xref: /openbmc/linux/drivers/net/ipvlan/ipvlan_main.c (revision b78412b8)
1 /* Copyright (c) 2014 Mahesh Bandewar <maheshb@google.com>
2  *
3  * This program is free software; you can redistribute it and/or
4  * modify it under the terms of the GNU General Public License as
5  * published by the Free Software Foundation; either version 2 of
6  * the License, or (at your option) any later version.
7  *
8  */
9 
10 #include "ipvlan.h"
11 
12 static unsigned int ipvlan_netid __read_mostly;
13 
14 struct ipvlan_netns {
15 	unsigned int ipvl_nf_hook_refcnt;
16 };
17 
18 static const struct nf_hook_ops ipvl_nfops[] = {
19 	{
20 		.hook     = ipvlan_nf_input,
21 		.pf       = NFPROTO_IPV4,
22 		.hooknum  = NF_INET_LOCAL_IN,
23 		.priority = INT_MAX,
24 	},
25 	{
26 		.hook     = ipvlan_nf_input,
27 		.pf       = NFPROTO_IPV6,
28 		.hooknum  = NF_INET_LOCAL_IN,
29 		.priority = INT_MAX,
30 	},
31 };
32 
33 static const struct l3mdev_ops ipvl_l3mdev_ops = {
34 	.l3mdev_l3_rcv = ipvlan_l3_rcv,
35 };
36 
37 static void ipvlan_adjust_mtu(struct ipvl_dev *ipvlan, struct net_device *dev)
38 {
39 	ipvlan->dev->mtu = dev->mtu;
40 }
41 
42 static int ipvlan_register_nf_hook(struct net *net)
43 {
44 	struct ipvlan_netns *vnet = net_generic(net, ipvlan_netid);
45 	int err = 0;
46 
47 	if (!vnet->ipvl_nf_hook_refcnt) {
48 		err = nf_register_net_hooks(net, ipvl_nfops,
49 					    ARRAY_SIZE(ipvl_nfops));
50 		if (!err)
51 			vnet->ipvl_nf_hook_refcnt = 1;
52 	} else {
53 		vnet->ipvl_nf_hook_refcnt++;
54 	}
55 
56 	return err;
57 }
58 
59 static void ipvlan_unregister_nf_hook(struct net *net)
60 {
61 	struct ipvlan_netns *vnet = net_generic(net, ipvlan_netid);
62 
63 	if (WARN_ON(!vnet->ipvl_nf_hook_refcnt))
64 		return;
65 
66 	vnet->ipvl_nf_hook_refcnt--;
67 	if (!vnet->ipvl_nf_hook_refcnt)
68 		nf_unregister_net_hooks(net, ipvl_nfops,
69 					ARRAY_SIZE(ipvl_nfops));
70 }
71 
72 static int ipvlan_set_port_mode(struct ipvl_port *port, u16 nval)
73 {
74 	struct ipvl_dev *ipvlan;
75 	struct net_device *mdev = port->dev;
76 	int err = 0;
77 
78 	ASSERT_RTNL();
79 	if (port->mode != nval) {
80 		if (nval == IPVLAN_MODE_L3S) {
81 			/* New mode is L3S */
82 			err = ipvlan_register_nf_hook(read_pnet(&port->pnet));
83 			if (!err) {
84 				mdev->l3mdev_ops = &ipvl_l3mdev_ops;
85 				mdev->priv_flags |= IFF_L3MDEV_MASTER;
86 			} else
87 				return err;
88 		} else if (port->mode == IPVLAN_MODE_L3S) {
89 			/* Old mode was L3S */
90 			mdev->priv_flags &= ~IFF_L3MDEV_MASTER;
91 			ipvlan_unregister_nf_hook(read_pnet(&port->pnet));
92 			mdev->l3mdev_ops = NULL;
93 		}
94 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
95 			if (nval == IPVLAN_MODE_L3 || nval == IPVLAN_MODE_L3S)
96 				ipvlan->dev->flags |= IFF_NOARP;
97 			else
98 				ipvlan->dev->flags &= ~IFF_NOARP;
99 		}
100 		port->mode = nval;
101 	}
102 	return err;
103 }
104 
105 static int ipvlan_port_create(struct net_device *dev)
106 {
107 	struct ipvl_port *port;
108 	int err, idx;
109 
110 	if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK) {
111 		netdev_err(dev, "Master is either lo or non-ether device\n");
112 		return -EINVAL;
113 	}
114 
115 	if (netdev_is_rx_handler_busy(dev)) {
116 		netdev_err(dev, "Device is already in use.\n");
117 		return -EBUSY;
118 	}
119 
120 	port = kzalloc(sizeof(struct ipvl_port), GFP_KERNEL);
121 	if (!port)
122 		return -ENOMEM;
123 
124 	write_pnet(&port->pnet, dev_net(dev));
125 	port->dev = dev;
126 	port->mode = IPVLAN_MODE_L3;
127 	INIT_LIST_HEAD(&port->ipvlans);
128 	for (idx = 0; idx < IPVLAN_HASH_SIZE; idx++)
129 		INIT_HLIST_HEAD(&port->hlhead[idx]);
130 
131 	skb_queue_head_init(&port->backlog);
132 	INIT_WORK(&port->wq, ipvlan_process_multicast);
133 	ida_init(&port->ida);
134 	port->dev_id_start = 1;
135 
136 	err = netdev_rx_handler_register(dev, ipvlan_handle_frame, port);
137 	if (err)
138 		goto err;
139 
140 	dev->priv_flags |= IFF_IPVLAN_MASTER;
141 	return 0;
142 
143 err:
144 	kfree(port);
145 	return err;
146 }
147 
148 static void ipvlan_port_destroy(struct net_device *dev)
149 {
150 	struct ipvl_port *port = ipvlan_port_get_rtnl(dev);
151 	struct sk_buff *skb;
152 
153 	dev->priv_flags &= ~IFF_IPVLAN_MASTER;
154 	if (port->mode == IPVLAN_MODE_L3S) {
155 		dev->priv_flags &= ~IFF_L3MDEV_MASTER;
156 		ipvlan_unregister_nf_hook(dev_net(dev));
157 		dev->l3mdev_ops = NULL;
158 	}
159 	netdev_rx_handler_unregister(dev);
160 	cancel_work_sync(&port->wq);
161 	while ((skb = __skb_dequeue(&port->backlog)) != NULL) {
162 		if (skb->dev)
163 			dev_put(skb->dev);
164 		kfree_skb(skb);
165 	}
166 	ida_destroy(&port->ida);
167 	kfree(port);
168 }
169 
170 #define IPVLAN_FEATURES \
171 	(NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
172 	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_GSO_ROBUST | \
173 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
174 	 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
175 
176 #define IPVLAN_STATE_MASK \
177 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
178 
179 static int ipvlan_init(struct net_device *dev)
180 {
181 	struct ipvl_dev *ipvlan = netdev_priv(dev);
182 	const struct net_device *phy_dev = ipvlan->phy_dev;
183 	struct ipvl_port *port = ipvlan->port;
184 
185 	dev->state = (dev->state & ~IPVLAN_STATE_MASK) |
186 		     (phy_dev->state & IPVLAN_STATE_MASK);
187 	dev->features = phy_dev->features & IPVLAN_FEATURES;
188 	dev->features |= NETIF_F_LLTX;
189 	dev->gso_max_size = phy_dev->gso_max_size;
190 	dev->gso_max_segs = phy_dev->gso_max_segs;
191 	dev->hard_header_len = phy_dev->hard_header_len;
192 
193 	netdev_lockdep_set_classes(dev);
194 
195 	ipvlan->pcpu_stats = netdev_alloc_pcpu_stats(struct ipvl_pcpu_stats);
196 	if (!ipvlan->pcpu_stats)
197 		return -ENOMEM;
198 
199 	port->count += 1;
200 
201 	return 0;
202 }
203 
204 static void ipvlan_uninit(struct net_device *dev)
205 {
206 	struct ipvl_dev *ipvlan = netdev_priv(dev);
207 	struct ipvl_port *port = ipvlan->port;
208 
209 	free_percpu(ipvlan->pcpu_stats);
210 
211 	port->count -= 1;
212 	if (!port->count)
213 		ipvlan_port_destroy(port->dev);
214 }
215 
216 static int ipvlan_open(struct net_device *dev)
217 {
218 	struct ipvl_dev *ipvlan = netdev_priv(dev);
219 	struct net_device *phy_dev = ipvlan->phy_dev;
220 	struct ipvl_addr *addr;
221 
222 	if (ipvlan->port->mode == IPVLAN_MODE_L3 ||
223 	    ipvlan->port->mode == IPVLAN_MODE_L3S)
224 		dev->flags |= IFF_NOARP;
225 	else
226 		dev->flags &= ~IFF_NOARP;
227 
228 	list_for_each_entry(addr, &ipvlan->addrs, anode)
229 		ipvlan_ht_addr_add(ipvlan, addr);
230 
231 	return dev_uc_add(phy_dev, phy_dev->dev_addr);
232 }
233 
234 static int ipvlan_stop(struct net_device *dev)
235 {
236 	struct ipvl_dev *ipvlan = netdev_priv(dev);
237 	struct net_device *phy_dev = ipvlan->phy_dev;
238 	struct ipvl_addr *addr;
239 
240 	dev_uc_unsync(phy_dev, dev);
241 	dev_mc_unsync(phy_dev, dev);
242 
243 	dev_uc_del(phy_dev, phy_dev->dev_addr);
244 
245 	list_for_each_entry(addr, &ipvlan->addrs, anode)
246 		ipvlan_ht_addr_del(addr);
247 
248 	return 0;
249 }
250 
251 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
252 				     struct net_device *dev)
253 {
254 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
255 	int skblen = skb->len;
256 	int ret;
257 
258 	ret = ipvlan_queue_xmit(skb, dev);
259 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
260 		struct ipvl_pcpu_stats *pcptr;
261 
262 		pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
263 
264 		u64_stats_update_begin(&pcptr->syncp);
265 		pcptr->tx_pkts++;
266 		pcptr->tx_bytes += skblen;
267 		u64_stats_update_end(&pcptr->syncp);
268 	} else {
269 		this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
270 	}
271 	return ret;
272 }
273 
274 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
275 					     netdev_features_t features)
276 {
277 	struct ipvl_dev *ipvlan = netdev_priv(dev);
278 
279 	return features & (ipvlan->sfeatures | ~IPVLAN_FEATURES);
280 }
281 
282 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
283 {
284 	struct ipvl_dev *ipvlan = netdev_priv(dev);
285 	struct net_device *phy_dev = ipvlan->phy_dev;
286 
287 	if (change & IFF_ALLMULTI)
288 		dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
289 }
290 
291 static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
292 {
293 	struct ipvl_dev *ipvlan = netdev_priv(dev);
294 
295 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
296 		bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
297 	} else {
298 		struct netdev_hw_addr *ha;
299 		DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
300 
301 		bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
302 		netdev_for_each_mc_addr(ha, dev)
303 			__set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
304 
305 		/* Turn-on broadcast bit irrespective of address family,
306 		 * since broadcast is deferred to a work-queue, hence no
307 		 * impact on fast-path processing.
308 		 */
309 		__set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
310 
311 		bitmap_copy(ipvlan->mac_filters, mc_filters,
312 			    IPVLAN_MAC_FILTER_SIZE);
313 	}
314 	dev_uc_sync(ipvlan->phy_dev, dev);
315 	dev_mc_sync(ipvlan->phy_dev, dev);
316 }
317 
318 static void ipvlan_get_stats64(struct net_device *dev,
319 			       struct rtnl_link_stats64 *s)
320 {
321 	struct ipvl_dev *ipvlan = netdev_priv(dev);
322 
323 	if (ipvlan->pcpu_stats) {
324 		struct ipvl_pcpu_stats *pcptr;
325 		u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
326 		u32 rx_errs = 0, tx_drps = 0;
327 		u32 strt;
328 		int idx;
329 
330 		for_each_possible_cpu(idx) {
331 			pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
332 			do {
333 				strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
334 				rx_pkts = pcptr->rx_pkts;
335 				rx_bytes = pcptr->rx_bytes;
336 				rx_mcast = pcptr->rx_mcast;
337 				tx_pkts = pcptr->tx_pkts;
338 				tx_bytes = pcptr->tx_bytes;
339 			} while (u64_stats_fetch_retry_irq(&pcptr->syncp,
340 							   strt));
341 
342 			s->rx_packets += rx_pkts;
343 			s->rx_bytes += rx_bytes;
344 			s->multicast += rx_mcast;
345 			s->tx_packets += tx_pkts;
346 			s->tx_bytes += tx_bytes;
347 
348 			/* u32 values are updated without syncp protection. */
349 			rx_errs += pcptr->rx_errs;
350 			tx_drps += pcptr->tx_drps;
351 		}
352 		s->rx_errors = rx_errs;
353 		s->rx_dropped = rx_errs;
354 		s->tx_dropped = tx_drps;
355 	}
356 }
357 
358 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
359 {
360 	struct ipvl_dev *ipvlan = netdev_priv(dev);
361 	struct net_device *phy_dev = ipvlan->phy_dev;
362 
363 	return vlan_vid_add(phy_dev, proto, vid);
364 }
365 
366 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
367 				   u16 vid)
368 {
369 	struct ipvl_dev *ipvlan = netdev_priv(dev);
370 	struct net_device *phy_dev = ipvlan->phy_dev;
371 
372 	vlan_vid_del(phy_dev, proto, vid);
373 	return 0;
374 }
375 
376 static int ipvlan_get_iflink(const struct net_device *dev)
377 {
378 	struct ipvl_dev *ipvlan = netdev_priv(dev);
379 
380 	return ipvlan->phy_dev->ifindex;
381 }
382 
383 static const struct net_device_ops ipvlan_netdev_ops = {
384 	.ndo_init		= ipvlan_init,
385 	.ndo_uninit		= ipvlan_uninit,
386 	.ndo_open		= ipvlan_open,
387 	.ndo_stop		= ipvlan_stop,
388 	.ndo_start_xmit		= ipvlan_start_xmit,
389 	.ndo_fix_features	= ipvlan_fix_features,
390 	.ndo_change_rx_flags	= ipvlan_change_rx_flags,
391 	.ndo_set_rx_mode	= ipvlan_set_multicast_mac_filter,
392 	.ndo_get_stats64	= ipvlan_get_stats64,
393 	.ndo_vlan_rx_add_vid	= ipvlan_vlan_rx_add_vid,
394 	.ndo_vlan_rx_kill_vid	= ipvlan_vlan_rx_kill_vid,
395 	.ndo_get_iflink		= ipvlan_get_iflink,
396 };
397 
398 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
399 			      unsigned short type, const void *daddr,
400 			      const void *saddr, unsigned len)
401 {
402 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
403 	struct net_device *phy_dev = ipvlan->phy_dev;
404 
405 	/* TODO Probably use a different field than dev_addr so that the
406 	 * mac-address on the virtual device is portable and can be carried
407 	 * while the packets use the mac-addr on the physical device.
408 	 */
409 	return dev_hard_header(skb, phy_dev, type, daddr,
410 			       saddr ? : dev->dev_addr, len);
411 }
412 
413 static const struct header_ops ipvlan_header_ops = {
414 	.create  	= ipvlan_hard_header,
415 	.parse		= eth_header_parse,
416 	.cache		= eth_header_cache,
417 	.cache_update	= eth_header_cache_update,
418 };
419 
420 static int ipvlan_ethtool_get_link_ksettings(struct net_device *dev,
421 					     struct ethtool_link_ksettings *cmd)
422 {
423 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
424 
425 	return __ethtool_get_link_ksettings(ipvlan->phy_dev, cmd);
426 }
427 
428 static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
429 				       struct ethtool_drvinfo *drvinfo)
430 {
431 	strlcpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
432 	strlcpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
433 }
434 
435 static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
436 {
437 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
438 
439 	return ipvlan->msg_enable;
440 }
441 
442 static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
443 {
444 	struct ipvl_dev *ipvlan = netdev_priv(dev);
445 
446 	ipvlan->msg_enable = value;
447 }
448 
449 static const struct ethtool_ops ipvlan_ethtool_ops = {
450 	.get_link	= ethtool_op_get_link,
451 	.get_link_ksettings	= ipvlan_ethtool_get_link_ksettings,
452 	.get_drvinfo	= ipvlan_ethtool_get_drvinfo,
453 	.get_msglevel	= ipvlan_ethtool_get_msglevel,
454 	.set_msglevel	= ipvlan_ethtool_set_msglevel,
455 };
456 
457 static int ipvlan_nl_changelink(struct net_device *dev,
458 				struct nlattr *tb[], struct nlattr *data[],
459 				struct netlink_ext_ack *extack)
460 {
461 	struct ipvl_dev *ipvlan = netdev_priv(dev);
462 	struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
463 	int err = 0;
464 
465 	if (data && data[IFLA_IPVLAN_MODE]) {
466 		u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
467 
468 		err = ipvlan_set_port_mode(port, nmode);
469 	}
470 	return err;
471 }
472 
473 static size_t ipvlan_nl_getsize(const struct net_device *dev)
474 {
475 	return (0
476 		+ nla_total_size(2) /* IFLA_IPVLAN_MODE */
477 		);
478 }
479 
480 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[],
481 			      struct netlink_ext_ack *extack)
482 {
483 	if (data && data[IFLA_IPVLAN_MODE]) {
484 		u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
485 
486 		if (mode < IPVLAN_MODE_L2 || mode >= IPVLAN_MODE_MAX)
487 			return -EINVAL;
488 	}
489 	return 0;
490 }
491 
492 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
493 			      const struct net_device *dev)
494 {
495 	struct ipvl_dev *ipvlan = netdev_priv(dev);
496 	struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
497 	int ret = -EINVAL;
498 
499 	if (!port)
500 		goto err;
501 
502 	ret = -EMSGSIZE;
503 	if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
504 		goto err;
505 
506 	return 0;
507 
508 err:
509 	return ret;
510 }
511 
512 int ipvlan_link_new(struct net *src_net, struct net_device *dev,
513 		    struct nlattr *tb[], struct nlattr *data[],
514 		    struct netlink_ext_ack *extack)
515 {
516 	struct ipvl_dev *ipvlan = netdev_priv(dev);
517 	struct ipvl_port *port;
518 	struct net_device *phy_dev;
519 	int err;
520 	u16 mode = IPVLAN_MODE_L3;
521 	bool create = false;
522 
523 	if (!tb[IFLA_LINK])
524 		return -EINVAL;
525 
526 	phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
527 	if (!phy_dev)
528 		return -ENODEV;
529 
530 	if (netif_is_ipvlan(phy_dev)) {
531 		struct ipvl_dev *tmp = netdev_priv(phy_dev);
532 
533 		phy_dev = tmp->phy_dev;
534 	} else if (!netif_is_ipvlan_port(phy_dev)) {
535 		err = ipvlan_port_create(phy_dev);
536 		if (err < 0)
537 			return err;
538 		create = true;
539 	}
540 
541 	if (data && data[IFLA_IPVLAN_MODE])
542 		mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
543 
544 	port = ipvlan_port_get_rtnl(phy_dev);
545 	ipvlan->phy_dev = phy_dev;
546 	ipvlan->dev = dev;
547 	ipvlan->port = port;
548 	ipvlan->sfeatures = IPVLAN_FEATURES;
549 	ipvlan_adjust_mtu(ipvlan, phy_dev);
550 	INIT_LIST_HEAD(&ipvlan->addrs);
551 
552 	/* If the port-id base is at the MAX value, then wrap it around and
553 	 * begin from 0x1 again. This may be due to a busy system where lots
554 	 * of slaves are getting created and deleted.
555 	 */
556 	if (port->dev_id_start == 0xFFFE)
557 		port->dev_id_start = 0x1;
558 
559 	/* Since L2 address is shared among all IPvlan slaves including
560 	 * master, use unique 16 bit dev-ids to diffentiate among them.
561 	 * Assign IDs between 0x1 and 0xFFFE (used by the master) to each
562 	 * slave link [see addrconf_ifid_eui48()].
563 	 */
564 	err = ida_simple_get(&port->ida, port->dev_id_start, 0xFFFE,
565 			     GFP_KERNEL);
566 	if (err < 0)
567 		err = ida_simple_get(&port->ida, 0x1, port->dev_id_start,
568 				     GFP_KERNEL);
569 	if (err < 0)
570 		goto destroy_ipvlan_port;
571 	dev->dev_id = err;
572 	/* Increment id-base to the next slot for the future assignment */
573 	port->dev_id_start = err + 1;
574 
575 	/* TODO Probably put random address here to be presented to the
576 	 * world but keep using the physical-dev address for the outgoing
577 	 * packets.
578 	 */
579 	memcpy(dev->dev_addr, phy_dev->dev_addr, ETH_ALEN);
580 
581 	dev->priv_flags |= IFF_IPVLAN_SLAVE;
582 
583 	err = register_netdevice(dev);
584 	if (err < 0)
585 		goto remove_ida;
586 
587 	err = netdev_upper_dev_link(phy_dev, dev);
588 	if (err) {
589 		goto unregister_netdev;
590 	}
591 	err = ipvlan_set_port_mode(port, mode);
592 	if (err) {
593 		goto unlink_netdev;
594 	}
595 
596 	list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
597 	netif_stacked_transfer_operstate(phy_dev, dev);
598 	return 0;
599 
600 unlink_netdev:
601 	netdev_upper_dev_unlink(phy_dev, dev);
602 unregister_netdev:
603 	unregister_netdevice(dev);
604 remove_ida:
605 	ida_simple_remove(&port->ida, dev->dev_id);
606 destroy_ipvlan_port:
607 	if (create)
608 		ipvlan_port_destroy(phy_dev);
609 	return err;
610 }
611 EXPORT_SYMBOL_GPL(ipvlan_link_new);
612 
613 void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
614 {
615 	struct ipvl_dev *ipvlan = netdev_priv(dev);
616 	struct ipvl_addr *addr, *next;
617 
618 	list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
619 		ipvlan_ht_addr_del(addr);
620 		list_del(&addr->anode);
621 		kfree_rcu(addr, rcu);
622 	}
623 
624 	ida_simple_remove(&ipvlan->port->ida, dev->dev_id);
625 	list_del_rcu(&ipvlan->pnode);
626 	unregister_netdevice_queue(dev, head);
627 	netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
628 }
629 EXPORT_SYMBOL_GPL(ipvlan_link_delete);
630 
631 void ipvlan_link_setup(struct net_device *dev)
632 {
633 	ether_setup(dev);
634 
635 	dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
636 	dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
637 	dev->netdev_ops = &ipvlan_netdev_ops;
638 	dev->needs_free_netdev = true;
639 	dev->header_ops = &ipvlan_header_ops;
640 	dev->ethtool_ops = &ipvlan_ethtool_ops;
641 }
642 EXPORT_SYMBOL_GPL(ipvlan_link_setup);
643 
644 static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
645 {
646 	[IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
647 };
648 
649 static struct rtnl_link_ops ipvlan_link_ops = {
650 	.kind		= "ipvlan",
651 	.priv_size	= sizeof(struct ipvl_dev),
652 
653 	.setup		= ipvlan_link_setup,
654 	.newlink	= ipvlan_link_new,
655 	.dellink	= ipvlan_link_delete,
656 };
657 
658 int ipvlan_link_register(struct rtnl_link_ops *ops)
659 {
660 	ops->get_size	= ipvlan_nl_getsize;
661 	ops->policy	= ipvlan_nl_policy;
662 	ops->validate	= ipvlan_nl_validate;
663 	ops->fill_info	= ipvlan_nl_fillinfo;
664 	ops->changelink = ipvlan_nl_changelink;
665 	ops->maxtype	= IFLA_IPVLAN_MAX;
666 	return rtnl_link_register(ops);
667 }
668 EXPORT_SYMBOL_GPL(ipvlan_link_register);
669 
670 static int ipvlan_device_event(struct notifier_block *unused,
671 			       unsigned long event, void *ptr)
672 {
673 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
674 	struct ipvl_dev *ipvlan, *next;
675 	struct ipvl_port *port;
676 	LIST_HEAD(lst_kill);
677 
678 	if (!netif_is_ipvlan_port(dev))
679 		return NOTIFY_DONE;
680 
681 	port = ipvlan_port_get_rtnl(dev);
682 
683 	switch (event) {
684 	case NETDEV_CHANGE:
685 		list_for_each_entry(ipvlan, &port->ipvlans, pnode)
686 			netif_stacked_transfer_operstate(ipvlan->phy_dev,
687 							 ipvlan->dev);
688 		break;
689 
690 	case NETDEV_REGISTER: {
691 		struct net *oldnet, *newnet = dev_net(dev);
692 		struct ipvlan_netns *old_vnet;
693 
694 		oldnet = read_pnet(&port->pnet);
695 		if (net_eq(newnet, oldnet))
696 			break;
697 
698 		write_pnet(&port->pnet, newnet);
699 
700 		old_vnet = net_generic(oldnet, ipvlan_netid);
701 		if (!old_vnet->ipvl_nf_hook_refcnt)
702 			break;
703 
704 		ipvlan_register_nf_hook(newnet);
705 		ipvlan_unregister_nf_hook(oldnet);
706 		break;
707 	}
708 	case NETDEV_UNREGISTER:
709 		if (dev->reg_state != NETREG_UNREGISTERING)
710 			break;
711 
712 		list_for_each_entry_safe(ipvlan, next, &port->ipvlans,
713 					 pnode)
714 			ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
715 							    &lst_kill);
716 		unregister_netdevice_many(&lst_kill);
717 		break;
718 
719 	case NETDEV_FEAT_CHANGE:
720 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
721 			ipvlan->dev->features = dev->features & IPVLAN_FEATURES;
722 			ipvlan->dev->gso_max_size = dev->gso_max_size;
723 			ipvlan->dev->gso_max_segs = dev->gso_max_segs;
724 			netdev_features_change(ipvlan->dev);
725 		}
726 		break;
727 
728 	case NETDEV_CHANGEMTU:
729 		list_for_each_entry(ipvlan, &port->ipvlans, pnode)
730 			ipvlan_adjust_mtu(ipvlan, dev);
731 		break;
732 
733 	case NETDEV_PRE_TYPE_CHANGE:
734 		/* Forbid underlying device to change its type. */
735 		return NOTIFY_BAD;
736 	}
737 	return NOTIFY_DONE;
738 }
739 
740 static int ipvlan_add_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
741 {
742 	struct ipvl_addr *addr;
743 
744 	addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
745 	if (!addr)
746 		return -ENOMEM;
747 
748 	addr->master = ipvlan;
749 	if (is_v6) {
750 		memcpy(&addr->ip6addr, iaddr, sizeof(struct in6_addr));
751 		addr->atype = IPVL_IPV6;
752 	} else {
753 		memcpy(&addr->ip4addr, iaddr, sizeof(struct in_addr));
754 		addr->atype = IPVL_IPV4;
755 	}
756 	list_add_tail(&addr->anode, &ipvlan->addrs);
757 
758 	/* If the interface is not up, the address will be added to the hash
759 	 * list by ipvlan_open.
760 	 */
761 	if (netif_running(ipvlan->dev))
762 		ipvlan_ht_addr_add(ipvlan, addr);
763 
764 	return 0;
765 }
766 
767 static void ipvlan_del_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
768 {
769 	struct ipvl_addr *addr;
770 
771 	addr = ipvlan_find_addr(ipvlan, iaddr, is_v6);
772 	if (!addr)
773 		return;
774 
775 	ipvlan_ht_addr_del(addr);
776 	list_del(&addr->anode);
777 	kfree_rcu(addr, rcu);
778 
779 	return;
780 }
781 
782 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
783 {
784 	if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true)) {
785 		netif_err(ipvlan, ifup, ipvlan->dev,
786 			  "Failed to add IPv6=%pI6c addr for %s intf\n",
787 			  ip6_addr, ipvlan->dev->name);
788 		return -EINVAL;
789 	}
790 
791 	return ipvlan_add_addr(ipvlan, ip6_addr, true);
792 }
793 
794 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
795 {
796 	return ipvlan_del_addr(ipvlan, ip6_addr, true);
797 }
798 
799 static int ipvlan_addr6_event(struct notifier_block *unused,
800 			      unsigned long event, void *ptr)
801 {
802 	struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
803 	struct net_device *dev = (struct net_device *)if6->idev->dev;
804 	struct ipvl_dev *ipvlan = netdev_priv(dev);
805 
806 	/* FIXME IPv6 autoconf calls us from bh without RTNL */
807 	if (in_softirq())
808 		return NOTIFY_DONE;
809 
810 	if (!netif_is_ipvlan(dev))
811 		return NOTIFY_DONE;
812 
813 	if (!ipvlan || !ipvlan->port)
814 		return NOTIFY_DONE;
815 
816 	switch (event) {
817 	case NETDEV_UP:
818 		if (ipvlan_add_addr6(ipvlan, &if6->addr))
819 			return NOTIFY_BAD;
820 		break;
821 
822 	case NETDEV_DOWN:
823 		ipvlan_del_addr6(ipvlan, &if6->addr);
824 		break;
825 	}
826 
827 	return NOTIFY_OK;
828 }
829 
830 static int ipvlan_addr6_validator_event(struct notifier_block *unused,
831 					unsigned long event, void *ptr)
832 {
833 	struct in6_validator_info *i6vi = (struct in6_validator_info *)ptr;
834 	struct net_device *dev = (struct net_device *)i6vi->i6vi_dev->dev;
835 	struct ipvl_dev *ipvlan = netdev_priv(dev);
836 
837 	/* FIXME IPv6 autoconf calls us from bh without RTNL */
838 	if (in_softirq())
839 		return NOTIFY_DONE;
840 
841 	if (!netif_is_ipvlan(dev))
842 		return NOTIFY_DONE;
843 
844 	if (!ipvlan || !ipvlan->port)
845 		return NOTIFY_DONE;
846 
847 	switch (event) {
848 	case NETDEV_UP:
849 		if (ipvlan_addr_busy(ipvlan->port, &i6vi->i6vi_addr, true))
850 			return notifier_from_errno(-EADDRINUSE);
851 		break;
852 	}
853 
854 	return NOTIFY_OK;
855 }
856 
857 static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
858 {
859 	if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false)) {
860 		netif_err(ipvlan, ifup, ipvlan->dev,
861 			  "Failed to add IPv4=%pI4 on %s intf.\n",
862 			  ip4_addr, ipvlan->dev->name);
863 		return -EINVAL;
864 	}
865 
866 	return ipvlan_add_addr(ipvlan, ip4_addr, false);
867 }
868 
869 static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
870 {
871 	return ipvlan_del_addr(ipvlan, ip4_addr, false);
872 }
873 
874 static int ipvlan_addr4_event(struct notifier_block *unused,
875 			      unsigned long event, void *ptr)
876 {
877 	struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
878 	struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
879 	struct ipvl_dev *ipvlan = netdev_priv(dev);
880 	struct in_addr ip4_addr;
881 
882 	if (!netif_is_ipvlan(dev))
883 		return NOTIFY_DONE;
884 
885 	if (!ipvlan || !ipvlan->port)
886 		return NOTIFY_DONE;
887 
888 	switch (event) {
889 	case NETDEV_UP:
890 		ip4_addr.s_addr = if4->ifa_address;
891 		if (ipvlan_add_addr4(ipvlan, &ip4_addr))
892 			return NOTIFY_BAD;
893 		break;
894 
895 	case NETDEV_DOWN:
896 		ip4_addr.s_addr = if4->ifa_address;
897 		ipvlan_del_addr4(ipvlan, &ip4_addr);
898 		break;
899 	}
900 
901 	return NOTIFY_OK;
902 }
903 
904 static int ipvlan_addr4_validator_event(struct notifier_block *unused,
905 					unsigned long event, void *ptr)
906 {
907 	struct in_validator_info *ivi = (struct in_validator_info *)ptr;
908 	struct net_device *dev = (struct net_device *)ivi->ivi_dev->dev;
909 	struct ipvl_dev *ipvlan = netdev_priv(dev);
910 
911 	if (!netif_is_ipvlan(dev))
912 		return NOTIFY_DONE;
913 
914 	if (!ipvlan || !ipvlan->port)
915 		return NOTIFY_DONE;
916 
917 	switch (event) {
918 	case NETDEV_UP:
919 		if (ipvlan_addr_busy(ipvlan->port, &ivi->ivi_addr, false))
920 			return notifier_from_errno(-EADDRINUSE);
921 		break;
922 	}
923 
924 	return NOTIFY_OK;
925 }
926 
927 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
928 	.notifier_call = ipvlan_addr4_event,
929 };
930 
931 static struct notifier_block ipvlan_addr4_vtor_notifier_block __read_mostly = {
932 	.notifier_call = ipvlan_addr4_validator_event,
933 };
934 
935 static struct notifier_block ipvlan_notifier_block __read_mostly = {
936 	.notifier_call = ipvlan_device_event,
937 };
938 
939 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
940 	.notifier_call = ipvlan_addr6_event,
941 };
942 
943 static struct notifier_block ipvlan_addr6_vtor_notifier_block __read_mostly = {
944 	.notifier_call = ipvlan_addr6_validator_event,
945 };
946 
947 static void ipvlan_ns_exit(struct net *net)
948 {
949 	struct ipvlan_netns *vnet = net_generic(net, ipvlan_netid);
950 
951 	if (WARN_ON_ONCE(vnet->ipvl_nf_hook_refcnt)) {
952 		vnet->ipvl_nf_hook_refcnt = 0;
953 		nf_unregister_net_hooks(net, ipvl_nfops,
954 					ARRAY_SIZE(ipvl_nfops));
955 	}
956 }
957 
958 static struct pernet_operations ipvlan_net_ops = {
959 	.id = &ipvlan_netid,
960 	.size = sizeof(struct ipvlan_netns),
961 	.exit = ipvlan_ns_exit,
962 };
963 
964 static int __init ipvlan_init_module(void)
965 {
966 	int err;
967 
968 	ipvlan_init_secret();
969 	register_netdevice_notifier(&ipvlan_notifier_block);
970 	register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
971 	register_inet6addr_validator_notifier(
972 	    &ipvlan_addr6_vtor_notifier_block);
973 	register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
974 	register_inetaddr_validator_notifier(&ipvlan_addr4_vtor_notifier_block);
975 
976 	err = register_pernet_subsys(&ipvlan_net_ops);
977 	if (err < 0)
978 		goto error;
979 
980 	err = ipvlan_link_register(&ipvlan_link_ops);
981 	if (err < 0) {
982 		unregister_pernet_subsys(&ipvlan_net_ops);
983 		goto error;
984 	}
985 
986 	return 0;
987 error:
988 	unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
989 	unregister_inetaddr_validator_notifier(
990 	    &ipvlan_addr4_vtor_notifier_block);
991 	unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
992 	unregister_inet6addr_validator_notifier(
993 	    &ipvlan_addr6_vtor_notifier_block);
994 	unregister_netdevice_notifier(&ipvlan_notifier_block);
995 	return err;
996 }
997 
998 static void __exit ipvlan_cleanup_module(void)
999 {
1000 	rtnl_link_unregister(&ipvlan_link_ops);
1001 	unregister_pernet_subsys(&ipvlan_net_ops);
1002 	unregister_netdevice_notifier(&ipvlan_notifier_block);
1003 	unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1004 	unregister_inetaddr_validator_notifier(
1005 	    &ipvlan_addr4_vtor_notifier_block);
1006 	unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1007 	unregister_inet6addr_validator_notifier(
1008 	    &ipvlan_addr6_vtor_notifier_block);
1009 }
1010 
1011 module_init(ipvlan_init_module);
1012 module_exit(ipvlan_cleanup_module);
1013 
1014 MODULE_LICENSE("GPL");
1015 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
1016 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
1017 MODULE_ALIAS_RTNL_LINK("ipvlan");
1018