xref: /openbmc/linux/drivers/net/ipvlan/ipvlan_main.c (revision d7b31359)
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 #if IS_ENABLED(CONFIG_IPV6)
26 	{
27 		.hook     = ipvlan_nf_input,
28 		.pf       = NFPROTO_IPV6,
29 		.hooknum  = NF_INET_LOCAL_IN,
30 		.priority = INT_MAX,
31 	},
32 #endif
33 };
34 
35 static const struct l3mdev_ops ipvl_l3mdev_ops = {
36 	.l3mdev_l3_rcv = ipvlan_l3_rcv,
37 };
38 
39 static void ipvlan_adjust_mtu(struct ipvl_dev *ipvlan, struct net_device *dev)
40 {
41 	ipvlan->dev->mtu = dev->mtu;
42 }
43 
44 static int ipvlan_register_nf_hook(struct net *net)
45 {
46 	struct ipvlan_netns *vnet = net_generic(net, ipvlan_netid);
47 	int err = 0;
48 
49 	if (!vnet->ipvl_nf_hook_refcnt) {
50 		err = nf_register_net_hooks(net, ipvl_nfops,
51 					    ARRAY_SIZE(ipvl_nfops));
52 		if (!err)
53 			vnet->ipvl_nf_hook_refcnt = 1;
54 	} else {
55 		vnet->ipvl_nf_hook_refcnt++;
56 	}
57 
58 	return err;
59 }
60 
61 static void ipvlan_unregister_nf_hook(struct net *net)
62 {
63 	struct ipvlan_netns *vnet = net_generic(net, ipvlan_netid);
64 
65 	if (WARN_ON(!vnet->ipvl_nf_hook_refcnt))
66 		return;
67 
68 	vnet->ipvl_nf_hook_refcnt--;
69 	if (!vnet->ipvl_nf_hook_refcnt)
70 		nf_unregister_net_hooks(net, ipvl_nfops,
71 					ARRAY_SIZE(ipvl_nfops));
72 }
73 
74 static int ipvlan_set_port_mode(struct ipvl_port *port, u16 nval)
75 {
76 	struct ipvl_dev *ipvlan;
77 	struct net_device *mdev = port->dev;
78 	unsigned int flags;
79 	int err;
80 
81 	ASSERT_RTNL();
82 	if (port->mode != nval) {
83 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
84 			flags = ipvlan->dev->flags;
85 			if (nval == IPVLAN_MODE_L3 || nval == IPVLAN_MODE_L3S) {
86 				err = dev_change_flags(ipvlan->dev,
87 						       flags | IFF_NOARP);
88 			} else {
89 				err = dev_change_flags(ipvlan->dev,
90 						       flags & ~IFF_NOARP);
91 			}
92 			if (unlikely(err))
93 				goto fail;
94 		}
95 		if (nval == IPVLAN_MODE_L3S) {
96 			/* New mode is L3S */
97 			err = ipvlan_register_nf_hook(read_pnet(&port->pnet));
98 			if (!err) {
99 				mdev->l3mdev_ops = &ipvl_l3mdev_ops;
100 				mdev->priv_flags |= IFF_L3MDEV_MASTER;
101 			} else
102 				goto fail;
103 		} else if (port->mode == IPVLAN_MODE_L3S) {
104 			/* Old mode was L3S */
105 			mdev->priv_flags &= ~IFF_L3MDEV_MASTER;
106 			ipvlan_unregister_nf_hook(read_pnet(&port->pnet));
107 			mdev->l3mdev_ops = NULL;
108 		}
109 		port->mode = nval;
110 	}
111 	return 0;
112 
113 fail:
114 	/* Undo the flags changes that have been done so far. */
115 	list_for_each_entry_continue_reverse(ipvlan, &port->ipvlans, pnode) {
116 		flags = ipvlan->dev->flags;
117 		if (port->mode == IPVLAN_MODE_L3 ||
118 		    port->mode == IPVLAN_MODE_L3S)
119 			dev_change_flags(ipvlan->dev, flags | IFF_NOARP);
120 		else
121 			dev_change_flags(ipvlan->dev, flags & ~IFF_NOARP);
122 	}
123 
124 	return err;
125 }
126 
127 static int ipvlan_port_create(struct net_device *dev)
128 {
129 	struct ipvl_port *port;
130 	int err, idx;
131 
132 	port = kzalloc(sizeof(struct ipvl_port), GFP_KERNEL);
133 	if (!port)
134 		return -ENOMEM;
135 
136 	write_pnet(&port->pnet, dev_net(dev));
137 	port->dev = dev;
138 	port->mode = IPVLAN_MODE_L3;
139 	INIT_LIST_HEAD(&port->ipvlans);
140 	for (idx = 0; idx < IPVLAN_HASH_SIZE; idx++)
141 		INIT_HLIST_HEAD(&port->hlhead[idx]);
142 
143 	skb_queue_head_init(&port->backlog);
144 	INIT_WORK(&port->wq, ipvlan_process_multicast);
145 	ida_init(&port->ida);
146 	port->dev_id_start = 1;
147 
148 	err = netdev_rx_handler_register(dev, ipvlan_handle_frame, port);
149 	if (err)
150 		goto err;
151 
152 	return 0;
153 
154 err:
155 	kfree(port);
156 	return err;
157 }
158 
159 static void ipvlan_port_destroy(struct net_device *dev)
160 {
161 	struct ipvl_port *port = ipvlan_port_get_rtnl(dev);
162 	struct sk_buff *skb;
163 
164 	if (port->mode == IPVLAN_MODE_L3S) {
165 		dev->priv_flags &= ~IFF_L3MDEV_MASTER;
166 		ipvlan_unregister_nf_hook(dev_net(dev));
167 		dev->l3mdev_ops = NULL;
168 	}
169 	netdev_rx_handler_unregister(dev);
170 	cancel_work_sync(&port->wq);
171 	while ((skb = __skb_dequeue(&port->backlog)) != NULL) {
172 		if (skb->dev)
173 			dev_put(skb->dev);
174 		kfree_skb(skb);
175 	}
176 	ida_destroy(&port->ida);
177 	kfree(port);
178 }
179 
180 #define IPVLAN_FEATURES \
181 	(NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
182 	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_GSO_ROBUST | \
183 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
184 	 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
185 
186 #define IPVLAN_STATE_MASK \
187 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
188 
189 static int ipvlan_init(struct net_device *dev)
190 {
191 	struct ipvl_dev *ipvlan = netdev_priv(dev);
192 	struct net_device *phy_dev = ipvlan->phy_dev;
193 	struct ipvl_port *port;
194 	int err;
195 
196 	dev->state = (dev->state & ~IPVLAN_STATE_MASK) |
197 		     (phy_dev->state & IPVLAN_STATE_MASK);
198 	dev->features = phy_dev->features & IPVLAN_FEATURES;
199 	dev->features |= NETIF_F_LLTX | NETIF_F_VLAN_CHALLENGED;
200 	dev->gso_max_size = phy_dev->gso_max_size;
201 	dev->gso_max_segs = phy_dev->gso_max_segs;
202 	dev->hard_header_len = phy_dev->hard_header_len;
203 
204 	netdev_lockdep_set_classes(dev);
205 
206 	ipvlan->pcpu_stats = netdev_alloc_pcpu_stats(struct ipvl_pcpu_stats);
207 	if (!ipvlan->pcpu_stats)
208 		return -ENOMEM;
209 
210 	if (!netif_is_ipvlan_port(phy_dev)) {
211 		err = ipvlan_port_create(phy_dev);
212 		if (err < 0) {
213 			free_percpu(ipvlan->pcpu_stats);
214 			return err;
215 		}
216 	}
217 	port = ipvlan_port_get_rtnl(phy_dev);
218 	port->count += 1;
219 	return 0;
220 }
221 
222 static void ipvlan_uninit(struct net_device *dev)
223 {
224 	struct ipvl_dev *ipvlan = netdev_priv(dev);
225 	struct net_device *phy_dev = ipvlan->phy_dev;
226 	struct ipvl_port *port;
227 
228 	free_percpu(ipvlan->pcpu_stats);
229 
230 	port = ipvlan_port_get_rtnl(phy_dev);
231 	port->count -= 1;
232 	if (!port->count)
233 		ipvlan_port_destroy(port->dev);
234 }
235 
236 static int ipvlan_open(struct net_device *dev)
237 {
238 	struct ipvl_dev *ipvlan = netdev_priv(dev);
239 	struct net_device *phy_dev = ipvlan->phy_dev;
240 	struct ipvl_addr *addr;
241 
242 	if (ipvlan->port->mode == IPVLAN_MODE_L3 ||
243 	    ipvlan->port->mode == IPVLAN_MODE_L3S)
244 		dev->flags |= IFF_NOARP;
245 	else
246 		dev->flags &= ~IFF_NOARP;
247 
248 	rcu_read_lock();
249 	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
250 		ipvlan_ht_addr_add(ipvlan, addr);
251 	rcu_read_unlock();
252 
253 	return dev_uc_add(phy_dev, phy_dev->dev_addr);
254 }
255 
256 static int ipvlan_stop(struct net_device *dev)
257 {
258 	struct ipvl_dev *ipvlan = netdev_priv(dev);
259 	struct net_device *phy_dev = ipvlan->phy_dev;
260 	struct ipvl_addr *addr;
261 
262 	dev_uc_unsync(phy_dev, dev);
263 	dev_mc_unsync(phy_dev, dev);
264 
265 	dev_uc_del(phy_dev, phy_dev->dev_addr);
266 
267 	rcu_read_lock();
268 	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
269 		ipvlan_ht_addr_del(addr);
270 	rcu_read_unlock();
271 
272 	return 0;
273 }
274 
275 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
276 				     struct net_device *dev)
277 {
278 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
279 	int skblen = skb->len;
280 	int ret;
281 
282 	ret = ipvlan_queue_xmit(skb, dev);
283 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
284 		struct ipvl_pcpu_stats *pcptr;
285 
286 		pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
287 
288 		u64_stats_update_begin(&pcptr->syncp);
289 		pcptr->tx_pkts++;
290 		pcptr->tx_bytes += skblen;
291 		u64_stats_update_end(&pcptr->syncp);
292 	} else {
293 		this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
294 	}
295 	return ret;
296 }
297 
298 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
299 					     netdev_features_t features)
300 {
301 	struct ipvl_dev *ipvlan = netdev_priv(dev);
302 
303 	return features & (ipvlan->sfeatures | ~IPVLAN_FEATURES);
304 }
305 
306 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
307 {
308 	struct ipvl_dev *ipvlan = netdev_priv(dev);
309 	struct net_device *phy_dev = ipvlan->phy_dev;
310 
311 	if (change & IFF_ALLMULTI)
312 		dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
313 }
314 
315 static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
316 {
317 	struct ipvl_dev *ipvlan = netdev_priv(dev);
318 
319 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
320 		bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
321 	} else {
322 		struct netdev_hw_addr *ha;
323 		DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
324 
325 		bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
326 		netdev_for_each_mc_addr(ha, dev)
327 			__set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
328 
329 		/* Turn-on broadcast bit irrespective of address family,
330 		 * since broadcast is deferred to a work-queue, hence no
331 		 * impact on fast-path processing.
332 		 */
333 		__set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
334 
335 		bitmap_copy(ipvlan->mac_filters, mc_filters,
336 			    IPVLAN_MAC_FILTER_SIZE);
337 	}
338 	dev_uc_sync(ipvlan->phy_dev, dev);
339 	dev_mc_sync(ipvlan->phy_dev, dev);
340 }
341 
342 static void ipvlan_get_stats64(struct net_device *dev,
343 			       struct rtnl_link_stats64 *s)
344 {
345 	struct ipvl_dev *ipvlan = netdev_priv(dev);
346 
347 	if (ipvlan->pcpu_stats) {
348 		struct ipvl_pcpu_stats *pcptr;
349 		u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
350 		u32 rx_errs = 0, tx_drps = 0;
351 		u32 strt;
352 		int idx;
353 
354 		for_each_possible_cpu(idx) {
355 			pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
356 			do {
357 				strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
358 				rx_pkts = pcptr->rx_pkts;
359 				rx_bytes = pcptr->rx_bytes;
360 				rx_mcast = pcptr->rx_mcast;
361 				tx_pkts = pcptr->tx_pkts;
362 				tx_bytes = pcptr->tx_bytes;
363 			} while (u64_stats_fetch_retry_irq(&pcptr->syncp,
364 							   strt));
365 
366 			s->rx_packets += rx_pkts;
367 			s->rx_bytes += rx_bytes;
368 			s->multicast += rx_mcast;
369 			s->tx_packets += tx_pkts;
370 			s->tx_bytes += tx_bytes;
371 
372 			/* u32 values are updated without syncp protection. */
373 			rx_errs += pcptr->rx_errs;
374 			tx_drps += pcptr->tx_drps;
375 		}
376 		s->rx_errors = rx_errs;
377 		s->rx_dropped = rx_errs;
378 		s->tx_dropped = tx_drps;
379 	}
380 }
381 
382 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
383 {
384 	struct ipvl_dev *ipvlan = netdev_priv(dev);
385 	struct net_device *phy_dev = ipvlan->phy_dev;
386 
387 	return vlan_vid_add(phy_dev, proto, vid);
388 }
389 
390 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
391 				   u16 vid)
392 {
393 	struct ipvl_dev *ipvlan = netdev_priv(dev);
394 	struct net_device *phy_dev = ipvlan->phy_dev;
395 
396 	vlan_vid_del(phy_dev, proto, vid);
397 	return 0;
398 }
399 
400 static int ipvlan_get_iflink(const struct net_device *dev)
401 {
402 	struct ipvl_dev *ipvlan = netdev_priv(dev);
403 
404 	return ipvlan->phy_dev->ifindex;
405 }
406 
407 static const struct net_device_ops ipvlan_netdev_ops = {
408 	.ndo_init		= ipvlan_init,
409 	.ndo_uninit		= ipvlan_uninit,
410 	.ndo_open		= ipvlan_open,
411 	.ndo_stop		= ipvlan_stop,
412 	.ndo_start_xmit		= ipvlan_start_xmit,
413 	.ndo_fix_features	= ipvlan_fix_features,
414 	.ndo_change_rx_flags	= ipvlan_change_rx_flags,
415 	.ndo_set_rx_mode	= ipvlan_set_multicast_mac_filter,
416 	.ndo_get_stats64	= ipvlan_get_stats64,
417 	.ndo_vlan_rx_add_vid	= ipvlan_vlan_rx_add_vid,
418 	.ndo_vlan_rx_kill_vid	= ipvlan_vlan_rx_kill_vid,
419 	.ndo_get_iflink		= ipvlan_get_iflink,
420 };
421 
422 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
423 			      unsigned short type, const void *daddr,
424 			      const void *saddr, unsigned len)
425 {
426 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
427 	struct net_device *phy_dev = ipvlan->phy_dev;
428 
429 	/* TODO Probably use a different field than dev_addr so that the
430 	 * mac-address on the virtual device is portable and can be carried
431 	 * while the packets use the mac-addr on the physical device.
432 	 */
433 	return dev_hard_header(skb, phy_dev, type, daddr,
434 			       saddr ? : phy_dev->dev_addr, len);
435 }
436 
437 static const struct header_ops ipvlan_header_ops = {
438 	.create  	= ipvlan_hard_header,
439 	.parse		= eth_header_parse,
440 	.cache		= eth_header_cache,
441 	.cache_update	= eth_header_cache_update,
442 };
443 
444 static bool netif_is_ipvlan(const struct net_device *dev)
445 {
446 	/* both ipvlan and ipvtap devices use the same netdev_ops */
447 	return dev->netdev_ops == &ipvlan_netdev_ops;
448 }
449 
450 static int ipvlan_ethtool_get_link_ksettings(struct net_device *dev,
451 					     struct ethtool_link_ksettings *cmd)
452 {
453 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
454 
455 	return __ethtool_get_link_ksettings(ipvlan->phy_dev, cmd);
456 }
457 
458 static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
459 				       struct ethtool_drvinfo *drvinfo)
460 {
461 	strlcpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
462 	strlcpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
463 }
464 
465 static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
466 {
467 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
468 
469 	return ipvlan->msg_enable;
470 }
471 
472 static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
473 {
474 	struct ipvl_dev *ipvlan = netdev_priv(dev);
475 
476 	ipvlan->msg_enable = value;
477 }
478 
479 static const struct ethtool_ops ipvlan_ethtool_ops = {
480 	.get_link	= ethtool_op_get_link,
481 	.get_link_ksettings	= ipvlan_ethtool_get_link_ksettings,
482 	.get_drvinfo	= ipvlan_ethtool_get_drvinfo,
483 	.get_msglevel	= ipvlan_ethtool_get_msglevel,
484 	.set_msglevel	= ipvlan_ethtool_set_msglevel,
485 };
486 
487 static int ipvlan_nl_changelink(struct net_device *dev,
488 				struct nlattr *tb[], struct nlattr *data[],
489 				struct netlink_ext_ack *extack)
490 {
491 	struct ipvl_dev *ipvlan = netdev_priv(dev);
492 	struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
493 	int err = 0;
494 
495 	if (!data)
496 		return 0;
497 
498 	if (data[IFLA_IPVLAN_MODE]) {
499 		u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
500 
501 		err = ipvlan_set_port_mode(port, nmode);
502 	}
503 
504 	if (!err && data[IFLA_IPVLAN_FLAGS]) {
505 		u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
506 
507 		if (flags & IPVLAN_F_PRIVATE)
508 			ipvlan_mark_private(port);
509 		else
510 			ipvlan_clear_private(port);
511 
512 		if (flags & IPVLAN_F_VEPA)
513 			ipvlan_mark_vepa(port);
514 		else
515 			ipvlan_clear_vepa(port);
516 	}
517 
518 	return err;
519 }
520 
521 static size_t ipvlan_nl_getsize(const struct net_device *dev)
522 {
523 	return (0
524 		+ nla_total_size(2) /* IFLA_IPVLAN_MODE */
525 		+ nla_total_size(2) /* IFLA_IPVLAN_FLAGS */
526 		);
527 }
528 
529 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[],
530 			      struct netlink_ext_ack *extack)
531 {
532 	if (!data)
533 		return 0;
534 
535 	if (data[IFLA_IPVLAN_MODE]) {
536 		u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
537 
538 		if (mode < IPVLAN_MODE_L2 || mode >= IPVLAN_MODE_MAX)
539 			return -EINVAL;
540 	}
541 	if (data[IFLA_IPVLAN_FLAGS]) {
542 		u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
543 
544 		/* Only two bits are used at this moment. */
545 		if (flags & ~(IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
546 			return -EINVAL;
547 		/* Also both flags can't be active at the same time. */
548 		if ((flags & (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA)) ==
549 		    (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
550 			return -EINVAL;
551 	}
552 
553 	return 0;
554 }
555 
556 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
557 			      const struct net_device *dev)
558 {
559 	struct ipvl_dev *ipvlan = netdev_priv(dev);
560 	struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
561 	int ret = -EINVAL;
562 
563 	if (!port)
564 		goto err;
565 
566 	ret = -EMSGSIZE;
567 	if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
568 		goto err;
569 	if (nla_put_u16(skb, IFLA_IPVLAN_FLAGS, port->flags))
570 		goto err;
571 
572 	return 0;
573 
574 err:
575 	return ret;
576 }
577 
578 int ipvlan_link_new(struct net *src_net, struct net_device *dev,
579 		    struct nlattr *tb[], struct nlattr *data[],
580 		    struct netlink_ext_ack *extack)
581 {
582 	struct ipvl_dev *ipvlan = netdev_priv(dev);
583 	struct ipvl_port *port;
584 	struct net_device *phy_dev;
585 	int err;
586 	u16 mode = IPVLAN_MODE_L3;
587 
588 	if (!tb[IFLA_LINK])
589 		return -EINVAL;
590 
591 	phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
592 	if (!phy_dev)
593 		return -ENODEV;
594 
595 	if (netif_is_ipvlan(phy_dev)) {
596 		struct ipvl_dev *tmp = netdev_priv(phy_dev);
597 
598 		phy_dev = tmp->phy_dev;
599 	} else if (!netif_is_ipvlan_port(phy_dev)) {
600 		/* Exit early if the underlying link is invalid or busy */
601 		if (phy_dev->type != ARPHRD_ETHER ||
602 		    phy_dev->flags & IFF_LOOPBACK) {
603 			netdev_err(phy_dev,
604 				   "Master is either lo or non-ether device\n");
605 			return -EINVAL;
606 		}
607 
608 		if (netdev_is_rx_handler_busy(phy_dev)) {
609 			netdev_err(phy_dev, "Device is already in use.\n");
610 			return -EBUSY;
611 		}
612 	}
613 
614 	ipvlan->phy_dev = phy_dev;
615 	ipvlan->dev = dev;
616 	ipvlan->sfeatures = IPVLAN_FEATURES;
617 	if (!tb[IFLA_MTU])
618 		ipvlan_adjust_mtu(ipvlan, phy_dev);
619 	INIT_LIST_HEAD(&ipvlan->addrs);
620 	spin_lock_init(&ipvlan->addrs_lock);
621 
622 	/* TODO Probably put random address here to be presented to the
623 	 * world but keep using the physical-dev address for the outgoing
624 	 * packets.
625 	 */
626 	memcpy(dev->dev_addr, phy_dev->dev_addr, ETH_ALEN);
627 
628 	dev->priv_flags |= IFF_NO_RX_HANDLER;
629 
630 	err = register_netdevice(dev);
631 	if (err < 0)
632 		return err;
633 
634 	/* ipvlan_init() would have created the port, if required */
635 	port = ipvlan_port_get_rtnl(phy_dev);
636 	ipvlan->port = port;
637 
638 	/* If the port-id base is at the MAX value, then wrap it around and
639 	 * begin from 0x1 again. This may be due to a busy system where lots
640 	 * of slaves are getting created and deleted.
641 	 */
642 	if (port->dev_id_start == 0xFFFE)
643 		port->dev_id_start = 0x1;
644 
645 	/* Since L2 address is shared among all IPvlan slaves including
646 	 * master, use unique 16 bit dev-ids to diffentiate among them.
647 	 * Assign IDs between 0x1 and 0xFFFE (used by the master) to each
648 	 * slave link [see addrconf_ifid_eui48()].
649 	 */
650 	err = ida_simple_get(&port->ida, port->dev_id_start, 0xFFFE,
651 			     GFP_KERNEL);
652 	if (err < 0)
653 		err = ida_simple_get(&port->ida, 0x1, port->dev_id_start,
654 				     GFP_KERNEL);
655 	if (err < 0)
656 		goto unregister_netdev;
657 	dev->dev_id = err;
658 
659 	/* Increment id-base to the next slot for the future assignment */
660 	port->dev_id_start = err + 1;
661 
662 	err = netdev_upper_dev_link(phy_dev, dev, extack);
663 	if (err)
664 		goto remove_ida;
665 
666 	/* Flags are per port and latest update overrides. User has
667 	 * to be consistent in setting it just like the mode attribute.
668 	 */
669 	if (data && data[IFLA_IPVLAN_FLAGS])
670 		port->flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
671 
672 	if (data && data[IFLA_IPVLAN_MODE])
673 		mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
674 
675 	err = ipvlan_set_port_mode(port, mode);
676 	if (err)
677 		goto unlink_netdev;
678 
679 	list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
680 	netif_stacked_transfer_operstate(phy_dev, dev);
681 	return 0;
682 
683 unlink_netdev:
684 	netdev_upper_dev_unlink(phy_dev, dev);
685 remove_ida:
686 	ida_simple_remove(&port->ida, dev->dev_id);
687 unregister_netdev:
688 	unregister_netdevice(dev);
689 	return err;
690 }
691 EXPORT_SYMBOL_GPL(ipvlan_link_new);
692 
693 void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
694 {
695 	struct ipvl_dev *ipvlan = netdev_priv(dev);
696 	struct ipvl_addr *addr, *next;
697 
698 	spin_lock_bh(&ipvlan->addrs_lock);
699 	list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
700 		ipvlan_ht_addr_del(addr);
701 		list_del_rcu(&addr->anode);
702 		kfree_rcu(addr, rcu);
703 	}
704 	spin_unlock_bh(&ipvlan->addrs_lock);
705 
706 	ida_simple_remove(&ipvlan->port->ida, dev->dev_id);
707 	list_del_rcu(&ipvlan->pnode);
708 	unregister_netdevice_queue(dev, head);
709 	netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
710 }
711 EXPORT_SYMBOL_GPL(ipvlan_link_delete);
712 
713 void ipvlan_link_setup(struct net_device *dev)
714 {
715 	ether_setup(dev);
716 
717 	dev->max_mtu = ETH_MAX_MTU;
718 	dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
719 	dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
720 	dev->netdev_ops = &ipvlan_netdev_ops;
721 	dev->needs_free_netdev = true;
722 	dev->header_ops = &ipvlan_header_ops;
723 	dev->ethtool_ops = &ipvlan_ethtool_ops;
724 }
725 EXPORT_SYMBOL_GPL(ipvlan_link_setup);
726 
727 static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
728 {
729 	[IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
730 	[IFLA_IPVLAN_FLAGS] = { .type = NLA_U16 },
731 };
732 
733 static struct rtnl_link_ops ipvlan_link_ops = {
734 	.kind		= "ipvlan",
735 	.priv_size	= sizeof(struct ipvl_dev),
736 
737 	.setup		= ipvlan_link_setup,
738 	.newlink	= ipvlan_link_new,
739 	.dellink	= ipvlan_link_delete,
740 };
741 
742 int ipvlan_link_register(struct rtnl_link_ops *ops)
743 {
744 	ops->get_size	= ipvlan_nl_getsize;
745 	ops->policy	= ipvlan_nl_policy;
746 	ops->validate	= ipvlan_nl_validate;
747 	ops->fill_info	= ipvlan_nl_fillinfo;
748 	ops->changelink = ipvlan_nl_changelink;
749 	ops->maxtype	= IFLA_IPVLAN_MAX;
750 	return rtnl_link_register(ops);
751 }
752 EXPORT_SYMBOL_GPL(ipvlan_link_register);
753 
754 static int ipvlan_device_event(struct notifier_block *unused,
755 			       unsigned long event, void *ptr)
756 {
757 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
758 	struct ipvl_dev *ipvlan, *next;
759 	struct ipvl_port *port;
760 	LIST_HEAD(lst_kill);
761 
762 	if (!netif_is_ipvlan_port(dev))
763 		return NOTIFY_DONE;
764 
765 	port = ipvlan_port_get_rtnl(dev);
766 
767 	switch (event) {
768 	case NETDEV_CHANGE:
769 		list_for_each_entry(ipvlan, &port->ipvlans, pnode)
770 			netif_stacked_transfer_operstate(ipvlan->phy_dev,
771 							 ipvlan->dev);
772 		break;
773 
774 	case NETDEV_REGISTER: {
775 		struct net *oldnet, *newnet = dev_net(dev);
776 		struct ipvlan_netns *old_vnet;
777 
778 		oldnet = read_pnet(&port->pnet);
779 		if (net_eq(newnet, oldnet))
780 			break;
781 
782 		write_pnet(&port->pnet, newnet);
783 
784 		old_vnet = net_generic(oldnet, ipvlan_netid);
785 		if (!old_vnet->ipvl_nf_hook_refcnt)
786 			break;
787 
788 		ipvlan_register_nf_hook(newnet);
789 		ipvlan_unregister_nf_hook(oldnet);
790 		break;
791 	}
792 	case NETDEV_UNREGISTER:
793 		if (dev->reg_state != NETREG_UNREGISTERING)
794 			break;
795 
796 		list_for_each_entry_safe(ipvlan, next, &port->ipvlans, pnode)
797 			ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
798 							    &lst_kill);
799 		unregister_netdevice_many(&lst_kill);
800 		break;
801 
802 	case NETDEV_FEAT_CHANGE:
803 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
804 			ipvlan->dev->features = dev->features & IPVLAN_FEATURES;
805 			ipvlan->dev->gso_max_size = dev->gso_max_size;
806 			ipvlan->dev->gso_max_segs = dev->gso_max_segs;
807 			netdev_features_change(ipvlan->dev);
808 		}
809 		break;
810 
811 	case NETDEV_CHANGEMTU:
812 		list_for_each_entry(ipvlan, &port->ipvlans, pnode)
813 			ipvlan_adjust_mtu(ipvlan, dev);
814 		break;
815 
816 	case NETDEV_CHANGEADDR:
817 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
818 			ether_addr_copy(ipvlan->dev->dev_addr, dev->dev_addr);
819 			call_netdevice_notifiers(NETDEV_CHANGEADDR, ipvlan->dev);
820 		}
821 		break;
822 
823 	case NETDEV_PRE_TYPE_CHANGE:
824 		/* Forbid underlying device to change its type. */
825 		return NOTIFY_BAD;
826 	}
827 	return NOTIFY_DONE;
828 }
829 
830 /* the caller must held the addrs lock */
831 static int ipvlan_add_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
832 {
833 	struct ipvl_addr *addr;
834 
835 	addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
836 	if (!addr)
837 		return -ENOMEM;
838 
839 	addr->master = ipvlan;
840 	if (!is_v6) {
841 		memcpy(&addr->ip4addr, iaddr, sizeof(struct in_addr));
842 		addr->atype = IPVL_IPV4;
843 #if IS_ENABLED(CONFIG_IPV6)
844 	} else {
845 		memcpy(&addr->ip6addr, iaddr, sizeof(struct in6_addr));
846 		addr->atype = IPVL_IPV6;
847 #endif
848 	}
849 
850 	list_add_tail_rcu(&addr->anode, &ipvlan->addrs);
851 
852 	/* If the interface is not up, the address will be added to the hash
853 	 * list by ipvlan_open.
854 	 */
855 	if (netif_running(ipvlan->dev))
856 		ipvlan_ht_addr_add(ipvlan, addr);
857 
858 	return 0;
859 }
860 
861 static void ipvlan_del_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
862 {
863 	struct ipvl_addr *addr;
864 
865 	spin_lock_bh(&ipvlan->addrs_lock);
866 	addr = ipvlan_find_addr(ipvlan, iaddr, is_v6);
867 	if (!addr) {
868 		spin_unlock_bh(&ipvlan->addrs_lock);
869 		return;
870 	}
871 
872 	ipvlan_ht_addr_del(addr);
873 	list_del_rcu(&addr->anode);
874 	spin_unlock_bh(&ipvlan->addrs_lock);
875 	kfree_rcu(addr, rcu);
876 }
877 
878 static bool ipvlan_is_valid_dev(const struct net_device *dev)
879 {
880 	struct ipvl_dev *ipvlan = netdev_priv(dev);
881 
882 	if (!netif_is_ipvlan(dev))
883 		return false;
884 
885 	if (!ipvlan || !ipvlan->port)
886 		return false;
887 
888 	return true;
889 }
890 
891 #if IS_ENABLED(CONFIG_IPV6)
892 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
893 {
894 	int ret = -EINVAL;
895 
896 	spin_lock_bh(&ipvlan->addrs_lock);
897 	if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true))
898 		netif_err(ipvlan, ifup, ipvlan->dev,
899 			  "Failed to add IPv6=%pI6c addr for %s intf\n",
900 			  ip6_addr, ipvlan->dev->name);
901 	else
902 		ret = ipvlan_add_addr(ipvlan, ip6_addr, true);
903 	spin_unlock_bh(&ipvlan->addrs_lock);
904 	return ret;
905 }
906 
907 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
908 {
909 	return ipvlan_del_addr(ipvlan, ip6_addr, true);
910 }
911 
912 static int ipvlan_addr6_event(struct notifier_block *unused,
913 			      unsigned long event, void *ptr)
914 {
915 	struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
916 	struct net_device *dev = (struct net_device *)if6->idev->dev;
917 	struct ipvl_dev *ipvlan = netdev_priv(dev);
918 
919 	if (!ipvlan_is_valid_dev(dev))
920 		return NOTIFY_DONE;
921 
922 	switch (event) {
923 	case NETDEV_UP:
924 		if (ipvlan_add_addr6(ipvlan, &if6->addr))
925 			return NOTIFY_BAD;
926 		break;
927 
928 	case NETDEV_DOWN:
929 		ipvlan_del_addr6(ipvlan, &if6->addr);
930 		break;
931 	}
932 
933 	return NOTIFY_OK;
934 }
935 
936 static int ipvlan_addr6_validator_event(struct notifier_block *unused,
937 					unsigned long event, void *ptr)
938 {
939 	struct in6_validator_info *i6vi = (struct in6_validator_info *)ptr;
940 	struct net_device *dev = (struct net_device *)i6vi->i6vi_dev->dev;
941 	struct ipvl_dev *ipvlan = netdev_priv(dev);
942 
943 	if (!ipvlan_is_valid_dev(dev))
944 		return NOTIFY_DONE;
945 
946 	switch (event) {
947 	case NETDEV_UP:
948 		if (ipvlan_addr_busy(ipvlan->port, &i6vi->i6vi_addr, true)) {
949 			NL_SET_ERR_MSG(i6vi->extack,
950 				       "Address already assigned to an ipvlan device");
951 			return notifier_from_errno(-EADDRINUSE);
952 		}
953 		break;
954 	}
955 
956 	return NOTIFY_OK;
957 }
958 #endif
959 
960 static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
961 {
962 	int ret = -EINVAL;
963 
964 	spin_lock_bh(&ipvlan->addrs_lock);
965 	if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false))
966 		netif_err(ipvlan, ifup, ipvlan->dev,
967 			  "Failed to add IPv4=%pI4 on %s intf.\n",
968 			  ip4_addr, ipvlan->dev->name);
969 	else
970 		ret = ipvlan_add_addr(ipvlan, ip4_addr, false);
971 	spin_unlock_bh(&ipvlan->addrs_lock);
972 	return ret;
973 }
974 
975 static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
976 {
977 	return ipvlan_del_addr(ipvlan, ip4_addr, false);
978 }
979 
980 static int ipvlan_addr4_event(struct notifier_block *unused,
981 			      unsigned long event, void *ptr)
982 {
983 	struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
984 	struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
985 	struct ipvl_dev *ipvlan = netdev_priv(dev);
986 	struct in_addr ip4_addr;
987 
988 	if (!ipvlan_is_valid_dev(dev))
989 		return NOTIFY_DONE;
990 
991 	switch (event) {
992 	case NETDEV_UP:
993 		ip4_addr.s_addr = if4->ifa_address;
994 		if (ipvlan_add_addr4(ipvlan, &ip4_addr))
995 			return NOTIFY_BAD;
996 		break;
997 
998 	case NETDEV_DOWN:
999 		ip4_addr.s_addr = if4->ifa_address;
1000 		ipvlan_del_addr4(ipvlan, &ip4_addr);
1001 		break;
1002 	}
1003 
1004 	return NOTIFY_OK;
1005 }
1006 
1007 static int ipvlan_addr4_validator_event(struct notifier_block *unused,
1008 					unsigned long event, void *ptr)
1009 {
1010 	struct in_validator_info *ivi = (struct in_validator_info *)ptr;
1011 	struct net_device *dev = (struct net_device *)ivi->ivi_dev->dev;
1012 	struct ipvl_dev *ipvlan = netdev_priv(dev);
1013 
1014 	if (!ipvlan_is_valid_dev(dev))
1015 		return NOTIFY_DONE;
1016 
1017 	switch (event) {
1018 	case NETDEV_UP:
1019 		if (ipvlan_addr_busy(ipvlan->port, &ivi->ivi_addr, false)) {
1020 			NL_SET_ERR_MSG(ivi->extack,
1021 				       "Address already assigned to an ipvlan device");
1022 			return notifier_from_errno(-EADDRINUSE);
1023 		}
1024 		break;
1025 	}
1026 
1027 	return NOTIFY_OK;
1028 }
1029 
1030 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
1031 	.notifier_call = ipvlan_addr4_event,
1032 };
1033 
1034 static struct notifier_block ipvlan_addr4_vtor_notifier_block __read_mostly = {
1035 	.notifier_call = ipvlan_addr4_validator_event,
1036 };
1037 
1038 static struct notifier_block ipvlan_notifier_block __read_mostly = {
1039 	.notifier_call = ipvlan_device_event,
1040 };
1041 
1042 #if IS_ENABLED(CONFIG_IPV6)
1043 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
1044 	.notifier_call = ipvlan_addr6_event,
1045 };
1046 
1047 static struct notifier_block ipvlan_addr6_vtor_notifier_block __read_mostly = {
1048 	.notifier_call = ipvlan_addr6_validator_event,
1049 };
1050 #endif
1051 
1052 static void ipvlan_ns_exit(struct net *net)
1053 {
1054 	struct ipvlan_netns *vnet = net_generic(net, ipvlan_netid);
1055 
1056 	if (WARN_ON_ONCE(vnet->ipvl_nf_hook_refcnt)) {
1057 		vnet->ipvl_nf_hook_refcnt = 0;
1058 		nf_unregister_net_hooks(net, ipvl_nfops,
1059 					ARRAY_SIZE(ipvl_nfops));
1060 	}
1061 }
1062 
1063 static struct pernet_operations ipvlan_net_ops = {
1064 	.id = &ipvlan_netid,
1065 	.size = sizeof(struct ipvlan_netns),
1066 	.exit = ipvlan_ns_exit,
1067 };
1068 
1069 static int __init ipvlan_init_module(void)
1070 {
1071 	int err;
1072 
1073 	ipvlan_init_secret();
1074 	register_netdevice_notifier(&ipvlan_notifier_block);
1075 #if IS_ENABLED(CONFIG_IPV6)
1076 	register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1077 	register_inet6addr_validator_notifier(
1078 	    &ipvlan_addr6_vtor_notifier_block);
1079 #endif
1080 	register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1081 	register_inetaddr_validator_notifier(&ipvlan_addr4_vtor_notifier_block);
1082 
1083 	err = register_pernet_subsys(&ipvlan_net_ops);
1084 	if (err < 0)
1085 		goto error;
1086 
1087 	err = ipvlan_link_register(&ipvlan_link_ops);
1088 	if (err < 0) {
1089 		unregister_pernet_subsys(&ipvlan_net_ops);
1090 		goto error;
1091 	}
1092 
1093 	return 0;
1094 error:
1095 	unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1096 	unregister_inetaddr_validator_notifier(
1097 	    &ipvlan_addr4_vtor_notifier_block);
1098 #if IS_ENABLED(CONFIG_IPV6)
1099 	unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1100 	unregister_inet6addr_validator_notifier(
1101 	    &ipvlan_addr6_vtor_notifier_block);
1102 #endif
1103 	unregister_netdevice_notifier(&ipvlan_notifier_block);
1104 	return err;
1105 }
1106 
1107 static void __exit ipvlan_cleanup_module(void)
1108 {
1109 	rtnl_link_unregister(&ipvlan_link_ops);
1110 	unregister_pernet_subsys(&ipvlan_net_ops);
1111 	unregister_netdevice_notifier(&ipvlan_notifier_block);
1112 	unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1113 	unregister_inetaddr_validator_notifier(
1114 	    &ipvlan_addr4_vtor_notifier_block);
1115 #if IS_ENABLED(CONFIG_IPV6)
1116 	unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1117 	unregister_inet6addr_validator_notifier(
1118 	    &ipvlan_addr6_vtor_notifier_block);
1119 #endif
1120 }
1121 
1122 module_init(ipvlan_init_module);
1123 module_exit(ipvlan_cleanup_module);
1124 
1125 MODULE_LICENSE("GPL");
1126 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
1127 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
1128 MODULE_ALIAS_RTNL_LINK("ipvlan");
1129