xref: /openbmc/linux/drivers/net/ipvlan/ipvlan_main.c (revision d003d772)
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 int ipvlan_set_port_mode(struct ipvl_port *port, u16 nval,
13 				struct netlink_ext_ack *extack)
14 {
15 	struct ipvl_dev *ipvlan;
16 	unsigned int flags;
17 	int err;
18 
19 	ASSERT_RTNL();
20 	if (port->mode != nval) {
21 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
22 			flags = ipvlan->dev->flags;
23 			if (nval == IPVLAN_MODE_L3 || nval == IPVLAN_MODE_L3S) {
24 				err = dev_change_flags(ipvlan->dev,
25 						       flags | IFF_NOARP,
26 						       extack);
27 			} else {
28 				err = dev_change_flags(ipvlan->dev,
29 						       flags & ~IFF_NOARP,
30 						       extack);
31 			}
32 			if (unlikely(err))
33 				goto fail;
34 		}
35 		if (nval == IPVLAN_MODE_L3S) {
36 			/* New mode is L3S */
37 			err = ipvlan_l3s_register(port);
38 			if (err)
39 				goto fail;
40 		} else if (port->mode == IPVLAN_MODE_L3S) {
41 			/* Old mode was L3S */
42 			ipvlan_l3s_unregister(port);
43 		}
44 		port->mode = nval;
45 	}
46 	return 0;
47 
48 fail:
49 	/* Undo the flags changes that have been done so far. */
50 	list_for_each_entry_continue_reverse(ipvlan, &port->ipvlans, pnode) {
51 		flags = ipvlan->dev->flags;
52 		if (port->mode == IPVLAN_MODE_L3 ||
53 		    port->mode == IPVLAN_MODE_L3S)
54 			dev_change_flags(ipvlan->dev, flags | IFF_NOARP,
55 					 NULL);
56 		else
57 			dev_change_flags(ipvlan->dev, flags & ~IFF_NOARP,
58 					 NULL);
59 	}
60 
61 	return err;
62 }
63 
64 static int ipvlan_port_create(struct net_device *dev)
65 {
66 	struct ipvl_port *port;
67 	int err, idx;
68 
69 	port = kzalloc(sizeof(struct ipvl_port), GFP_KERNEL);
70 	if (!port)
71 		return -ENOMEM;
72 
73 	write_pnet(&port->pnet, dev_net(dev));
74 	port->dev = dev;
75 	port->mode = IPVLAN_MODE_L3;
76 	INIT_LIST_HEAD(&port->ipvlans);
77 	for (idx = 0; idx < IPVLAN_HASH_SIZE; idx++)
78 		INIT_HLIST_HEAD(&port->hlhead[idx]);
79 
80 	skb_queue_head_init(&port->backlog);
81 	INIT_WORK(&port->wq, ipvlan_process_multicast);
82 	ida_init(&port->ida);
83 	port->dev_id_start = 1;
84 
85 	err = netdev_rx_handler_register(dev, ipvlan_handle_frame, port);
86 	if (err)
87 		goto err;
88 
89 	return 0;
90 
91 err:
92 	kfree(port);
93 	return err;
94 }
95 
96 static void ipvlan_port_destroy(struct net_device *dev)
97 {
98 	struct ipvl_port *port = ipvlan_port_get_rtnl(dev);
99 	struct sk_buff *skb;
100 
101 	if (port->mode == IPVLAN_MODE_L3S)
102 		ipvlan_l3s_unregister(port);
103 	netdev_rx_handler_unregister(dev);
104 	cancel_work_sync(&port->wq);
105 	while ((skb = __skb_dequeue(&port->backlog)) != NULL) {
106 		if (skb->dev)
107 			dev_put(skb->dev);
108 		kfree_skb(skb);
109 	}
110 	ida_destroy(&port->ida);
111 	kfree(port);
112 }
113 
114 #define IPVLAN_FEATURES \
115 	(NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
116 	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_GSO_ROBUST | \
117 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
118 	 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
119 
120 #define IPVLAN_STATE_MASK \
121 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
122 
123 static int ipvlan_init(struct net_device *dev)
124 {
125 	struct ipvl_dev *ipvlan = netdev_priv(dev);
126 	struct net_device *phy_dev = ipvlan->phy_dev;
127 	struct ipvl_port *port;
128 	int err;
129 
130 	dev->state = (dev->state & ~IPVLAN_STATE_MASK) |
131 		     (phy_dev->state & IPVLAN_STATE_MASK);
132 	dev->features = phy_dev->features & IPVLAN_FEATURES;
133 	dev->features |= NETIF_F_LLTX | NETIF_F_VLAN_CHALLENGED;
134 	dev->gso_max_size = phy_dev->gso_max_size;
135 	dev->gso_max_segs = phy_dev->gso_max_segs;
136 	dev->hard_header_len = phy_dev->hard_header_len;
137 
138 	netdev_lockdep_set_classes(dev);
139 
140 	ipvlan->pcpu_stats = netdev_alloc_pcpu_stats(struct ipvl_pcpu_stats);
141 	if (!ipvlan->pcpu_stats)
142 		return -ENOMEM;
143 
144 	if (!netif_is_ipvlan_port(phy_dev)) {
145 		err = ipvlan_port_create(phy_dev);
146 		if (err < 0) {
147 			free_percpu(ipvlan->pcpu_stats);
148 			return err;
149 		}
150 	}
151 	port = ipvlan_port_get_rtnl(phy_dev);
152 	port->count += 1;
153 	return 0;
154 }
155 
156 static void ipvlan_uninit(struct net_device *dev)
157 {
158 	struct ipvl_dev *ipvlan = netdev_priv(dev);
159 	struct net_device *phy_dev = ipvlan->phy_dev;
160 	struct ipvl_port *port;
161 
162 	free_percpu(ipvlan->pcpu_stats);
163 
164 	port = ipvlan_port_get_rtnl(phy_dev);
165 	port->count -= 1;
166 	if (!port->count)
167 		ipvlan_port_destroy(port->dev);
168 }
169 
170 static int ipvlan_open(struct net_device *dev)
171 {
172 	struct ipvl_dev *ipvlan = netdev_priv(dev);
173 	struct net_device *phy_dev = ipvlan->phy_dev;
174 	struct ipvl_addr *addr;
175 
176 	if (ipvlan->port->mode == IPVLAN_MODE_L3 ||
177 	    ipvlan->port->mode == IPVLAN_MODE_L3S)
178 		dev->flags |= IFF_NOARP;
179 	else
180 		dev->flags &= ~IFF_NOARP;
181 
182 	rcu_read_lock();
183 	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
184 		ipvlan_ht_addr_add(ipvlan, addr);
185 	rcu_read_unlock();
186 
187 	return dev_uc_add(phy_dev, phy_dev->dev_addr);
188 }
189 
190 static int ipvlan_stop(struct net_device *dev)
191 {
192 	struct ipvl_dev *ipvlan = netdev_priv(dev);
193 	struct net_device *phy_dev = ipvlan->phy_dev;
194 	struct ipvl_addr *addr;
195 
196 	dev_uc_unsync(phy_dev, dev);
197 	dev_mc_unsync(phy_dev, dev);
198 
199 	dev_uc_del(phy_dev, phy_dev->dev_addr);
200 
201 	rcu_read_lock();
202 	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
203 		ipvlan_ht_addr_del(addr);
204 	rcu_read_unlock();
205 
206 	return 0;
207 }
208 
209 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
210 				     struct net_device *dev)
211 {
212 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
213 	int skblen = skb->len;
214 	int ret;
215 
216 	ret = ipvlan_queue_xmit(skb, dev);
217 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
218 		struct ipvl_pcpu_stats *pcptr;
219 
220 		pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
221 
222 		u64_stats_update_begin(&pcptr->syncp);
223 		pcptr->tx_pkts++;
224 		pcptr->tx_bytes += skblen;
225 		u64_stats_update_end(&pcptr->syncp);
226 	} else {
227 		this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
228 	}
229 	return ret;
230 }
231 
232 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
233 					     netdev_features_t features)
234 {
235 	struct ipvl_dev *ipvlan = netdev_priv(dev);
236 
237 	return features & (ipvlan->sfeatures | ~IPVLAN_FEATURES);
238 }
239 
240 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
241 {
242 	struct ipvl_dev *ipvlan = netdev_priv(dev);
243 	struct net_device *phy_dev = ipvlan->phy_dev;
244 
245 	if (change & IFF_ALLMULTI)
246 		dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
247 }
248 
249 static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
250 {
251 	struct ipvl_dev *ipvlan = netdev_priv(dev);
252 
253 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
254 		bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
255 	} else {
256 		struct netdev_hw_addr *ha;
257 		DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
258 
259 		bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
260 		netdev_for_each_mc_addr(ha, dev)
261 			__set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
262 
263 		/* Turn-on broadcast bit irrespective of address family,
264 		 * since broadcast is deferred to a work-queue, hence no
265 		 * impact on fast-path processing.
266 		 */
267 		__set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
268 
269 		bitmap_copy(ipvlan->mac_filters, mc_filters,
270 			    IPVLAN_MAC_FILTER_SIZE);
271 	}
272 	dev_uc_sync(ipvlan->phy_dev, dev);
273 	dev_mc_sync(ipvlan->phy_dev, dev);
274 }
275 
276 static void ipvlan_get_stats64(struct net_device *dev,
277 			       struct rtnl_link_stats64 *s)
278 {
279 	struct ipvl_dev *ipvlan = netdev_priv(dev);
280 
281 	if (ipvlan->pcpu_stats) {
282 		struct ipvl_pcpu_stats *pcptr;
283 		u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
284 		u32 rx_errs = 0, tx_drps = 0;
285 		u32 strt;
286 		int idx;
287 
288 		for_each_possible_cpu(idx) {
289 			pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
290 			do {
291 				strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
292 				rx_pkts = pcptr->rx_pkts;
293 				rx_bytes = pcptr->rx_bytes;
294 				rx_mcast = pcptr->rx_mcast;
295 				tx_pkts = pcptr->tx_pkts;
296 				tx_bytes = pcptr->tx_bytes;
297 			} while (u64_stats_fetch_retry_irq(&pcptr->syncp,
298 							   strt));
299 
300 			s->rx_packets += rx_pkts;
301 			s->rx_bytes += rx_bytes;
302 			s->multicast += rx_mcast;
303 			s->tx_packets += tx_pkts;
304 			s->tx_bytes += tx_bytes;
305 
306 			/* u32 values are updated without syncp protection. */
307 			rx_errs += pcptr->rx_errs;
308 			tx_drps += pcptr->tx_drps;
309 		}
310 		s->rx_errors = rx_errs;
311 		s->rx_dropped = rx_errs;
312 		s->tx_dropped = tx_drps;
313 	}
314 }
315 
316 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
317 {
318 	struct ipvl_dev *ipvlan = netdev_priv(dev);
319 	struct net_device *phy_dev = ipvlan->phy_dev;
320 
321 	return vlan_vid_add(phy_dev, proto, vid);
322 }
323 
324 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
325 				   u16 vid)
326 {
327 	struct ipvl_dev *ipvlan = netdev_priv(dev);
328 	struct net_device *phy_dev = ipvlan->phy_dev;
329 
330 	vlan_vid_del(phy_dev, proto, vid);
331 	return 0;
332 }
333 
334 static int ipvlan_get_iflink(const struct net_device *dev)
335 {
336 	struct ipvl_dev *ipvlan = netdev_priv(dev);
337 
338 	return ipvlan->phy_dev->ifindex;
339 }
340 
341 static const struct net_device_ops ipvlan_netdev_ops = {
342 	.ndo_init		= ipvlan_init,
343 	.ndo_uninit		= ipvlan_uninit,
344 	.ndo_open		= ipvlan_open,
345 	.ndo_stop		= ipvlan_stop,
346 	.ndo_start_xmit		= ipvlan_start_xmit,
347 	.ndo_fix_features	= ipvlan_fix_features,
348 	.ndo_change_rx_flags	= ipvlan_change_rx_flags,
349 	.ndo_set_rx_mode	= ipvlan_set_multicast_mac_filter,
350 	.ndo_get_stats64	= ipvlan_get_stats64,
351 	.ndo_vlan_rx_add_vid	= ipvlan_vlan_rx_add_vid,
352 	.ndo_vlan_rx_kill_vid	= ipvlan_vlan_rx_kill_vid,
353 	.ndo_get_iflink		= ipvlan_get_iflink,
354 };
355 
356 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
357 			      unsigned short type, const void *daddr,
358 			      const void *saddr, unsigned len)
359 {
360 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
361 	struct net_device *phy_dev = ipvlan->phy_dev;
362 
363 	/* TODO Probably use a different field than dev_addr so that the
364 	 * mac-address on the virtual device is portable and can be carried
365 	 * while the packets use the mac-addr on the physical device.
366 	 */
367 	return dev_hard_header(skb, phy_dev, type, daddr,
368 			       saddr ? : phy_dev->dev_addr, len);
369 }
370 
371 static const struct header_ops ipvlan_header_ops = {
372 	.create  	= ipvlan_hard_header,
373 	.parse		= eth_header_parse,
374 	.cache		= eth_header_cache,
375 	.cache_update	= eth_header_cache_update,
376 };
377 
378 static void ipvlan_adjust_mtu(struct ipvl_dev *ipvlan, struct net_device *dev)
379 {
380 	ipvlan->dev->mtu = dev->mtu;
381 }
382 
383 static bool netif_is_ipvlan(const struct net_device *dev)
384 {
385 	/* both ipvlan and ipvtap devices use the same netdev_ops */
386 	return dev->netdev_ops == &ipvlan_netdev_ops;
387 }
388 
389 static int ipvlan_ethtool_get_link_ksettings(struct net_device *dev,
390 					     struct ethtool_link_ksettings *cmd)
391 {
392 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
393 
394 	return __ethtool_get_link_ksettings(ipvlan->phy_dev, cmd);
395 }
396 
397 static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
398 				       struct ethtool_drvinfo *drvinfo)
399 {
400 	strlcpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
401 	strlcpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
402 }
403 
404 static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
405 {
406 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
407 
408 	return ipvlan->msg_enable;
409 }
410 
411 static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
412 {
413 	struct ipvl_dev *ipvlan = netdev_priv(dev);
414 
415 	ipvlan->msg_enable = value;
416 }
417 
418 static const struct ethtool_ops ipvlan_ethtool_ops = {
419 	.get_link	= ethtool_op_get_link,
420 	.get_link_ksettings	= ipvlan_ethtool_get_link_ksettings,
421 	.get_drvinfo	= ipvlan_ethtool_get_drvinfo,
422 	.get_msglevel	= ipvlan_ethtool_get_msglevel,
423 	.set_msglevel	= ipvlan_ethtool_set_msglevel,
424 };
425 
426 static int ipvlan_nl_changelink(struct net_device *dev,
427 				struct nlattr *tb[], struct nlattr *data[],
428 				struct netlink_ext_ack *extack)
429 {
430 	struct ipvl_dev *ipvlan = netdev_priv(dev);
431 	struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
432 	int err = 0;
433 
434 	if (!data)
435 		return 0;
436 	if (!ns_capable(dev_net(ipvlan->phy_dev)->user_ns, CAP_NET_ADMIN))
437 		return -EPERM;
438 
439 	if (data[IFLA_IPVLAN_MODE]) {
440 		u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
441 
442 		err = ipvlan_set_port_mode(port, nmode, extack);
443 	}
444 
445 	if (!err && data[IFLA_IPVLAN_FLAGS]) {
446 		u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
447 
448 		if (flags & IPVLAN_F_PRIVATE)
449 			ipvlan_mark_private(port);
450 		else
451 			ipvlan_clear_private(port);
452 
453 		if (flags & IPVLAN_F_VEPA)
454 			ipvlan_mark_vepa(port);
455 		else
456 			ipvlan_clear_vepa(port);
457 	}
458 
459 	return err;
460 }
461 
462 static size_t ipvlan_nl_getsize(const struct net_device *dev)
463 {
464 	return (0
465 		+ nla_total_size(2) /* IFLA_IPVLAN_MODE */
466 		+ nla_total_size(2) /* IFLA_IPVLAN_FLAGS */
467 		);
468 }
469 
470 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[],
471 			      struct netlink_ext_ack *extack)
472 {
473 	if (!data)
474 		return 0;
475 
476 	if (data[IFLA_IPVLAN_MODE]) {
477 		u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
478 
479 		if (mode >= IPVLAN_MODE_MAX)
480 			return -EINVAL;
481 	}
482 	if (data[IFLA_IPVLAN_FLAGS]) {
483 		u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
484 
485 		/* Only two bits are used at this moment. */
486 		if (flags & ~(IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
487 			return -EINVAL;
488 		/* Also both flags can't be active at the same time. */
489 		if ((flags & (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA)) ==
490 		    (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
491 			return -EINVAL;
492 	}
493 
494 	return 0;
495 }
496 
497 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
498 			      const struct net_device *dev)
499 {
500 	struct ipvl_dev *ipvlan = netdev_priv(dev);
501 	struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
502 	int ret = -EINVAL;
503 
504 	if (!port)
505 		goto err;
506 
507 	ret = -EMSGSIZE;
508 	if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
509 		goto err;
510 	if (nla_put_u16(skb, IFLA_IPVLAN_FLAGS, port->flags))
511 		goto err;
512 
513 	return 0;
514 
515 err:
516 	return ret;
517 }
518 
519 int ipvlan_link_new(struct net *src_net, struct net_device *dev,
520 		    struct nlattr *tb[], struct nlattr *data[],
521 		    struct netlink_ext_ack *extack)
522 {
523 	struct ipvl_dev *ipvlan = netdev_priv(dev);
524 	struct ipvl_port *port;
525 	struct net_device *phy_dev;
526 	int err;
527 	u16 mode = IPVLAN_MODE_L3;
528 
529 	if (!tb[IFLA_LINK])
530 		return -EINVAL;
531 
532 	phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
533 	if (!phy_dev)
534 		return -ENODEV;
535 
536 	if (netif_is_ipvlan(phy_dev)) {
537 		struct ipvl_dev *tmp = netdev_priv(phy_dev);
538 
539 		phy_dev = tmp->phy_dev;
540 		if (!ns_capable(dev_net(phy_dev)->user_ns, CAP_NET_ADMIN))
541 			return -EPERM;
542 	} else if (!netif_is_ipvlan_port(phy_dev)) {
543 		/* Exit early if the underlying link is invalid or busy */
544 		if (phy_dev->type != ARPHRD_ETHER ||
545 		    phy_dev->flags & IFF_LOOPBACK) {
546 			netdev_err(phy_dev,
547 				   "Master is either lo or non-ether device\n");
548 			return -EINVAL;
549 		}
550 
551 		if (netdev_is_rx_handler_busy(phy_dev)) {
552 			netdev_err(phy_dev, "Device is already in use.\n");
553 			return -EBUSY;
554 		}
555 	}
556 
557 	ipvlan->phy_dev = phy_dev;
558 	ipvlan->dev = dev;
559 	ipvlan->sfeatures = IPVLAN_FEATURES;
560 	if (!tb[IFLA_MTU])
561 		ipvlan_adjust_mtu(ipvlan, phy_dev);
562 	INIT_LIST_HEAD(&ipvlan->addrs);
563 	spin_lock_init(&ipvlan->addrs_lock);
564 
565 	/* TODO Probably put random address here to be presented to the
566 	 * world but keep using the physical-dev address for the outgoing
567 	 * packets.
568 	 */
569 	memcpy(dev->dev_addr, phy_dev->dev_addr, ETH_ALEN);
570 
571 	dev->priv_flags |= IFF_NO_RX_HANDLER;
572 
573 	err = register_netdevice(dev);
574 	if (err < 0)
575 		return err;
576 
577 	/* ipvlan_init() would have created the port, if required */
578 	port = ipvlan_port_get_rtnl(phy_dev);
579 	ipvlan->port = port;
580 
581 	/* If the port-id base is at the MAX value, then wrap it around and
582 	 * begin from 0x1 again. This may be due to a busy system where lots
583 	 * of slaves are getting created and deleted.
584 	 */
585 	if (port->dev_id_start == 0xFFFE)
586 		port->dev_id_start = 0x1;
587 
588 	/* Since L2 address is shared among all IPvlan slaves including
589 	 * master, use unique 16 bit dev-ids to diffentiate among them.
590 	 * Assign IDs between 0x1 and 0xFFFE (used by the master) to each
591 	 * slave link [see addrconf_ifid_eui48()].
592 	 */
593 	err = ida_simple_get(&port->ida, port->dev_id_start, 0xFFFE,
594 			     GFP_KERNEL);
595 	if (err < 0)
596 		err = ida_simple_get(&port->ida, 0x1, port->dev_id_start,
597 				     GFP_KERNEL);
598 	if (err < 0)
599 		goto unregister_netdev;
600 	dev->dev_id = err;
601 
602 	/* Increment id-base to the next slot for the future assignment */
603 	port->dev_id_start = err + 1;
604 
605 	err = netdev_upper_dev_link(phy_dev, dev, extack);
606 	if (err)
607 		goto remove_ida;
608 
609 	/* Flags are per port and latest update overrides. User has
610 	 * to be consistent in setting it just like the mode attribute.
611 	 */
612 	if (data && data[IFLA_IPVLAN_FLAGS])
613 		port->flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
614 
615 	if (data && data[IFLA_IPVLAN_MODE])
616 		mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
617 
618 	err = ipvlan_set_port_mode(port, mode, extack);
619 	if (err)
620 		goto unlink_netdev;
621 
622 	list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
623 	netif_stacked_transfer_operstate(phy_dev, dev);
624 	return 0;
625 
626 unlink_netdev:
627 	netdev_upper_dev_unlink(phy_dev, dev);
628 remove_ida:
629 	ida_simple_remove(&port->ida, dev->dev_id);
630 unregister_netdev:
631 	unregister_netdevice(dev);
632 	return err;
633 }
634 EXPORT_SYMBOL_GPL(ipvlan_link_new);
635 
636 void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
637 {
638 	struct ipvl_dev *ipvlan = netdev_priv(dev);
639 	struct ipvl_addr *addr, *next;
640 
641 	spin_lock_bh(&ipvlan->addrs_lock);
642 	list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
643 		ipvlan_ht_addr_del(addr);
644 		list_del_rcu(&addr->anode);
645 		kfree_rcu(addr, rcu);
646 	}
647 	spin_unlock_bh(&ipvlan->addrs_lock);
648 
649 	ida_simple_remove(&ipvlan->port->ida, dev->dev_id);
650 	list_del_rcu(&ipvlan->pnode);
651 	unregister_netdevice_queue(dev, head);
652 	netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
653 }
654 EXPORT_SYMBOL_GPL(ipvlan_link_delete);
655 
656 void ipvlan_link_setup(struct net_device *dev)
657 {
658 	ether_setup(dev);
659 
660 	dev->max_mtu = ETH_MAX_MTU;
661 	dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
662 	dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
663 	dev->netdev_ops = &ipvlan_netdev_ops;
664 	dev->needs_free_netdev = true;
665 	dev->header_ops = &ipvlan_header_ops;
666 	dev->ethtool_ops = &ipvlan_ethtool_ops;
667 }
668 EXPORT_SYMBOL_GPL(ipvlan_link_setup);
669 
670 static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
671 {
672 	[IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
673 	[IFLA_IPVLAN_FLAGS] = { .type = NLA_U16 },
674 };
675 
676 static struct rtnl_link_ops ipvlan_link_ops = {
677 	.kind		= "ipvlan",
678 	.priv_size	= sizeof(struct ipvl_dev),
679 
680 	.setup		= ipvlan_link_setup,
681 	.newlink	= ipvlan_link_new,
682 	.dellink	= ipvlan_link_delete,
683 };
684 
685 int ipvlan_link_register(struct rtnl_link_ops *ops)
686 {
687 	ops->get_size	= ipvlan_nl_getsize;
688 	ops->policy	= ipvlan_nl_policy;
689 	ops->validate	= ipvlan_nl_validate;
690 	ops->fill_info	= ipvlan_nl_fillinfo;
691 	ops->changelink = ipvlan_nl_changelink;
692 	ops->maxtype	= IFLA_IPVLAN_MAX;
693 	return rtnl_link_register(ops);
694 }
695 EXPORT_SYMBOL_GPL(ipvlan_link_register);
696 
697 static int ipvlan_device_event(struct notifier_block *unused,
698 			       unsigned long event, void *ptr)
699 {
700 	struct netlink_ext_ack *extack = netdev_notifier_info_to_extack(ptr);
701 	struct netdev_notifier_pre_changeaddr_info *prechaddr_info;
702 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
703 	struct ipvl_dev *ipvlan, *next;
704 	struct ipvl_port *port;
705 	LIST_HEAD(lst_kill);
706 	int err;
707 
708 	if (!netif_is_ipvlan_port(dev))
709 		return NOTIFY_DONE;
710 
711 	port = ipvlan_port_get_rtnl(dev);
712 
713 	switch (event) {
714 	case NETDEV_CHANGE:
715 		list_for_each_entry(ipvlan, &port->ipvlans, pnode)
716 			netif_stacked_transfer_operstate(ipvlan->phy_dev,
717 							 ipvlan->dev);
718 		break;
719 
720 	case NETDEV_REGISTER: {
721 		struct net *oldnet, *newnet = dev_net(dev);
722 
723 		oldnet = read_pnet(&port->pnet);
724 		if (net_eq(newnet, oldnet))
725 			break;
726 
727 		write_pnet(&port->pnet, newnet);
728 
729 		ipvlan_migrate_l3s_hook(oldnet, newnet);
730 		break;
731 	}
732 	case NETDEV_UNREGISTER:
733 		if (dev->reg_state != NETREG_UNREGISTERING)
734 			break;
735 
736 		list_for_each_entry_safe(ipvlan, next, &port->ipvlans, pnode)
737 			ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
738 							    &lst_kill);
739 		unregister_netdevice_many(&lst_kill);
740 		break;
741 
742 	case NETDEV_FEAT_CHANGE:
743 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
744 			ipvlan->dev->features = dev->features & IPVLAN_FEATURES;
745 			ipvlan->dev->gso_max_size = dev->gso_max_size;
746 			ipvlan->dev->gso_max_segs = dev->gso_max_segs;
747 			netdev_features_change(ipvlan->dev);
748 		}
749 		break;
750 
751 	case NETDEV_CHANGEMTU:
752 		list_for_each_entry(ipvlan, &port->ipvlans, pnode)
753 			ipvlan_adjust_mtu(ipvlan, dev);
754 		break;
755 
756 	case NETDEV_PRE_CHANGEADDR:
757 		prechaddr_info = ptr;
758 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
759 			err = dev_pre_changeaddr_notify(ipvlan->dev,
760 						    prechaddr_info->dev_addr,
761 						    extack);
762 			if (err)
763 				return notifier_from_errno(err);
764 		}
765 		break;
766 
767 	case NETDEV_CHANGEADDR:
768 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
769 			ether_addr_copy(ipvlan->dev->dev_addr, dev->dev_addr);
770 			call_netdevice_notifiers(NETDEV_CHANGEADDR, ipvlan->dev);
771 		}
772 		break;
773 
774 	case NETDEV_PRE_TYPE_CHANGE:
775 		/* Forbid underlying device to change its type. */
776 		return NOTIFY_BAD;
777 	}
778 	return NOTIFY_DONE;
779 }
780 
781 /* the caller must held the addrs lock */
782 static int ipvlan_add_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
783 {
784 	struct ipvl_addr *addr;
785 
786 	addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
787 	if (!addr)
788 		return -ENOMEM;
789 
790 	addr->master = ipvlan;
791 	if (!is_v6) {
792 		memcpy(&addr->ip4addr, iaddr, sizeof(struct in_addr));
793 		addr->atype = IPVL_IPV4;
794 #if IS_ENABLED(CONFIG_IPV6)
795 	} else {
796 		memcpy(&addr->ip6addr, iaddr, sizeof(struct in6_addr));
797 		addr->atype = IPVL_IPV6;
798 #endif
799 	}
800 
801 	list_add_tail_rcu(&addr->anode, &ipvlan->addrs);
802 
803 	/* If the interface is not up, the address will be added to the hash
804 	 * list by ipvlan_open.
805 	 */
806 	if (netif_running(ipvlan->dev))
807 		ipvlan_ht_addr_add(ipvlan, addr);
808 
809 	return 0;
810 }
811 
812 static void ipvlan_del_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
813 {
814 	struct ipvl_addr *addr;
815 
816 	spin_lock_bh(&ipvlan->addrs_lock);
817 	addr = ipvlan_find_addr(ipvlan, iaddr, is_v6);
818 	if (!addr) {
819 		spin_unlock_bh(&ipvlan->addrs_lock);
820 		return;
821 	}
822 
823 	ipvlan_ht_addr_del(addr);
824 	list_del_rcu(&addr->anode);
825 	spin_unlock_bh(&ipvlan->addrs_lock);
826 	kfree_rcu(addr, rcu);
827 }
828 
829 static bool ipvlan_is_valid_dev(const struct net_device *dev)
830 {
831 	struct ipvl_dev *ipvlan = netdev_priv(dev);
832 
833 	if (!netif_is_ipvlan(dev))
834 		return false;
835 
836 	if (!ipvlan || !ipvlan->port)
837 		return false;
838 
839 	return true;
840 }
841 
842 #if IS_ENABLED(CONFIG_IPV6)
843 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
844 {
845 	int ret = -EINVAL;
846 
847 	spin_lock_bh(&ipvlan->addrs_lock);
848 	if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true))
849 		netif_err(ipvlan, ifup, ipvlan->dev,
850 			  "Failed to add IPv6=%pI6c addr for %s intf\n",
851 			  ip6_addr, ipvlan->dev->name);
852 	else
853 		ret = ipvlan_add_addr(ipvlan, ip6_addr, true);
854 	spin_unlock_bh(&ipvlan->addrs_lock);
855 	return ret;
856 }
857 
858 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
859 {
860 	return ipvlan_del_addr(ipvlan, ip6_addr, true);
861 }
862 
863 static int ipvlan_addr6_event(struct notifier_block *unused,
864 			      unsigned long event, void *ptr)
865 {
866 	struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
867 	struct net_device *dev = (struct net_device *)if6->idev->dev;
868 	struct ipvl_dev *ipvlan = netdev_priv(dev);
869 
870 	if (!ipvlan_is_valid_dev(dev))
871 		return NOTIFY_DONE;
872 
873 	switch (event) {
874 	case NETDEV_UP:
875 		if (ipvlan_add_addr6(ipvlan, &if6->addr))
876 			return NOTIFY_BAD;
877 		break;
878 
879 	case NETDEV_DOWN:
880 		ipvlan_del_addr6(ipvlan, &if6->addr);
881 		break;
882 	}
883 
884 	return NOTIFY_OK;
885 }
886 
887 static int ipvlan_addr6_validator_event(struct notifier_block *unused,
888 					unsigned long event, void *ptr)
889 {
890 	struct in6_validator_info *i6vi = (struct in6_validator_info *)ptr;
891 	struct net_device *dev = (struct net_device *)i6vi->i6vi_dev->dev;
892 	struct ipvl_dev *ipvlan = netdev_priv(dev);
893 
894 	if (!ipvlan_is_valid_dev(dev))
895 		return NOTIFY_DONE;
896 
897 	switch (event) {
898 	case NETDEV_UP:
899 		if (ipvlan_addr_busy(ipvlan->port, &i6vi->i6vi_addr, true)) {
900 			NL_SET_ERR_MSG(i6vi->extack,
901 				       "Address already assigned to an ipvlan device");
902 			return notifier_from_errno(-EADDRINUSE);
903 		}
904 		break;
905 	}
906 
907 	return NOTIFY_OK;
908 }
909 #endif
910 
911 static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
912 {
913 	int ret = -EINVAL;
914 
915 	spin_lock_bh(&ipvlan->addrs_lock);
916 	if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false))
917 		netif_err(ipvlan, ifup, ipvlan->dev,
918 			  "Failed to add IPv4=%pI4 on %s intf.\n",
919 			  ip4_addr, ipvlan->dev->name);
920 	else
921 		ret = ipvlan_add_addr(ipvlan, ip4_addr, false);
922 	spin_unlock_bh(&ipvlan->addrs_lock);
923 	return ret;
924 }
925 
926 static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
927 {
928 	return ipvlan_del_addr(ipvlan, ip4_addr, false);
929 }
930 
931 static int ipvlan_addr4_event(struct notifier_block *unused,
932 			      unsigned long event, void *ptr)
933 {
934 	struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
935 	struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
936 	struct ipvl_dev *ipvlan = netdev_priv(dev);
937 	struct in_addr ip4_addr;
938 
939 	if (!ipvlan_is_valid_dev(dev))
940 		return NOTIFY_DONE;
941 
942 	switch (event) {
943 	case NETDEV_UP:
944 		ip4_addr.s_addr = if4->ifa_address;
945 		if (ipvlan_add_addr4(ipvlan, &ip4_addr))
946 			return NOTIFY_BAD;
947 		break;
948 
949 	case NETDEV_DOWN:
950 		ip4_addr.s_addr = if4->ifa_address;
951 		ipvlan_del_addr4(ipvlan, &ip4_addr);
952 		break;
953 	}
954 
955 	return NOTIFY_OK;
956 }
957 
958 static int ipvlan_addr4_validator_event(struct notifier_block *unused,
959 					unsigned long event, void *ptr)
960 {
961 	struct in_validator_info *ivi = (struct in_validator_info *)ptr;
962 	struct net_device *dev = (struct net_device *)ivi->ivi_dev->dev;
963 	struct ipvl_dev *ipvlan = netdev_priv(dev);
964 
965 	if (!ipvlan_is_valid_dev(dev))
966 		return NOTIFY_DONE;
967 
968 	switch (event) {
969 	case NETDEV_UP:
970 		if (ipvlan_addr_busy(ipvlan->port, &ivi->ivi_addr, false)) {
971 			NL_SET_ERR_MSG(ivi->extack,
972 				       "Address already assigned to an ipvlan device");
973 			return notifier_from_errno(-EADDRINUSE);
974 		}
975 		break;
976 	}
977 
978 	return NOTIFY_OK;
979 }
980 
981 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
982 	.notifier_call = ipvlan_addr4_event,
983 };
984 
985 static struct notifier_block ipvlan_addr4_vtor_notifier_block __read_mostly = {
986 	.notifier_call = ipvlan_addr4_validator_event,
987 };
988 
989 static struct notifier_block ipvlan_notifier_block __read_mostly = {
990 	.notifier_call = ipvlan_device_event,
991 };
992 
993 #if IS_ENABLED(CONFIG_IPV6)
994 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
995 	.notifier_call = ipvlan_addr6_event,
996 };
997 
998 static struct notifier_block ipvlan_addr6_vtor_notifier_block __read_mostly = {
999 	.notifier_call = ipvlan_addr6_validator_event,
1000 };
1001 #endif
1002 
1003 static int __init ipvlan_init_module(void)
1004 {
1005 	int err;
1006 
1007 	ipvlan_init_secret();
1008 	register_netdevice_notifier(&ipvlan_notifier_block);
1009 #if IS_ENABLED(CONFIG_IPV6)
1010 	register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1011 	register_inet6addr_validator_notifier(
1012 	    &ipvlan_addr6_vtor_notifier_block);
1013 #endif
1014 	register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1015 	register_inetaddr_validator_notifier(&ipvlan_addr4_vtor_notifier_block);
1016 
1017 	err = ipvlan_l3s_init();
1018 	if (err < 0)
1019 		goto error;
1020 
1021 	err = ipvlan_link_register(&ipvlan_link_ops);
1022 	if (err < 0) {
1023 		ipvlan_l3s_cleanup();
1024 		goto error;
1025 	}
1026 
1027 	return 0;
1028 error:
1029 	unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1030 	unregister_inetaddr_validator_notifier(
1031 	    &ipvlan_addr4_vtor_notifier_block);
1032 #if IS_ENABLED(CONFIG_IPV6)
1033 	unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1034 	unregister_inet6addr_validator_notifier(
1035 	    &ipvlan_addr6_vtor_notifier_block);
1036 #endif
1037 	unregister_netdevice_notifier(&ipvlan_notifier_block);
1038 	return err;
1039 }
1040 
1041 static void __exit ipvlan_cleanup_module(void)
1042 {
1043 	rtnl_link_unregister(&ipvlan_link_ops);
1044 	ipvlan_l3s_cleanup();
1045 	unregister_netdevice_notifier(&ipvlan_notifier_block);
1046 	unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1047 	unregister_inetaddr_validator_notifier(
1048 	    &ipvlan_addr4_vtor_notifier_block);
1049 #if IS_ENABLED(CONFIG_IPV6)
1050 	unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1051 	unregister_inet6addr_validator_notifier(
1052 	    &ipvlan_addr6_vtor_notifier_block);
1053 #endif
1054 }
1055 
1056 module_init(ipvlan_init_module);
1057 module_exit(ipvlan_cleanup_module);
1058 
1059 MODULE_LICENSE("GPL");
1060 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
1061 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
1062 MODULE_ALIAS_RTNL_LINK("ipvlan");
1063