xref: /openbmc/linux/drivers/net/ipvlan/ipvlan_main.c (revision 4419617e)
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_TSO | NETIF_F_GSO_ROBUST | \
112 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | 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->gso_max_size = phy_dev->gso_max_size;
130 	dev->gso_max_segs = phy_dev->gso_max_segs;
131 	dev->hard_header_len = phy_dev->hard_header_len;
132 
133 	netdev_lockdep_set_classes(dev);
134 
135 	ipvlan->pcpu_stats = netdev_alloc_pcpu_stats(struct ipvl_pcpu_stats);
136 	if (!ipvlan->pcpu_stats)
137 		return -ENOMEM;
138 
139 	if (!netif_is_ipvlan_port(phy_dev)) {
140 		err = ipvlan_port_create(phy_dev);
141 		if (err < 0) {
142 			free_percpu(ipvlan->pcpu_stats);
143 			return err;
144 		}
145 	}
146 	port = ipvlan_port_get_rtnl(phy_dev);
147 	port->count += 1;
148 	return 0;
149 }
150 
151 static void ipvlan_uninit(struct net_device *dev)
152 {
153 	struct ipvl_dev *ipvlan = netdev_priv(dev);
154 	struct net_device *phy_dev = ipvlan->phy_dev;
155 	struct ipvl_port *port;
156 
157 	free_percpu(ipvlan->pcpu_stats);
158 
159 	port = ipvlan_port_get_rtnl(phy_dev);
160 	port->count -= 1;
161 	if (!port->count)
162 		ipvlan_port_destroy(port->dev);
163 }
164 
165 static int ipvlan_open(struct net_device *dev)
166 {
167 	struct ipvl_dev *ipvlan = netdev_priv(dev);
168 	struct net_device *phy_dev = ipvlan->phy_dev;
169 	struct ipvl_addr *addr;
170 
171 	if (ipvlan->port->mode == IPVLAN_MODE_L3 ||
172 	    ipvlan->port->mode == IPVLAN_MODE_L3S)
173 		dev->flags |= IFF_NOARP;
174 	else
175 		dev->flags &= ~IFF_NOARP;
176 
177 	rcu_read_lock();
178 	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
179 		ipvlan_ht_addr_add(ipvlan, addr);
180 	rcu_read_unlock();
181 
182 	return dev_uc_add(phy_dev, phy_dev->dev_addr);
183 }
184 
185 static int ipvlan_stop(struct net_device *dev)
186 {
187 	struct ipvl_dev *ipvlan = netdev_priv(dev);
188 	struct net_device *phy_dev = ipvlan->phy_dev;
189 	struct ipvl_addr *addr;
190 
191 	dev_uc_unsync(phy_dev, dev);
192 	dev_mc_unsync(phy_dev, dev);
193 
194 	dev_uc_del(phy_dev, phy_dev->dev_addr);
195 
196 	rcu_read_lock();
197 	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
198 		ipvlan_ht_addr_del(addr);
199 	rcu_read_unlock();
200 
201 	return 0;
202 }
203 
204 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
205 				     struct net_device *dev)
206 {
207 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
208 	int skblen = skb->len;
209 	int ret;
210 
211 	ret = ipvlan_queue_xmit(skb, dev);
212 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
213 		struct ipvl_pcpu_stats *pcptr;
214 
215 		pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
216 
217 		u64_stats_update_begin(&pcptr->syncp);
218 		pcptr->tx_pkts++;
219 		pcptr->tx_bytes += skblen;
220 		u64_stats_update_end(&pcptr->syncp);
221 	} else {
222 		this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
223 	}
224 	return ret;
225 }
226 
227 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
228 					     netdev_features_t features)
229 {
230 	struct ipvl_dev *ipvlan = netdev_priv(dev);
231 
232 	return features & (ipvlan->sfeatures | ~IPVLAN_FEATURES);
233 }
234 
235 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
236 {
237 	struct ipvl_dev *ipvlan = netdev_priv(dev);
238 	struct net_device *phy_dev = ipvlan->phy_dev;
239 
240 	if (change & IFF_ALLMULTI)
241 		dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
242 }
243 
244 static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
245 {
246 	struct ipvl_dev *ipvlan = netdev_priv(dev);
247 
248 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
249 		bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
250 	} else {
251 		struct netdev_hw_addr *ha;
252 		DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
253 
254 		bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
255 		netdev_for_each_mc_addr(ha, dev)
256 			__set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
257 
258 		/* Turn-on broadcast bit irrespective of address family,
259 		 * since broadcast is deferred to a work-queue, hence no
260 		 * impact on fast-path processing.
261 		 */
262 		__set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
263 
264 		bitmap_copy(ipvlan->mac_filters, mc_filters,
265 			    IPVLAN_MAC_FILTER_SIZE);
266 	}
267 	dev_uc_sync(ipvlan->phy_dev, dev);
268 	dev_mc_sync(ipvlan->phy_dev, dev);
269 }
270 
271 static void ipvlan_get_stats64(struct net_device *dev,
272 			       struct rtnl_link_stats64 *s)
273 {
274 	struct ipvl_dev *ipvlan = netdev_priv(dev);
275 
276 	if (ipvlan->pcpu_stats) {
277 		struct ipvl_pcpu_stats *pcptr;
278 		u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
279 		u32 rx_errs = 0, tx_drps = 0;
280 		u32 strt;
281 		int idx;
282 
283 		for_each_possible_cpu(idx) {
284 			pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
285 			do {
286 				strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
287 				rx_pkts = pcptr->rx_pkts;
288 				rx_bytes = pcptr->rx_bytes;
289 				rx_mcast = pcptr->rx_mcast;
290 				tx_pkts = pcptr->tx_pkts;
291 				tx_bytes = pcptr->tx_bytes;
292 			} while (u64_stats_fetch_retry_irq(&pcptr->syncp,
293 							   strt));
294 
295 			s->rx_packets += rx_pkts;
296 			s->rx_bytes += rx_bytes;
297 			s->multicast += rx_mcast;
298 			s->tx_packets += tx_pkts;
299 			s->tx_bytes += tx_bytes;
300 
301 			/* u32 values are updated without syncp protection. */
302 			rx_errs += pcptr->rx_errs;
303 			tx_drps += pcptr->tx_drps;
304 		}
305 		s->rx_errors = rx_errs;
306 		s->rx_dropped = rx_errs;
307 		s->tx_dropped = tx_drps;
308 	}
309 }
310 
311 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
312 {
313 	struct ipvl_dev *ipvlan = netdev_priv(dev);
314 	struct net_device *phy_dev = ipvlan->phy_dev;
315 
316 	return vlan_vid_add(phy_dev, proto, vid);
317 }
318 
319 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
320 				   u16 vid)
321 {
322 	struct ipvl_dev *ipvlan = netdev_priv(dev);
323 	struct net_device *phy_dev = ipvlan->phy_dev;
324 
325 	vlan_vid_del(phy_dev, proto, vid);
326 	return 0;
327 }
328 
329 static int ipvlan_get_iflink(const struct net_device *dev)
330 {
331 	struct ipvl_dev *ipvlan = netdev_priv(dev);
332 
333 	return ipvlan->phy_dev->ifindex;
334 }
335 
336 static const struct net_device_ops ipvlan_netdev_ops = {
337 	.ndo_init		= ipvlan_init,
338 	.ndo_uninit		= ipvlan_uninit,
339 	.ndo_open		= ipvlan_open,
340 	.ndo_stop		= ipvlan_stop,
341 	.ndo_start_xmit		= ipvlan_start_xmit,
342 	.ndo_fix_features	= ipvlan_fix_features,
343 	.ndo_change_rx_flags	= ipvlan_change_rx_flags,
344 	.ndo_set_rx_mode	= ipvlan_set_multicast_mac_filter,
345 	.ndo_get_stats64	= ipvlan_get_stats64,
346 	.ndo_vlan_rx_add_vid	= ipvlan_vlan_rx_add_vid,
347 	.ndo_vlan_rx_kill_vid	= ipvlan_vlan_rx_kill_vid,
348 	.ndo_get_iflink		= ipvlan_get_iflink,
349 };
350 
351 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
352 			      unsigned short type, const void *daddr,
353 			      const void *saddr, unsigned len)
354 {
355 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
356 	struct net_device *phy_dev = ipvlan->phy_dev;
357 
358 	/* TODO Probably use a different field than dev_addr so that the
359 	 * mac-address on the virtual device is portable and can be carried
360 	 * while the packets use the mac-addr on the physical device.
361 	 */
362 	return dev_hard_header(skb, phy_dev, type, daddr,
363 			       saddr ? : phy_dev->dev_addr, len);
364 }
365 
366 static const struct header_ops ipvlan_header_ops = {
367 	.create  	= ipvlan_hard_header,
368 	.parse		= eth_header_parse,
369 	.cache		= eth_header_cache,
370 	.cache_update	= eth_header_cache_update,
371 };
372 
373 static void ipvlan_adjust_mtu(struct ipvl_dev *ipvlan, struct net_device *dev)
374 {
375 	ipvlan->dev->mtu = dev->mtu;
376 }
377 
378 static bool netif_is_ipvlan(const struct net_device *dev)
379 {
380 	/* both ipvlan and ipvtap devices use the same netdev_ops */
381 	return dev->netdev_ops == &ipvlan_netdev_ops;
382 }
383 
384 static int ipvlan_ethtool_get_link_ksettings(struct net_device *dev,
385 					     struct ethtool_link_ksettings *cmd)
386 {
387 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
388 
389 	return __ethtool_get_link_ksettings(ipvlan->phy_dev, cmd);
390 }
391 
392 static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
393 				       struct ethtool_drvinfo *drvinfo)
394 {
395 	strlcpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
396 	strlcpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
397 }
398 
399 static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
400 {
401 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
402 
403 	return ipvlan->msg_enable;
404 }
405 
406 static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
407 {
408 	struct ipvl_dev *ipvlan = netdev_priv(dev);
409 
410 	ipvlan->msg_enable = value;
411 }
412 
413 static const struct ethtool_ops ipvlan_ethtool_ops = {
414 	.get_link	= ethtool_op_get_link,
415 	.get_link_ksettings	= ipvlan_ethtool_get_link_ksettings,
416 	.get_drvinfo	= ipvlan_ethtool_get_drvinfo,
417 	.get_msglevel	= ipvlan_ethtool_get_msglevel,
418 	.set_msglevel	= ipvlan_ethtool_set_msglevel,
419 };
420 
421 static int ipvlan_nl_changelink(struct net_device *dev,
422 				struct nlattr *tb[], struct nlattr *data[],
423 				struct netlink_ext_ack *extack)
424 {
425 	struct ipvl_dev *ipvlan = netdev_priv(dev);
426 	struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
427 	int err = 0;
428 
429 	if (!data)
430 		return 0;
431 	if (!ns_capable(dev_net(ipvlan->phy_dev)->user_ns, CAP_NET_ADMIN))
432 		return -EPERM;
433 
434 	if (data[IFLA_IPVLAN_MODE]) {
435 		u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
436 
437 		err = ipvlan_set_port_mode(port, nmode, extack);
438 	}
439 
440 	if (!err && data[IFLA_IPVLAN_FLAGS]) {
441 		u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
442 
443 		if (flags & IPVLAN_F_PRIVATE)
444 			ipvlan_mark_private(port);
445 		else
446 			ipvlan_clear_private(port);
447 
448 		if (flags & IPVLAN_F_VEPA)
449 			ipvlan_mark_vepa(port);
450 		else
451 			ipvlan_clear_vepa(port);
452 	}
453 
454 	return err;
455 }
456 
457 static size_t ipvlan_nl_getsize(const struct net_device *dev)
458 {
459 	return (0
460 		+ nla_total_size(2) /* IFLA_IPVLAN_MODE */
461 		+ nla_total_size(2) /* IFLA_IPVLAN_FLAGS */
462 		);
463 }
464 
465 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[],
466 			      struct netlink_ext_ack *extack)
467 {
468 	if (!data)
469 		return 0;
470 
471 	if (data[IFLA_IPVLAN_MODE]) {
472 		u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
473 
474 		if (mode >= IPVLAN_MODE_MAX)
475 			return -EINVAL;
476 	}
477 	if (data[IFLA_IPVLAN_FLAGS]) {
478 		u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
479 
480 		/* Only two bits are used at this moment. */
481 		if (flags & ~(IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
482 			return -EINVAL;
483 		/* Also both flags can't be active at the same time. */
484 		if ((flags & (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA)) ==
485 		    (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
486 			return -EINVAL;
487 	}
488 
489 	return 0;
490 }
491 
492 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
493 			      const struct net_device *dev)
494 {
495 	struct ipvl_dev *ipvlan = netdev_priv(dev);
496 	struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
497 	int ret = -EINVAL;
498 
499 	if (!port)
500 		goto err;
501 
502 	ret = -EMSGSIZE;
503 	if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
504 		goto err;
505 	if (nla_put_u16(skb, IFLA_IPVLAN_FLAGS, port->flags))
506 		goto err;
507 
508 	return 0;
509 
510 err:
511 	return ret;
512 }
513 
514 int ipvlan_link_new(struct net *src_net, struct net_device *dev,
515 		    struct nlattr *tb[], struct nlattr *data[],
516 		    struct netlink_ext_ack *extack)
517 {
518 	struct ipvl_dev *ipvlan = netdev_priv(dev);
519 	struct ipvl_port *port;
520 	struct net_device *phy_dev;
521 	int err;
522 	u16 mode = IPVLAN_MODE_L3;
523 
524 	if (!tb[IFLA_LINK])
525 		return -EINVAL;
526 
527 	phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
528 	if (!phy_dev)
529 		return -ENODEV;
530 
531 	if (netif_is_ipvlan(phy_dev)) {
532 		struct ipvl_dev *tmp = netdev_priv(phy_dev);
533 
534 		phy_dev = tmp->phy_dev;
535 		if (!ns_capable(dev_net(phy_dev)->user_ns, CAP_NET_ADMIN))
536 			return -EPERM;
537 	} else if (!netif_is_ipvlan_port(phy_dev)) {
538 		/* Exit early if the underlying link is invalid or busy */
539 		if (phy_dev->type != ARPHRD_ETHER ||
540 		    phy_dev->flags & IFF_LOOPBACK) {
541 			netdev_err(phy_dev,
542 				   "Master is either lo or non-ether device\n");
543 			return -EINVAL;
544 		}
545 
546 		if (netdev_is_rx_handler_busy(phy_dev)) {
547 			netdev_err(phy_dev, "Device is already in use.\n");
548 			return -EBUSY;
549 		}
550 	}
551 
552 	ipvlan->phy_dev = phy_dev;
553 	ipvlan->dev = dev;
554 	ipvlan->sfeatures = IPVLAN_FEATURES;
555 	if (!tb[IFLA_MTU])
556 		ipvlan_adjust_mtu(ipvlan, phy_dev);
557 	INIT_LIST_HEAD(&ipvlan->addrs);
558 	spin_lock_init(&ipvlan->addrs_lock);
559 
560 	/* TODO Probably put random address here to be presented to the
561 	 * world but keep using the physical-dev address for the outgoing
562 	 * packets.
563 	 */
564 	memcpy(dev->dev_addr, phy_dev->dev_addr, ETH_ALEN);
565 
566 	dev->priv_flags |= IFF_NO_RX_HANDLER;
567 
568 	err = register_netdevice(dev);
569 	if (err < 0)
570 		return err;
571 
572 	/* ipvlan_init() would have created the port, if required */
573 	port = ipvlan_port_get_rtnl(phy_dev);
574 	ipvlan->port = port;
575 
576 	/* If the port-id base is at the MAX value, then wrap it around and
577 	 * begin from 0x1 again. This may be due to a busy system where lots
578 	 * of slaves are getting created and deleted.
579 	 */
580 	if (port->dev_id_start == 0xFFFE)
581 		port->dev_id_start = 0x1;
582 
583 	/* Since L2 address is shared among all IPvlan slaves including
584 	 * master, use unique 16 bit dev-ids to diffentiate among them.
585 	 * Assign IDs between 0x1 and 0xFFFE (used by the master) to each
586 	 * slave link [see addrconf_ifid_eui48()].
587 	 */
588 	err = ida_simple_get(&port->ida, port->dev_id_start, 0xFFFE,
589 			     GFP_KERNEL);
590 	if (err < 0)
591 		err = ida_simple_get(&port->ida, 0x1, port->dev_id_start,
592 				     GFP_KERNEL);
593 	if (err < 0)
594 		goto unregister_netdev;
595 	dev->dev_id = err;
596 
597 	/* Increment id-base to the next slot for the future assignment */
598 	port->dev_id_start = err + 1;
599 
600 	err = netdev_upper_dev_link(phy_dev, dev, extack);
601 	if (err)
602 		goto remove_ida;
603 
604 	/* Flags are per port and latest update overrides. User has
605 	 * to be consistent in setting it just like the mode attribute.
606 	 */
607 	if (data && data[IFLA_IPVLAN_FLAGS])
608 		port->flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
609 
610 	if (data && data[IFLA_IPVLAN_MODE])
611 		mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
612 
613 	err = ipvlan_set_port_mode(port, mode, extack);
614 	if (err)
615 		goto unlink_netdev;
616 
617 	list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
618 	netif_stacked_transfer_operstate(phy_dev, dev);
619 	return 0;
620 
621 unlink_netdev:
622 	netdev_upper_dev_unlink(phy_dev, dev);
623 remove_ida:
624 	ida_simple_remove(&port->ida, dev->dev_id);
625 unregister_netdev:
626 	unregister_netdevice(dev);
627 	return err;
628 }
629 EXPORT_SYMBOL_GPL(ipvlan_link_new);
630 
631 void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
632 {
633 	struct ipvl_dev *ipvlan = netdev_priv(dev);
634 	struct ipvl_addr *addr, *next;
635 
636 	spin_lock_bh(&ipvlan->addrs_lock);
637 	list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
638 		ipvlan_ht_addr_del(addr);
639 		list_del_rcu(&addr->anode);
640 		kfree_rcu(addr, rcu);
641 	}
642 	spin_unlock_bh(&ipvlan->addrs_lock);
643 
644 	ida_simple_remove(&ipvlan->port->ida, dev->dev_id);
645 	list_del_rcu(&ipvlan->pnode);
646 	unregister_netdevice_queue(dev, head);
647 	netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
648 }
649 EXPORT_SYMBOL_GPL(ipvlan_link_delete);
650 
651 void ipvlan_link_setup(struct net_device *dev)
652 {
653 	ether_setup(dev);
654 
655 	dev->max_mtu = ETH_MAX_MTU;
656 	dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
657 	dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
658 	dev->netdev_ops = &ipvlan_netdev_ops;
659 	dev->needs_free_netdev = true;
660 	dev->header_ops = &ipvlan_header_ops;
661 	dev->ethtool_ops = &ipvlan_ethtool_ops;
662 }
663 EXPORT_SYMBOL_GPL(ipvlan_link_setup);
664 
665 static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
666 {
667 	[IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
668 	[IFLA_IPVLAN_FLAGS] = { .type = NLA_U16 },
669 };
670 
671 static struct rtnl_link_ops ipvlan_link_ops = {
672 	.kind		= "ipvlan",
673 	.priv_size	= sizeof(struct ipvl_dev),
674 
675 	.setup		= ipvlan_link_setup,
676 	.newlink	= ipvlan_link_new,
677 	.dellink	= ipvlan_link_delete,
678 };
679 
680 int ipvlan_link_register(struct rtnl_link_ops *ops)
681 {
682 	ops->get_size	= ipvlan_nl_getsize;
683 	ops->policy	= ipvlan_nl_policy;
684 	ops->validate	= ipvlan_nl_validate;
685 	ops->fill_info	= ipvlan_nl_fillinfo;
686 	ops->changelink = ipvlan_nl_changelink;
687 	ops->maxtype	= IFLA_IPVLAN_MAX;
688 	return rtnl_link_register(ops);
689 }
690 EXPORT_SYMBOL_GPL(ipvlan_link_register);
691 
692 static int ipvlan_device_event(struct notifier_block *unused,
693 			       unsigned long event, void *ptr)
694 {
695 	struct netlink_ext_ack *extack = netdev_notifier_info_to_extack(ptr);
696 	struct netdev_notifier_pre_changeaddr_info *prechaddr_info;
697 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
698 	struct ipvl_dev *ipvlan, *next;
699 	struct ipvl_port *port;
700 	LIST_HEAD(lst_kill);
701 	int err;
702 
703 	if (!netif_is_ipvlan_port(dev))
704 		return NOTIFY_DONE;
705 
706 	port = ipvlan_port_get_rtnl(dev);
707 
708 	switch (event) {
709 	case NETDEV_CHANGE:
710 		list_for_each_entry(ipvlan, &port->ipvlans, pnode)
711 			netif_stacked_transfer_operstate(ipvlan->phy_dev,
712 							 ipvlan->dev);
713 		break;
714 
715 	case NETDEV_REGISTER: {
716 		struct net *oldnet, *newnet = dev_net(dev);
717 
718 		oldnet = read_pnet(&port->pnet);
719 		if (net_eq(newnet, oldnet))
720 			break;
721 
722 		write_pnet(&port->pnet, newnet);
723 
724 		ipvlan_migrate_l3s_hook(oldnet, newnet);
725 		break;
726 	}
727 	case NETDEV_UNREGISTER:
728 		if (dev->reg_state != NETREG_UNREGISTERING)
729 			break;
730 
731 		list_for_each_entry_safe(ipvlan, next, &port->ipvlans, pnode)
732 			ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
733 							    &lst_kill);
734 		unregister_netdevice_many(&lst_kill);
735 		break;
736 
737 	case NETDEV_FEAT_CHANGE:
738 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
739 			ipvlan->dev->features = dev->features & IPVLAN_FEATURES;
740 			ipvlan->dev->gso_max_size = dev->gso_max_size;
741 			ipvlan->dev->gso_max_segs = dev->gso_max_segs;
742 			netdev_features_change(ipvlan->dev);
743 		}
744 		break;
745 
746 	case NETDEV_CHANGEMTU:
747 		list_for_each_entry(ipvlan, &port->ipvlans, pnode)
748 			ipvlan_adjust_mtu(ipvlan, dev);
749 		break;
750 
751 	case NETDEV_PRE_CHANGEADDR:
752 		prechaddr_info = ptr;
753 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
754 			err = dev_pre_changeaddr_notify(ipvlan->dev,
755 						    prechaddr_info->dev_addr,
756 						    extack);
757 			if (err)
758 				return notifier_from_errno(err);
759 		}
760 		break;
761 
762 	case NETDEV_CHANGEADDR:
763 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
764 			ether_addr_copy(ipvlan->dev->dev_addr, dev->dev_addr);
765 			call_netdevice_notifiers(NETDEV_CHANGEADDR, ipvlan->dev);
766 		}
767 		break;
768 
769 	case NETDEV_PRE_TYPE_CHANGE:
770 		/* Forbid underlying device to change its type. */
771 		return NOTIFY_BAD;
772 	}
773 	return NOTIFY_DONE;
774 }
775 
776 /* the caller must held the addrs lock */
777 static int ipvlan_add_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
778 {
779 	struct ipvl_addr *addr;
780 
781 	addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
782 	if (!addr)
783 		return -ENOMEM;
784 
785 	addr->master = ipvlan;
786 	if (!is_v6) {
787 		memcpy(&addr->ip4addr, iaddr, sizeof(struct in_addr));
788 		addr->atype = IPVL_IPV4;
789 #if IS_ENABLED(CONFIG_IPV6)
790 	} else {
791 		memcpy(&addr->ip6addr, iaddr, sizeof(struct in6_addr));
792 		addr->atype = IPVL_IPV6;
793 #endif
794 	}
795 
796 	list_add_tail_rcu(&addr->anode, &ipvlan->addrs);
797 
798 	/* If the interface is not up, the address will be added to the hash
799 	 * list by ipvlan_open.
800 	 */
801 	if (netif_running(ipvlan->dev))
802 		ipvlan_ht_addr_add(ipvlan, addr);
803 
804 	return 0;
805 }
806 
807 static void ipvlan_del_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
808 {
809 	struct ipvl_addr *addr;
810 
811 	spin_lock_bh(&ipvlan->addrs_lock);
812 	addr = ipvlan_find_addr(ipvlan, iaddr, is_v6);
813 	if (!addr) {
814 		spin_unlock_bh(&ipvlan->addrs_lock);
815 		return;
816 	}
817 
818 	ipvlan_ht_addr_del(addr);
819 	list_del_rcu(&addr->anode);
820 	spin_unlock_bh(&ipvlan->addrs_lock);
821 	kfree_rcu(addr, rcu);
822 }
823 
824 static bool ipvlan_is_valid_dev(const struct net_device *dev)
825 {
826 	struct ipvl_dev *ipvlan = netdev_priv(dev);
827 
828 	if (!netif_is_ipvlan(dev))
829 		return false;
830 
831 	if (!ipvlan || !ipvlan->port)
832 		return false;
833 
834 	return true;
835 }
836 
837 #if IS_ENABLED(CONFIG_IPV6)
838 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
839 {
840 	int ret = -EINVAL;
841 
842 	spin_lock_bh(&ipvlan->addrs_lock);
843 	if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true))
844 		netif_err(ipvlan, ifup, ipvlan->dev,
845 			  "Failed to add IPv6=%pI6c addr for %s intf\n",
846 			  ip6_addr, ipvlan->dev->name);
847 	else
848 		ret = ipvlan_add_addr(ipvlan, ip6_addr, true);
849 	spin_unlock_bh(&ipvlan->addrs_lock);
850 	return ret;
851 }
852 
853 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
854 {
855 	return ipvlan_del_addr(ipvlan, ip6_addr, true);
856 }
857 
858 static int ipvlan_addr6_event(struct notifier_block *unused,
859 			      unsigned long event, void *ptr)
860 {
861 	struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
862 	struct net_device *dev = (struct net_device *)if6->idev->dev;
863 	struct ipvl_dev *ipvlan = netdev_priv(dev);
864 
865 	if (!ipvlan_is_valid_dev(dev))
866 		return NOTIFY_DONE;
867 
868 	switch (event) {
869 	case NETDEV_UP:
870 		if (ipvlan_add_addr6(ipvlan, &if6->addr))
871 			return NOTIFY_BAD;
872 		break;
873 
874 	case NETDEV_DOWN:
875 		ipvlan_del_addr6(ipvlan, &if6->addr);
876 		break;
877 	}
878 
879 	return NOTIFY_OK;
880 }
881 
882 static int ipvlan_addr6_validator_event(struct notifier_block *unused,
883 					unsigned long event, void *ptr)
884 {
885 	struct in6_validator_info *i6vi = (struct in6_validator_info *)ptr;
886 	struct net_device *dev = (struct net_device *)i6vi->i6vi_dev->dev;
887 	struct ipvl_dev *ipvlan = netdev_priv(dev);
888 
889 	if (!ipvlan_is_valid_dev(dev))
890 		return NOTIFY_DONE;
891 
892 	switch (event) {
893 	case NETDEV_UP:
894 		if (ipvlan_addr_busy(ipvlan->port, &i6vi->i6vi_addr, true)) {
895 			NL_SET_ERR_MSG(i6vi->extack,
896 				       "Address already assigned to an ipvlan device");
897 			return notifier_from_errno(-EADDRINUSE);
898 		}
899 		break;
900 	}
901 
902 	return NOTIFY_OK;
903 }
904 #endif
905 
906 static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
907 {
908 	int ret = -EINVAL;
909 
910 	spin_lock_bh(&ipvlan->addrs_lock);
911 	if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false))
912 		netif_err(ipvlan, ifup, ipvlan->dev,
913 			  "Failed to add IPv4=%pI4 on %s intf.\n",
914 			  ip4_addr, ipvlan->dev->name);
915 	else
916 		ret = ipvlan_add_addr(ipvlan, ip4_addr, false);
917 	spin_unlock_bh(&ipvlan->addrs_lock);
918 	return ret;
919 }
920 
921 static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
922 {
923 	return ipvlan_del_addr(ipvlan, ip4_addr, false);
924 }
925 
926 static int ipvlan_addr4_event(struct notifier_block *unused,
927 			      unsigned long event, void *ptr)
928 {
929 	struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
930 	struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
931 	struct ipvl_dev *ipvlan = netdev_priv(dev);
932 	struct in_addr ip4_addr;
933 
934 	if (!ipvlan_is_valid_dev(dev))
935 		return NOTIFY_DONE;
936 
937 	switch (event) {
938 	case NETDEV_UP:
939 		ip4_addr.s_addr = if4->ifa_address;
940 		if (ipvlan_add_addr4(ipvlan, &ip4_addr))
941 			return NOTIFY_BAD;
942 		break;
943 
944 	case NETDEV_DOWN:
945 		ip4_addr.s_addr = if4->ifa_address;
946 		ipvlan_del_addr4(ipvlan, &ip4_addr);
947 		break;
948 	}
949 
950 	return NOTIFY_OK;
951 }
952 
953 static int ipvlan_addr4_validator_event(struct notifier_block *unused,
954 					unsigned long event, void *ptr)
955 {
956 	struct in_validator_info *ivi = (struct in_validator_info *)ptr;
957 	struct net_device *dev = (struct net_device *)ivi->ivi_dev->dev;
958 	struct ipvl_dev *ipvlan = netdev_priv(dev);
959 
960 	if (!ipvlan_is_valid_dev(dev))
961 		return NOTIFY_DONE;
962 
963 	switch (event) {
964 	case NETDEV_UP:
965 		if (ipvlan_addr_busy(ipvlan->port, &ivi->ivi_addr, false)) {
966 			NL_SET_ERR_MSG(ivi->extack,
967 				       "Address already assigned to an ipvlan device");
968 			return notifier_from_errno(-EADDRINUSE);
969 		}
970 		break;
971 	}
972 
973 	return NOTIFY_OK;
974 }
975 
976 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
977 	.notifier_call = ipvlan_addr4_event,
978 };
979 
980 static struct notifier_block ipvlan_addr4_vtor_notifier_block __read_mostly = {
981 	.notifier_call = ipvlan_addr4_validator_event,
982 };
983 
984 static struct notifier_block ipvlan_notifier_block __read_mostly = {
985 	.notifier_call = ipvlan_device_event,
986 };
987 
988 #if IS_ENABLED(CONFIG_IPV6)
989 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
990 	.notifier_call = ipvlan_addr6_event,
991 };
992 
993 static struct notifier_block ipvlan_addr6_vtor_notifier_block __read_mostly = {
994 	.notifier_call = ipvlan_addr6_validator_event,
995 };
996 #endif
997 
998 static int __init ipvlan_init_module(void)
999 {
1000 	int err;
1001 
1002 	ipvlan_init_secret();
1003 	register_netdevice_notifier(&ipvlan_notifier_block);
1004 #if IS_ENABLED(CONFIG_IPV6)
1005 	register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1006 	register_inet6addr_validator_notifier(
1007 	    &ipvlan_addr6_vtor_notifier_block);
1008 #endif
1009 	register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1010 	register_inetaddr_validator_notifier(&ipvlan_addr4_vtor_notifier_block);
1011 
1012 	err = ipvlan_l3s_init();
1013 	if (err < 0)
1014 		goto error;
1015 
1016 	err = ipvlan_link_register(&ipvlan_link_ops);
1017 	if (err < 0) {
1018 		ipvlan_l3s_cleanup();
1019 		goto error;
1020 	}
1021 
1022 	return 0;
1023 error:
1024 	unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1025 	unregister_inetaddr_validator_notifier(
1026 	    &ipvlan_addr4_vtor_notifier_block);
1027 #if IS_ENABLED(CONFIG_IPV6)
1028 	unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1029 	unregister_inet6addr_validator_notifier(
1030 	    &ipvlan_addr6_vtor_notifier_block);
1031 #endif
1032 	unregister_netdevice_notifier(&ipvlan_notifier_block);
1033 	return err;
1034 }
1035 
1036 static void __exit ipvlan_cleanup_module(void)
1037 {
1038 	rtnl_link_unregister(&ipvlan_link_ops);
1039 	ipvlan_l3s_cleanup();
1040 	unregister_netdevice_notifier(&ipvlan_notifier_block);
1041 	unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1042 	unregister_inetaddr_validator_notifier(
1043 	    &ipvlan_addr4_vtor_notifier_block);
1044 #if IS_ENABLED(CONFIG_IPV6)
1045 	unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1046 	unregister_inet6addr_validator_notifier(
1047 	    &ipvlan_addr6_vtor_notifier_block);
1048 #endif
1049 }
1050 
1051 module_init(ipvlan_init_module);
1052 module_exit(ipvlan_cleanup_module);
1053 
1054 MODULE_LICENSE("GPL");
1055 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
1056 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
1057 MODULE_ALIAS_RTNL_LINK("ipvlan");
1058