xref: /openbmc/linux/drivers/net/ipvlan/ipvlan_main.c (revision d9e32672)
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->hw_enc_features |= dev->features;
130 	dev->gso_max_size = phy_dev->gso_max_size;
131 	dev->gso_max_segs = phy_dev->gso_max_segs;
132 	dev->hard_header_len = phy_dev->hard_header_len;
133 
134 	netdev_lockdep_set_classes(dev);
135 
136 	ipvlan->pcpu_stats = netdev_alloc_pcpu_stats(struct ipvl_pcpu_stats);
137 	if (!ipvlan->pcpu_stats)
138 		return -ENOMEM;
139 
140 	if (!netif_is_ipvlan_port(phy_dev)) {
141 		err = ipvlan_port_create(phy_dev);
142 		if (err < 0) {
143 			free_percpu(ipvlan->pcpu_stats);
144 			return err;
145 		}
146 	}
147 	port = ipvlan_port_get_rtnl(phy_dev);
148 	port->count += 1;
149 	return 0;
150 }
151 
152 static void ipvlan_uninit(struct net_device *dev)
153 {
154 	struct ipvl_dev *ipvlan = netdev_priv(dev);
155 	struct net_device *phy_dev = ipvlan->phy_dev;
156 	struct ipvl_port *port;
157 
158 	free_percpu(ipvlan->pcpu_stats);
159 
160 	port = ipvlan_port_get_rtnl(phy_dev);
161 	port->count -= 1;
162 	if (!port->count)
163 		ipvlan_port_destroy(port->dev);
164 }
165 
166 static int ipvlan_open(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_addr *addr;
171 
172 	if (ipvlan->port->mode == IPVLAN_MODE_L3 ||
173 	    ipvlan->port->mode == IPVLAN_MODE_L3S)
174 		dev->flags |= IFF_NOARP;
175 	else
176 		dev->flags &= ~IFF_NOARP;
177 
178 	rcu_read_lock();
179 	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
180 		ipvlan_ht_addr_add(ipvlan, addr);
181 	rcu_read_unlock();
182 
183 	return dev_uc_add(phy_dev, phy_dev->dev_addr);
184 }
185 
186 static int ipvlan_stop(struct net_device *dev)
187 {
188 	struct ipvl_dev *ipvlan = netdev_priv(dev);
189 	struct net_device *phy_dev = ipvlan->phy_dev;
190 	struct ipvl_addr *addr;
191 
192 	dev_uc_unsync(phy_dev, dev);
193 	dev_mc_unsync(phy_dev, dev);
194 
195 	dev_uc_del(phy_dev, phy_dev->dev_addr);
196 
197 	rcu_read_lock();
198 	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
199 		ipvlan_ht_addr_del(addr);
200 	rcu_read_unlock();
201 
202 	return 0;
203 }
204 
205 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
206 				     struct net_device *dev)
207 {
208 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
209 	int skblen = skb->len;
210 	int ret;
211 
212 	ret = ipvlan_queue_xmit(skb, dev);
213 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
214 		struct ipvl_pcpu_stats *pcptr;
215 
216 		pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
217 
218 		u64_stats_update_begin(&pcptr->syncp);
219 		pcptr->tx_pkts++;
220 		pcptr->tx_bytes += skblen;
221 		u64_stats_update_end(&pcptr->syncp);
222 	} else {
223 		this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
224 	}
225 	return ret;
226 }
227 
228 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
229 					     netdev_features_t features)
230 {
231 	struct ipvl_dev *ipvlan = netdev_priv(dev);
232 
233 	return features & (ipvlan->sfeatures | ~IPVLAN_FEATURES);
234 }
235 
236 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
237 {
238 	struct ipvl_dev *ipvlan = netdev_priv(dev);
239 	struct net_device *phy_dev = ipvlan->phy_dev;
240 
241 	if (change & IFF_ALLMULTI)
242 		dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
243 }
244 
245 static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
246 {
247 	struct ipvl_dev *ipvlan = netdev_priv(dev);
248 
249 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
250 		bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
251 	} else {
252 		struct netdev_hw_addr *ha;
253 		DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
254 
255 		bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
256 		netdev_for_each_mc_addr(ha, dev)
257 			__set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
258 
259 		/* Turn-on broadcast bit irrespective of address family,
260 		 * since broadcast is deferred to a work-queue, hence no
261 		 * impact on fast-path processing.
262 		 */
263 		__set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
264 
265 		bitmap_copy(ipvlan->mac_filters, mc_filters,
266 			    IPVLAN_MAC_FILTER_SIZE);
267 	}
268 	dev_uc_sync(ipvlan->phy_dev, dev);
269 	dev_mc_sync(ipvlan->phy_dev, dev);
270 }
271 
272 static void ipvlan_get_stats64(struct net_device *dev,
273 			       struct rtnl_link_stats64 *s)
274 {
275 	struct ipvl_dev *ipvlan = netdev_priv(dev);
276 
277 	if (ipvlan->pcpu_stats) {
278 		struct ipvl_pcpu_stats *pcptr;
279 		u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
280 		u32 rx_errs = 0, tx_drps = 0;
281 		u32 strt;
282 		int idx;
283 
284 		for_each_possible_cpu(idx) {
285 			pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
286 			do {
287 				strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
288 				rx_pkts = pcptr->rx_pkts;
289 				rx_bytes = pcptr->rx_bytes;
290 				rx_mcast = pcptr->rx_mcast;
291 				tx_pkts = pcptr->tx_pkts;
292 				tx_bytes = pcptr->tx_bytes;
293 			} while (u64_stats_fetch_retry_irq(&pcptr->syncp,
294 							   strt));
295 
296 			s->rx_packets += rx_pkts;
297 			s->rx_bytes += rx_bytes;
298 			s->multicast += rx_mcast;
299 			s->tx_packets += tx_pkts;
300 			s->tx_bytes += tx_bytes;
301 
302 			/* u32 values are updated without syncp protection. */
303 			rx_errs += pcptr->rx_errs;
304 			tx_drps += pcptr->tx_drps;
305 		}
306 		s->rx_errors = rx_errs;
307 		s->rx_dropped = rx_errs;
308 		s->tx_dropped = tx_drps;
309 	}
310 }
311 
312 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
313 {
314 	struct ipvl_dev *ipvlan = netdev_priv(dev);
315 	struct net_device *phy_dev = ipvlan->phy_dev;
316 
317 	return vlan_vid_add(phy_dev, proto, vid);
318 }
319 
320 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
321 				   u16 vid)
322 {
323 	struct ipvl_dev *ipvlan = netdev_priv(dev);
324 	struct net_device *phy_dev = ipvlan->phy_dev;
325 
326 	vlan_vid_del(phy_dev, proto, vid);
327 	return 0;
328 }
329 
330 static int ipvlan_get_iflink(const struct net_device *dev)
331 {
332 	struct ipvl_dev *ipvlan = netdev_priv(dev);
333 
334 	return ipvlan->phy_dev->ifindex;
335 }
336 
337 static const struct net_device_ops ipvlan_netdev_ops = {
338 	.ndo_init		= ipvlan_init,
339 	.ndo_uninit		= ipvlan_uninit,
340 	.ndo_open		= ipvlan_open,
341 	.ndo_stop		= ipvlan_stop,
342 	.ndo_start_xmit		= ipvlan_start_xmit,
343 	.ndo_fix_features	= ipvlan_fix_features,
344 	.ndo_change_rx_flags	= ipvlan_change_rx_flags,
345 	.ndo_set_rx_mode	= ipvlan_set_multicast_mac_filter,
346 	.ndo_get_stats64	= ipvlan_get_stats64,
347 	.ndo_vlan_rx_add_vid	= ipvlan_vlan_rx_add_vid,
348 	.ndo_vlan_rx_kill_vid	= ipvlan_vlan_rx_kill_vid,
349 	.ndo_get_iflink		= ipvlan_get_iflink,
350 };
351 
352 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
353 			      unsigned short type, const void *daddr,
354 			      const void *saddr, unsigned len)
355 {
356 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
357 	struct net_device *phy_dev = ipvlan->phy_dev;
358 
359 	/* TODO Probably use a different field than dev_addr so that the
360 	 * mac-address on the virtual device is portable and can be carried
361 	 * while the packets use the mac-addr on the physical device.
362 	 */
363 	return dev_hard_header(skb, phy_dev, type, daddr,
364 			       saddr ? : phy_dev->dev_addr, len);
365 }
366 
367 static const struct header_ops ipvlan_header_ops = {
368 	.create  	= ipvlan_hard_header,
369 	.parse		= eth_header_parse,
370 	.cache		= eth_header_cache,
371 	.cache_update	= eth_header_cache_update,
372 };
373 
374 static void ipvlan_adjust_mtu(struct ipvl_dev *ipvlan, struct net_device *dev)
375 {
376 	ipvlan->dev->mtu = dev->mtu;
377 }
378 
379 static bool netif_is_ipvlan(const struct net_device *dev)
380 {
381 	/* both ipvlan and ipvtap devices use the same netdev_ops */
382 	return dev->netdev_ops == &ipvlan_netdev_ops;
383 }
384 
385 static int ipvlan_ethtool_get_link_ksettings(struct net_device *dev,
386 					     struct ethtool_link_ksettings *cmd)
387 {
388 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
389 
390 	return __ethtool_get_link_ksettings(ipvlan->phy_dev, cmd);
391 }
392 
393 static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
394 				       struct ethtool_drvinfo *drvinfo)
395 {
396 	strlcpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
397 	strlcpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
398 }
399 
400 static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
401 {
402 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
403 
404 	return ipvlan->msg_enable;
405 }
406 
407 static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
408 {
409 	struct ipvl_dev *ipvlan = netdev_priv(dev);
410 
411 	ipvlan->msg_enable = value;
412 }
413 
414 static const struct ethtool_ops ipvlan_ethtool_ops = {
415 	.get_link	= ethtool_op_get_link,
416 	.get_link_ksettings	= ipvlan_ethtool_get_link_ksettings,
417 	.get_drvinfo	= ipvlan_ethtool_get_drvinfo,
418 	.get_msglevel	= ipvlan_ethtool_get_msglevel,
419 	.set_msglevel	= ipvlan_ethtool_set_msglevel,
420 };
421 
422 static int ipvlan_nl_changelink(struct net_device *dev,
423 				struct nlattr *tb[], struct nlattr *data[],
424 				struct netlink_ext_ack *extack)
425 {
426 	struct ipvl_dev *ipvlan = netdev_priv(dev);
427 	struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
428 	int err = 0;
429 
430 	if (!data)
431 		return 0;
432 	if (!ns_capable(dev_net(ipvlan->phy_dev)->user_ns, CAP_NET_ADMIN))
433 		return -EPERM;
434 
435 	if (data[IFLA_IPVLAN_MODE]) {
436 		u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
437 
438 		err = ipvlan_set_port_mode(port, nmode, extack);
439 	}
440 
441 	if (!err && data[IFLA_IPVLAN_FLAGS]) {
442 		u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
443 
444 		if (flags & IPVLAN_F_PRIVATE)
445 			ipvlan_mark_private(port);
446 		else
447 			ipvlan_clear_private(port);
448 
449 		if (flags & IPVLAN_F_VEPA)
450 			ipvlan_mark_vepa(port);
451 		else
452 			ipvlan_clear_vepa(port);
453 	}
454 
455 	return err;
456 }
457 
458 static size_t ipvlan_nl_getsize(const struct net_device *dev)
459 {
460 	return (0
461 		+ nla_total_size(2) /* IFLA_IPVLAN_MODE */
462 		+ nla_total_size(2) /* IFLA_IPVLAN_FLAGS */
463 		);
464 }
465 
466 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[],
467 			      struct netlink_ext_ack *extack)
468 {
469 	if (!data)
470 		return 0;
471 
472 	if (data[IFLA_IPVLAN_MODE]) {
473 		u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
474 
475 		if (mode >= IPVLAN_MODE_MAX)
476 			return -EINVAL;
477 	}
478 	if (data[IFLA_IPVLAN_FLAGS]) {
479 		u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
480 
481 		/* Only two bits are used at this moment. */
482 		if (flags & ~(IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
483 			return -EINVAL;
484 		/* Also both flags can't be active at the same time. */
485 		if ((flags & (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA)) ==
486 		    (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
487 			return -EINVAL;
488 	}
489 
490 	return 0;
491 }
492 
493 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
494 			      const struct net_device *dev)
495 {
496 	struct ipvl_dev *ipvlan = netdev_priv(dev);
497 	struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
498 	int ret = -EINVAL;
499 
500 	if (!port)
501 		goto err;
502 
503 	ret = -EMSGSIZE;
504 	if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
505 		goto err;
506 	if (nla_put_u16(skb, IFLA_IPVLAN_FLAGS, port->flags))
507 		goto err;
508 
509 	return 0;
510 
511 err:
512 	return ret;
513 }
514 
515 int ipvlan_link_new(struct net *src_net, struct net_device *dev,
516 		    struct nlattr *tb[], struct nlattr *data[],
517 		    struct netlink_ext_ack *extack)
518 {
519 	struct ipvl_dev *ipvlan = netdev_priv(dev);
520 	struct ipvl_port *port;
521 	struct net_device *phy_dev;
522 	int err;
523 	u16 mode = IPVLAN_MODE_L3;
524 
525 	if (!tb[IFLA_LINK])
526 		return -EINVAL;
527 
528 	phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
529 	if (!phy_dev)
530 		return -ENODEV;
531 
532 	if (netif_is_ipvlan(phy_dev)) {
533 		struct ipvl_dev *tmp = netdev_priv(phy_dev);
534 
535 		phy_dev = tmp->phy_dev;
536 		if (!ns_capable(dev_net(phy_dev)->user_ns, CAP_NET_ADMIN))
537 			return -EPERM;
538 	} else if (!netif_is_ipvlan_port(phy_dev)) {
539 		/* Exit early if the underlying link is invalid or busy */
540 		if (phy_dev->type != ARPHRD_ETHER ||
541 		    phy_dev->flags & IFF_LOOPBACK) {
542 			netdev_err(phy_dev,
543 				   "Master is either lo or non-ether device\n");
544 			return -EINVAL;
545 		}
546 
547 		if (netdev_is_rx_handler_busy(phy_dev)) {
548 			netdev_err(phy_dev, "Device is already in use.\n");
549 			return -EBUSY;
550 		}
551 	}
552 
553 	ipvlan->phy_dev = phy_dev;
554 	ipvlan->dev = dev;
555 	ipvlan->sfeatures = IPVLAN_FEATURES;
556 	if (!tb[IFLA_MTU])
557 		ipvlan_adjust_mtu(ipvlan, phy_dev);
558 	INIT_LIST_HEAD(&ipvlan->addrs);
559 	spin_lock_init(&ipvlan->addrs_lock);
560 
561 	/* TODO Probably put random address here to be presented to the
562 	 * world but keep using the physical-dev address for the outgoing
563 	 * packets.
564 	 */
565 	memcpy(dev->dev_addr, phy_dev->dev_addr, ETH_ALEN);
566 
567 	dev->priv_flags |= IFF_NO_RX_HANDLER;
568 
569 	err = register_netdevice(dev);
570 	if (err < 0)
571 		return err;
572 
573 	/* ipvlan_init() would have created the port, if required */
574 	port = ipvlan_port_get_rtnl(phy_dev);
575 	ipvlan->port = port;
576 
577 	/* If the port-id base is at the MAX value, then wrap it around and
578 	 * begin from 0x1 again. This may be due to a busy system where lots
579 	 * of slaves are getting created and deleted.
580 	 */
581 	if (port->dev_id_start == 0xFFFE)
582 		port->dev_id_start = 0x1;
583 
584 	/* Since L2 address is shared among all IPvlan slaves including
585 	 * master, use unique 16 bit dev-ids to diffentiate among them.
586 	 * Assign IDs between 0x1 and 0xFFFE (used by the master) to each
587 	 * slave link [see addrconf_ifid_eui48()].
588 	 */
589 	err = ida_simple_get(&port->ida, port->dev_id_start, 0xFFFE,
590 			     GFP_KERNEL);
591 	if (err < 0)
592 		err = ida_simple_get(&port->ida, 0x1, port->dev_id_start,
593 				     GFP_KERNEL);
594 	if (err < 0)
595 		goto unregister_netdev;
596 	dev->dev_id = err;
597 
598 	/* Increment id-base to the next slot for the future assignment */
599 	port->dev_id_start = err + 1;
600 
601 	err = netdev_upper_dev_link(phy_dev, dev, extack);
602 	if (err)
603 		goto remove_ida;
604 
605 	/* Flags are per port and latest update overrides. User has
606 	 * to be consistent in setting it just like the mode attribute.
607 	 */
608 	if (data && data[IFLA_IPVLAN_FLAGS])
609 		port->flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
610 
611 	if (data && data[IFLA_IPVLAN_MODE])
612 		mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
613 
614 	err = ipvlan_set_port_mode(port, mode, extack);
615 	if (err)
616 		goto unlink_netdev;
617 
618 	list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
619 	netif_stacked_transfer_operstate(phy_dev, dev);
620 	return 0;
621 
622 unlink_netdev:
623 	netdev_upper_dev_unlink(phy_dev, dev);
624 remove_ida:
625 	ida_simple_remove(&port->ida, dev->dev_id);
626 unregister_netdev:
627 	unregister_netdevice(dev);
628 	return err;
629 }
630 EXPORT_SYMBOL_GPL(ipvlan_link_new);
631 
632 void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
633 {
634 	struct ipvl_dev *ipvlan = netdev_priv(dev);
635 	struct ipvl_addr *addr, *next;
636 
637 	spin_lock_bh(&ipvlan->addrs_lock);
638 	list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
639 		ipvlan_ht_addr_del(addr);
640 		list_del_rcu(&addr->anode);
641 		kfree_rcu(addr, rcu);
642 	}
643 	spin_unlock_bh(&ipvlan->addrs_lock);
644 
645 	ida_simple_remove(&ipvlan->port->ida, dev->dev_id);
646 	list_del_rcu(&ipvlan->pnode);
647 	unregister_netdevice_queue(dev, head);
648 	netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
649 }
650 EXPORT_SYMBOL_GPL(ipvlan_link_delete);
651 
652 void ipvlan_link_setup(struct net_device *dev)
653 {
654 	ether_setup(dev);
655 
656 	dev->max_mtu = ETH_MAX_MTU;
657 	dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
658 	dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
659 	dev->netdev_ops = &ipvlan_netdev_ops;
660 	dev->needs_free_netdev = true;
661 	dev->header_ops = &ipvlan_header_ops;
662 	dev->ethtool_ops = &ipvlan_ethtool_ops;
663 }
664 EXPORT_SYMBOL_GPL(ipvlan_link_setup);
665 
666 static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
667 {
668 	[IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
669 	[IFLA_IPVLAN_FLAGS] = { .type = NLA_U16 },
670 };
671 
672 static struct rtnl_link_ops ipvlan_link_ops = {
673 	.kind		= "ipvlan",
674 	.priv_size	= sizeof(struct ipvl_dev),
675 
676 	.setup		= ipvlan_link_setup,
677 	.newlink	= ipvlan_link_new,
678 	.dellink	= ipvlan_link_delete,
679 };
680 
681 int ipvlan_link_register(struct rtnl_link_ops *ops)
682 {
683 	ops->get_size	= ipvlan_nl_getsize;
684 	ops->policy	= ipvlan_nl_policy;
685 	ops->validate	= ipvlan_nl_validate;
686 	ops->fill_info	= ipvlan_nl_fillinfo;
687 	ops->changelink = ipvlan_nl_changelink;
688 	ops->maxtype	= IFLA_IPVLAN_MAX;
689 	return rtnl_link_register(ops);
690 }
691 EXPORT_SYMBOL_GPL(ipvlan_link_register);
692 
693 static int ipvlan_device_event(struct notifier_block *unused,
694 			       unsigned long event, void *ptr)
695 {
696 	struct netlink_ext_ack *extack = netdev_notifier_info_to_extack(ptr);
697 	struct netdev_notifier_pre_changeaddr_info *prechaddr_info;
698 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
699 	struct ipvl_dev *ipvlan, *next;
700 	struct ipvl_port *port;
701 	LIST_HEAD(lst_kill);
702 	int err;
703 
704 	if (!netif_is_ipvlan_port(dev))
705 		return NOTIFY_DONE;
706 
707 	port = ipvlan_port_get_rtnl(dev);
708 
709 	switch (event) {
710 	case NETDEV_CHANGE:
711 		list_for_each_entry(ipvlan, &port->ipvlans, pnode)
712 			netif_stacked_transfer_operstate(ipvlan->phy_dev,
713 							 ipvlan->dev);
714 		break;
715 
716 	case NETDEV_REGISTER: {
717 		struct net *oldnet, *newnet = dev_net(dev);
718 
719 		oldnet = read_pnet(&port->pnet);
720 		if (net_eq(newnet, oldnet))
721 			break;
722 
723 		write_pnet(&port->pnet, newnet);
724 
725 		ipvlan_migrate_l3s_hook(oldnet, newnet);
726 		break;
727 	}
728 	case NETDEV_UNREGISTER:
729 		if (dev->reg_state != NETREG_UNREGISTERING)
730 			break;
731 
732 		list_for_each_entry_safe(ipvlan, next, &port->ipvlans, pnode)
733 			ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
734 							    &lst_kill);
735 		unregister_netdevice_many(&lst_kill);
736 		break;
737 
738 	case NETDEV_FEAT_CHANGE:
739 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
740 			ipvlan->dev->features = dev->features & IPVLAN_FEATURES;
741 			ipvlan->dev->gso_max_size = dev->gso_max_size;
742 			ipvlan->dev->gso_max_segs = dev->gso_max_segs;
743 			netdev_features_change(ipvlan->dev);
744 		}
745 		break;
746 
747 	case NETDEV_CHANGEMTU:
748 		list_for_each_entry(ipvlan, &port->ipvlans, pnode)
749 			ipvlan_adjust_mtu(ipvlan, dev);
750 		break;
751 
752 	case NETDEV_PRE_CHANGEADDR:
753 		prechaddr_info = ptr;
754 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
755 			err = dev_pre_changeaddr_notify(ipvlan->dev,
756 						    prechaddr_info->dev_addr,
757 						    extack);
758 			if (err)
759 				return notifier_from_errno(err);
760 		}
761 		break;
762 
763 	case NETDEV_CHANGEADDR:
764 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
765 			ether_addr_copy(ipvlan->dev->dev_addr, dev->dev_addr);
766 			call_netdevice_notifiers(NETDEV_CHANGEADDR, ipvlan->dev);
767 		}
768 		break;
769 
770 	case NETDEV_PRE_TYPE_CHANGE:
771 		/* Forbid underlying device to change its type. */
772 		return NOTIFY_BAD;
773 	}
774 	return NOTIFY_DONE;
775 }
776 
777 /* the caller must held the addrs lock */
778 static int ipvlan_add_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
779 {
780 	struct ipvl_addr *addr;
781 
782 	addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
783 	if (!addr)
784 		return -ENOMEM;
785 
786 	addr->master = ipvlan;
787 	if (!is_v6) {
788 		memcpy(&addr->ip4addr, iaddr, sizeof(struct in_addr));
789 		addr->atype = IPVL_IPV4;
790 #if IS_ENABLED(CONFIG_IPV6)
791 	} else {
792 		memcpy(&addr->ip6addr, iaddr, sizeof(struct in6_addr));
793 		addr->atype = IPVL_IPV6;
794 #endif
795 	}
796 
797 	list_add_tail_rcu(&addr->anode, &ipvlan->addrs);
798 
799 	/* If the interface is not up, the address will be added to the hash
800 	 * list by ipvlan_open.
801 	 */
802 	if (netif_running(ipvlan->dev))
803 		ipvlan_ht_addr_add(ipvlan, addr);
804 
805 	return 0;
806 }
807 
808 static void ipvlan_del_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
809 {
810 	struct ipvl_addr *addr;
811 
812 	spin_lock_bh(&ipvlan->addrs_lock);
813 	addr = ipvlan_find_addr(ipvlan, iaddr, is_v6);
814 	if (!addr) {
815 		spin_unlock_bh(&ipvlan->addrs_lock);
816 		return;
817 	}
818 
819 	ipvlan_ht_addr_del(addr);
820 	list_del_rcu(&addr->anode);
821 	spin_unlock_bh(&ipvlan->addrs_lock);
822 	kfree_rcu(addr, rcu);
823 }
824 
825 static bool ipvlan_is_valid_dev(const struct net_device *dev)
826 {
827 	struct ipvl_dev *ipvlan = netdev_priv(dev);
828 
829 	if (!netif_is_ipvlan(dev))
830 		return false;
831 
832 	if (!ipvlan || !ipvlan->port)
833 		return false;
834 
835 	return true;
836 }
837 
838 #if IS_ENABLED(CONFIG_IPV6)
839 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
840 {
841 	int ret = -EINVAL;
842 
843 	spin_lock_bh(&ipvlan->addrs_lock);
844 	if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true))
845 		netif_err(ipvlan, ifup, ipvlan->dev,
846 			  "Failed to add IPv6=%pI6c addr for %s intf\n",
847 			  ip6_addr, ipvlan->dev->name);
848 	else
849 		ret = ipvlan_add_addr(ipvlan, ip6_addr, true);
850 	spin_unlock_bh(&ipvlan->addrs_lock);
851 	return ret;
852 }
853 
854 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
855 {
856 	return ipvlan_del_addr(ipvlan, ip6_addr, true);
857 }
858 
859 static int ipvlan_addr6_event(struct notifier_block *unused,
860 			      unsigned long event, void *ptr)
861 {
862 	struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
863 	struct net_device *dev = (struct net_device *)if6->idev->dev;
864 	struct ipvl_dev *ipvlan = netdev_priv(dev);
865 
866 	if (!ipvlan_is_valid_dev(dev))
867 		return NOTIFY_DONE;
868 
869 	switch (event) {
870 	case NETDEV_UP:
871 		if (ipvlan_add_addr6(ipvlan, &if6->addr))
872 			return NOTIFY_BAD;
873 		break;
874 
875 	case NETDEV_DOWN:
876 		ipvlan_del_addr6(ipvlan, &if6->addr);
877 		break;
878 	}
879 
880 	return NOTIFY_OK;
881 }
882 
883 static int ipvlan_addr6_validator_event(struct notifier_block *unused,
884 					unsigned long event, void *ptr)
885 {
886 	struct in6_validator_info *i6vi = (struct in6_validator_info *)ptr;
887 	struct net_device *dev = (struct net_device *)i6vi->i6vi_dev->dev;
888 	struct ipvl_dev *ipvlan = netdev_priv(dev);
889 
890 	if (!ipvlan_is_valid_dev(dev))
891 		return NOTIFY_DONE;
892 
893 	switch (event) {
894 	case NETDEV_UP:
895 		if (ipvlan_addr_busy(ipvlan->port, &i6vi->i6vi_addr, true)) {
896 			NL_SET_ERR_MSG(i6vi->extack,
897 				       "Address already assigned to an ipvlan device");
898 			return notifier_from_errno(-EADDRINUSE);
899 		}
900 		break;
901 	}
902 
903 	return NOTIFY_OK;
904 }
905 #endif
906 
907 static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
908 {
909 	int ret = -EINVAL;
910 
911 	spin_lock_bh(&ipvlan->addrs_lock);
912 	if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false))
913 		netif_err(ipvlan, ifup, ipvlan->dev,
914 			  "Failed to add IPv4=%pI4 on %s intf.\n",
915 			  ip4_addr, ipvlan->dev->name);
916 	else
917 		ret = ipvlan_add_addr(ipvlan, ip4_addr, false);
918 	spin_unlock_bh(&ipvlan->addrs_lock);
919 	return ret;
920 }
921 
922 static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
923 {
924 	return ipvlan_del_addr(ipvlan, ip4_addr, false);
925 }
926 
927 static int ipvlan_addr4_event(struct notifier_block *unused,
928 			      unsigned long event, void *ptr)
929 {
930 	struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
931 	struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
932 	struct ipvl_dev *ipvlan = netdev_priv(dev);
933 	struct in_addr ip4_addr;
934 
935 	if (!ipvlan_is_valid_dev(dev))
936 		return NOTIFY_DONE;
937 
938 	switch (event) {
939 	case NETDEV_UP:
940 		ip4_addr.s_addr = if4->ifa_address;
941 		if (ipvlan_add_addr4(ipvlan, &ip4_addr))
942 			return NOTIFY_BAD;
943 		break;
944 
945 	case NETDEV_DOWN:
946 		ip4_addr.s_addr = if4->ifa_address;
947 		ipvlan_del_addr4(ipvlan, &ip4_addr);
948 		break;
949 	}
950 
951 	return NOTIFY_OK;
952 }
953 
954 static int ipvlan_addr4_validator_event(struct notifier_block *unused,
955 					unsigned long event, void *ptr)
956 {
957 	struct in_validator_info *ivi = (struct in_validator_info *)ptr;
958 	struct net_device *dev = (struct net_device *)ivi->ivi_dev->dev;
959 	struct ipvl_dev *ipvlan = netdev_priv(dev);
960 
961 	if (!ipvlan_is_valid_dev(dev))
962 		return NOTIFY_DONE;
963 
964 	switch (event) {
965 	case NETDEV_UP:
966 		if (ipvlan_addr_busy(ipvlan->port, &ivi->ivi_addr, false)) {
967 			NL_SET_ERR_MSG(ivi->extack,
968 				       "Address already assigned to an ipvlan device");
969 			return notifier_from_errno(-EADDRINUSE);
970 		}
971 		break;
972 	}
973 
974 	return NOTIFY_OK;
975 }
976 
977 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
978 	.notifier_call = ipvlan_addr4_event,
979 };
980 
981 static struct notifier_block ipvlan_addr4_vtor_notifier_block __read_mostly = {
982 	.notifier_call = ipvlan_addr4_validator_event,
983 };
984 
985 static struct notifier_block ipvlan_notifier_block __read_mostly = {
986 	.notifier_call = ipvlan_device_event,
987 };
988 
989 #if IS_ENABLED(CONFIG_IPV6)
990 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
991 	.notifier_call = ipvlan_addr6_event,
992 };
993 
994 static struct notifier_block ipvlan_addr6_vtor_notifier_block __read_mostly = {
995 	.notifier_call = ipvlan_addr6_validator_event,
996 };
997 #endif
998 
999 static int __init ipvlan_init_module(void)
1000 {
1001 	int err;
1002 
1003 	ipvlan_init_secret();
1004 	register_netdevice_notifier(&ipvlan_notifier_block);
1005 #if IS_ENABLED(CONFIG_IPV6)
1006 	register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1007 	register_inet6addr_validator_notifier(
1008 	    &ipvlan_addr6_vtor_notifier_block);
1009 #endif
1010 	register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1011 	register_inetaddr_validator_notifier(&ipvlan_addr4_vtor_notifier_block);
1012 
1013 	err = ipvlan_l3s_init();
1014 	if (err < 0)
1015 		goto error;
1016 
1017 	err = ipvlan_link_register(&ipvlan_link_ops);
1018 	if (err < 0) {
1019 		ipvlan_l3s_cleanup();
1020 		goto error;
1021 	}
1022 
1023 	return 0;
1024 error:
1025 	unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1026 	unregister_inetaddr_validator_notifier(
1027 	    &ipvlan_addr4_vtor_notifier_block);
1028 #if IS_ENABLED(CONFIG_IPV6)
1029 	unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1030 	unregister_inet6addr_validator_notifier(
1031 	    &ipvlan_addr6_vtor_notifier_block);
1032 #endif
1033 	unregister_netdevice_notifier(&ipvlan_notifier_block);
1034 	return err;
1035 }
1036 
1037 static void __exit ipvlan_cleanup_module(void)
1038 {
1039 	rtnl_link_unregister(&ipvlan_link_ops);
1040 	ipvlan_l3s_cleanup();
1041 	unregister_netdevice_notifier(&ipvlan_notifier_block);
1042 	unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1043 	unregister_inetaddr_validator_notifier(
1044 	    &ipvlan_addr4_vtor_notifier_block);
1045 #if IS_ENABLED(CONFIG_IPV6)
1046 	unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1047 	unregister_inet6addr_validator_notifier(
1048 	    &ipvlan_addr6_vtor_notifier_block);
1049 #endif
1050 }
1051 
1052 module_init(ipvlan_init_module);
1053 module_exit(ipvlan_cleanup_module);
1054 
1055 MODULE_LICENSE("GPL");
1056 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
1057 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
1058 MODULE_ALIAS_RTNL_LINK("ipvlan");
1059