xref: /openbmc/linux/drivers/net/ipvlan/ipvlan_main.c (revision dc6a81c3)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Copyright (c) 2014 Mahesh Bandewar <maheshb@google.com>
3  */
4 
5 #include "ipvlan.h"
6 
7 static int ipvlan_set_port_mode(struct ipvl_port *port, u16 nval,
8 				struct netlink_ext_ack *extack)
9 {
10 	struct ipvl_dev *ipvlan;
11 	unsigned int flags;
12 	int err;
13 
14 	ASSERT_RTNL();
15 	if (port->mode != nval) {
16 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
17 			flags = ipvlan->dev->flags;
18 			if (nval == IPVLAN_MODE_L3 || nval == IPVLAN_MODE_L3S) {
19 				err = dev_change_flags(ipvlan->dev,
20 						       flags | IFF_NOARP,
21 						       extack);
22 			} else {
23 				err = dev_change_flags(ipvlan->dev,
24 						       flags & ~IFF_NOARP,
25 						       extack);
26 			}
27 			if (unlikely(err))
28 				goto fail;
29 		}
30 		if (nval == IPVLAN_MODE_L3S) {
31 			/* New mode is L3S */
32 			err = ipvlan_l3s_register(port);
33 			if (err)
34 				goto fail;
35 		} else if (port->mode == IPVLAN_MODE_L3S) {
36 			/* Old mode was L3S */
37 			ipvlan_l3s_unregister(port);
38 		}
39 		port->mode = nval;
40 	}
41 	return 0;
42 
43 fail:
44 	/* Undo the flags changes that have been done so far. */
45 	list_for_each_entry_continue_reverse(ipvlan, &port->ipvlans, pnode) {
46 		flags = ipvlan->dev->flags;
47 		if (port->mode == IPVLAN_MODE_L3 ||
48 		    port->mode == IPVLAN_MODE_L3S)
49 			dev_change_flags(ipvlan->dev, flags | IFF_NOARP,
50 					 NULL);
51 		else
52 			dev_change_flags(ipvlan->dev, flags & ~IFF_NOARP,
53 					 NULL);
54 	}
55 
56 	return err;
57 }
58 
59 static int ipvlan_port_create(struct net_device *dev)
60 {
61 	struct ipvl_port *port;
62 	int err, idx;
63 
64 	port = kzalloc(sizeof(struct ipvl_port), GFP_KERNEL);
65 	if (!port)
66 		return -ENOMEM;
67 
68 	write_pnet(&port->pnet, dev_net(dev));
69 	port->dev = dev;
70 	port->mode = IPVLAN_MODE_L3;
71 	INIT_LIST_HEAD(&port->ipvlans);
72 	for (idx = 0; idx < IPVLAN_HASH_SIZE; idx++)
73 		INIT_HLIST_HEAD(&port->hlhead[idx]);
74 
75 	skb_queue_head_init(&port->backlog);
76 	INIT_WORK(&port->wq, ipvlan_process_multicast);
77 	ida_init(&port->ida);
78 	port->dev_id_start = 1;
79 
80 	err = netdev_rx_handler_register(dev, ipvlan_handle_frame, port);
81 	if (err)
82 		goto err;
83 
84 	return 0;
85 
86 err:
87 	kfree(port);
88 	return err;
89 }
90 
91 static void ipvlan_port_destroy(struct net_device *dev)
92 {
93 	struct ipvl_port *port = ipvlan_port_get_rtnl(dev);
94 	struct sk_buff *skb;
95 
96 	if (port->mode == IPVLAN_MODE_L3S)
97 		ipvlan_l3s_unregister(port);
98 	netdev_rx_handler_unregister(dev);
99 	cancel_work_sync(&port->wq);
100 	while ((skb = __skb_dequeue(&port->backlog)) != NULL) {
101 		if (skb->dev)
102 			dev_put(skb->dev);
103 		kfree_skb(skb);
104 	}
105 	ida_destroy(&port->ida);
106 	kfree(port);
107 }
108 
109 #define IPVLAN_FEATURES \
110 	(NETIF_F_SG | NETIF_F_CSUM_MASK | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
111 	 NETIF_F_GSO | NETIF_F_ALL_TSO | NETIF_F_GSO_ROBUST | \
112 	 NETIF_F_GRO | NETIF_F_RXCSUM | \
113 	 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
114 
115 #define IPVLAN_STATE_MASK \
116 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
117 
118 static int ipvlan_init(struct net_device *dev)
119 {
120 	struct ipvl_dev *ipvlan = netdev_priv(dev);
121 	struct net_device *phy_dev = ipvlan->phy_dev;
122 	struct ipvl_port *port;
123 	int err;
124 
125 	dev->state = (dev->state & ~IPVLAN_STATE_MASK) |
126 		     (phy_dev->state & IPVLAN_STATE_MASK);
127 	dev->features = phy_dev->features & IPVLAN_FEATURES;
128 	dev->features |= NETIF_F_LLTX | NETIF_F_VLAN_CHALLENGED;
129 	dev->hw_enc_features |= dev->features;
130 	dev->gso_max_size = phy_dev->gso_max_size;
131 	dev->gso_max_segs = phy_dev->gso_max_segs;
132 	dev->hard_header_len = phy_dev->hard_header_len;
133 
134 	ipvlan->pcpu_stats = netdev_alloc_pcpu_stats(struct ipvl_pcpu_stats);
135 	if (!ipvlan->pcpu_stats)
136 		return -ENOMEM;
137 
138 	if (!netif_is_ipvlan_port(phy_dev)) {
139 		err = ipvlan_port_create(phy_dev);
140 		if (err < 0) {
141 			free_percpu(ipvlan->pcpu_stats);
142 			return err;
143 		}
144 	}
145 	port = ipvlan_port_get_rtnl(phy_dev);
146 	port->count += 1;
147 	return 0;
148 }
149 
150 static void ipvlan_uninit(struct net_device *dev)
151 {
152 	struct ipvl_dev *ipvlan = netdev_priv(dev);
153 	struct net_device *phy_dev = ipvlan->phy_dev;
154 	struct ipvl_port *port;
155 
156 	free_percpu(ipvlan->pcpu_stats);
157 
158 	port = ipvlan_port_get_rtnl(phy_dev);
159 	port->count -= 1;
160 	if (!port->count)
161 		ipvlan_port_destroy(port->dev);
162 }
163 
164 static int ipvlan_open(struct net_device *dev)
165 {
166 	struct ipvl_dev *ipvlan = netdev_priv(dev);
167 	struct net_device *phy_dev = ipvlan->phy_dev;
168 	struct ipvl_addr *addr;
169 
170 	if (ipvlan->port->mode == IPVLAN_MODE_L3 ||
171 	    ipvlan->port->mode == IPVLAN_MODE_L3S)
172 		dev->flags |= IFF_NOARP;
173 	else
174 		dev->flags &= ~IFF_NOARP;
175 
176 	rcu_read_lock();
177 	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
178 		ipvlan_ht_addr_add(ipvlan, addr);
179 	rcu_read_unlock();
180 
181 	return dev_uc_add(phy_dev, phy_dev->dev_addr);
182 }
183 
184 static int ipvlan_stop(struct net_device *dev)
185 {
186 	struct ipvl_dev *ipvlan = netdev_priv(dev);
187 	struct net_device *phy_dev = ipvlan->phy_dev;
188 	struct ipvl_addr *addr;
189 
190 	dev_uc_unsync(phy_dev, dev);
191 	dev_mc_unsync(phy_dev, dev);
192 
193 	dev_uc_del(phy_dev, phy_dev->dev_addr);
194 
195 	rcu_read_lock();
196 	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
197 		ipvlan_ht_addr_del(addr);
198 	rcu_read_unlock();
199 
200 	return 0;
201 }
202 
203 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
204 				     struct net_device *dev)
205 {
206 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
207 	int skblen = skb->len;
208 	int ret;
209 
210 	ret = ipvlan_queue_xmit(skb, dev);
211 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
212 		struct ipvl_pcpu_stats *pcptr;
213 
214 		pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
215 
216 		u64_stats_update_begin(&pcptr->syncp);
217 		pcptr->tx_pkts++;
218 		pcptr->tx_bytes += skblen;
219 		u64_stats_update_end(&pcptr->syncp);
220 	} else {
221 		this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
222 	}
223 	return ret;
224 }
225 
226 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
227 					     netdev_features_t features)
228 {
229 	struct ipvl_dev *ipvlan = netdev_priv(dev);
230 
231 	return features & (ipvlan->sfeatures | ~IPVLAN_FEATURES);
232 }
233 
234 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
235 {
236 	struct ipvl_dev *ipvlan = netdev_priv(dev);
237 	struct net_device *phy_dev = ipvlan->phy_dev;
238 
239 	if (change & IFF_ALLMULTI)
240 		dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
241 }
242 
243 static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
244 {
245 	struct ipvl_dev *ipvlan = netdev_priv(dev);
246 
247 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
248 		bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
249 	} else {
250 		struct netdev_hw_addr *ha;
251 		DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
252 
253 		bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
254 		netdev_for_each_mc_addr(ha, dev)
255 			__set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
256 
257 		/* Turn-on broadcast bit irrespective of address family,
258 		 * since broadcast is deferred to a work-queue, hence no
259 		 * impact on fast-path processing.
260 		 */
261 		__set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
262 
263 		bitmap_copy(ipvlan->mac_filters, mc_filters,
264 			    IPVLAN_MAC_FILTER_SIZE);
265 	}
266 	dev_uc_sync(ipvlan->phy_dev, dev);
267 	dev_mc_sync(ipvlan->phy_dev, dev);
268 }
269 
270 static void ipvlan_get_stats64(struct net_device *dev,
271 			       struct rtnl_link_stats64 *s)
272 {
273 	struct ipvl_dev *ipvlan = netdev_priv(dev);
274 
275 	if (ipvlan->pcpu_stats) {
276 		struct ipvl_pcpu_stats *pcptr;
277 		u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
278 		u32 rx_errs = 0, tx_drps = 0;
279 		u32 strt;
280 		int idx;
281 
282 		for_each_possible_cpu(idx) {
283 			pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
284 			do {
285 				strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
286 				rx_pkts = pcptr->rx_pkts;
287 				rx_bytes = pcptr->rx_bytes;
288 				rx_mcast = pcptr->rx_mcast;
289 				tx_pkts = pcptr->tx_pkts;
290 				tx_bytes = pcptr->tx_bytes;
291 			} while (u64_stats_fetch_retry_irq(&pcptr->syncp,
292 							   strt));
293 
294 			s->rx_packets += rx_pkts;
295 			s->rx_bytes += rx_bytes;
296 			s->multicast += rx_mcast;
297 			s->tx_packets += tx_pkts;
298 			s->tx_bytes += tx_bytes;
299 
300 			/* u32 values are updated without syncp protection. */
301 			rx_errs += pcptr->rx_errs;
302 			tx_drps += pcptr->tx_drps;
303 		}
304 		s->rx_errors = rx_errs;
305 		s->rx_dropped = rx_errs;
306 		s->tx_dropped = tx_drps;
307 	}
308 }
309 
310 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
311 {
312 	struct ipvl_dev *ipvlan = netdev_priv(dev);
313 	struct net_device *phy_dev = ipvlan->phy_dev;
314 
315 	return vlan_vid_add(phy_dev, proto, vid);
316 }
317 
318 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
319 				   u16 vid)
320 {
321 	struct ipvl_dev *ipvlan = netdev_priv(dev);
322 	struct net_device *phy_dev = ipvlan->phy_dev;
323 
324 	vlan_vid_del(phy_dev, proto, vid);
325 	return 0;
326 }
327 
328 static int ipvlan_get_iflink(const struct net_device *dev)
329 {
330 	struct ipvl_dev *ipvlan = netdev_priv(dev);
331 
332 	return ipvlan->phy_dev->ifindex;
333 }
334 
335 static const struct net_device_ops ipvlan_netdev_ops = {
336 	.ndo_init		= ipvlan_init,
337 	.ndo_uninit		= ipvlan_uninit,
338 	.ndo_open		= ipvlan_open,
339 	.ndo_stop		= ipvlan_stop,
340 	.ndo_start_xmit		= ipvlan_start_xmit,
341 	.ndo_fix_features	= ipvlan_fix_features,
342 	.ndo_change_rx_flags	= ipvlan_change_rx_flags,
343 	.ndo_set_rx_mode	= ipvlan_set_multicast_mac_filter,
344 	.ndo_get_stats64	= ipvlan_get_stats64,
345 	.ndo_vlan_rx_add_vid	= ipvlan_vlan_rx_add_vid,
346 	.ndo_vlan_rx_kill_vid	= ipvlan_vlan_rx_kill_vid,
347 	.ndo_get_iflink		= ipvlan_get_iflink,
348 };
349 
350 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
351 			      unsigned short type, const void *daddr,
352 			      const void *saddr, unsigned len)
353 {
354 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
355 	struct net_device *phy_dev = ipvlan->phy_dev;
356 
357 	/* TODO Probably use a different field than dev_addr so that the
358 	 * mac-address on the virtual device is portable and can be carried
359 	 * while the packets use the mac-addr on the physical device.
360 	 */
361 	return dev_hard_header(skb, phy_dev, type, daddr,
362 			       saddr ? : phy_dev->dev_addr, len);
363 }
364 
365 static const struct header_ops ipvlan_header_ops = {
366 	.create  	= ipvlan_hard_header,
367 	.parse		= eth_header_parse,
368 	.cache		= eth_header_cache,
369 	.cache_update	= eth_header_cache_update,
370 };
371 
372 static void ipvlan_adjust_mtu(struct ipvl_dev *ipvlan, struct net_device *dev)
373 {
374 	ipvlan->dev->mtu = dev->mtu;
375 }
376 
377 static bool netif_is_ipvlan(const struct net_device *dev)
378 {
379 	/* both ipvlan and ipvtap devices use the same netdev_ops */
380 	return dev->netdev_ops == &ipvlan_netdev_ops;
381 }
382 
383 static int ipvlan_ethtool_get_link_ksettings(struct net_device *dev,
384 					     struct ethtool_link_ksettings *cmd)
385 {
386 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
387 
388 	return __ethtool_get_link_ksettings(ipvlan->phy_dev, cmd);
389 }
390 
391 static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
392 				       struct ethtool_drvinfo *drvinfo)
393 {
394 	strlcpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
395 	strlcpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
396 }
397 
398 static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
399 {
400 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
401 
402 	return ipvlan->msg_enable;
403 }
404 
405 static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
406 {
407 	struct ipvl_dev *ipvlan = netdev_priv(dev);
408 
409 	ipvlan->msg_enable = value;
410 }
411 
412 static const struct ethtool_ops ipvlan_ethtool_ops = {
413 	.get_link	= ethtool_op_get_link,
414 	.get_link_ksettings	= ipvlan_ethtool_get_link_ksettings,
415 	.get_drvinfo	= ipvlan_ethtool_get_drvinfo,
416 	.get_msglevel	= ipvlan_ethtool_get_msglevel,
417 	.set_msglevel	= ipvlan_ethtool_set_msglevel,
418 };
419 
420 static int ipvlan_nl_changelink(struct net_device *dev,
421 				struct nlattr *tb[], struct nlattr *data[],
422 				struct netlink_ext_ack *extack)
423 {
424 	struct ipvl_dev *ipvlan = netdev_priv(dev);
425 	struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
426 	int err = 0;
427 
428 	if (!data)
429 		return 0;
430 	if (!ns_capable(dev_net(ipvlan->phy_dev)->user_ns, CAP_NET_ADMIN))
431 		return -EPERM;
432 
433 	if (data[IFLA_IPVLAN_MODE]) {
434 		u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
435 
436 		err = ipvlan_set_port_mode(port, nmode, extack);
437 	}
438 
439 	if (!err && data[IFLA_IPVLAN_FLAGS]) {
440 		u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
441 
442 		if (flags & IPVLAN_F_PRIVATE)
443 			ipvlan_mark_private(port);
444 		else
445 			ipvlan_clear_private(port);
446 
447 		if (flags & IPVLAN_F_VEPA)
448 			ipvlan_mark_vepa(port);
449 		else
450 			ipvlan_clear_vepa(port);
451 	}
452 
453 	return err;
454 }
455 
456 static size_t ipvlan_nl_getsize(const struct net_device *dev)
457 {
458 	return (0
459 		+ nla_total_size(2) /* IFLA_IPVLAN_MODE */
460 		+ nla_total_size(2) /* IFLA_IPVLAN_FLAGS */
461 		);
462 }
463 
464 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[],
465 			      struct netlink_ext_ack *extack)
466 {
467 	if (!data)
468 		return 0;
469 
470 	if (data[IFLA_IPVLAN_MODE]) {
471 		u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
472 
473 		if (mode >= IPVLAN_MODE_MAX)
474 			return -EINVAL;
475 	}
476 	if (data[IFLA_IPVLAN_FLAGS]) {
477 		u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
478 
479 		/* Only two bits are used at this moment. */
480 		if (flags & ~(IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
481 			return -EINVAL;
482 		/* Also both flags can't be active at the same time. */
483 		if ((flags & (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA)) ==
484 		    (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
485 			return -EINVAL;
486 	}
487 
488 	return 0;
489 }
490 
491 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
492 			      const struct net_device *dev)
493 {
494 	struct ipvl_dev *ipvlan = netdev_priv(dev);
495 	struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
496 	int ret = -EINVAL;
497 
498 	if (!port)
499 		goto err;
500 
501 	ret = -EMSGSIZE;
502 	if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
503 		goto err;
504 	if (nla_put_u16(skb, IFLA_IPVLAN_FLAGS, port->flags))
505 		goto err;
506 
507 	return 0;
508 
509 err:
510 	return ret;
511 }
512 
513 int ipvlan_link_new(struct net *src_net, struct net_device *dev,
514 		    struct nlattr *tb[], struct nlattr *data[],
515 		    struct netlink_ext_ack *extack)
516 {
517 	struct ipvl_dev *ipvlan = netdev_priv(dev);
518 	struct ipvl_port *port;
519 	struct net_device *phy_dev;
520 	int err;
521 	u16 mode = IPVLAN_MODE_L3;
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 		if (!ns_capable(dev_net(phy_dev)->user_ns, CAP_NET_ADMIN))
535 			return -EPERM;
536 	} else if (!netif_is_ipvlan_port(phy_dev)) {
537 		/* Exit early if the underlying link is invalid or busy */
538 		if (phy_dev->type != ARPHRD_ETHER ||
539 		    phy_dev->flags & IFF_LOOPBACK) {
540 			netdev_err(phy_dev,
541 				   "Master is either lo or non-ether device\n");
542 			return -EINVAL;
543 		}
544 
545 		if (netdev_is_rx_handler_busy(phy_dev)) {
546 			netdev_err(phy_dev, "Device is already in use.\n");
547 			return -EBUSY;
548 		}
549 	}
550 
551 	ipvlan->phy_dev = phy_dev;
552 	ipvlan->dev = dev;
553 	ipvlan->sfeatures = IPVLAN_FEATURES;
554 	if (!tb[IFLA_MTU])
555 		ipvlan_adjust_mtu(ipvlan, phy_dev);
556 	INIT_LIST_HEAD(&ipvlan->addrs);
557 	spin_lock_init(&ipvlan->addrs_lock);
558 
559 	/* TODO Probably put random address here to be presented to the
560 	 * world but keep using the physical-dev address for the outgoing
561 	 * packets.
562 	 */
563 	memcpy(dev->dev_addr, phy_dev->dev_addr, ETH_ALEN);
564 
565 	dev->priv_flags |= IFF_NO_RX_HANDLER;
566 
567 	err = register_netdevice(dev);
568 	if (err < 0)
569 		return err;
570 
571 	/* ipvlan_init() would have created the port, if required */
572 	port = ipvlan_port_get_rtnl(phy_dev);
573 	ipvlan->port = port;
574 
575 	/* If the port-id base is at the MAX value, then wrap it around and
576 	 * begin from 0x1 again. This may be due to a busy system where lots
577 	 * of slaves are getting created and deleted.
578 	 */
579 	if (port->dev_id_start == 0xFFFE)
580 		port->dev_id_start = 0x1;
581 
582 	/* Since L2 address is shared among all IPvlan slaves including
583 	 * master, use unique 16 bit dev-ids to diffentiate among them.
584 	 * Assign IDs between 0x1 and 0xFFFE (used by the master) to each
585 	 * slave link [see addrconf_ifid_eui48()].
586 	 */
587 	err = ida_simple_get(&port->ida, port->dev_id_start, 0xFFFE,
588 			     GFP_KERNEL);
589 	if (err < 0)
590 		err = ida_simple_get(&port->ida, 0x1, port->dev_id_start,
591 				     GFP_KERNEL);
592 	if (err < 0)
593 		goto unregister_netdev;
594 	dev->dev_id = err;
595 
596 	/* Increment id-base to the next slot for the future assignment */
597 	port->dev_id_start = err + 1;
598 
599 	err = netdev_upper_dev_link(phy_dev, dev, extack);
600 	if (err)
601 		goto remove_ida;
602 
603 	/* Flags are per port and latest update overrides. User has
604 	 * to be consistent in setting it just like the mode attribute.
605 	 */
606 	if (data && data[IFLA_IPVLAN_FLAGS])
607 		port->flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
608 
609 	if (data && data[IFLA_IPVLAN_MODE])
610 		mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
611 
612 	err = ipvlan_set_port_mode(port, mode, extack);
613 	if (err)
614 		goto unlink_netdev;
615 
616 	list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
617 	netif_stacked_transfer_operstate(phy_dev, dev);
618 	return 0;
619 
620 unlink_netdev:
621 	netdev_upper_dev_unlink(phy_dev, dev);
622 remove_ida:
623 	ida_simple_remove(&port->ida, dev->dev_id);
624 unregister_netdev:
625 	unregister_netdevice(dev);
626 	return err;
627 }
628 EXPORT_SYMBOL_GPL(ipvlan_link_new);
629 
630 void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
631 {
632 	struct ipvl_dev *ipvlan = netdev_priv(dev);
633 	struct ipvl_addr *addr, *next;
634 
635 	spin_lock_bh(&ipvlan->addrs_lock);
636 	list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
637 		ipvlan_ht_addr_del(addr);
638 		list_del_rcu(&addr->anode);
639 		kfree_rcu(addr, rcu);
640 	}
641 	spin_unlock_bh(&ipvlan->addrs_lock);
642 
643 	ida_simple_remove(&ipvlan->port->ida, dev->dev_id);
644 	list_del_rcu(&ipvlan->pnode);
645 	unregister_netdevice_queue(dev, head);
646 	netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
647 }
648 EXPORT_SYMBOL_GPL(ipvlan_link_delete);
649 
650 void ipvlan_link_setup(struct net_device *dev)
651 {
652 	ether_setup(dev);
653 
654 	dev->max_mtu = ETH_MAX_MTU;
655 	dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
656 	dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
657 	dev->netdev_ops = &ipvlan_netdev_ops;
658 	dev->needs_free_netdev = true;
659 	dev->header_ops = &ipvlan_header_ops;
660 	dev->ethtool_ops = &ipvlan_ethtool_ops;
661 }
662 EXPORT_SYMBOL_GPL(ipvlan_link_setup);
663 
664 static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
665 {
666 	[IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
667 	[IFLA_IPVLAN_FLAGS] = { .type = NLA_U16 },
668 };
669 
670 static struct rtnl_link_ops ipvlan_link_ops = {
671 	.kind		= "ipvlan",
672 	.priv_size	= sizeof(struct ipvl_dev),
673 
674 	.setup		= ipvlan_link_setup,
675 	.newlink	= ipvlan_link_new,
676 	.dellink	= ipvlan_link_delete,
677 };
678 
679 int ipvlan_link_register(struct rtnl_link_ops *ops)
680 {
681 	ops->get_size	= ipvlan_nl_getsize;
682 	ops->policy	= ipvlan_nl_policy;
683 	ops->validate	= ipvlan_nl_validate;
684 	ops->fill_info	= ipvlan_nl_fillinfo;
685 	ops->changelink = ipvlan_nl_changelink;
686 	ops->maxtype	= IFLA_IPVLAN_MAX;
687 	return rtnl_link_register(ops);
688 }
689 EXPORT_SYMBOL_GPL(ipvlan_link_register);
690 
691 static int ipvlan_device_event(struct notifier_block *unused,
692 			       unsigned long event, void *ptr)
693 {
694 	struct netlink_ext_ack *extack = netdev_notifier_info_to_extack(ptr);
695 	struct netdev_notifier_pre_changeaddr_info *prechaddr_info;
696 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
697 	struct ipvl_dev *ipvlan, *next;
698 	struct ipvl_port *port;
699 	LIST_HEAD(lst_kill);
700 	int err;
701 
702 	if (!netif_is_ipvlan_port(dev))
703 		return NOTIFY_DONE;
704 
705 	port = ipvlan_port_get_rtnl(dev);
706 
707 	switch (event) {
708 	case NETDEV_CHANGE:
709 		list_for_each_entry(ipvlan, &port->ipvlans, pnode)
710 			netif_stacked_transfer_operstate(ipvlan->phy_dev,
711 							 ipvlan->dev);
712 		break;
713 
714 	case NETDEV_REGISTER: {
715 		struct net *oldnet, *newnet = dev_net(dev);
716 
717 		oldnet = read_pnet(&port->pnet);
718 		if (net_eq(newnet, oldnet))
719 			break;
720 
721 		write_pnet(&port->pnet, newnet);
722 
723 		ipvlan_migrate_l3s_hook(oldnet, newnet);
724 		break;
725 	}
726 	case NETDEV_UNREGISTER:
727 		if (dev->reg_state != NETREG_UNREGISTERING)
728 			break;
729 
730 		list_for_each_entry_safe(ipvlan, next, &port->ipvlans, pnode)
731 			ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
732 							    &lst_kill);
733 		unregister_netdevice_many(&lst_kill);
734 		break;
735 
736 	case NETDEV_FEAT_CHANGE:
737 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
738 			ipvlan->dev->features = dev->features & IPVLAN_FEATURES;
739 			ipvlan->dev->gso_max_size = dev->gso_max_size;
740 			ipvlan->dev->gso_max_segs = dev->gso_max_segs;
741 			netdev_features_change(ipvlan->dev);
742 		}
743 		break;
744 
745 	case NETDEV_CHANGEMTU:
746 		list_for_each_entry(ipvlan, &port->ipvlans, pnode)
747 			ipvlan_adjust_mtu(ipvlan, dev);
748 		break;
749 
750 	case NETDEV_PRE_CHANGEADDR:
751 		prechaddr_info = ptr;
752 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
753 			err = dev_pre_changeaddr_notify(ipvlan->dev,
754 						    prechaddr_info->dev_addr,
755 						    extack);
756 			if (err)
757 				return notifier_from_errno(err);
758 		}
759 		break;
760 
761 	case NETDEV_CHANGEADDR:
762 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
763 			ether_addr_copy(ipvlan->dev->dev_addr, dev->dev_addr);
764 			call_netdevice_notifiers(NETDEV_CHANGEADDR, ipvlan->dev);
765 		}
766 		break;
767 
768 	case NETDEV_PRE_TYPE_CHANGE:
769 		/* Forbid underlying device to change its type. */
770 		return NOTIFY_BAD;
771 	}
772 	return NOTIFY_DONE;
773 }
774 
775 /* the caller must held the addrs lock */
776 static int ipvlan_add_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
777 {
778 	struct ipvl_addr *addr;
779 
780 	addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
781 	if (!addr)
782 		return -ENOMEM;
783 
784 	addr->master = ipvlan;
785 	if (!is_v6) {
786 		memcpy(&addr->ip4addr, iaddr, sizeof(struct in_addr));
787 		addr->atype = IPVL_IPV4;
788 #if IS_ENABLED(CONFIG_IPV6)
789 	} else {
790 		memcpy(&addr->ip6addr, iaddr, sizeof(struct in6_addr));
791 		addr->atype = IPVL_IPV6;
792 #endif
793 	}
794 
795 	list_add_tail_rcu(&addr->anode, &ipvlan->addrs);
796 
797 	/* If the interface is not up, the address will be added to the hash
798 	 * list by ipvlan_open.
799 	 */
800 	if (netif_running(ipvlan->dev))
801 		ipvlan_ht_addr_add(ipvlan, addr);
802 
803 	return 0;
804 }
805 
806 static void ipvlan_del_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
807 {
808 	struct ipvl_addr *addr;
809 
810 	spin_lock_bh(&ipvlan->addrs_lock);
811 	addr = ipvlan_find_addr(ipvlan, iaddr, is_v6);
812 	if (!addr) {
813 		spin_unlock_bh(&ipvlan->addrs_lock);
814 		return;
815 	}
816 
817 	ipvlan_ht_addr_del(addr);
818 	list_del_rcu(&addr->anode);
819 	spin_unlock_bh(&ipvlan->addrs_lock);
820 	kfree_rcu(addr, rcu);
821 }
822 
823 static bool ipvlan_is_valid_dev(const struct net_device *dev)
824 {
825 	struct ipvl_dev *ipvlan = netdev_priv(dev);
826 
827 	if (!netif_is_ipvlan(dev))
828 		return false;
829 
830 	if (!ipvlan || !ipvlan->port)
831 		return false;
832 
833 	return true;
834 }
835 
836 #if IS_ENABLED(CONFIG_IPV6)
837 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
838 {
839 	int ret = -EINVAL;
840 
841 	spin_lock_bh(&ipvlan->addrs_lock);
842 	if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true))
843 		netif_err(ipvlan, ifup, ipvlan->dev,
844 			  "Failed to add IPv6=%pI6c addr for %s intf\n",
845 			  ip6_addr, ipvlan->dev->name);
846 	else
847 		ret = ipvlan_add_addr(ipvlan, ip6_addr, true);
848 	spin_unlock_bh(&ipvlan->addrs_lock);
849 	return ret;
850 }
851 
852 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
853 {
854 	return ipvlan_del_addr(ipvlan, ip6_addr, true);
855 }
856 
857 static int ipvlan_addr6_event(struct notifier_block *unused,
858 			      unsigned long event, void *ptr)
859 {
860 	struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
861 	struct net_device *dev = (struct net_device *)if6->idev->dev;
862 	struct ipvl_dev *ipvlan = netdev_priv(dev);
863 
864 	if (!ipvlan_is_valid_dev(dev))
865 		return NOTIFY_DONE;
866 
867 	switch (event) {
868 	case NETDEV_UP:
869 		if (ipvlan_add_addr6(ipvlan, &if6->addr))
870 			return NOTIFY_BAD;
871 		break;
872 
873 	case NETDEV_DOWN:
874 		ipvlan_del_addr6(ipvlan, &if6->addr);
875 		break;
876 	}
877 
878 	return NOTIFY_OK;
879 }
880 
881 static int ipvlan_addr6_validator_event(struct notifier_block *unused,
882 					unsigned long event, void *ptr)
883 {
884 	struct in6_validator_info *i6vi = (struct in6_validator_info *)ptr;
885 	struct net_device *dev = (struct net_device *)i6vi->i6vi_dev->dev;
886 	struct ipvl_dev *ipvlan = netdev_priv(dev);
887 
888 	if (!ipvlan_is_valid_dev(dev))
889 		return NOTIFY_DONE;
890 
891 	switch (event) {
892 	case NETDEV_UP:
893 		if (ipvlan_addr_busy(ipvlan->port, &i6vi->i6vi_addr, true)) {
894 			NL_SET_ERR_MSG(i6vi->extack,
895 				       "Address already assigned to an ipvlan device");
896 			return notifier_from_errno(-EADDRINUSE);
897 		}
898 		break;
899 	}
900 
901 	return NOTIFY_OK;
902 }
903 #endif
904 
905 static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
906 {
907 	int ret = -EINVAL;
908 
909 	spin_lock_bh(&ipvlan->addrs_lock);
910 	if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false))
911 		netif_err(ipvlan, ifup, ipvlan->dev,
912 			  "Failed to add IPv4=%pI4 on %s intf.\n",
913 			  ip4_addr, ipvlan->dev->name);
914 	else
915 		ret = ipvlan_add_addr(ipvlan, ip4_addr, false);
916 	spin_unlock_bh(&ipvlan->addrs_lock);
917 	return ret;
918 }
919 
920 static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
921 {
922 	return ipvlan_del_addr(ipvlan, ip4_addr, false);
923 }
924 
925 static int ipvlan_addr4_event(struct notifier_block *unused,
926 			      unsigned long event, void *ptr)
927 {
928 	struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
929 	struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
930 	struct ipvl_dev *ipvlan = netdev_priv(dev);
931 	struct in_addr ip4_addr;
932 
933 	if (!ipvlan_is_valid_dev(dev))
934 		return NOTIFY_DONE;
935 
936 	switch (event) {
937 	case NETDEV_UP:
938 		ip4_addr.s_addr = if4->ifa_address;
939 		if (ipvlan_add_addr4(ipvlan, &ip4_addr))
940 			return NOTIFY_BAD;
941 		break;
942 
943 	case NETDEV_DOWN:
944 		ip4_addr.s_addr = if4->ifa_address;
945 		ipvlan_del_addr4(ipvlan, &ip4_addr);
946 		break;
947 	}
948 
949 	return NOTIFY_OK;
950 }
951 
952 static int ipvlan_addr4_validator_event(struct notifier_block *unused,
953 					unsigned long event, void *ptr)
954 {
955 	struct in_validator_info *ivi = (struct in_validator_info *)ptr;
956 	struct net_device *dev = (struct net_device *)ivi->ivi_dev->dev;
957 	struct ipvl_dev *ipvlan = netdev_priv(dev);
958 
959 	if (!ipvlan_is_valid_dev(dev))
960 		return NOTIFY_DONE;
961 
962 	switch (event) {
963 	case NETDEV_UP:
964 		if (ipvlan_addr_busy(ipvlan->port, &ivi->ivi_addr, false)) {
965 			NL_SET_ERR_MSG(ivi->extack,
966 				       "Address already assigned to an ipvlan device");
967 			return notifier_from_errno(-EADDRINUSE);
968 		}
969 		break;
970 	}
971 
972 	return NOTIFY_OK;
973 }
974 
975 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
976 	.notifier_call = ipvlan_addr4_event,
977 };
978 
979 static struct notifier_block ipvlan_addr4_vtor_notifier_block __read_mostly = {
980 	.notifier_call = ipvlan_addr4_validator_event,
981 };
982 
983 static struct notifier_block ipvlan_notifier_block __read_mostly = {
984 	.notifier_call = ipvlan_device_event,
985 };
986 
987 #if IS_ENABLED(CONFIG_IPV6)
988 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
989 	.notifier_call = ipvlan_addr6_event,
990 };
991 
992 static struct notifier_block ipvlan_addr6_vtor_notifier_block __read_mostly = {
993 	.notifier_call = ipvlan_addr6_validator_event,
994 };
995 #endif
996 
997 static int __init ipvlan_init_module(void)
998 {
999 	int err;
1000 
1001 	ipvlan_init_secret();
1002 	register_netdevice_notifier(&ipvlan_notifier_block);
1003 #if IS_ENABLED(CONFIG_IPV6)
1004 	register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1005 	register_inet6addr_validator_notifier(
1006 	    &ipvlan_addr6_vtor_notifier_block);
1007 #endif
1008 	register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1009 	register_inetaddr_validator_notifier(&ipvlan_addr4_vtor_notifier_block);
1010 
1011 	err = ipvlan_l3s_init();
1012 	if (err < 0)
1013 		goto error;
1014 
1015 	err = ipvlan_link_register(&ipvlan_link_ops);
1016 	if (err < 0) {
1017 		ipvlan_l3s_cleanup();
1018 		goto error;
1019 	}
1020 
1021 	return 0;
1022 error:
1023 	unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1024 	unregister_inetaddr_validator_notifier(
1025 	    &ipvlan_addr4_vtor_notifier_block);
1026 #if IS_ENABLED(CONFIG_IPV6)
1027 	unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1028 	unregister_inet6addr_validator_notifier(
1029 	    &ipvlan_addr6_vtor_notifier_block);
1030 #endif
1031 	unregister_netdevice_notifier(&ipvlan_notifier_block);
1032 	return err;
1033 }
1034 
1035 static void __exit ipvlan_cleanup_module(void)
1036 {
1037 	rtnl_link_unregister(&ipvlan_link_ops);
1038 	ipvlan_l3s_cleanup();
1039 	unregister_netdevice_notifier(&ipvlan_notifier_block);
1040 	unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1041 	unregister_inetaddr_validator_notifier(
1042 	    &ipvlan_addr4_vtor_notifier_block);
1043 #if IS_ENABLED(CONFIG_IPV6)
1044 	unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1045 	unregister_inet6addr_validator_notifier(
1046 	    &ipvlan_addr6_vtor_notifier_block);
1047 #endif
1048 }
1049 
1050 module_init(ipvlan_init_module);
1051 module_exit(ipvlan_cleanup_module);
1052 
1053 MODULE_LICENSE("GPL");
1054 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
1055 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
1056 MODULE_ALIAS_RTNL_LINK("ipvlan");
1057