xref: /openbmc/linux/drivers/net/ipvlan/ipvlan_main.c (revision ca637c0e)
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 	return 0;
87 
88 err:
89 	kfree(port);
90 	return err;
91 }
92 
93 static void ipvlan_port_destroy(struct net_device *dev)
94 {
95 	struct ipvl_port *port = ipvlan_port_get_rtnl(dev);
96 	struct sk_buff *skb;
97 
98 	if (port->mode == IPVLAN_MODE_L3S)
99 		ipvlan_l3s_unregister(port);
100 	netdev_rx_handler_unregister(dev);
101 	cancel_work_sync(&port->wq);
102 	while ((skb = __skb_dequeue(&port->backlog)) != NULL) {
103 		dev_put(skb->dev);
104 		kfree_skb(skb);
105 	}
106 	ida_destroy(&port->ida);
107 	kfree(port);
108 }
109 
110 #define IPVLAN_ALWAYS_ON_OFLOADS \
111 	(NETIF_F_SG | NETIF_F_HW_CSUM | \
112 	 NETIF_F_GSO_ROBUST | NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL)
113 
114 #define IPVLAN_ALWAYS_ON \
115 	(IPVLAN_ALWAYS_ON_OFLOADS | NETIF_F_LLTX | NETIF_F_VLAN_CHALLENGED)
116 
117 #define IPVLAN_FEATURES \
118 	(NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
119 	 NETIF_F_GSO | NETIF_F_ALL_TSO | NETIF_F_GSO_ROBUST | \
120 	 NETIF_F_GRO | NETIF_F_RXCSUM | \
121 	 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
122 
123 	/* NETIF_F_GSO_ENCAP_ALL NETIF_F_GSO_SOFTWARE Newly added */
124 
125 #define IPVLAN_STATE_MASK \
126 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
127 
128 static int ipvlan_init(struct net_device *dev)
129 {
130 	struct ipvl_dev *ipvlan = netdev_priv(dev);
131 	struct net_device *phy_dev = ipvlan->phy_dev;
132 	struct ipvl_port *port;
133 	int err;
134 
135 	dev->state = (dev->state & ~IPVLAN_STATE_MASK) |
136 		     (phy_dev->state & IPVLAN_STATE_MASK);
137 	dev->features = phy_dev->features & IPVLAN_FEATURES;
138 	dev->features |= IPVLAN_ALWAYS_ON;
139 	dev->vlan_features = phy_dev->vlan_features & IPVLAN_FEATURES;
140 	dev->vlan_features |= IPVLAN_ALWAYS_ON_OFLOADS;
141 	dev->hw_enc_features |= dev->features;
142 	netif_inherit_tso_max(dev, phy_dev);
143 	dev->hard_header_len = phy_dev->hard_header_len;
144 
145 	netdev_lockdep_set_classes(dev);
146 
147 	ipvlan->pcpu_stats = netdev_alloc_pcpu_stats(struct ipvl_pcpu_stats);
148 	if (!ipvlan->pcpu_stats)
149 		return -ENOMEM;
150 
151 	if (!netif_is_ipvlan_port(phy_dev)) {
152 		err = ipvlan_port_create(phy_dev);
153 		if (err < 0) {
154 			free_percpu(ipvlan->pcpu_stats);
155 			return err;
156 		}
157 	}
158 	port = ipvlan_port_get_rtnl(phy_dev);
159 	port->count += 1;
160 	return 0;
161 }
162 
163 static void ipvlan_uninit(struct net_device *dev)
164 {
165 	struct ipvl_dev *ipvlan = netdev_priv(dev);
166 	struct net_device *phy_dev = ipvlan->phy_dev;
167 	struct ipvl_port *port;
168 
169 	free_percpu(ipvlan->pcpu_stats);
170 
171 	port = ipvlan_port_get_rtnl(phy_dev);
172 	port->count -= 1;
173 	if (!port->count)
174 		ipvlan_port_destroy(port->dev);
175 }
176 
177 static int ipvlan_open(struct net_device *dev)
178 {
179 	struct ipvl_dev *ipvlan = netdev_priv(dev);
180 	struct ipvl_addr *addr;
181 
182 	if (ipvlan->port->mode == IPVLAN_MODE_L3 ||
183 	    ipvlan->port->mode == IPVLAN_MODE_L3S)
184 		dev->flags |= IFF_NOARP;
185 	else
186 		dev->flags &= ~IFF_NOARP;
187 
188 	rcu_read_lock();
189 	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
190 		ipvlan_ht_addr_add(ipvlan, addr);
191 	rcu_read_unlock();
192 
193 	return 0;
194 }
195 
196 static int ipvlan_stop(struct net_device *dev)
197 {
198 	struct ipvl_dev *ipvlan = netdev_priv(dev);
199 	struct net_device *phy_dev = ipvlan->phy_dev;
200 	struct ipvl_addr *addr;
201 
202 	dev_uc_unsync(phy_dev, dev);
203 	dev_mc_unsync(phy_dev, dev);
204 
205 	rcu_read_lock();
206 	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
207 		ipvlan_ht_addr_del(addr);
208 	rcu_read_unlock();
209 
210 	return 0;
211 }
212 
213 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
214 				     struct net_device *dev)
215 {
216 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
217 	int skblen = skb->len;
218 	int ret;
219 
220 	ret = ipvlan_queue_xmit(skb, dev);
221 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
222 		struct ipvl_pcpu_stats *pcptr;
223 
224 		pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
225 
226 		u64_stats_update_begin(&pcptr->syncp);
227 		u64_stats_inc(&pcptr->tx_pkts);
228 		u64_stats_add(&pcptr->tx_bytes, skblen);
229 		u64_stats_update_end(&pcptr->syncp);
230 	} else {
231 		this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
232 	}
233 	return ret;
234 }
235 
236 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
237 					     netdev_features_t features)
238 {
239 	struct ipvl_dev *ipvlan = netdev_priv(dev);
240 
241 	features |= NETIF_F_ALL_FOR_ALL;
242 	features &= (ipvlan->sfeatures | ~IPVLAN_FEATURES);
243 	features = netdev_increment_features(ipvlan->phy_dev->features,
244 					     features, features);
245 	features |= IPVLAN_ALWAYS_ON;
246 	features &= (IPVLAN_FEATURES | IPVLAN_ALWAYS_ON);
247 
248 	return features;
249 }
250 
251 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
252 {
253 	struct ipvl_dev *ipvlan = netdev_priv(dev);
254 	struct net_device *phy_dev = ipvlan->phy_dev;
255 
256 	if (change & IFF_ALLMULTI)
257 		dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
258 }
259 
260 static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
261 {
262 	struct ipvl_dev *ipvlan = netdev_priv(dev);
263 
264 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
265 		bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
266 	} else {
267 		struct netdev_hw_addr *ha;
268 		DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
269 
270 		bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
271 		netdev_for_each_mc_addr(ha, dev)
272 			__set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
273 
274 		/* Turn-on broadcast bit irrespective of address family,
275 		 * since broadcast is deferred to a work-queue, hence no
276 		 * impact on fast-path processing.
277 		 */
278 		__set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
279 
280 		bitmap_copy(ipvlan->mac_filters, mc_filters,
281 			    IPVLAN_MAC_FILTER_SIZE);
282 	}
283 	dev_uc_sync(ipvlan->phy_dev, dev);
284 	dev_mc_sync(ipvlan->phy_dev, dev);
285 }
286 
287 static void ipvlan_get_stats64(struct net_device *dev,
288 			       struct rtnl_link_stats64 *s)
289 {
290 	struct ipvl_dev *ipvlan = netdev_priv(dev);
291 
292 	if (ipvlan->pcpu_stats) {
293 		struct ipvl_pcpu_stats *pcptr;
294 		u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
295 		u32 rx_errs = 0, tx_drps = 0;
296 		u32 strt;
297 		int idx;
298 
299 		for_each_possible_cpu(idx) {
300 			pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
301 			do {
302 				strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
303 				rx_pkts = u64_stats_read(&pcptr->rx_pkts);
304 				rx_bytes = u64_stats_read(&pcptr->rx_bytes);
305 				rx_mcast = u64_stats_read(&pcptr->rx_mcast);
306 				tx_pkts = u64_stats_read(&pcptr->tx_pkts);
307 				tx_bytes = u64_stats_read(&pcptr->tx_bytes);
308 			} while (u64_stats_fetch_retry_irq(&pcptr->syncp,
309 							   strt));
310 
311 			s->rx_packets += rx_pkts;
312 			s->rx_bytes += rx_bytes;
313 			s->multicast += rx_mcast;
314 			s->tx_packets += tx_pkts;
315 			s->tx_bytes += tx_bytes;
316 
317 			/* u32 values are updated without syncp protection. */
318 			rx_errs += READ_ONCE(pcptr->rx_errs);
319 			tx_drps += READ_ONCE(pcptr->tx_drps);
320 		}
321 		s->rx_errors = rx_errs;
322 		s->rx_dropped = rx_errs;
323 		s->tx_dropped = tx_drps;
324 	}
325 }
326 
327 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
328 {
329 	struct ipvl_dev *ipvlan = netdev_priv(dev);
330 	struct net_device *phy_dev = ipvlan->phy_dev;
331 
332 	return vlan_vid_add(phy_dev, proto, vid);
333 }
334 
335 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
336 				   u16 vid)
337 {
338 	struct ipvl_dev *ipvlan = netdev_priv(dev);
339 	struct net_device *phy_dev = ipvlan->phy_dev;
340 
341 	vlan_vid_del(phy_dev, proto, vid);
342 	return 0;
343 }
344 
345 static int ipvlan_get_iflink(const struct net_device *dev)
346 {
347 	struct ipvl_dev *ipvlan = netdev_priv(dev);
348 
349 	return ipvlan->phy_dev->ifindex;
350 }
351 
352 static const struct net_device_ops ipvlan_netdev_ops = {
353 	.ndo_init		= ipvlan_init,
354 	.ndo_uninit		= ipvlan_uninit,
355 	.ndo_open		= ipvlan_open,
356 	.ndo_stop		= ipvlan_stop,
357 	.ndo_start_xmit		= ipvlan_start_xmit,
358 	.ndo_fix_features	= ipvlan_fix_features,
359 	.ndo_change_rx_flags	= ipvlan_change_rx_flags,
360 	.ndo_set_rx_mode	= ipvlan_set_multicast_mac_filter,
361 	.ndo_get_stats64	= ipvlan_get_stats64,
362 	.ndo_vlan_rx_add_vid	= ipvlan_vlan_rx_add_vid,
363 	.ndo_vlan_rx_kill_vid	= ipvlan_vlan_rx_kill_vid,
364 	.ndo_get_iflink		= ipvlan_get_iflink,
365 };
366 
367 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
368 			      unsigned short type, const void *daddr,
369 			      const void *saddr, unsigned len)
370 {
371 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
372 	struct net_device *phy_dev = ipvlan->phy_dev;
373 
374 	/* TODO Probably use a different field than dev_addr so that the
375 	 * mac-address on the virtual device is portable and can be carried
376 	 * while the packets use the mac-addr on the physical device.
377 	 */
378 	return dev_hard_header(skb, phy_dev, type, daddr,
379 			       saddr ? : phy_dev->dev_addr, len);
380 }
381 
382 static const struct header_ops ipvlan_header_ops = {
383 	.create  	= ipvlan_hard_header,
384 	.parse		= eth_header_parse,
385 	.cache		= eth_header_cache,
386 	.cache_update	= eth_header_cache_update,
387 };
388 
389 static void ipvlan_adjust_mtu(struct ipvl_dev *ipvlan, struct net_device *dev)
390 {
391 	ipvlan->dev->mtu = dev->mtu;
392 }
393 
394 static bool netif_is_ipvlan(const struct net_device *dev)
395 {
396 	/* both ipvlan and ipvtap devices use the same netdev_ops */
397 	return dev->netdev_ops == &ipvlan_netdev_ops;
398 }
399 
400 static int ipvlan_ethtool_get_link_ksettings(struct net_device *dev,
401 					     struct ethtool_link_ksettings *cmd)
402 {
403 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
404 
405 	return __ethtool_get_link_ksettings(ipvlan->phy_dev, cmd);
406 }
407 
408 static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
409 				       struct ethtool_drvinfo *drvinfo)
410 {
411 	strscpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
412 	strscpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
413 }
414 
415 static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
416 {
417 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
418 
419 	return ipvlan->msg_enable;
420 }
421 
422 static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
423 {
424 	struct ipvl_dev *ipvlan = netdev_priv(dev);
425 
426 	ipvlan->msg_enable = value;
427 }
428 
429 static const struct ethtool_ops ipvlan_ethtool_ops = {
430 	.get_link	= ethtool_op_get_link,
431 	.get_link_ksettings	= ipvlan_ethtool_get_link_ksettings,
432 	.get_drvinfo	= ipvlan_ethtool_get_drvinfo,
433 	.get_msglevel	= ipvlan_ethtool_get_msglevel,
434 	.set_msglevel	= ipvlan_ethtool_set_msglevel,
435 };
436 
437 static int ipvlan_nl_changelink(struct net_device *dev,
438 				struct nlattr *tb[], struct nlattr *data[],
439 				struct netlink_ext_ack *extack)
440 {
441 	struct ipvl_dev *ipvlan = netdev_priv(dev);
442 	struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
443 	int err = 0;
444 
445 	if (!data)
446 		return 0;
447 	if (!ns_capable(dev_net(ipvlan->phy_dev)->user_ns, CAP_NET_ADMIN))
448 		return -EPERM;
449 
450 	if (data[IFLA_IPVLAN_MODE]) {
451 		u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
452 
453 		err = ipvlan_set_port_mode(port, nmode, extack);
454 	}
455 
456 	if (!err && data[IFLA_IPVLAN_FLAGS]) {
457 		u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
458 
459 		if (flags & IPVLAN_F_PRIVATE)
460 			ipvlan_mark_private(port);
461 		else
462 			ipvlan_clear_private(port);
463 
464 		if (flags & IPVLAN_F_VEPA)
465 			ipvlan_mark_vepa(port);
466 		else
467 			ipvlan_clear_vepa(port);
468 	}
469 
470 	return err;
471 }
472 
473 static size_t ipvlan_nl_getsize(const struct net_device *dev)
474 {
475 	return (0
476 		+ nla_total_size(2) /* IFLA_IPVLAN_MODE */
477 		+ nla_total_size(2) /* IFLA_IPVLAN_FLAGS */
478 		);
479 }
480 
481 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[],
482 			      struct netlink_ext_ack *extack)
483 {
484 	if (!data)
485 		return 0;
486 
487 	if (data[IFLA_IPVLAN_MODE]) {
488 		u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
489 
490 		if (mode >= IPVLAN_MODE_MAX)
491 			return -EINVAL;
492 	}
493 	if (data[IFLA_IPVLAN_FLAGS]) {
494 		u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
495 
496 		/* Only two bits are used at this moment. */
497 		if (flags & ~(IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
498 			return -EINVAL;
499 		/* Also both flags can't be active at the same time. */
500 		if ((flags & (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA)) ==
501 		    (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
502 			return -EINVAL;
503 	}
504 
505 	return 0;
506 }
507 
508 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
509 			      const struct net_device *dev)
510 {
511 	struct ipvl_dev *ipvlan = netdev_priv(dev);
512 	struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
513 	int ret = -EINVAL;
514 
515 	if (!port)
516 		goto err;
517 
518 	ret = -EMSGSIZE;
519 	if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
520 		goto err;
521 	if (nla_put_u16(skb, IFLA_IPVLAN_FLAGS, port->flags))
522 		goto err;
523 
524 	return 0;
525 
526 err:
527 	return ret;
528 }
529 
530 int ipvlan_link_new(struct net *src_net, struct net_device *dev,
531 		    struct nlattr *tb[], struct nlattr *data[],
532 		    struct netlink_ext_ack *extack)
533 {
534 	struct ipvl_dev *ipvlan = netdev_priv(dev);
535 	struct ipvl_port *port;
536 	struct net_device *phy_dev;
537 	int err;
538 	u16 mode = IPVLAN_MODE_L3;
539 
540 	if (!tb[IFLA_LINK])
541 		return -EINVAL;
542 
543 	phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
544 	if (!phy_dev)
545 		return -ENODEV;
546 
547 	if (netif_is_ipvlan(phy_dev)) {
548 		struct ipvl_dev *tmp = netdev_priv(phy_dev);
549 
550 		phy_dev = tmp->phy_dev;
551 		if (!ns_capable(dev_net(phy_dev)->user_ns, CAP_NET_ADMIN))
552 			return -EPERM;
553 	} else if (!netif_is_ipvlan_port(phy_dev)) {
554 		/* Exit early if the underlying link is invalid or busy */
555 		if (phy_dev->type != ARPHRD_ETHER ||
556 		    phy_dev->flags & IFF_LOOPBACK) {
557 			netdev_err(phy_dev,
558 				   "Master is either lo or non-ether device\n");
559 			return -EINVAL;
560 		}
561 
562 		if (netdev_is_rx_handler_busy(phy_dev)) {
563 			netdev_err(phy_dev, "Device is already in use.\n");
564 			return -EBUSY;
565 		}
566 	}
567 
568 	ipvlan->phy_dev = phy_dev;
569 	ipvlan->dev = dev;
570 	ipvlan->sfeatures = IPVLAN_FEATURES;
571 	if (!tb[IFLA_MTU])
572 		ipvlan_adjust_mtu(ipvlan, phy_dev);
573 	INIT_LIST_HEAD(&ipvlan->addrs);
574 	spin_lock_init(&ipvlan->addrs_lock);
575 
576 	/* TODO Probably put random address here to be presented to the
577 	 * world but keep using the physical-dev address for the outgoing
578 	 * packets.
579 	 */
580 	eth_hw_addr_set(dev, phy_dev->dev_addr);
581 
582 	dev->priv_flags |= IFF_NO_RX_HANDLER;
583 
584 	err = register_netdevice(dev);
585 	if (err < 0)
586 		return err;
587 
588 	/* ipvlan_init() would have created the port, if required */
589 	port = ipvlan_port_get_rtnl(phy_dev);
590 	ipvlan->port = port;
591 
592 	/* If the port-id base is at the MAX value, then wrap it around and
593 	 * begin from 0x1 again. This may be due to a busy system where lots
594 	 * of slaves are getting created and deleted.
595 	 */
596 	if (port->dev_id_start == 0xFFFE)
597 		port->dev_id_start = 0x1;
598 
599 	/* Since L2 address is shared among all IPvlan slaves including
600 	 * master, use unique 16 bit dev-ids to diffentiate among them.
601 	 * Assign IDs between 0x1 and 0xFFFE (used by the master) to each
602 	 * slave link [see addrconf_ifid_eui48()].
603 	 */
604 	err = ida_simple_get(&port->ida, port->dev_id_start, 0xFFFE,
605 			     GFP_KERNEL);
606 	if (err < 0)
607 		err = ida_simple_get(&port->ida, 0x1, port->dev_id_start,
608 				     GFP_KERNEL);
609 	if (err < 0)
610 		goto unregister_netdev;
611 	dev->dev_id = err;
612 
613 	/* Increment id-base to the next slot for the future assignment */
614 	port->dev_id_start = err + 1;
615 
616 	err = netdev_upper_dev_link(phy_dev, dev, extack);
617 	if (err)
618 		goto remove_ida;
619 
620 	/* Flags are per port and latest update overrides. User has
621 	 * to be consistent in setting it just like the mode attribute.
622 	 */
623 	if (data && data[IFLA_IPVLAN_FLAGS])
624 		port->flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
625 
626 	if (data && data[IFLA_IPVLAN_MODE])
627 		mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
628 
629 	err = ipvlan_set_port_mode(port, mode, extack);
630 	if (err)
631 		goto unlink_netdev;
632 
633 	list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
634 	netif_stacked_transfer_operstate(phy_dev, dev);
635 	return 0;
636 
637 unlink_netdev:
638 	netdev_upper_dev_unlink(phy_dev, dev);
639 remove_ida:
640 	ida_simple_remove(&port->ida, dev->dev_id);
641 unregister_netdev:
642 	unregister_netdevice(dev);
643 	return err;
644 }
645 EXPORT_SYMBOL_GPL(ipvlan_link_new);
646 
647 void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
648 {
649 	struct ipvl_dev *ipvlan = netdev_priv(dev);
650 	struct ipvl_addr *addr, *next;
651 
652 	spin_lock_bh(&ipvlan->addrs_lock);
653 	list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
654 		ipvlan_ht_addr_del(addr);
655 		list_del_rcu(&addr->anode);
656 		kfree_rcu(addr, rcu);
657 	}
658 	spin_unlock_bh(&ipvlan->addrs_lock);
659 
660 	ida_simple_remove(&ipvlan->port->ida, dev->dev_id);
661 	list_del_rcu(&ipvlan->pnode);
662 	unregister_netdevice_queue(dev, head);
663 	netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
664 }
665 EXPORT_SYMBOL_GPL(ipvlan_link_delete);
666 
667 void ipvlan_link_setup(struct net_device *dev)
668 {
669 	ether_setup(dev);
670 
671 	dev->max_mtu = ETH_MAX_MTU;
672 	dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
673 	dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
674 	dev->netdev_ops = &ipvlan_netdev_ops;
675 	dev->needs_free_netdev = true;
676 	dev->header_ops = &ipvlan_header_ops;
677 	dev->ethtool_ops = &ipvlan_ethtool_ops;
678 }
679 EXPORT_SYMBOL_GPL(ipvlan_link_setup);
680 
681 static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
682 {
683 	[IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
684 	[IFLA_IPVLAN_FLAGS] = { .type = NLA_U16 },
685 };
686 
687 static struct net *ipvlan_get_link_net(const struct net_device *dev)
688 {
689 	struct ipvl_dev *ipvlan = netdev_priv(dev);
690 
691 	return dev_net(ipvlan->phy_dev);
692 }
693 
694 static struct rtnl_link_ops ipvlan_link_ops = {
695 	.kind		= "ipvlan",
696 	.priv_size	= sizeof(struct ipvl_dev),
697 
698 	.setup		= ipvlan_link_setup,
699 	.newlink	= ipvlan_link_new,
700 	.dellink	= ipvlan_link_delete,
701 	.get_link_net   = ipvlan_get_link_net,
702 };
703 
704 int ipvlan_link_register(struct rtnl_link_ops *ops)
705 {
706 	ops->get_size	= ipvlan_nl_getsize;
707 	ops->policy	= ipvlan_nl_policy;
708 	ops->validate	= ipvlan_nl_validate;
709 	ops->fill_info	= ipvlan_nl_fillinfo;
710 	ops->changelink = ipvlan_nl_changelink;
711 	ops->maxtype	= IFLA_IPVLAN_MAX;
712 	return rtnl_link_register(ops);
713 }
714 EXPORT_SYMBOL_GPL(ipvlan_link_register);
715 
716 static int ipvlan_device_event(struct notifier_block *unused,
717 			       unsigned long event, void *ptr)
718 {
719 	struct netlink_ext_ack *extack = netdev_notifier_info_to_extack(ptr);
720 	struct netdev_notifier_pre_changeaddr_info *prechaddr_info;
721 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
722 	struct ipvl_dev *ipvlan, *next;
723 	struct ipvl_port *port;
724 	LIST_HEAD(lst_kill);
725 	int err;
726 
727 	if (!netif_is_ipvlan_port(dev))
728 		return NOTIFY_DONE;
729 
730 	port = ipvlan_port_get_rtnl(dev);
731 
732 	switch (event) {
733 	case NETDEV_UP:
734 	case NETDEV_CHANGE:
735 		list_for_each_entry(ipvlan, &port->ipvlans, pnode)
736 			netif_stacked_transfer_operstate(ipvlan->phy_dev,
737 							 ipvlan->dev);
738 		break;
739 
740 	case NETDEV_REGISTER: {
741 		struct net *oldnet, *newnet = dev_net(dev);
742 
743 		oldnet = read_pnet(&port->pnet);
744 		if (net_eq(newnet, oldnet))
745 			break;
746 
747 		write_pnet(&port->pnet, newnet);
748 
749 		ipvlan_migrate_l3s_hook(oldnet, newnet);
750 		break;
751 	}
752 	case NETDEV_UNREGISTER:
753 		if (dev->reg_state != NETREG_UNREGISTERING)
754 			break;
755 
756 		list_for_each_entry_safe(ipvlan, next, &port->ipvlans, pnode)
757 			ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
758 							    &lst_kill);
759 		unregister_netdevice_many(&lst_kill);
760 		break;
761 
762 	case NETDEV_FEAT_CHANGE:
763 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
764 			netif_inherit_tso_max(ipvlan->dev, dev);
765 			netdev_update_features(ipvlan->dev);
766 		}
767 		break;
768 
769 	case NETDEV_CHANGEMTU:
770 		list_for_each_entry(ipvlan, &port->ipvlans, pnode)
771 			ipvlan_adjust_mtu(ipvlan, dev);
772 		break;
773 
774 	case NETDEV_PRE_CHANGEADDR:
775 		prechaddr_info = ptr;
776 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
777 			err = dev_pre_changeaddr_notify(ipvlan->dev,
778 						    prechaddr_info->dev_addr,
779 						    extack);
780 			if (err)
781 				return notifier_from_errno(err);
782 		}
783 		break;
784 
785 	case NETDEV_CHANGEADDR:
786 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
787 			eth_hw_addr_set(ipvlan->dev, dev->dev_addr);
788 			call_netdevice_notifiers(NETDEV_CHANGEADDR, ipvlan->dev);
789 		}
790 		break;
791 
792 	case NETDEV_PRE_TYPE_CHANGE:
793 		/* Forbid underlying device to change its type. */
794 		return NOTIFY_BAD;
795 	}
796 	return NOTIFY_DONE;
797 }
798 
799 /* the caller must held the addrs lock */
800 static int ipvlan_add_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
801 {
802 	struct ipvl_addr *addr;
803 
804 	addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
805 	if (!addr)
806 		return -ENOMEM;
807 
808 	addr->master = ipvlan;
809 	if (!is_v6) {
810 		memcpy(&addr->ip4addr, iaddr, sizeof(struct in_addr));
811 		addr->atype = IPVL_IPV4;
812 #if IS_ENABLED(CONFIG_IPV6)
813 	} else {
814 		memcpy(&addr->ip6addr, iaddr, sizeof(struct in6_addr));
815 		addr->atype = IPVL_IPV6;
816 #endif
817 	}
818 
819 	list_add_tail_rcu(&addr->anode, &ipvlan->addrs);
820 
821 	/* If the interface is not up, the address will be added to the hash
822 	 * list by ipvlan_open.
823 	 */
824 	if (netif_running(ipvlan->dev))
825 		ipvlan_ht_addr_add(ipvlan, addr);
826 
827 	return 0;
828 }
829 
830 static void ipvlan_del_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
831 {
832 	struct ipvl_addr *addr;
833 
834 	spin_lock_bh(&ipvlan->addrs_lock);
835 	addr = ipvlan_find_addr(ipvlan, iaddr, is_v6);
836 	if (!addr) {
837 		spin_unlock_bh(&ipvlan->addrs_lock);
838 		return;
839 	}
840 
841 	ipvlan_ht_addr_del(addr);
842 	list_del_rcu(&addr->anode);
843 	spin_unlock_bh(&ipvlan->addrs_lock);
844 	kfree_rcu(addr, rcu);
845 }
846 
847 static bool ipvlan_is_valid_dev(const struct net_device *dev)
848 {
849 	struct ipvl_dev *ipvlan = netdev_priv(dev);
850 
851 	if (!netif_is_ipvlan(dev))
852 		return false;
853 
854 	if (!ipvlan || !ipvlan->port)
855 		return false;
856 
857 	return true;
858 }
859 
860 #if IS_ENABLED(CONFIG_IPV6)
861 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
862 {
863 	int ret = -EINVAL;
864 
865 	spin_lock_bh(&ipvlan->addrs_lock);
866 	if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true))
867 		netif_err(ipvlan, ifup, ipvlan->dev,
868 			  "Failed to add IPv6=%pI6c addr for %s intf\n",
869 			  ip6_addr, ipvlan->dev->name);
870 	else
871 		ret = ipvlan_add_addr(ipvlan, ip6_addr, true);
872 	spin_unlock_bh(&ipvlan->addrs_lock);
873 	return ret;
874 }
875 
876 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
877 {
878 	return ipvlan_del_addr(ipvlan, ip6_addr, true);
879 }
880 
881 static int ipvlan_addr6_event(struct notifier_block *unused,
882 			      unsigned long event, void *ptr)
883 {
884 	struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
885 	struct net_device *dev = (struct net_device *)if6->idev->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_add_addr6(ipvlan, &if6->addr))
894 			return NOTIFY_BAD;
895 		break;
896 
897 	case NETDEV_DOWN:
898 		ipvlan_del_addr6(ipvlan, &if6->addr);
899 		break;
900 	}
901 
902 	return NOTIFY_OK;
903 }
904 
905 static int ipvlan_addr6_validator_event(struct notifier_block *unused,
906 					unsigned long event, void *ptr)
907 {
908 	struct in6_validator_info *i6vi = (struct in6_validator_info *)ptr;
909 	struct net_device *dev = (struct net_device *)i6vi->i6vi_dev->dev;
910 	struct ipvl_dev *ipvlan = netdev_priv(dev);
911 
912 	if (!ipvlan_is_valid_dev(dev))
913 		return NOTIFY_DONE;
914 
915 	switch (event) {
916 	case NETDEV_UP:
917 		if (ipvlan_addr_busy(ipvlan->port, &i6vi->i6vi_addr, true)) {
918 			NL_SET_ERR_MSG(i6vi->extack,
919 				       "Address already assigned to an ipvlan device");
920 			return notifier_from_errno(-EADDRINUSE);
921 		}
922 		break;
923 	}
924 
925 	return NOTIFY_OK;
926 }
927 #endif
928 
929 static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
930 {
931 	int ret = -EINVAL;
932 
933 	spin_lock_bh(&ipvlan->addrs_lock);
934 	if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false))
935 		netif_err(ipvlan, ifup, ipvlan->dev,
936 			  "Failed to add IPv4=%pI4 on %s intf.\n",
937 			  ip4_addr, ipvlan->dev->name);
938 	else
939 		ret = ipvlan_add_addr(ipvlan, ip4_addr, false);
940 	spin_unlock_bh(&ipvlan->addrs_lock);
941 	return ret;
942 }
943 
944 static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
945 {
946 	return ipvlan_del_addr(ipvlan, ip4_addr, false);
947 }
948 
949 static int ipvlan_addr4_event(struct notifier_block *unused,
950 			      unsigned long event, void *ptr)
951 {
952 	struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
953 	struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
954 	struct ipvl_dev *ipvlan = netdev_priv(dev);
955 	struct in_addr ip4_addr;
956 
957 	if (!ipvlan_is_valid_dev(dev))
958 		return NOTIFY_DONE;
959 
960 	switch (event) {
961 	case NETDEV_UP:
962 		ip4_addr.s_addr = if4->ifa_address;
963 		if (ipvlan_add_addr4(ipvlan, &ip4_addr))
964 			return NOTIFY_BAD;
965 		break;
966 
967 	case NETDEV_DOWN:
968 		ip4_addr.s_addr = if4->ifa_address;
969 		ipvlan_del_addr4(ipvlan, &ip4_addr);
970 		break;
971 	}
972 
973 	return NOTIFY_OK;
974 }
975 
976 static int ipvlan_addr4_validator_event(struct notifier_block *unused,
977 					unsigned long event, void *ptr)
978 {
979 	struct in_validator_info *ivi = (struct in_validator_info *)ptr;
980 	struct net_device *dev = (struct net_device *)ivi->ivi_dev->dev;
981 	struct ipvl_dev *ipvlan = netdev_priv(dev);
982 
983 	if (!ipvlan_is_valid_dev(dev))
984 		return NOTIFY_DONE;
985 
986 	switch (event) {
987 	case NETDEV_UP:
988 		if (ipvlan_addr_busy(ipvlan->port, &ivi->ivi_addr, false)) {
989 			NL_SET_ERR_MSG(ivi->extack,
990 				       "Address already assigned to an ipvlan device");
991 			return notifier_from_errno(-EADDRINUSE);
992 		}
993 		break;
994 	}
995 
996 	return NOTIFY_OK;
997 }
998 
999 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
1000 	.notifier_call = ipvlan_addr4_event,
1001 };
1002 
1003 static struct notifier_block ipvlan_addr4_vtor_notifier_block __read_mostly = {
1004 	.notifier_call = ipvlan_addr4_validator_event,
1005 };
1006 
1007 static struct notifier_block ipvlan_notifier_block __read_mostly = {
1008 	.notifier_call = ipvlan_device_event,
1009 };
1010 
1011 #if IS_ENABLED(CONFIG_IPV6)
1012 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
1013 	.notifier_call = ipvlan_addr6_event,
1014 };
1015 
1016 static struct notifier_block ipvlan_addr6_vtor_notifier_block __read_mostly = {
1017 	.notifier_call = ipvlan_addr6_validator_event,
1018 };
1019 #endif
1020 
1021 static int __init ipvlan_init_module(void)
1022 {
1023 	int err;
1024 
1025 	ipvlan_init_secret();
1026 	register_netdevice_notifier(&ipvlan_notifier_block);
1027 #if IS_ENABLED(CONFIG_IPV6)
1028 	register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1029 	register_inet6addr_validator_notifier(
1030 	    &ipvlan_addr6_vtor_notifier_block);
1031 #endif
1032 	register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1033 	register_inetaddr_validator_notifier(&ipvlan_addr4_vtor_notifier_block);
1034 
1035 	err = ipvlan_l3s_init();
1036 	if (err < 0)
1037 		goto error;
1038 
1039 	err = ipvlan_link_register(&ipvlan_link_ops);
1040 	if (err < 0) {
1041 		ipvlan_l3s_cleanup();
1042 		goto error;
1043 	}
1044 
1045 	return 0;
1046 error:
1047 	unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1048 	unregister_inetaddr_validator_notifier(
1049 	    &ipvlan_addr4_vtor_notifier_block);
1050 #if IS_ENABLED(CONFIG_IPV6)
1051 	unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1052 	unregister_inet6addr_validator_notifier(
1053 	    &ipvlan_addr6_vtor_notifier_block);
1054 #endif
1055 	unregister_netdevice_notifier(&ipvlan_notifier_block);
1056 	return err;
1057 }
1058 
1059 static void __exit ipvlan_cleanup_module(void)
1060 {
1061 	rtnl_link_unregister(&ipvlan_link_ops);
1062 	ipvlan_l3s_cleanup();
1063 	unregister_netdevice_notifier(&ipvlan_notifier_block);
1064 	unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1065 	unregister_inetaddr_validator_notifier(
1066 	    &ipvlan_addr4_vtor_notifier_block);
1067 #if IS_ENABLED(CONFIG_IPV6)
1068 	unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1069 	unregister_inet6addr_validator_notifier(
1070 	    &ipvlan_addr6_vtor_notifier_block);
1071 #endif
1072 }
1073 
1074 module_init(ipvlan_init_module);
1075 module_exit(ipvlan_cleanup_module);
1076 
1077 MODULE_LICENSE("GPL");
1078 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
1079 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
1080 MODULE_ALIAS_RTNL_LINK("ipvlan");
1081