xref: /linux/drivers/net/ipvlan/ipvlan_main.c (revision 55d0969c451159cff86949b38c39171cab962069)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Copyright (c) 2014 Mahesh Bandewar <maheshb@google.com>
3  */
4 
5 #include <linux/ethtool.h>
6 
7 #include "ipvlan.h"
8 
9 static int ipvlan_set_port_mode(struct ipvl_port *port, u16 nval,
10 				struct netlink_ext_ack *extack)
11 {
12 	struct ipvl_dev *ipvlan;
13 	unsigned int flags;
14 	int err;
15 
16 	ASSERT_RTNL();
17 	if (port->mode != nval) {
18 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
19 			flags = ipvlan->dev->flags;
20 			if (nval == IPVLAN_MODE_L3 || nval == IPVLAN_MODE_L3S) {
21 				err = dev_change_flags(ipvlan->dev,
22 						       flags | IFF_NOARP,
23 						       extack);
24 			} else {
25 				err = dev_change_flags(ipvlan->dev,
26 						       flags & ~IFF_NOARP,
27 						       extack);
28 			}
29 			if (unlikely(err))
30 				goto fail;
31 		}
32 		if (nval == IPVLAN_MODE_L3S) {
33 			/* New mode is L3S */
34 			err = ipvlan_l3s_register(port);
35 			if (err)
36 				goto fail;
37 		} else if (port->mode == IPVLAN_MODE_L3S) {
38 			/* Old mode was L3S */
39 			ipvlan_l3s_unregister(port);
40 		}
41 		port->mode = nval;
42 	}
43 	return 0;
44 
45 fail:
46 	/* Undo the flags changes that have been done so far. */
47 	list_for_each_entry_continue_reverse(ipvlan, &port->ipvlans, pnode) {
48 		flags = ipvlan->dev->flags;
49 		if (port->mode == IPVLAN_MODE_L3 ||
50 		    port->mode == IPVLAN_MODE_L3S)
51 			dev_change_flags(ipvlan->dev, flags | IFF_NOARP,
52 					 NULL);
53 		else
54 			dev_change_flags(ipvlan->dev, flags & ~IFF_NOARP,
55 					 NULL);
56 	}
57 
58 	return err;
59 }
60 
61 static int ipvlan_port_create(struct net_device *dev)
62 {
63 	struct ipvl_port *port;
64 	int err, idx;
65 
66 	port = kzalloc(sizeof(struct ipvl_port), GFP_KERNEL);
67 	if (!port)
68 		return -ENOMEM;
69 
70 	write_pnet(&port->pnet, dev_net(dev));
71 	port->dev = dev;
72 	port->mode = IPVLAN_MODE_L3;
73 	INIT_LIST_HEAD(&port->ipvlans);
74 	for (idx = 0; idx < IPVLAN_HASH_SIZE; idx++)
75 		INIT_HLIST_HEAD(&port->hlhead[idx]);
76 
77 	skb_queue_head_init(&port->backlog);
78 	INIT_WORK(&port->wq, ipvlan_process_multicast);
79 	ida_init(&port->ida);
80 	port->dev_id_start = 1;
81 
82 	err = netdev_rx_handler_register(dev, ipvlan_handle_frame, port);
83 	if (err)
84 		goto err;
85 
86 	netdev_hold(dev, &port->dev_tracker, GFP_KERNEL);
87 	return 0;
88 
89 err:
90 	kfree(port);
91 	return err;
92 }
93 
94 static void ipvlan_port_destroy(struct net_device *dev)
95 {
96 	struct ipvl_port *port = ipvlan_port_get_rtnl(dev);
97 	struct sk_buff *skb;
98 
99 	netdev_put(dev, &port->dev_tracker);
100 	if (port->mode == IPVLAN_MODE_L3S)
101 		ipvlan_l3s_unregister(port);
102 	netdev_rx_handler_unregister(dev);
103 	cancel_work_sync(&port->wq);
104 	while ((skb = __skb_dequeue(&port->backlog)) != NULL) {
105 		dev_put(skb->dev);
106 		kfree_skb(skb);
107 	}
108 	ida_destroy(&port->ida);
109 	kfree(port);
110 }
111 
112 #define IPVLAN_ALWAYS_ON_OFLOADS \
113 	(NETIF_F_SG | NETIF_F_HW_CSUM | \
114 	 NETIF_F_GSO_ROBUST | NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL)
115 
116 #define IPVLAN_ALWAYS_ON \
117 	(IPVLAN_ALWAYS_ON_OFLOADS | NETIF_F_VLAN_CHALLENGED)
118 
119 #define IPVLAN_FEATURES \
120 	(NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
121 	 NETIF_F_GSO | NETIF_F_ALL_TSO | NETIF_F_GSO_ROBUST | \
122 	 NETIF_F_GRO | NETIF_F_RXCSUM | \
123 	 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
124 
125 	/* NETIF_F_GSO_ENCAP_ALL NETIF_F_GSO_SOFTWARE Newly added */
126 
127 #define IPVLAN_STATE_MASK \
128 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
129 
130 static int ipvlan_init(struct net_device *dev)
131 {
132 	struct ipvl_dev *ipvlan = netdev_priv(dev);
133 	struct net_device *phy_dev = ipvlan->phy_dev;
134 	struct ipvl_port *port;
135 	int err;
136 
137 	dev->state = (dev->state & ~IPVLAN_STATE_MASK) |
138 		     (phy_dev->state & IPVLAN_STATE_MASK);
139 	dev->features = phy_dev->features & IPVLAN_FEATURES;
140 	dev->features |= IPVLAN_ALWAYS_ON;
141 	dev->vlan_features = phy_dev->vlan_features & IPVLAN_FEATURES;
142 	dev->vlan_features |= IPVLAN_ALWAYS_ON_OFLOADS;
143 	dev->hw_enc_features |= dev->features;
144 	dev->lltx = true;
145 	netif_inherit_tso_max(dev, phy_dev);
146 	dev->hard_header_len = phy_dev->hard_header_len;
147 
148 	netdev_lockdep_set_classes(dev);
149 
150 	ipvlan->pcpu_stats = netdev_alloc_pcpu_stats(struct ipvl_pcpu_stats);
151 	if (!ipvlan->pcpu_stats)
152 		return -ENOMEM;
153 
154 	if (!netif_is_ipvlan_port(phy_dev)) {
155 		err = ipvlan_port_create(phy_dev);
156 		if (err < 0) {
157 			free_percpu(ipvlan->pcpu_stats);
158 			return err;
159 		}
160 	}
161 	port = ipvlan_port_get_rtnl(phy_dev);
162 	port->count += 1;
163 	return 0;
164 }
165 
166 static void ipvlan_uninit(struct net_device *dev)
167 {
168 	struct ipvl_dev *ipvlan = netdev_priv(dev);
169 	struct net_device *phy_dev = ipvlan->phy_dev;
170 	struct ipvl_port *port;
171 
172 	free_percpu(ipvlan->pcpu_stats);
173 
174 	port = ipvlan_port_get_rtnl(phy_dev);
175 	port->count -= 1;
176 	if (!port->count)
177 		ipvlan_port_destroy(port->dev);
178 }
179 
180 static int ipvlan_open(struct net_device *dev)
181 {
182 	struct ipvl_dev *ipvlan = netdev_priv(dev);
183 	struct ipvl_addr *addr;
184 
185 	if (ipvlan->port->mode == IPVLAN_MODE_L3 ||
186 	    ipvlan->port->mode == IPVLAN_MODE_L3S)
187 		dev->flags |= IFF_NOARP;
188 	else
189 		dev->flags &= ~IFF_NOARP;
190 
191 	rcu_read_lock();
192 	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
193 		ipvlan_ht_addr_add(ipvlan, addr);
194 	rcu_read_unlock();
195 
196 	return 0;
197 }
198 
199 static int ipvlan_stop(struct net_device *dev)
200 {
201 	struct ipvl_dev *ipvlan = netdev_priv(dev);
202 	struct net_device *phy_dev = ipvlan->phy_dev;
203 	struct ipvl_addr *addr;
204 
205 	dev_uc_unsync(phy_dev, dev);
206 	dev_mc_unsync(phy_dev, dev);
207 
208 	rcu_read_lock();
209 	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
210 		ipvlan_ht_addr_del(addr);
211 	rcu_read_unlock();
212 
213 	return 0;
214 }
215 
216 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
217 				     struct net_device *dev)
218 {
219 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
220 	int skblen = skb->len;
221 	int ret;
222 
223 	ret = ipvlan_queue_xmit(skb, dev);
224 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
225 		struct ipvl_pcpu_stats *pcptr;
226 
227 		pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
228 
229 		u64_stats_update_begin(&pcptr->syncp);
230 		u64_stats_inc(&pcptr->tx_pkts);
231 		u64_stats_add(&pcptr->tx_bytes, skblen);
232 		u64_stats_update_end(&pcptr->syncp);
233 	} else {
234 		this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
235 	}
236 	return ret;
237 }
238 
239 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
240 					     netdev_features_t features)
241 {
242 	struct ipvl_dev *ipvlan = netdev_priv(dev);
243 
244 	features |= NETIF_F_ALL_FOR_ALL;
245 	features &= (ipvlan->sfeatures | ~IPVLAN_FEATURES);
246 	features = netdev_increment_features(ipvlan->phy_dev->features,
247 					     features, features);
248 	features |= IPVLAN_ALWAYS_ON;
249 	features &= (IPVLAN_FEATURES | IPVLAN_ALWAYS_ON);
250 
251 	return features;
252 }
253 
254 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
255 {
256 	struct ipvl_dev *ipvlan = netdev_priv(dev);
257 	struct net_device *phy_dev = ipvlan->phy_dev;
258 
259 	if (change & IFF_ALLMULTI)
260 		dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
261 }
262 
263 static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
264 {
265 	struct ipvl_dev *ipvlan = netdev_priv(dev);
266 
267 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
268 		bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
269 	} else {
270 		struct netdev_hw_addr *ha;
271 		DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
272 
273 		bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
274 		netdev_for_each_mc_addr(ha, dev)
275 			__set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
276 
277 		/* Turn-on broadcast bit irrespective of address family,
278 		 * since broadcast is deferred to a work-queue, hence no
279 		 * impact on fast-path processing.
280 		 */
281 		__set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
282 
283 		bitmap_copy(ipvlan->mac_filters, mc_filters,
284 			    IPVLAN_MAC_FILTER_SIZE);
285 	}
286 	dev_uc_sync(ipvlan->phy_dev, dev);
287 	dev_mc_sync(ipvlan->phy_dev, dev);
288 }
289 
290 static void ipvlan_get_stats64(struct net_device *dev,
291 			       struct rtnl_link_stats64 *s)
292 {
293 	struct ipvl_dev *ipvlan = netdev_priv(dev);
294 
295 	if (ipvlan->pcpu_stats) {
296 		struct ipvl_pcpu_stats *pcptr;
297 		u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
298 		u32 rx_errs = 0, tx_drps = 0;
299 		u32 strt;
300 		int idx;
301 
302 		for_each_possible_cpu(idx) {
303 			pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
304 			do {
305 				strt = u64_stats_fetch_begin(&pcptr->syncp);
306 				rx_pkts = u64_stats_read(&pcptr->rx_pkts);
307 				rx_bytes = u64_stats_read(&pcptr->rx_bytes);
308 				rx_mcast = u64_stats_read(&pcptr->rx_mcast);
309 				tx_pkts = u64_stats_read(&pcptr->tx_pkts);
310 				tx_bytes = u64_stats_read(&pcptr->tx_bytes);
311 			} while (u64_stats_fetch_retry(&pcptr->syncp,
312 							   strt));
313 
314 			s->rx_packets += rx_pkts;
315 			s->rx_bytes += rx_bytes;
316 			s->multicast += rx_mcast;
317 			s->tx_packets += tx_pkts;
318 			s->tx_bytes += tx_bytes;
319 
320 			/* u32 values are updated without syncp protection. */
321 			rx_errs += READ_ONCE(pcptr->rx_errs);
322 			tx_drps += READ_ONCE(pcptr->tx_drps);
323 		}
324 		s->rx_errors = rx_errs;
325 		s->rx_dropped = rx_errs;
326 		s->tx_dropped = tx_drps;
327 	}
328 	s->tx_errors = DEV_STATS_READ(dev, tx_errors);
329 }
330 
331 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
332 {
333 	struct ipvl_dev *ipvlan = netdev_priv(dev);
334 	struct net_device *phy_dev = ipvlan->phy_dev;
335 
336 	return vlan_vid_add(phy_dev, proto, vid);
337 }
338 
339 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
340 				   u16 vid)
341 {
342 	struct ipvl_dev *ipvlan = netdev_priv(dev);
343 	struct net_device *phy_dev = ipvlan->phy_dev;
344 
345 	vlan_vid_del(phy_dev, proto, vid);
346 	return 0;
347 }
348 
349 static int ipvlan_get_iflink(const struct net_device *dev)
350 {
351 	struct ipvl_dev *ipvlan = netdev_priv(dev);
352 
353 	return READ_ONCE(ipvlan->phy_dev->ifindex);
354 }
355 
356 static const struct net_device_ops ipvlan_netdev_ops = {
357 	.ndo_init		= ipvlan_init,
358 	.ndo_uninit		= ipvlan_uninit,
359 	.ndo_open		= ipvlan_open,
360 	.ndo_stop		= ipvlan_stop,
361 	.ndo_start_xmit		= ipvlan_start_xmit,
362 	.ndo_fix_features	= ipvlan_fix_features,
363 	.ndo_change_rx_flags	= ipvlan_change_rx_flags,
364 	.ndo_set_rx_mode	= ipvlan_set_multicast_mac_filter,
365 	.ndo_get_stats64	= ipvlan_get_stats64,
366 	.ndo_vlan_rx_add_vid	= ipvlan_vlan_rx_add_vid,
367 	.ndo_vlan_rx_kill_vid	= ipvlan_vlan_rx_kill_vid,
368 	.ndo_get_iflink		= ipvlan_get_iflink,
369 };
370 
371 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
372 			      unsigned short type, const void *daddr,
373 			      const void *saddr, unsigned len)
374 {
375 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
376 	struct net_device *phy_dev = ipvlan->phy_dev;
377 
378 	/* TODO Probably use a different field than dev_addr so that the
379 	 * mac-address on the virtual device is portable and can be carried
380 	 * while the packets use the mac-addr on the physical device.
381 	 */
382 	return dev_hard_header(skb, phy_dev, type, daddr,
383 			       saddr ? : phy_dev->dev_addr, len);
384 }
385 
386 static const struct header_ops ipvlan_header_ops = {
387 	.create  	= ipvlan_hard_header,
388 	.parse		= eth_header_parse,
389 	.cache		= eth_header_cache,
390 	.cache_update	= eth_header_cache_update,
391 	.parse_protocol	= eth_header_parse_protocol,
392 };
393 
394 static void ipvlan_adjust_mtu(struct ipvl_dev *ipvlan, struct net_device *dev)
395 {
396 	ipvlan->dev->mtu = dev->mtu;
397 }
398 
399 static bool netif_is_ipvlan(const struct net_device *dev)
400 {
401 	/* both ipvlan and ipvtap devices use the same netdev_ops */
402 	return dev->netdev_ops == &ipvlan_netdev_ops;
403 }
404 
405 static int ipvlan_ethtool_get_link_ksettings(struct net_device *dev,
406 					     struct ethtool_link_ksettings *cmd)
407 {
408 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
409 
410 	return __ethtool_get_link_ksettings(ipvlan->phy_dev, cmd);
411 }
412 
413 static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
414 				       struct ethtool_drvinfo *drvinfo)
415 {
416 	strscpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
417 	strscpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
418 }
419 
420 static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
421 {
422 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
423 
424 	return ipvlan->msg_enable;
425 }
426 
427 static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
428 {
429 	struct ipvl_dev *ipvlan = netdev_priv(dev);
430 
431 	ipvlan->msg_enable = value;
432 }
433 
434 static const struct ethtool_ops ipvlan_ethtool_ops = {
435 	.get_link	= ethtool_op_get_link,
436 	.get_link_ksettings	= ipvlan_ethtool_get_link_ksettings,
437 	.get_drvinfo	= ipvlan_ethtool_get_drvinfo,
438 	.get_msglevel	= ipvlan_ethtool_get_msglevel,
439 	.set_msglevel	= ipvlan_ethtool_set_msglevel,
440 };
441 
442 static int ipvlan_nl_changelink(struct net_device *dev,
443 				struct nlattr *tb[], struct nlattr *data[],
444 				struct netlink_ext_ack *extack)
445 {
446 	struct ipvl_dev *ipvlan = netdev_priv(dev);
447 	struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
448 	int err = 0;
449 
450 	if (!data)
451 		return 0;
452 	if (!ns_capable(dev_net(ipvlan->phy_dev)->user_ns, CAP_NET_ADMIN))
453 		return -EPERM;
454 
455 	if (data[IFLA_IPVLAN_MODE]) {
456 		u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
457 
458 		err = ipvlan_set_port_mode(port, nmode, extack);
459 	}
460 
461 	if (!err && data[IFLA_IPVLAN_FLAGS]) {
462 		u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
463 
464 		if (flags & IPVLAN_F_PRIVATE)
465 			ipvlan_mark_private(port);
466 		else
467 			ipvlan_clear_private(port);
468 
469 		if (flags & IPVLAN_F_VEPA)
470 			ipvlan_mark_vepa(port);
471 		else
472 			ipvlan_clear_vepa(port);
473 	}
474 
475 	return err;
476 }
477 
478 static size_t ipvlan_nl_getsize(const struct net_device *dev)
479 {
480 	return (0
481 		+ nla_total_size(2) /* IFLA_IPVLAN_MODE */
482 		+ nla_total_size(2) /* IFLA_IPVLAN_FLAGS */
483 		);
484 }
485 
486 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[],
487 			      struct netlink_ext_ack *extack)
488 {
489 	if (!data)
490 		return 0;
491 
492 	if (data[IFLA_IPVLAN_MODE]) {
493 		u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
494 
495 		if (mode >= IPVLAN_MODE_MAX)
496 			return -EINVAL;
497 	}
498 	if (data[IFLA_IPVLAN_FLAGS]) {
499 		u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
500 
501 		/* Only two bits are used at this moment. */
502 		if (flags & ~(IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
503 			return -EINVAL;
504 		/* Also both flags can't be active at the same time. */
505 		if ((flags & (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA)) ==
506 		    (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
507 			return -EINVAL;
508 	}
509 
510 	return 0;
511 }
512 
513 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
514 			      const struct net_device *dev)
515 {
516 	struct ipvl_dev *ipvlan = netdev_priv(dev);
517 	struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
518 	int ret = -EINVAL;
519 
520 	if (!port)
521 		goto err;
522 
523 	ret = -EMSGSIZE;
524 	if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
525 		goto err;
526 	if (nla_put_u16(skb, IFLA_IPVLAN_FLAGS, port->flags))
527 		goto err;
528 
529 	return 0;
530 
531 err:
532 	return ret;
533 }
534 
535 int ipvlan_link_new(struct net *src_net, struct net_device *dev,
536 		    struct nlattr *tb[], struct nlattr *data[],
537 		    struct netlink_ext_ack *extack)
538 {
539 	struct ipvl_dev *ipvlan = netdev_priv(dev);
540 	struct ipvl_port *port;
541 	struct net_device *phy_dev;
542 	int err;
543 	u16 mode = IPVLAN_MODE_L3;
544 
545 	if (!tb[IFLA_LINK])
546 		return -EINVAL;
547 
548 	phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
549 	if (!phy_dev)
550 		return -ENODEV;
551 
552 	if (netif_is_ipvlan(phy_dev)) {
553 		struct ipvl_dev *tmp = netdev_priv(phy_dev);
554 
555 		phy_dev = tmp->phy_dev;
556 		if (!ns_capable(dev_net(phy_dev)->user_ns, CAP_NET_ADMIN))
557 			return -EPERM;
558 	} else if (!netif_is_ipvlan_port(phy_dev)) {
559 		/* Exit early if the underlying link is invalid or busy */
560 		if (phy_dev->type != ARPHRD_ETHER ||
561 		    phy_dev->flags & IFF_LOOPBACK) {
562 			netdev_err(phy_dev,
563 				   "Master is either lo or non-ether device\n");
564 			return -EINVAL;
565 		}
566 
567 		if (netdev_is_rx_handler_busy(phy_dev)) {
568 			netdev_err(phy_dev, "Device is already in use.\n");
569 			return -EBUSY;
570 		}
571 	}
572 
573 	ipvlan->phy_dev = phy_dev;
574 	ipvlan->dev = dev;
575 	ipvlan->sfeatures = IPVLAN_FEATURES;
576 	if (!tb[IFLA_MTU])
577 		ipvlan_adjust_mtu(ipvlan, phy_dev);
578 	INIT_LIST_HEAD(&ipvlan->addrs);
579 	spin_lock_init(&ipvlan->addrs_lock);
580 
581 	/* TODO Probably put random address here to be presented to the
582 	 * world but keep using the physical-dev address for the outgoing
583 	 * packets.
584 	 */
585 	eth_hw_addr_set(dev, phy_dev->dev_addr);
586 
587 	dev->priv_flags |= IFF_NO_RX_HANDLER;
588 
589 	err = register_netdevice(dev);
590 	if (err < 0)
591 		return err;
592 
593 	/* ipvlan_init() would have created the port, if required */
594 	port = ipvlan_port_get_rtnl(phy_dev);
595 	ipvlan->port = port;
596 
597 	/* If the port-id base is at the MAX value, then wrap it around and
598 	 * begin from 0x1 again. This may be due to a busy system where lots
599 	 * of slaves are getting created and deleted.
600 	 */
601 	if (port->dev_id_start == 0xFFFE)
602 		port->dev_id_start = 0x1;
603 
604 	/* Since L2 address is shared among all IPvlan slaves including
605 	 * master, use unique 16 bit dev-ids to differentiate among them.
606 	 * Assign IDs between 0x1 and 0xFFFE (used by the master) to each
607 	 * slave link [see addrconf_ifid_eui48()].
608 	 */
609 	err = ida_alloc_range(&port->ida, port->dev_id_start, 0xFFFD,
610 			      GFP_KERNEL);
611 	if (err < 0)
612 		err = ida_alloc_range(&port->ida, 0x1, port->dev_id_start - 1,
613 				      GFP_KERNEL);
614 	if (err < 0)
615 		goto unregister_netdev;
616 	dev->dev_id = err;
617 
618 	/* Increment id-base to the next slot for the future assignment */
619 	port->dev_id_start = err + 1;
620 
621 	err = netdev_upper_dev_link(phy_dev, dev, extack);
622 	if (err)
623 		goto remove_ida;
624 
625 	/* Flags are per port and latest update overrides. User has
626 	 * to be consistent in setting it just like the mode attribute.
627 	 */
628 	if (data && data[IFLA_IPVLAN_FLAGS])
629 		port->flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
630 
631 	if (data && data[IFLA_IPVLAN_MODE])
632 		mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
633 
634 	err = ipvlan_set_port_mode(port, mode, extack);
635 	if (err)
636 		goto unlink_netdev;
637 
638 	list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
639 	netif_stacked_transfer_operstate(phy_dev, dev);
640 	return 0;
641 
642 unlink_netdev:
643 	netdev_upper_dev_unlink(phy_dev, dev);
644 remove_ida:
645 	ida_free(&port->ida, dev->dev_id);
646 unregister_netdev:
647 	unregister_netdevice(dev);
648 	return err;
649 }
650 EXPORT_SYMBOL_GPL(ipvlan_link_new);
651 
652 void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
653 {
654 	struct ipvl_dev *ipvlan = netdev_priv(dev);
655 	struct ipvl_addr *addr, *next;
656 
657 	spin_lock_bh(&ipvlan->addrs_lock);
658 	list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
659 		ipvlan_ht_addr_del(addr);
660 		list_del_rcu(&addr->anode);
661 		kfree_rcu(addr, rcu);
662 	}
663 	spin_unlock_bh(&ipvlan->addrs_lock);
664 
665 	ida_free(&ipvlan->port->ida, dev->dev_id);
666 	list_del_rcu(&ipvlan->pnode);
667 	unregister_netdevice_queue(dev, head);
668 	netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
669 }
670 EXPORT_SYMBOL_GPL(ipvlan_link_delete);
671 
672 void ipvlan_link_setup(struct net_device *dev)
673 {
674 	ether_setup(dev);
675 
676 	dev->max_mtu = ETH_MAX_MTU;
677 	dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
678 	dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
679 	dev->netdev_ops = &ipvlan_netdev_ops;
680 	dev->needs_free_netdev = true;
681 	dev->header_ops = &ipvlan_header_ops;
682 	dev->ethtool_ops = &ipvlan_ethtool_ops;
683 }
684 EXPORT_SYMBOL_GPL(ipvlan_link_setup);
685 
686 static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
687 {
688 	[IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
689 	[IFLA_IPVLAN_FLAGS] = { .type = NLA_U16 },
690 };
691 
692 static struct net *ipvlan_get_link_net(const struct net_device *dev)
693 {
694 	struct ipvl_dev *ipvlan = netdev_priv(dev);
695 
696 	return dev_net(ipvlan->phy_dev);
697 }
698 
699 static struct rtnl_link_ops ipvlan_link_ops = {
700 	.kind		= "ipvlan",
701 	.priv_size	= sizeof(struct ipvl_dev),
702 
703 	.setup		= ipvlan_link_setup,
704 	.newlink	= ipvlan_link_new,
705 	.dellink	= ipvlan_link_delete,
706 	.get_link_net   = ipvlan_get_link_net,
707 };
708 
709 int ipvlan_link_register(struct rtnl_link_ops *ops)
710 {
711 	ops->get_size	= ipvlan_nl_getsize;
712 	ops->policy	= ipvlan_nl_policy;
713 	ops->validate	= ipvlan_nl_validate;
714 	ops->fill_info	= ipvlan_nl_fillinfo;
715 	ops->changelink = ipvlan_nl_changelink;
716 	ops->maxtype	= IFLA_IPVLAN_MAX;
717 	return rtnl_link_register(ops);
718 }
719 EXPORT_SYMBOL_GPL(ipvlan_link_register);
720 
721 static int ipvlan_device_event(struct notifier_block *unused,
722 			       unsigned long event, void *ptr)
723 {
724 	struct netlink_ext_ack *extack = netdev_notifier_info_to_extack(ptr);
725 	struct netdev_notifier_pre_changeaddr_info *prechaddr_info;
726 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
727 	struct ipvl_dev *ipvlan, *next;
728 	struct ipvl_port *port;
729 	LIST_HEAD(lst_kill);
730 	int err;
731 
732 	if (!netif_is_ipvlan_port(dev))
733 		return NOTIFY_DONE;
734 
735 	port = ipvlan_port_get_rtnl(dev);
736 
737 	switch (event) {
738 	case NETDEV_UP:
739 	case NETDEV_DOWN:
740 	case NETDEV_CHANGE:
741 		list_for_each_entry(ipvlan, &port->ipvlans, pnode)
742 			netif_stacked_transfer_operstate(ipvlan->phy_dev,
743 							 ipvlan->dev);
744 		break;
745 
746 	case NETDEV_REGISTER: {
747 		struct net *oldnet, *newnet = dev_net(dev);
748 
749 		oldnet = read_pnet(&port->pnet);
750 		if (net_eq(newnet, oldnet))
751 			break;
752 
753 		write_pnet(&port->pnet, newnet);
754 
755 		if (port->mode == IPVLAN_MODE_L3S)
756 			ipvlan_migrate_l3s_hook(oldnet, newnet);
757 		break;
758 	}
759 	case NETDEV_UNREGISTER:
760 		if (dev->reg_state != NETREG_UNREGISTERING)
761 			break;
762 
763 		list_for_each_entry_safe(ipvlan, next, &port->ipvlans, pnode)
764 			ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
765 							    &lst_kill);
766 		unregister_netdevice_many(&lst_kill);
767 		break;
768 
769 	case NETDEV_FEAT_CHANGE:
770 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
771 			netif_inherit_tso_max(ipvlan->dev, dev);
772 			netdev_update_features(ipvlan->dev);
773 		}
774 		break;
775 
776 	case NETDEV_CHANGEMTU:
777 		list_for_each_entry(ipvlan, &port->ipvlans, pnode)
778 			ipvlan_adjust_mtu(ipvlan, dev);
779 		break;
780 
781 	case NETDEV_PRE_CHANGEADDR:
782 		prechaddr_info = ptr;
783 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
784 			err = dev_pre_changeaddr_notify(ipvlan->dev,
785 						    prechaddr_info->dev_addr,
786 						    extack);
787 			if (err)
788 				return notifier_from_errno(err);
789 		}
790 		break;
791 
792 	case NETDEV_CHANGEADDR:
793 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
794 			eth_hw_addr_set(ipvlan->dev, dev->dev_addr);
795 			call_netdevice_notifiers(NETDEV_CHANGEADDR, ipvlan->dev);
796 		}
797 		break;
798 
799 	case NETDEV_PRE_TYPE_CHANGE:
800 		/* Forbid underlying device to change its type. */
801 		return NOTIFY_BAD;
802 	}
803 	return NOTIFY_DONE;
804 }
805 
806 /* the caller must held the addrs lock */
807 static int ipvlan_add_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
808 {
809 	struct ipvl_addr *addr;
810 
811 	addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
812 	if (!addr)
813 		return -ENOMEM;
814 
815 	addr->master = ipvlan;
816 	if (!is_v6) {
817 		memcpy(&addr->ip4addr, iaddr, sizeof(struct in_addr));
818 		addr->atype = IPVL_IPV4;
819 #if IS_ENABLED(CONFIG_IPV6)
820 	} else {
821 		memcpy(&addr->ip6addr, iaddr, sizeof(struct in6_addr));
822 		addr->atype = IPVL_IPV6;
823 #endif
824 	}
825 
826 	list_add_tail_rcu(&addr->anode, &ipvlan->addrs);
827 
828 	/* If the interface is not up, the address will be added to the hash
829 	 * list by ipvlan_open.
830 	 */
831 	if (netif_running(ipvlan->dev))
832 		ipvlan_ht_addr_add(ipvlan, addr);
833 
834 	return 0;
835 }
836 
837 static void ipvlan_del_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
838 {
839 	struct ipvl_addr *addr;
840 
841 	spin_lock_bh(&ipvlan->addrs_lock);
842 	addr = ipvlan_find_addr(ipvlan, iaddr, is_v6);
843 	if (!addr) {
844 		spin_unlock_bh(&ipvlan->addrs_lock);
845 		return;
846 	}
847 
848 	ipvlan_ht_addr_del(addr);
849 	list_del_rcu(&addr->anode);
850 	spin_unlock_bh(&ipvlan->addrs_lock);
851 	kfree_rcu(addr, rcu);
852 }
853 
854 static bool ipvlan_is_valid_dev(const struct net_device *dev)
855 {
856 	struct ipvl_dev *ipvlan = netdev_priv(dev);
857 
858 	if (!netif_is_ipvlan(dev))
859 		return false;
860 
861 	if (!ipvlan || !ipvlan->port)
862 		return false;
863 
864 	return true;
865 }
866 
867 #if IS_ENABLED(CONFIG_IPV6)
868 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
869 {
870 	int ret = -EINVAL;
871 
872 	spin_lock_bh(&ipvlan->addrs_lock);
873 	if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true))
874 		netif_err(ipvlan, ifup, ipvlan->dev,
875 			  "Failed to add IPv6=%pI6c addr for %s intf\n",
876 			  ip6_addr, ipvlan->dev->name);
877 	else
878 		ret = ipvlan_add_addr(ipvlan, ip6_addr, true);
879 	spin_unlock_bh(&ipvlan->addrs_lock);
880 	return ret;
881 }
882 
883 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
884 {
885 	return ipvlan_del_addr(ipvlan, ip6_addr, true);
886 }
887 
888 static int ipvlan_addr6_event(struct notifier_block *unused,
889 			      unsigned long event, void *ptr)
890 {
891 	struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
892 	struct net_device *dev = (struct net_device *)if6->idev->dev;
893 	struct ipvl_dev *ipvlan = netdev_priv(dev);
894 
895 	if (!ipvlan_is_valid_dev(dev))
896 		return NOTIFY_DONE;
897 
898 	switch (event) {
899 	case NETDEV_UP:
900 		if (ipvlan_add_addr6(ipvlan, &if6->addr))
901 			return NOTIFY_BAD;
902 		break;
903 
904 	case NETDEV_DOWN:
905 		ipvlan_del_addr6(ipvlan, &if6->addr);
906 		break;
907 	}
908 
909 	return NOTIFY_OK;
910 }
911 
912 static int ipvlan_addr6_validator_event(struct notifier_block *unused,
913 					unsigned long event, void *ptr)
914 {
915 	struct in6_validator_info *i6vi = (struct in6_validator_info *)ptr;
916 	struct net_device *dev = (struct net_device *)i6vi->i6vi_dev->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_addr_busy(ipvlan->port, &i6vi->i6vi_addr, true)) {
925 			NL_SET_ERR_MSG(i6vi->extack,
926 				       "Address already assigned to an ipvlan device");
927 			return notifier_from_errno(-EADDRINUSE);
928 		}
929 		break;
930 	}
931 
932 	return NOTIFY_OK;
933 }
934 #endif
935 
936 static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
937 {
938 	int ret = -EINVAL;
939 
940 	spin_lock_bh(&ipvlan->addrs_lock);
941 	if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false))
942 		netif_err(ipvlan, ifup, ipvlan->dev,
943 			  "Failed to add IPv4=%pI4 on %s intf.\n",
944 			  ip4_addr, ipvlan->dev->name);
945 	else
946 		ret = ipvlan_add_addr(ipvlan, ip4_addr, false);
947 	spin_unlock_bh(&ipvlan->addrs_lock);
948 	return ret;
949 }
950 
951 static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
952 {
953 	return ipvlan_del_addr(ipvlan, ip4_addr, false);
954 }
955 
956 static int ipvlan_addr4_event(struct notifier_block *unused,
957 			      unsigned long event, void *ptr)
958 {
959 	struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
960 	struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
961 	struct ipvl_dev *ipvlan = netdev_priv(dev);
962 	struct in_addr ip4_addr;
963 
964 	if (!ipvlan_is_valid_dev(dev))
965 		return NOTIFY_DONE;
966 
967 	switch (event) {
968 	case NETDEV_UP:
969 		ip4_addr.s_addr = if4->ifa_address;
970 		if (ipvlan_add_addr4(ipvlan, &ip4_addr))
971 			return NOTIFY_BAD;
972 		break;
973 
974 	case NETDEV_DOWN:
975 		ip4_addr.s_addr = if4->ifa_address;
976 		ipvlan_del_addr4(ipvlan, &ip4_addr);
977 		break;
978 	}
979 
980 	return NOTIFY_OK;
981 }
982 
983 static int ipvlan_addr4_validator_event(struct notifier_block *unused,
984 					unsigned long event, void *ptr)
985 {
986 	struct in_validator_info *ivi = (struct in_validator_info *)ptr;
987 	struct net_device *dev = (struct net_device *)ivi->ivi_dev->dev;
988 	struct ipvl_dev *ipvlan = netdev_priv(dev);
989 
990 	if (!ipvlan_is_valid_dev(dev))
991 		return NOTIFY_DONE;
992 
993 	switch (event) {
994 	case NETDEV_UP:
995 		if (ipvlan_addr_busy(ipvlan->port, &ivi->ivi_addr, false)) {
996 			NL_SET_ERR_MSG(ivi->extack,
997 				       "Address already assigned to an ipvlan device");
998 			return notifier_from_errno(-EADDRINUSE);
999 		}
1000 		break;
1001 	}
1002 
1003 	return NOTIFY_OK;
1004 }
1005 
1006 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
1007 	.notifier_call = ipvlan_addr4_event,
1008 };
1009 
1010 static struct notifier_block ipvlan_addr4_vtor_notifier_block __read_mostly = {
1011 	.notifier_call = ipvlan_addr4_validator_event,
1012 };
1013 
1014 static struct notifier_block ipvlan_notifier_block __read_mostly = {
1015 	.notifier_call = ipvlan_device_event,
1016 };
1017 
1018 #if IS_ENABLED(CONFIG_IPV6)
1019 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
1020 	.notifier_call = ipvlan_addr6_event,
1021 };
1022 
1023 static struct notifier_block ipvlan_addr6_vtor_notifier_block __read_mostly = {
1024 	.notifier_call = ipvlan_addr6_validator_event,
1025 };
1026 #endif
1027 
1028 static int __init ipvlan_init_module(void)
1029 {
1030 	int err;
1031 
1032 	ipvlan_init_secret();
1033 	register_netdevice_notifier(&ipvlan_notifier_block);
1034 #if IS_ENABLED(CONFIG_IPV6)
1035 	register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1036 	register_inet6addr_validator_notifier(
1037 	    &ipvlan_addr6_vtor_notifier_block);
1038 #endif
1039 	register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1040 	register_inetaddr_validator_notifier(&ipvlan_addr4_vtor_notifier_block);
1041 
1042 	err = ipvlan_l3s_init();
1043 	if (err < 0)
1044 		goto error;
1045 
1046 	err = ipvlan_link_register(&ipvlan_link_ops);
1047 	if (err < 0) {
1048 		ipvlan_l3s_cleanup();
1049 		goto error;
1050 	}
1051 
1052 	return 0;
1053 error:
1054 	unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1055 	unregister_inetaddr_validator_notifier(
1056 	    &ipvlan_addr4_vtor_notifier_block);
1057 #if IS_ENABLED(CONFIG_IPV6)
1058 	unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1059 	unregister_inet6addr_validator_notifier(
1060 	    &ipvlan_addr6_vtor_notifier_block);
1061 #endif
1062 	unregister_netdevice_notifier(&ipvlan_notifier_block);
1063 	return err;
1064 }
1065 
1066 static void __exit ipvlan_cleanup_module(void)
1067 {
1068 	rtnl_link_unregister(&ipvlan_link_ops);
1069 	ipvlan_l3s_cleanup();
1070 	unregister_netdevice_notifier(&ipvlan_notifier_block);
1071 	unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1072 	unregister_inetaddr_validator_notifier(
1073 	    &ipvlan_addr4_vtor_notifier_block);
1074 #if IS_ENABLED(CONFIG_IPV6)
1075 	unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1076 	unregister_inet6addr_validator_notifier(
1077 	    &ipvlan_addr6_vtor_notifier_block);
1078 #endif
1079 }
1080 
1081 module_init(ipvlan_init_module);
1082 module_exit(ipvlan_cleanup_module);
1083 
1084 MODULE_LICENSE("GPL");
1085 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
1086 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
1087 MODULE_ALIAS_RTNL_LINK("ipvlan");
1088