xref: /openbmc/linux/drivers/net/ipvlan/ipvlan_main.c (revision 8b030a57)
1 /* Copyright (c) 2014 Mahesh Bandewar <maheshb@google.com>
2  *
3  * This program is free software; you can redistribute it and/or
4  * modify it under the terms of the GNU General Public License as
5  * published by the Free Software Foundation; either version 2 of
6  * the License, or (at your option) any later version.
7  *
8  */
9 
10 #include "ipvlan.h"
11 
12 static unsigned int ipvlan_netid __read_mostly;
13 
14 struct ipvlan_netns {
15 	unsigned int ipvl_nf_hook_refcnt;
16 };
17 
18 static const struct nf_hook_ops ipvl_nfops[] = {
19 	{
20 		.hook     = ipvlan_nf_input,
21 		.pf       = NFPROTO_IPV4,
22 		.hooknum  = NF_INET_LOCAL_IN,
23 		.priority = INT_MAX,
24 	},
25 #if IS_ENABLED(CONFIG_IPV6)
26 	{
27 		.hook     = ipvlan_nf_input,
28 		.pf       = NFPROTO_IPV6,
29 		.hooknum  = NF_INET_LOCAL_IN,
30 		.priority = INT_MAX,
31 	},
32 #endif
33 };
34 
35 static const struct l3mdev_ops ipvl_l3mdev_ops = {
36 	.l3mdev_l3_rcv = ipvlan_l3_rcv,
37 };
38 
39 static void ipvlan_adjust_mtu(struct ipvl_dev *ipvlan, struct net_device *dev)
40 {
41 	ipvlan->dev->mtu = dev->mtu;
42 }
43 
44 static int ipvlan_register_nf_hook(struct net *net)
45 {
46 	struct ipvlan_netns *vnet = net_generic(net, ipvlan_netid);
47 	int err = 0;
48 
49 	if (!vnet->ipvl_nf_hook_refcnt) {
50 		err = nf_register_net_hooks(net, ipvl_nfops,
51 					    ARRAY_SIZE(ipvl_nfops));
52 		if (!err)
53 			vnet->ipvl_nf_hook_refcnt = 1;
54 	} else {
55 		vnet->ipvl_nf_hook_refcnt++;
56 	}
57 
58 	return err;
59 }
60 
61 static void ipvlan_unregister_nf_hook(struct net *net)
62 {
63 	struct ipvlan_netns *vnet = net_generic(net, ipvlan_netid);
64 
65 	if (WARN_ON(!vnet->ipvl_nf_hook_refcnt))
66 		return;
67 
68 	vnet->ipvl_nf_hook_refcnt--;
69 	if (!vnet->ipvl_nf_hook_refcnt)
70 		nf_unregister_net_hooks(net, ipvl_nfops,
71 					ARRAY_SIZE(ipvl_nfops));
72 }
73 
74 static int ipvlan_set_port_mode(struct ipvl_port *port, u16 nval,
75 				struct netlink_ext_ack *extack)
76 {
77 	struct ipvl_dev *ipvlan;
78 	struct net_device *mdev = port->dev;
79 	unsigned int flags;
80 	int err;
81 
82 	ASSERT_RTNL();
83 	if (port->mode != nval) {
84 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
85 			flags = ipvlan->dev->flags;
86 			if (nval == IPVLAN_MODE_L3 || nval == IPVLAN_MODE_L3S) {
87 				err = dev_change_flags(ipvlan->dev,
88 						       flags | IFF_NOARP,
89 						       extack);
90 			} else {
91 				err = dev_change_flags(ipvlan->dev,
92 						       flags & ~IFF_NOARP,
93 						       extack);
94 			}
95 			if (unlikely(err))
96 				goto fail;
97 		}
98 		if (nval == IPVLAN_MODE_L3S) {
99 			/* New mode is L3S */
100 			err = ipvlan_register_nf_hook(read_pnet(&port->pnet));
101 			if (!err) {
102 				mdev->l3mdev_ops = &ipvl_l3mdev_ops;
103 				mdev->priv_flags |= IFF_L3MDEV_RX_HANDLER;
104 			} else
105 				goto fail;
106 		} else if (port->mode == IPVLAN_MODE_L3S) {
107 			/* Old mode was L3S */
108 			mdev->priv_flags &= ~IFF_L3MDEV_RX_HANDLER;
109 			ipvlan_unregister_nf_hook(read_pnet(&port->pnet));
110 			mdev->l3mdev_ops = NULL;
111 		}
112 		port->mode = nval;
113 	}
114 	return 0;
115 
116 fail:
117 	/* Undo the flags changes that have been done so far. */
118 	list_for_each_entry_continue_reverse(ipvlan, &port->ipvlans, pnode) {
119 		flags = ipvlan->dev->flags;
120 		if (port->mode == IPVLAN_MODE_L3 ||
121 		    port->mode == IPVLAN_MODE_L3S)
122 			dev_change_flags(ipvlan->dev, flags | IFF_NOARP,
123 					 NULL);
124 		else
125 			dev_change_flags(ipvlan->dev, flags & ~IFF_NOARP,
126 					 NULL);
127 	}
128 
129 	return err;
130 }
131 
132 static int ipvlan_port_create(struct net_device *dev)
133 {
134 	struct ipvl_port *port;
135 	int err, idx;
136 
137 	port = kzalloc(sizeof(struct ipvl_port), GFP_KERNEL);
138 	if (!port)
139 		return -ENOMEM;
140 
141 	write_pnet(&port->pnet, dev_net(dev));
142 	port->dev = dev;
143 	port->mode = IPVLAN_MODE_L3;
144 	INIT_LIST_HEAD(&port->ipvlans);
145 	for (idx = 0; idx < IPVLAN_HASH_SIZE; idx++)
146 		INIT_HLIST_HEAD(&port->hlhead[idx]);
147 
148 	skb_queue_head_init(&port->backlog);
149 	INIT_WORK(&port->wq, ipvlan_process_multicast);
150 	ida_init(&port->ida);
151 	port->dev_id_start = 1;
152 
153 	err = netdev_rx_handler_register(dev, ipvlan_handle_frame, port);
154 	if (err)
155 		goto err;
156 
157 	return 0;
158 
159 err:
160 	kfree(port);
161 	return err;
162 }
163 
164 static void ipvlan_port_destroy(struct net_device *dev)
165 {
166 	struct ipvl_port *port = ipvlan_port_get_rtnl(dev);
167 	struct sk_buff *skb;
168 
169 	if (port->mode == IPVLAN_MODE_L3S) {
170 		dev->priv_flags &= ~IFF_L3MDEV_RX_HANDLER;
171 		ipvlan_unregister_nf_hook(dev_net(dev));
172 		dev->l3mdev_ops = NULL;
173 	}
174 	netdev_rx_handler_unregister(dev);
175 	cancel_work_sync(&port->wq);
176 	while ((skb = __skb_dequeue(&port->backlog)) != NULL) {
177 		if (skb->dev)
178 			dev_put(skb->dev);
179 		kfree_skb(skb);
180 	}
181 	ida_destroy(&port->ida);
182 	kfree(port);
183 }
184 
185 #define IPVLAN_FEATURES \
186 	(NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
187 	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_GSO_ROBUST | \
188 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
189 	 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
190 
191 #define IPVLAN_STATE_MASK \
192 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
193 
194 static int ipvlan_init(struct net_device *dev)
195 {
196 	struct ipvl_dev *ipvlan = netdev_priv(dev);
197 	struct net_device *phy_dev = ipvlan->phy_dev;
198 	struct ipvl_port *port;
199 	int err;
200 
201 	dev->state = (dev->state & ~IPVLAN_STATE_MASK) |
202 		     (phy_dev->state & IPVLAN_STATE_MASK);
203 	dev->features = phy_dev->features & IPVLAN_FEATURES;
204 	dev->features |= NETIF_F_LLTX | NETIF_F_VLAN_CHALLENGED;
205 	dev->gso_max_size = phy_dev->gso_max_size;
206 	dev->gso_max_segs = phy_dev->gso_max_segs;
207 	dev->hard_header_len = phy_dev->hard_header_len;
208 
209 	netdev_lockdep_set_classes(dev);
210 
211 	ipvlan->pcpu_stats = netdev_alloc_pcpu_stats(struct ipvl_pcpu_stats);
212 	if (!ipvlan->pcpu_stats)
213 		return -ENOMEM;
214 
215 	if (!netif_is_ipvlan_port(phy_dev)) {
216 		err = ipvlan_port_create(phy_dev);
217 		if (err < 0) {
218 			free_percpu(ipvlan->pcpu_stats);
219 			return err;
220 		}
221 	}
222 	port = ipvlan_port_get_rtnl(phy_dev);
223 	port->count += 1;
224 	return 0;
225 }
226 
227 static void ipvlan_uninit(struct net_device *dev)
228 {
229 	struct ipvl_dev *ipvlan = netdev_priv(dev);
230 	struct net_device *phy_dev = ipvlan->phy_dev;
231 	struct ipvl_port *port;
232 
233 	free_percpu(ipvlan->pcpu_stats);
234 
235 	port = ipvlan_port_get_rtnl(phy_dev);
236 	port->count -= 1;
237 	if (!port->count)
238 		ipvlan_port_destroy(port->dev);
239 }
240 
241 static int ipvlan_open(struct net_device *dev)
242 {
243 	struct ipvl_dev *ipvlan = netdev_priv(dev);
244 	struct net_device *phy_dev = ipvlan->phy_dev;
245 	struct ipvl_addr *addr;
246 
247 	if (ipvlan->port->mode == IPVLAN_MODE_L3 ||
248 	    ipvlan->port->mode == IPVLAN_MODE_L3S)
249 		dev->flags |= IFF_NOARP;
250 	else
251 		dev->flags &= ~IFF_NOARP;
252 
253 	rcu_read_lock();
254 	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
255 		ipvlan_ht_addr_add(ipvlan, addr);
256 	rcu_read_unlock();
257 
258 	return dev_uc_add(phy_dev, phy_dev->dev_addr);
259 }
260 
261 static int ipvlan_stop(struct net_device *dev)
262 {
263 	struct ipvl_dev *ipvlan = netdev_priv(dev);
264 	struct net_device *phy_dev = ipvlan->phy_dev;
265 	struct ipvl_addr *addr;
266 
267 	dev_uc_unsync(phy_dev, dev);
268 	dev_mc_unsync(phy_dev, dev);
269 
270 	dev_uc_del(phy_dev, phy_dev->dev_addr);
271 
272 	rcu_read_lock();
273 	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
274 		ipvlan_ht_addr_del(addr);
275 	rcu_read_unlock();
276 
277 	return 0;
278 }
279 
280 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
281 				     struct net_device *dev)
282 {
283 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
284 	int skblen = skb->len;
285 	int ret;
286 
287 	ret = ipvlan_queue_xmit(skb, dev);
288 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
289 		struct ipvl_pcpu_stats *pcptr;
290 
291 		pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
292 
293 		u64_stats_update_begin(&pcptr->syncp);
294 		pcptr->tx_pkts++;
295 		pcptr->tx_bytes += skblen;
296 		u64_stats_update_end(&pcptr->syncp);
297 	} else {
298 		this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
299 	}
300 	return ret;
301 }
302 
303 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
304 					     netdev_features_t features)
305 {
306 	struct ipvl_dev *ipvlan = netdev_priv(dev);
307 
308 	return features & (ipvlan->sfeatures | ~IPVLAN_FEATURES);
309 }
310 
311 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
312 {
313 	struct ipvl_dev *ipvlan = netdev_priv(dev);
314 	struct net_device *phy_dev = ipvlan->phy_dev;
315 
316 	if (change & IFF_ALLMULTI)
317 		dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
318 }
319 
320 static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
321 {
322 	struct ipvl_dev *ipvlan = netdev_priv(dev);
323 
324 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
325 		bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
326 	} else {
327 		struct netdev_hw_addr *ha;
328 		DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
329 
330 		bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
331 		netdev_for_each_mc_addr(ha, dev)
332 			__set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
333 
334 		/* Turn-on broadcast bit irrespective of address family,
335 		 * since broadcast is deferred to a work-queue, hence no
336 		 * impact on fast-path processing.
337 		 */
338 		__set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
339 
340 		bitmap_copy(ipvlan->mac_filters, mc_filters,
341 			    IPVLAN_MAC_FILTER_SIZE);
342 	}
343 	dev_uc_sync(ipvlan->phy_dev, dev);
344 	dev_mc_sync(ipvlan->phy_dev, dev);
345 }
346 
347 static void ipvlan_get_stats64(struct net_device *dev,
348 			       struct rtnl_link_stats64 *s)
349 {
350 	struct ipvl_dev *ipvlan = netdev_priv(dev);
351 
352 	if (ipvlan->pcpu_stats) {
353 		struct ipvl_pcpu_stats *pcptr;
354 		u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
355 		u32 rx_errs = 0, tx_drps = 0;
356 		u32 strt;
357 		int idx;
358 
359 		for_each_possible_cpu(idx) {
360 			pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
361 			do {
362 				strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
363 				rx_pkts = pcptr->rx_pkts;
364 				rx_bytes = pcptr->rx_bytes;
365 				rx_mcast = pcptr->rx_mcast;
366 				tx_pkts = pcptr->tx_pkts;
367 				tx_bytes = pcptr->tx_bytes;
368 			} while (u64_stats_fetch_retry_irq(&pcptr->syncp,
369 							   strt));
370 
371 			s->rx_packets += rx_pkts;
372 			s->rx_bytes += rx_bytes;
373 			s->multicast += rx_mcast;
374 			s->tx_packets += tx_pkts;
375 			s->tx_bytes += tx_bytes;
376 
377 			/* u32 values are updated without syncp protection. */
378 			rx_errs += pcptr->rx_errs;
379 			tx_drps += pcptr->tx_drps;
380 		}
381 		s->rx_errors = rx_errs;
382 		s->rx_dropped = rx_errs;
383 		s->tx_dropped = tx_drps;
384 	}
385 }
386 
387 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
388 {
389 	struct ipvl_dev *ipvlan = netdev_priv(dev);
390 	struct net_device *phy_dev = ipvlan->phy_dev;
391 
392 	return vlan_vid_add(phy_dev, proto, vid);
393 }
394 
395 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
396 				   u16 vid)
397 {
398 	struct ipvl_dev *ipvlan = netdev_priv(dev);
399 	struct net_device *phy_dev = ipvlan->phy_dev;
400 
401 	vlan_vid_del(phy_dev, proto, vid);
402 	return 0;
403 }
404 
405 static int ipvlan_get_iflink(const struct net_device *dev)
406 {
407 	struct ipvl_dev *ipvlan = netdev_priv(dev);
408 
409 	return ipvlan->phy_dev->ifindex;
410 }
411 
412 static const struct net_device_ops ipvlan_netdev_ops = {
413 	.ndo_init		= ipvlan_init,
414 	.ndo_uninit		= ipvlan_uninit,
415 	.ndo_open		= ipvlan_open,
416 	.ndo_stop		= ipvlan_stop,
417 	.ndo_start_xmit		= ipvlan_start_xmit,
418 	.ndo_fix_features	= ipvlan_fix_features,
419 	.ndo_change_rx_flags	= ipvlan_change_rx_flags,
420 	.ndo_set_rx_mode	= ipvlan_set_multicast_mac_filter,
421 	.ndo_get_stats64	= ipvlan_get_stats64,
422 	.ndo_vlan_rx_add_vid	= ipvlan_vlan_rx_add_vid,
423 	.ndo_vlan_rx_kill_vid	= ipvlan_vlan_rx_kill_vid,
424 	.ndo_get_iflink		= ipvlan_get_iflink,
425 };
426 
427 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
428 			      unsigned short type, const void *daddr,
429 			      const void *saddr, unsigned len)
430 {
431 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
432 	struct net_device *phy_dev = ipvlan->phy_dev;
433 
434 	/* TODO Probably use a different field than dev_addr so that the
435 	 * mac-address on the virtual device is portable and can be carried
436 	 * while the packets use the mac-addr on the physical device.
437 	 */
438 	return dev_hard_header(skb, phy_dev, type, daddr,
439 			       saddr ? : phy_dev->dev_addr, len);
440 }
441 
442 static const struct header_ops ipvlan_header_ops = {
443 	.create  	= ipvlan_hard_header,
444 	.parse		= eth_header_parse,
445 	.cache		= eth_header_cache,
446 	.cache_update	= eth_header_cache_update,
447 };
448 
449 static bool netif_is_ipvlan(const struct net_device *dev)
450 {
451 	/* both ipvlan and ipvtap devices use the same netdev_ops */
452 	return dev->netdev_ops == &ipvlan_netdev_ops;
453 }
454 
455 static int ipvlan_ethtool_get_link_ksettings(struct net_device *dev,
456 					     struct ethtool_link_ksettings *cmd)
457 {
458 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
459 
460 	return __ethtool_get_link_ksettings(ipvlan->phy_dev, cmd);
461 }
462 
463 static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
464 				       struct ethtool_drvinfo *drvinfo)
465 {
466 	strlcpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
467 	strlcpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
468 }
469 
470 static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
471 {
472 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
473 
474 	return ipvlan->msg_enable;
475 }
476 
477 static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
478 {
479 	struct ipvl_dev *ipvlan = netdev_priv(dev);
480 
481 	ipvlan->msg_enable = value;
482 }
483 
484 static const struct ethtool_ops ipvlan_ethtool_ops = {
485 	.get_link	= ethtool_op_get_link,
486 	.get_link_ksettings	= ipvlan_ethtool_get_link_ksettings,
487 	.get_drvinfo	= ipvlan_ethtool_get_drvinfo,
488 	.get_msglevel	= ipvlan_ethtool_get_msglevel,
489 	.set_msglevel	= ipvlan_ethtool_set_msglevel,
490 };
491 
492 static int ipvlan_nl_changelink(struct net_device *dev,
493 				struct nlattr *tb[], struct nlattr *data[],
494 				struct netlink_ext_ack *extack)
495 {
496 	struct ipvl_dev *ipvlan = netdev_priv(dev);
497 	struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
498 	int err = 0;
499 
500 	if (!data)
501 		return 0;
502 	if (!ns_capable(dev_net(ipvlan->phy_dev)->user_ns, CAP_NET_ADMIN))
503 		return -EPERM;
504 
505 	if (data[IFLA_IPVLAN_MODE]) {
506 		u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
507 
508 		err = ipvlan_set_port_mode(port, nmode, extack);
509 	}
510 
511 	if (!err && data[IFLA_IPVLAN_FLAGS]) {
512 		u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
513 
514 		if (flags & IPVLAN_F_PRIVATE)
515 			ipvlan_mark_private(port);
516 		else
517 			ipvlan_clear_private(port);
518 
519 		if (flags & IPVLAN_F_VEPA)
520 			ipvlan_mark_vepa(port);
521 		else
522 			ipvlan_clear_vepa(port);
523 	}
524 
525 	return err;
526 }
527 
528 static size_t ipvlan_nl_getsize(const struct net_device *dev)
529 {
530 	return (0
531 		+ nla_total_size(2) /* IFLA_IPVLAN_MODE */
532 		+ nla_total_size(2) /* IFLA_IPVLAN_FLAGS */
533 		);
534 }
535 
536 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[],
537 			      struct netlink_ext_ack *extack)
538 {
539 	if (!data)
540 		return 0;
541 
542 	if (data[IFLA_IPVLAN_MODE]) {
543 		u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
544 
545 		if (mode >= IPVLAN_MODE_MAX)
546 			return -EINVAL;
547 	}
548 	if (data[IFLA_IPVLAN_FLAGS]) {
549 		u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
550 
551 		/* Only two bits are used at this moment. */
552 		if (flags & ~(IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
553 			return -EINVAL;
554 		/* Also both flags can't be active at the same time. */
555 		if ((flags & (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA)) ==
556 		    (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
557 			return -EINVAL;
558 	}
559 
560 	return 0;
561 }
562 
563 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
564 			      const struct net_device *dev)
565 {
566 	struct ipvl_dev *ipvlan = netdev_priv(dev);
567 	struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
568 	int ret = -EINVAL;
569 
570 	if (!port)
571 		goto err;
572 
573 	ret = -EMSGSIZE;
574 	if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
575 		goto err;
576 	if (nla_put_u16(skb, IFLA_IPVLAN_FLAGS, port->flags))
577 		goto err;
578 
579 	return 0;
580 
581 err:
582 	return ret;
583 }
584 
585 int ipvlan_link_new(struct net *src_net, struct net_device *dev,
586 		    struct nlattr *tb[], struct nlattr *data[],
587 		    struct netlink_ext_ack *extack)
588 {
589 	struct ipvl_dev *ipvlan = netdev_priv(dev);
590 	struct ipvl_port *port;
591 	struct net_device *phy_dev;
592 	int err;
593 	u16 mode = IPVLAN_MODE_L3;
594 
595 	if (!tb[IFLA_LINK])
596 		return -EINVAL;
597 
598 	phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
599 	if (!phy_dev)
600 		return -ENODEV;
601 
602 	if (netif_is_ipvlan(phy_dev)) {
603 		struct ipvl_dev *tmp = netdev_priv(phy_dev);
604 
605 		phy_dev = tmp->phy_dev;
606 		if (!ns_capable(dev_net(phy_dev)->user_ns, CAP_NET_ADMIN))
607 			return -EPERM;
608 	} else if (!netif_is_ipvlan_port(phy_dev)) {
609 		/* Exit early if the underlying link is invalid or busy */
610 		if (phy_dev->type != ARPHRD_ETHER ||
611 		    phy_dev->flags & IFF_LOOPBACK) {
612 			netdev_err(phy_dev,
613 				   "Master is either lo or non-ether device\n");
614 			return -EINVAL;
615 		}
616 
617 		if (netdev_is_rx_handler_busy(phy_dev)) {
618 			netdev_err(phy_dev, "Device is already in use.\n");
619 			return -EBUSY;
620 		}
621 	}
622 
623 	ipvlan->phy_dev = phy_dev;
624 	ipvlan->dev = dev;
625 	ipvlan->sfeatures = IPVLAN_FEATURES;
626 	if (!tb[IFLA_MTU])
627 		ipvlan_adjust_mtu(ipvlan, phy_dev);
628 	INIT_LIST_HEAD(&ipvlan->addrs);
629 	spin_lock_init(&ipvlan->addrs_lock);
630 
631 	/* TODO Probably put random address here to be presented to the
632 	 * world but keep using the physical-dev address for the outgoing
633 	 * packets.
634 	 */
635 	memcpy(dev->dev_addr, phy_dev->dev_addr, ETH_ALEN);
636 
637 	dev->priv_flags |= IFF_NO_RX_HANDLER;
638 
639 	err = register_netdevice(dev);
640 	if (err < 0)
641 		return err;
642 
643 	/* ipvlan_init() would have created the port, if required */
644 	port = ipvlan_port_get_rtnl(phy_dev);
645 	ipvlan->port = port;
646 
647 	/* If the port-id base is at the MAX value, then wrap it around and
648 	 * begin from 0x1 again. This may be due to a busy system where lots
649 	 * of slaves are getting created and deleted.
650 	 */
651 	if (port->dev_id_start == 0xFFFE)
652 		port->dev_id_start = 0x1;
653 
654 	/* Since L2 address is shared among all IPvlan slaves including
655 	 * master, use unique 16 bit dev-ids to diffentiate among them.
656 	 * Assign IDs between 0x1 and 0xFFFE (used by the master) to each
657 	 * slave link [see addrconf_ifid_eui48()].
658 	 */
659 	err = ida_simple_get(&port->ida, port->dev_id_start, 0xFFFE,
660 			     GFP_KERNEL);
661 	if (err < 0)
662 		err = ida_simple_get(&port->ida, 0x1, port->dev_id_start,
663 				     GFP_KERNEL);
664 	if (err < 0)
665 		goto unregister_netdev;
666 	dev->dev_id = err;
667 
668 	/* Increment id-base to the next slot for the future assignment */
669 	port->dev_id_start = err + 1;
670 
671 	err = netdev_upper_dev_link(phy_dev, dev, extack);
672 	if (err)
673 		goto remove_ida;
674 
675 	/* Flags are per port and latest update overrides. User has
676 	 * to be consistent in setting it just like the mode attribute.
677 	 */
678 	if (data && data[IFLA_IPVLAN_FLAGS])
679 		port->flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
680 
681 	if (data && data[IFLA_IPVLAN_MODE])
682 		mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
683 
684 	err = ipvlan_set_port_mode(port, mode, extack);
685 	if (err)
686 		goto unlink_netdev;
687 
688 	list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
689 	netif_stacked_transfer_operstate(phy_dev, dev);
690 	return 0;
691 
692 unlink_netdev:
693 	netdev_upper_dev_unlink(phy_dev, dev);
694 remove_ida:
695 	ida_simple_remove(&port->ida, dev->dev_id);
696 unregister_netdev:
697 	unregister_netdevice(dev);
698 	return err;
699 }
700 EXPORT_SYMBOL_GPL(ipvlan_link_new);
701 
702 void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
703 {
704 	struct ipvl_dev *ipvlan = netdev_priv(dev);
705 	struct ipvl_addr *addr, *next;
706 
707 	spin_lock_bh(&ipvlan->addrs_lock);
708 	list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
709 		ipvlan_ht_addr_del(addr);
710 		list_del_rcu(&addr->anode);
711 		kfree_rcu(addr, rcu);
712 	}
713 	spin_unlock_bh(&ipvlan->addrs_lock);
714 
715 	ida_simple_remove(&ipvlan->port->ida, dev->dev_id);
716 	list_del_rcu(&ipvlan->pnode);
717 	unregister_netdevice_queue(dev, head);
718 	netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
719 }
720 EXPORT_SYMBOL_GPL(ipvlan_link_delete);
721 
722 void ipvlan_link_setup(struct net_device *dev)
723 {
724 	ether_setup(dev);
725 
726 	dev->max_mtu = ETH_MAX_MTU;
727 	dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
728 	dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
729 	dev->netdev_ops = &ipvlan_netdev_ops;
730 	dev->needs_free_netdev = true;
731 	dev->header_ops = &ipvlan_header_ops;
732 	dev->ethtool_ops = &ipvlan_ethtool_ops;
733 }
734 EXPORT_SYMBOL_GPL(ipvlan_link_setup);
735 
736 static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
737 {
738 	[IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
739 	[IFLA_IPVLAN_FLAGS] = { .type = NLA_U16 },
740 };
741 
742 static struct rtnl_link_ops ipvlan_link_ops = {
743 	.kind		= "ipvlan",
744 	.priv_size	= sizeof(struct ipvl_dev),
745 
746 	.setup		= ipvlan_link_setup,
747 	.newlink	= ipvlan_link_new,
748 	.dellink	= ipvlan_link_delete,
749 };
750 
751 int ipvlan_link_register(struct rtnl_link_ops *ops)
752 {
753 	ops->get_size	= ipvlan_nl_getsize;
754 	ops->policy	= ipvlan_nl_policy;
755 	ops->validate	= ipvlan_nl_validate;
756 	ops->fill_info	= ipvlan_nl_fillinfo;
757 	ops->changelink = ipvlan_nl_changelink;
758 	ops->maxtype	= IFLA_IPVLAN_MAX;
759 	return rtnl_link_register(ops);
760 }
761 EXPORT_SYMBOL_GPL(ipvlan_link_register);
762 
763 static int ipvlan_device_event(struct notifier_block *unused,
764 			       unsigned long event, void *ptr)
765 {
766 	struct netlink_ext_ack *extack = netdev_notifier_info_to_extack(ptr);
767 	struct netdev_notifier_pre_changeaddr_info *prechaddr_info;
768 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
769 	struct ipvl_dev *ipvlan, *next;
770 	struct ipvl_port *port;
771 	LIST_HEAD(lst_kill);
772 	int err;
773 
774 	if (!netif_is_ipvlan_port(dev))
775 		return NOTIFY_DONE;
776 
777 	port = ipvlan_port_get_rtnl(dev);
778 
779 	switch (event) {
780 	case NETDEV_CHANGE:
781 		list_for_each_entry(ipvlan, &port->ipvlans, pnode)
782 			netif_stacked_transfer_operstate(ipvlan->phy_dev,
783 							 ipvlan->dev);
784 		break;
785 
786 	case NETDEV_REGISTER: {
787 		struct net *oldnet, *newnet = dev_net(dev);
788 		struct ipvlan_netns *old_vnet;
789 
790 		oldnet = read_pnet(&port->pnet);
791 		if (net_eq(newnet, oldnet))
792 			break;
793 
794 		write_pnet(&port->pnet, newnet);
795 
796 		old_vnet = net_generic(oldnet, ipvlan_netid);
797 		if (!old_vnet->ipvl_nf_hook_refcnt)
798 			break;
799 
800 		ipvlan_register_nf_hook(newnet);
801 		ipvlan_unregister_nf_hook(oldnet);
802 		break;
803 	}
804 	case NETDEV_UNREGISTER:
805 		if (dev->reg_state != NETREG_UNREGISTERING)
806 			break;
807 
808 		list_for_each_entry_safe(ipvlan, next, &port->ipvlans, pnode)
809 			ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
810 							    &lst_kill);
811 		unregister_netdevice_many(&lst_kill);
812 		break;
813 
814 	case NETDEV_FEAT_CHANGE:
815 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
816 			ipvlan->dev->features = dev->features & IPVLAN_FEATURES;
817 			ipvlan->dev->gso_max_size = dev->gso_max_size;
818 			ipvlan->dev->gso_max_segs = dev->gso_max_segs;
819 			netdev_features_change(ipvlan->dev);
820 		}
821 		break;
822 
823 	case NETDEV_CHANGEMTU:
824 		list_for_each_entry(ipvlan, &port->ipvlans, pnode)
825 			ipvlan_adjust_mtu(ipvlan, dev);
826 		break;
827 
828 	case NETDEV_PRE_CHANGEADDR:
829 		prechaddr_info = ptr;
830 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
831 			err = dev_pre_changeaddr_notify(ipvlan->dev,
832 						    prechaddr_info->dev_addr,
833 						    extack);
834 			if (err)
835 				return notifier_from_errno(err);
836 		}
837 		break;
838 
839 	case NETDEV_CHANGEADDR:
840 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
841 			ether_addr_copy(ipvlan->dev->dev_addr, dev->dev_addr);
842 			call_netdevice_notifiers(NETDEV_CHANGEADDR, ipvlan->dev);
843 		}
844 		break;
845 
846 	case NETDEV_PRE_TYPE_CHANGE:
847 		/* Forbid underlying device to change its type. */
848 		return NOTIFY_BAD;
849 	}
850 	return NOTIFY_DONE;
851 }
852 
853 /* the caller must held the addrs lock */
854 static int ipvlan_add_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
855 {
856 	struct ipvl_addr *addr;
857 
858 	addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
859 	if (!addr)
860 		return -ENOMEM;
861 
862 	addr->master = ipvlan;
863 	if (!is_v6) {
864 		memcpy(&addr->ip4addr, iaddr, sizeof(struct in_addr));
865 		addr->atype = IPVL_IPV4;
866 #if IS_ENABLED(CONFIG_IPV6)
867 	} else {
868 		memcpy(&addr->ip6addr, iaddr, sizeof(struct in6_addr));
869 		addr->atype = IPVL_IPV6;
870 #endif
871 	}
872 
873 	list_add_tail_rcu(&addr->anode, &ipvlan->addrs);
874 
875 	/* If the interface is not up, the address will be added to the hash
876 	 * list by ipvlan_open.
877 	 */
878 	if (netif_running(ipvlan->dev))
879 		ipvlan_ht_addr_add(ipvlan, addr);
880 
881 	return 0;
882 }
883 
884 static void ipvlan_del_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
885 {
886 	struct ipvl_addr *addr;
887 
888 	spin_lock_bh(&ipvlan->addrs_lock);
889 	addr = ipvlan_find_addr(ipvlan, iaddr, is_v6);
890 	if (!addr) {
891 		spin_unlock_bh(&ipvlan->addrs_lock);
892 		return;
893 	}
894 
895 	ipvlan_ht_addr_del(addr);
896 	list_del_rcu(&addr->anode);
897 	spin_unlock_bh(&ipvlan->addrs_lock);
898 	kfree_rcu(addr, rcu);
899 }
900 
901 static bool ipvlan_is_valid_dev(const struct net_device *dev)
902 {
903 	struct ipvl_dev *ipvlan = netdev_priv(dev);
904 
905 	if (!netif_is_ipvlan(dev))
906 		return false;
907 
908 	if (!ipvlan || !ipvlan->port)
909 		return false;
910 
911 	return true;
912 }
913 
914 #if IS_ENABLED(CONFIG_IPV6)
915 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
916 {
917 	int ret = -EINVAL;
918 
919 	spin_lock_bh(&ipvlan->addrs_lock);
920 	if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true))
921 		netif_err(ipvlan, ifup, ipvlan->dev,
922 			  "Failed to add IPv6=%pI6c addr for %s intf\n",
923 			  ip6_addr, ipvlan->dev->name);
924 	else
925 		ret = ipvlan_add_addr(ipvlan, ip6_addr, true);
926 	spin_unlock_bh(&ipvlan->addrs_lock);
927 	return ret;
928 }
929 
930 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
931 {
932 	return ipvlan_del_addr(ipvlan, ip6_addr, true);
933 }
934 
935 static int ipvlan_addr6_event(struct notifier_block *unused,
936 			      unsigned long event, void *ptr)
937 {
938 	struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
939 	struct net_device *dev = (struct net_device *)if6->idev->dev;
940 	struct ipvl_dev *ipvlan = netdev_priv(dev);
941 
942 	if (!ipvlan_is_valid_dev(dev))
943 		return NOTIFY_DONE;
944 
945 	switch (event) {
946 	case NETDEV_UP:
947 		if (ipvlan_add_addr6(ipvlan, &if6->addr))
948 			return NOTIFY_BAD;
949 		break;
950 
951 	case NETDEV_DOWN:
952 		ipvlan_del_addr6(ipvlan, &if6->addr);
953 		break;
954 	}
955 
956 	return NOTIFY_OK;
957 }
958 
959 static int ipvlan_addr6_validator_event(struct notifier_block *unused,
960 					unsigned long event, void *ptr)
961 {
962 	struct in6_validator_info *i6vi = (struct in6_validator_info *)ptr;
963 	struct net_device *dev = (struct net_device *)i6vi->i6vi_dev->dev;
964 	struct ipvl_dev *ipvlan = netdev_priv(dev);
965 
966 	if (!ipvlan_is_valid_dev(dev))
967 		return NOTIFY_DONE;
968 
969 	switch (event) {
970 	case NETDEV_UP:
971 		if (ipvlan_addr_busy(ipvlan->port, &i6vi->i6vi_addr, true)) {
972 			NL_SET_ERR_MSG(i6vi->extack,
973 				       "Address already assigned to an ipvlan device");
974 			return notifier_from_errno(-EADDRINUSE);
975 		}
976 		break;
977 	}
978 
979 	return NOTIFY_OK;
980 }
981 #endif
982 
983 static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
984 {
985 	int ret = -EINVAL;
986 
987 	spin_lock_bh(&ipvlan->addrs_lock);
988 	if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false))
989 		netif_err(ipvlan, ifup, ipvlan->dev,
990 			  "Failed to add IPv4=%pI4 on %s intf.\n",
991 			  ip4_addr, ipvlan->dev->name);
992 	else
993 		ret = ipvlan_add_addr(ipvlan, ip4_addr, false);
994 	spin_unlock_bh(&ipvlan->addrs_lock);
995 	return ret;
996 }
997 
998 static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
999 {
1000 	return ipvlan_del_addr(ipvlan, ip4_addr, false);
1001 }
1002 
1003 static int ipvlan_addr4_event(struct notifier_block *unused,
1004 			      unsigned long event, void *ptr)
1005 {
1006 	struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
1007 	struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
1008 	struct ipvl_dev *ipvlan = netdev_priv(dev);
1009 	struct in_addr ip4_addr;
1010 
1011 	if (!ipvlan_is_valid_dev(dev))
1012 		return NOTIFY_DONE;
1013 
1014 	switch (event) {
1015 	case NETDEV_UP:
1016 		ip4_addr.s_addr = if4->ifa_address;
1017 		if (ipvlan_add_addr4(ipvlan, &ip4_addr))
1018 			return NOTIFY_BAD;
1019 		break;
1020 
1021 	case NETDEV_DOWN:
1022 		ip4_addr.s_addr = if4->ifa_address;
1023 		ipvlan_del_addr4(ipvlan, &ip4_addr);
1024 		break;
1025 	}
1026 
1027 	return NOTIFY_OK;
1028 }
1029 
1030 static int ipvlan_addr4_validator_event(struct notifier_block *unused,
1031 					unsigned long event, void *ptr)
1032 {
1033 	struct in_validator_info *ivi = (struct in_validator_info *)ptr;
1034 	struct net_device *dev = (struct net_device *)ivi->ivi_dev->dev;
1035 	struct ipvl_dev *ipvlan = netdev_priv(dev);
1036 
1037 	if (!ipvlan_is_valid_dev(dev))
1038 		return NOTIFY_DONE;
1039 
1040 	switch (event) {
1041 	case NETDEV_UP:
1042 		if (ipvlan_addr_busy(ipvlan->port, &ivi->ivi_addr, false)) {
1043 			NL_SET_ERR_MSG(ivi->extack,
1044 				       "Address already assigned to an ipvlan device");
1045 			return notifier_from_errno(-EADDRINUSE);
1046 		}
1047 		break;
1048 	}
1049 
1050 	return NOTIFY_OK;
1051 }
1052 
1053 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
1054 	.notifier_call = ipvlan_addr4_event,
1055 };
1056 
1057 static struct notifier_block ipvlan_addr4_vtor_notifier_block __read_mostly = {
1058 	.notifier_call = ipvlan_addr4_validator_event,
1059 };
1060 
1061 static struct notifier_block ipvlan_notifier_block __read_mostly = {
1062 	.notifier_call = ipvlan_device_event,
1063 };
1064 
1065 #if IS_ENABLED(CONFIG_IPV6)
1066 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
1067 	.notifier_call = ipvlan_addr6_event,
1068 };
1069 
1070 static struct notifier_block ipvlan_addr6_vtor_notifier_block __read_mostly = {
1071 	.notifier_call = ipvlan_addr6_validator_event,
1072 };
1073 #endif
1074 
1075 static void ipvlan_ns_exit(struct net *net)
1076 {
1077 	struct ipvlan_netns *vnet = net_generic(net, ipvlan_netid);
1078 
1079 	if (WARN_ON_ONCE(vnet->ipvl_nf_hook_refcnt)) {
1080 		vnet->ipvl_nf_hook_refcnt = 0;
1081 		nf_unregister_net_hooks(net, ipvl_nfops,
1082 					ARRAY_SIZE(ipvl_nfops));
1083 	}
1084 }
1085 
1086 static struct pernet_operations ipvlan_net_ops = {
1087 	.id = &ipvlan_netid,
1088 	.size = sizeof(struct ipvlan_netns),
1089 	.exit = ipvlan_ns_exit,
1090 };
1091 
1092 static int __init ipvlan_init_module(void)
1093 {
1094 	int err;
1095 
1096 	ipvlan_init_secret();
1097 	register_netdevice_notifier(&ipvlan_notifier_block);
1098 #if IS_ENABLED(CONFIG_IPV6)
1099 	register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1100 	register_inet6addr_validator_notifier(
1101 	    &ipvlan_addr6_vtor_notifier_block);
1102 #endif
1103 	register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1104 	register_inetaddr_validator_notifier(&ipvlan_addr4_vtor_notifier_block);
1105 
1106 	err = register_pernet_subsys(&ipvlan_net_ops);
1107 	if (err < 0)
1108 		goto error;
1109 
1110 	err = ipvlan_link_register(&ipvlan_link_ops);
1111 	if (err < 0) {
1112 		unregister_pernet_subsys(&ipvlan_net_ops);
1113 		goto error;
1114 	}
1115 
1116 	return 0;
1117 error:
1118 	unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1119 	unregister_inetaddr_validator_notifier(
1120 	    &ipvlan_addr4_vtor_notifier_block);
1121 #if IS_ENABLED(CONFIG_IPV6)
1122 	unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1123 	unregister_inet6addr_validator_notifier(
1124 	    &ipvlan_addr6_vtor_notifier_block);
1125 #endif
1126 	unregister_netdevice_notifier(&ipvlan_notifier_block);
1127 	return err;
1128 }
1129 
1130 static void __exit ipvlan_cleanup_module(void)
1131 {
1132 	rtnl_link_unregister(&ipvlan_link_ops);
1133 	unregister_pernet_subsys(&ipvlan_net_ops);
1134 	unregister_netdevice_notifier(&ipvlan_notifier_block);
1135 	unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1136 	unregister_inetaddr_validator_notifier(
1137 	    &ipvlan_addr4_vtor_notifier_block);
1138 #if IS_ENABLED(CONFIG_IPV6)
1139 	unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1140 	unregister_inet6addr_validator_notifier(
1141 	    &ipvlan_addr6_vtor_notifier_block);
1142 #endif
1143 }
1144 
1145 module_init(ipvlan_init_module);
1146 module_exit(ipvlan_cleanup_module);
1147 
1148 MODULE_LICENSE("GPL");
1149 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
1150 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
1151 MODULE_ALIAS_RTNL_LINK("ipvlan");
1152