu_vstats {
27 	u64			packets;
28 	u64			bytes;
29 	struct u64_stats_sync	syncp;
30 };
31 
32 struct veth_priv {
33 	struct net_device __rcu	*peer;
34 	atomic64_t		dropped;
35 	unsigned		requested_headroom;
36 };
37 
38 /*
39  * ethtool interface
40  */
41 
42 static struct {
43 	const char string[ETH_GSTRING_LEN];
44 } ethtool_stats_keys[] = {
45 	{ "peer_ifindex" },
46 };
47 
48 static int veth_get_link_ksettings(struct net_device *dev,
49 				   struct ethtool_link_ksettings *cmd)
50 {
51 	cmd->base.speed		= SPEED_10000;
52 	cmd->base.duplex	= DUPLEX_FULL;
53 	cmd->base.port		= PORT_TP;
54 	cmd->base.autoneg	= AUTONEG_DISABLE;
55 	return 0;
56 }
57 
58 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
59 {
60 	strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
61 	strlcpy(info->version, DRV_VERSION, sizeof(info->version));
62 }
63 
64 static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
65 {
66 	switch(stringset) {
67 	case ETH_SS_STATS:
68 		memcpy(buf, &ethtool_stats_keys, sizeof(ethtool_stats_keys));
69 		break;
70 	}
71 }
72 
73 static int veth_get_sset_count(struct net_device *dev, int sset)
74 {
75 	switch (sset) {
76 	case ETH_SS_STATS:
77 		return ARRAY_SIZE(ethtool_stats_keys);
78 	default:
79 		return -EOPNOTSUPP;
80 	}
81 }
82 
83 static void veth_get_ethtool_stats(struct net_device *dev,
84 		struct ethtool_stats *stats, u64 *data)
85 {
86 	struct veth_priv *priv = netdev_priv(dev);
87 	struct net_device *peer = rtnl_dereference(priv->peer);
88 
89 	data[0] = peer ? peer->ifindex : 0;
90 }
91 
92 static const struct ethtool_ops veth_ethtool_ops = {
93 	.get_drvinfo		= veth_get_drvinfo,
94 	.get_link		= ethtool_op_get_link,
95 	.get_strings		= veth_get_strings,
96 	.get_sset_count		= veth_get_sset_count,
97 	.get_ethtool_stats	= veth_get_ethtool_stats,
98 	.get_link_ksettings	= veth_get_link_ksettings,
99 };
100 
101 static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
102 {
103 	struct veth_priv *priv = netdev_priv(dev);
104 	struct net_device *rcv;
105 	int length = skb->len;
106 
107 	rcu_read_lock();
108 	rcv = rcu_dereference(priv->peer);
109 	if (unlikely(!rcv)) {
110 		kfree_skb(skb);
111 		goto drop;
112 	}
113 
114 	if (likely(dev_forward_skb(rcv, skb) == NET_RX_SUCCESS)) {
115 		struct pcpu_vstats *stats = this_cpu_ptr(dev->vstats);
116 
117 		u64_stats_update_begin(&stats->syncp);
118 		stats->bytes += length;
119 		stats->packets++;
120 		u64_stats_update_end(&stats->syncp);
121 	} else {
122 drop:
123 		atomic64_inc(&priv->dropped);
124 	}
125 	rcu_read_unlock();
126 	return NETDEV_TX_OK;
127 }
128 
129 /*
130  * general routines
131  */
132 
133 static u64 veth_stats_one(struct pcpu_vstats *result, struct net_device *dev)
134 {
135 	struct veth_priv *priv = netdev_priv(dev);
136 	int cpu;
137 
138 	result->packets = 0;
139 	result->bytes = 0;
140 	for_each_possible_cpu(cpu) {
141 		struct pcpu_vstats *stats = per_cpu_ptr(dev->vstats, cpu);
142 		u64 packets, bytes;
143 		unsigned int start;
144 
145 		do {
146 			start = u64_stats_fetch_begin_irq(&stats->syncp);
147 			packets = stats->packets;
148 			bytes = stats->bytes;
149 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
150 		result->packets += packets;
151 		result->bytes += bytes;
152 	}
153 	return atomic64_read(&priv->dropped);
154 }
155 
156 static void veth_get_stats64(struct net_device *dev,
157 			     struct rtnl_link_stats64 *tot)
158 {
159 	struct veth_priv *priv = netdev_priv(dev);
160 	struct net_device *peer;
161 	struct pcpu_vstats one;
162 
163 	tot->tx_dropped = veth_stats_one(&one, dev);
164 	tot->tx_bytes = one.bytes;
165 	tot->tx_packets = one.packets;
166 
167 	rcu_read_lock();
168 	peer = rcu_dereference(priv->peer);
169 	if (peer) {
170 		tot->rx_dropped = veth_stats_one(&one, peer);
171 		tot->rx_bytes = one.bytes;
172 		tot->rx_packets = one.packets;
173 	}
174 	rcu_read_unlock();
175 }
176 
177 /* fake multicast ability */
178 static void veth_set_multicast_list(struct net_device *dev)
179 {
180 }
181 
182 static int veth_open(struct net_device *dev)
183 {
184 	struct veth_priv *priv = netdev_priv(dev);
185 	struct net_device *peer = rtnl_dereference(priv->peer);
186 
187 	if (!peer)
188 		return -ENOTCONN;
189 
190 	if (peer->flags & IFF_UP) {
191 		netif_carrier_on(dev);
192 		netif_carrier_on(peer);
193 	}
194 	return 0;
195 }
196 
197 static int veth_close(struct net_device *dev)
198 {
199 	struct veth_priv *priv = netdev_priv(dev);
200 	struct net_device *peer = rtnl_dereference(priv->peer);
201 
202 	netif_carrier_off(dev);
203 	if (peer)
204 		netif_carrier_off(peer);
205 
206 	return 0;
207 }
208 
209 static int is_valid_veth_mtu(int mtu)
210 {
211 	return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU;
212 }
213 
214 static int veth_dev_init(struct net_device *dev)
215 {
216 	dev->vstats = netdev_alloc_pcpu_stats(struct pcpu_vstats);
217 	if (!dev->vstats)
218 		return -ENOMEM;
219 	return 0;
220 }
221 
222 static void veth_dev_free(struct net_device *dev)
223 {
224 	free_percpu(dev->vstats);
225 	free_netdev(dev);
226 }
227 
228 #ifdef CONFIG_NET_POLL_CONTROLLER
229 static void veth_poll_controller(struct net_device *dev)
230 {
231 	/* veth only receives frames when its peer sends one
232 	 * Since it's a synchronous operation, we are guaranteed
233 	 * never to have pending data when we poll for it so
234 	 * there is nothing to do here.
235 	 *
236 	 * We need this though so netpoll recognizes us as an interface that
237 	 * supports polling, which enables bridge devices in virt setups to
238 	 * still use netconsole
239 	 */
240 }
241 #endif	/* CONFIG_NET_POLL_CONTROLLER */
242 
243 static int veth_get_iflink(const struct net_device *dev)
244 {
245 	struct veth_priv *priv = netdev_priv(dev);
246 	struct net_device *peer;
247 	int iflink;
248 
249 	rcu_read_lock();
250 	peer = rcu_dereference(priv->peer);
251 	iflink = peer ? peer->ifindex : 0;
252 	rcu_read_unlock();
253 
254 	return iflink;
255 }
256 
257 static void veth_set_rx_headroom(struct net_device *dev, int new_hr)
258 {
259 	struct veth_priv *peer_priv, *priv = netdev_priv(dev);
260 	struct net_device *peer;
261 
262 	if (new_hr < 0)
263 		new_hr = 0;
264 
265 	rcu_read_lock();
266 	peer = rcu_dereference(priv->peer);
267 	if (unlikely(!peer))
268 		goto out;
269 
270 	peer_priv = netdev_priv(peer);
271 	priv->requested_headroom = new_hr;
272 	new_hr = max(priv->requested_headroom, peer_priv->requested_headroom);
273 	dev->needed_headroom = new_hr;
274 	peer->needed_headroom = new_hr;
275 
276 out:
277 	rcu_read_unlock();
278 }
279 
280 static const struct net_device_ops veth_netdev_ops = {
281 	.ndo_init            = veth_dev_init,
282 	.ndo_open            = veth_open,
283 	.ndo_stop            = veth_close,
284 	.ndo_start_xmit      = veth_xmit,
285 	.ndo_get_stats64     = veth_get_stats64,
286 	.ndo_set_rx_mode     = veth_set_multicast_list,
287 	.ndo_set_mac_address = eth_mac_addr,
288 #ifdef CONFIG_NET_POLL_CONTROLLER
289 	.ndo_poll_controller	= veth_poll_controller,
290 #endif
291 	.ndo_get_iflink		= veth_get_iflink,
292 	.ndo_features_check	= passthru_features_check,
293 	.ndo_set_rx_headroom	= veth_set_rx_headroom,
294 };
295 
296 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \
297 		       NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \
298 		       NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \
299 		       NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \
300 		       NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX )
301 
302 static void veth_setup(struct net_device *dev)
303 {
304 	ether_setup(dev);
305 
306 	dev->priv_flags &= ~IFF_TX_SKB_SHARING;
307 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
308 	dev->priv_flags |= IFF_NO_QUEUE;
309 	dev->priv_flags |= IFF_PHONY_HEADROOM;
310 
311 	dev->netdev_ops = &veth_netdev_ops;
312 	dev->ethtool_ops = &veth_ethtool_ops;
313 	dev->features |= NETIF_F_LLTX;
314 	dev->features |= VETH_FEATURES;
315 	dev->vlan_features = dev->features &
316 			     ~(NETIF_F_HW_VLAN_CTAG_TX |
317 			       NETIF_F_HW_VLAN_STAG_TX |
318 			       NETIF_F_HW_VLAN_CTAG_RX |
319 			       NETIF_F_HW_VLAN_STAG_RX);
320 	dev->destructor = veth_dev_free;
321 	dev->max_mtu = ETH_MAX_MTU;
322 
323 	dev->hw_features = VETH_FEATURES;
324 	dev->hw_enc_features = VETH_FEATURES;
325 	dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE;
326 }
327 
328 /*
329  * netlink interface
330  */
331 
332 static int veth_validate(struct nlattr *tb[], struct nlattr *data[])
333 {
334 	if (tb[IFLA_ADDRESS]) {
335 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
336 			return -EINVAL;
337 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
338 			return -EADDRNOTAVAIL;
339 	}
340 	if (tb[IFLA_MTU]) {
341 		if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU])))
342 			return -EINVAL;
343 	}
344 	return 0;
345 }
346 
347 static struct rtnl_link_ops veth_link_ops;
348 
349 static int veth_newlink(struct net *src_net, struct net_device *dev,
350 			 struct nlattr *tb[], struct nlattr *data[])
351 {
352 	int err;
353 	struct net_device *peer;
354 	struct veth_priv *priv;
355 	char ifname[IFNAMSIZ];
356 	struct nlattr *peer_tb[IFLA_MAX + 1], **tbp;
357 	unsigned char name_assign_type;
358 	struct ifinfomsg *ifmp;
359 	struct net *net;
360 
361 	/*
362 	 * create and register peer first
363 	 */
364 	if (data != NULL && data[VETH_INFO_PEER] != NULL) {
365 		struct nlattr *nla_peer;
366 
367 		nla_peer = data[VETH_INFO_PEER];
368 		ifmp = nla_data(nla_peer);
369 		err = rtnl_nla_parse_ifla(peer_tb,
370 					  nla_data(nla_peer) + sizeof(struct ifinfomsg),
371 					  nla_len(nla_peer) - sizeof(struct ifinfomsg),
372 					  NULL);
373 		if (err < 0)
374 			return err;
375 
376 		err = veth_validate(peer_tb, NULL);
377 		if (err < 0)
378 			return err;
379 
380 		tbp = peer_tb;
381 	} else {
382 		ifmp = NULL;
383 		tbp = tb;
384 	}
385 
386 	if (tbp[IFLA_IFNAME]) {
387 		nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ);
388 		name_assign_type = NET_NAME_USER;
389 	} else {
390 		snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d");
391 		name_assign_type = NET_NAME_ENUM;
392 	}
393 
394 	net = rtnl_link_get_net(src_net, tbp);
395 	if (IS_ERR(net))
396 		return PTR_ERR(net);
397 
398 	peer = rtnl_create_link(net, ifname, name_assign_type,
399 				&veth_link_ops, tbp);
400 	if (IS_ERR(peer)) {
401 		put_net(net);
402 		return PTR_ERR(peer);
403 	}
404 
405 	if (tbp[IFLA_ADDRESS] == NULL)
406 		eth_hw_addr_random(peer);
407 
408 	if (ifmp && (dev->ifindex != 0))
409 		peer->ifindex = ifmp->ifi_index;
410 
411 	err = register_netdevice(peer);
412 	put_net(net);
413 	net = NULL;
414 	if (err < 0)
415 		goto err_register_peer;
416 
417 	netif_carrier_off(peer);
418 
419 	err = rtnl_configure_link(peer, ifmp);
420 	if (err < 0)
421 		goto err_configure_peer;
422 
423 	/*
424 	 * register dev last
425 	 *
426 	 * note, that since we've registered new device the dev's name
427 	 * should be re-allocated
428 	 */
429 
430 	if (tb[IFLA_ADDRESS] == NULL)
431 		eth_hw_addr_random(dev);
432 
433 	if (tb[IFLA_IFNAME])
434 		nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ);
435 	else
436 		snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d");
437 
438 	err = register_netdevice(dev);
439 	if (err < 0)
440 		goto err_register_dev;
441 
442 	netif_carrier_off(dev);
443 
444 	/*
445 	 * tie the deviced together
446 	 */
447 
448 	priv = netdev_priv(dev);
449 	rcu_assign_pointer(priv->peer, peer);
450 
451 	priv = netdev_priv(peer);
452 	rcu_assign_pointer(priv->peer, dev);
453 	return 0;
454 
455 err_register_dev:
456 	/* nothing to do */
457 err_configure_peer:
458 	unregister_netdevice(peer);
459 	return err;
460 
461 err_register_peer:
462 	free_netdev(peer);
463 	return err;
464 }
465 
466 static void veth_dellink(struct net_device *dev, struct list_head *head)
467 {
468 	struct veth_priv *priv;
469 	struct net_device *peer;
470 
471 	priv = netdev_priv(dev);
472 	peer = rtnl_dereference(priv->peer);
473 
474 	/* Note : dellink() is called from default_device_exit_batch(),
475 	 * before a rcu_synchronize() point. The devices are guaranteed
476 	 * not being freed before one RCU grace period.
477 	 */
478 	RCU_INIT_POINTER(priv->peer, NULL);
479 	unregister_netdevice_queue(dev, head);
480 
481 	if (peer) {
482 		priv = netdev_priv(peer);
483 		RCU_INIT_POINTER(priv->peer, NULL);
484 		unregister_netdevice_queue(peer, head);
485 	}
486 }
487 
488 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = {
489 	[VETH_INFO_PEER]	= { .len = sizeof(struct ifinfomsg) },
490 };
491 
492 static struct net *veth_get_link_net(const struct net_device *dev)
493 {
494 	struct veth_priv *priv = netdev_priv(dev);
495 	struct net_device *peer = rtnl_dereference(priv->peer);
496 
497 	return peer ? dev_net(peer) : dev_net(dev);
498 }
499 
500 static struct rtnl_link_ops veth_link_ops = {
501 	.kind		= DRV_NAME,
502 	.priv_size	= sizeof(struct veth_priv),
503 	.setup		= veth_setup,
504 	.validate	= veth_validate,
505 	.newlink	= veth_newlink,
506 	.dellink	= veth_dellink,
507 	.policy		= veth_policy,
508 	.maxtype	= VETH_INFO_MAX,
509 	.get_link_net	= veth_get_link_net,
510 };
511 
512 /*
513  * init/fini
514  */
515 
516 static __init int veth_init(void)
517 {
518 	return rtnl_link_register(&veth_link_ops);
519 }
520 
521 static __exit void veth_exit(void)
522 {
523 	rtnl_link_unregister(&veth_link_ops);
524 }
525 
526 module_init(veth_init);
527 module_exit(veth_exit);
528 
529 MODULE_DESCRIPTION("Virtual Ethernet Tunnel");
530 MODULE_LICENSE("GPL v2");
531 MODULE_ALIAS_RTNL_LINK(DRV_NAME);
532