xref: /openbmc/linux/drivers/net/ipvlan/ipvlan_main.c (revision c4a11bf4)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Copyright (c) 2014 Mahesh Bandewar <maheshb@google.com>
3  */
4 
5 #include <linux/ethtool.h>
6 
7 #include "ipvlan.h"
8 
9 static int ipvlan_set_port_mode(struct ipvl_port *port, u16 nval,
10 				struct netlink_ext_ack *extack)
11 {
12 	struct ipvl_dev *ipvlan;
13 	unsigned int flags;
14 	int err;
15 
16 	ASSERT_RTNL();
17 	if (port->mode != nval) {
18 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
19 			flags = ipvlan->dev->flags;
20 			if (nval == IPVLAN_MODE_L3 || nval == IPVLAN_MODE_L3S) {
21 				err = dev_change_flags(ipvlan->dev,
22 						       flags | IFF_NOARP,
23 						       extack);
24 			} else {
25 				err = dev_change_flags(ipvlan->dev,
26 						       flags & ~IFF_NOARP,
27 						       extack);
28 			}
29 			if (unlikely(err))
30 				goto fail;
31 		}
32 		if (nval == IPVLAN_MODE_L3S) {
33 			/* New mode is L3S */
34 			err = ipvlan_l3s_register(port);
35 			if (err)
36 				goto fail;
37 		} else if (port->mode == IPVLAN_MODE_L3S) {
38 			/* Old mode was L3S */
39 			ipvlan_l3s_unregister(port);
40 		}
41 		port->mode = nval;
42 	}
43 	return 0;
44 
45 fail:
46 	/* Undo the flags changes that have been done so far. */
47 	list_for_each_entry_continue_reverse(ipvlan, &port->ipvlans, pnode) {
48 		flags = ipvlan->dev->flags;
49 		if (port->mode == IPVLAN_MODE_L3 ||
50 		    port->mode == IPVLAN_MODE_L3S)
51 			dev_change_flags(ipvlan->dev, flags | IFF_NOARP,
52 					 NULL);
53 		else
54 			dev_change_flags(ipvlan->dev, flags & ~IFF_NOARP,
55 					 NULL);
56 	}
57 
58 	return err;
59 }
60 
61 static int ipvlan_port_create(struct net_device *dev)
62 {
63 	struct ipvl_port *port;
64 	int err, idx;
65 
66 	port = kzalloc(sizeof(struct ipvl_port), GFP_KERNEL);
67 	if (!port)
68 		return -ENOMEM;
69 
70 	write_pnet(&port->pnet, dev_net(dev));
71 	port->dev = dev;
72 	port->mode = IPVLAN_MODE_L3;
73 	INIT_LIST_HEAD(&port->ipvlans);
74 	for (idx = 0; idx < IPVLAN_HASH_SIZE; idx++)
75 		INIT_HLIST_HEAD(&port->hlhead[idx]);
76 
77 	skb_queue_head_init(&port->backlog);
78 	INIT_WORK(&port->wq, ipvlan_process_multicast);
79 	ida_init(&port->ida);
80 	port->dev_id_start = 1;
81 
82 	err = netdev_rx_handler_register(dev, ipvlan_handle_frame, port);
83 	if (err)
84 		goto err;
85 
86 	return 0;
87 
88 err:
89 	kfree(port);
90 	return err;
91 }
92 
93 static void ipvlan_port_destroy(struct net_device *dev)
94 {
95 	struct ipvl_port *port = ipvlan_port_get_rtnl(dev);
96 	struct sk_buff *skb;
97 
98 	if (port->mode == IPVLAN_MODE_L3S)
99 		ipvlan_l3s_unregister(port);
100 	netdev_rx_handler_unregister(dev);
101 	cancel_work_sync(&port->wq);
102 	while ((skb = __skb_dequeue(&port->backlog)) != NULL) {
103 		if (skb->dev)
104 			dev_put(skb->dev);
105 		kfree_skb(skb);
106 	}
107 	ida_destroy(&port->ida);
108 	kfree(port);
109 }
110 
111 #define IPVLAN_ALWAYS_ON_OFLOADS \
112 	(NETIF_F_SG | NETIF_F_HW_CSUM | \
113 	 NETIF_F_GSO_ROBUST | NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL)
114 
115 #define IPVLAN_ALWAYS_ON \
116 	(IPVLAN_ALWAYS_ON_OFLOADS | NETIF_F_LLTX | NETIF_F_VLAN_CHALLENGED)
117 
118 #define IPVLAN_FEATURES \
119 	(NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
120 	 NETIF_F_GSO | NETIF_F_ALL_TSO | NETIF_F_GSO_ROBUST | \
121 	 NETIF_F_GRO | NETIF_F_RXCSUM | \
122 	 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
123 
124 	/* NETIF_F_GSO_ENCAP_ALL NETIF_F_GSO_SOFTWARE Newly added */
125 
126 #define IPVLAN_STATE_MASK \
127 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
128 
129 static int ipvlan_init(struct net_device *dev)
130 {
131 	struct ipvl_dev *ipvlan = netdev_priv(dev);
132 	struct net_device *phy_dev = ipvlan->phy_dev;
133 	struct ipvl_port *port;
134 	int err;
135 
136 	dev->state = (dev->state & ~IPVLAN_STATE_MASK) |
137 		     (phy_dev->state & IPVLAN_STATE_MASK);
138 	dev->features = phy_dev->features & IPVLAN_FEATURES;
139 	dev->features |= IPVLAN_ALWAYS_ON;
140 	dev->vlan_features = phy_dev->vlan_features & IPVLAN_FEATURES;
141 	dev->vlan_features |= IPVLAN_ALWAYS_ON_OFLOADS;
142 	dev->hw_enc_features |= dev->features;
143 	dev->gso_max_size = phy_dev->gso_max_size;
144 	dev->gso_max_segs = phy_dev->gso_max_segs;
145 	dev->hard_header_len = phy_dev->hard_header_len;
146 
147 	netdev_lockdep_set_classes(dev);
148 
149 	ipvlan->pcpu_stats = netdev_alloc_pcpu_stats(struct ipvl_pcpu_stats);
150 	if (!ipvlan->pcpu_stats)
151 		return -ENOMEM;
152 
153 	if (!netif_is_ipvlan_port(phy_dev)) {
154 		err = ipvlan_port_create(phy_dev);
155 		if (err < 0) {
156 			free_percpu(ipvlan->pcpu_stats);
157 			return err;
158 		}
159 	}
160 	port = ipvlan_port_get_rtnl(phy_dev);
161 	port->count += 1;
162 	return 0;
163 }
164 
165 static void ipvlan_uninit(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_port *port;
170 
171 	free_percpu(ipvlan->pcpu_stats);
172 
173 	port = ipvlan_port_get_rtnl(phy_dev);
174 	port->count -= 1;
175 	if (!port->count)
176 		ipvlan_port_destroy(port->dev);
177 }
178 
179 static int ipvlan_open(struct net_device *dev)
180 {
181 	struct ipvl_dev *ipvlan = netdev_priv(dev);
182 	struct ipvl_addr *addr;
183 
184 	if (ipvlan->port->mode == IPVLAN_MODE_L3 ||
185 	    ipvlan->port->mode == IPVLAN_MODE_L3S)
186 		dev->flags |= IFF_NOARP;
187 	else
188 		dev->flags &= ~IFF_NOARP;
189 
190 	rcu_read_lock();
191 	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
192 		ipvlan_ht_addr_add(ipvlan, addr);
193 	rcu_read_unlock();
194 
195 	return 0;
196 }
197 
198 static int ipvlan_stop(struct net_device *dev)
199 {
200 	struct ipvl_dev *ipvlan = netdev_priv(dev);
201 	struct net_device *phy_dev = ipvlan->phy_dev;
202 	struct ipvl_addr *addr;
203 
204 	dev_uc_unsync(phy_dev, dev);
205 	dev_mc_unsync(phy_dev, dev);
206 
207 	rcu_read_lock();
208 	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
209 		ipvlan_ht_addr_del(addr);
210 	rcu_read_unlock();
211 
212 	return 0;
213 }
214 
215 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
216 				     struct net_device *dev)
217 {
218 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
219 	int skblen = skb->len;
220 	int ret;
221 
222 	ret = ipvlan_queue_xmit(skb, dev);
223 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
224 		struct ipvl_pcpu_stats *pcptr;
225 
226 		pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
227 
228 		u64_stats_update_begin(&pcptr->syncp);
229 		pcptr->tx_pkts++;
230 		pcptr->tx_bytes += skblen;
231 		u64_stats_update_end(&pcptr->syncp);
232 	} else {
233 		this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
234 	}
235 	return ret;
236 }
237 
238 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
239 					     netdev_features_t features)
240 {
241 	struct ipvl_dev *ipvlan = netdev_priv(dev);
242 
243 	features |= NETIF_F_ALL_FOR_ALL;
244 	features &= (ipvlan->sfeatures | ~IPVLAN_FEATURES);
245 	features = netdev_increment_features(ipvlan->phy_dev->features,
246 					     features, features);
247 	features |= IPVLAN_ALWAYS_ON;
248 	features &= (IPVLAN_FEATURES | IPVLAN_ALWAYS_ON);
249 
250 	return features;
251 }
252 
253 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
254 {
255 	struct ipvl_dev *ipvlan = netdev_priv(dev);
256 	struct net_device *phy_dev = ipvlan->phy_dev;
257 
258 	if (change & IFF_ALLMULTI)
259 		dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
260 }
261 
262 static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
263 {
264 	struct ipvl_dev *ipvlan = netdev_priv(dev);
265 
266 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
267 		bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
268 	} else {
269 		struct netdev_hw_addr *ha;
270 		DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
271 
272 		bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
273 		netdev_for_each_mc_addr(ha, dev)
274 			__set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
275 
276 		/* Turn-on broadcast bit irrespective of address family,
277 		 * since broadcast is deferred to a work-queue, hence no
278 		 * impact on fast-path processing.
279 		 */
280 		__set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
281 
282 		bitmap_copy(ipvlan->mac_filters, mc_filters,
283 			    IPVLAN_MAC_FILTER_SIZE);
284 	}
285 	dev_uc_sync(ipvlan->phy_dev, dev);
286 	dev_mc_sync(ipvlan->phy_dev, dev);
287 }
288 
289 static void ipvlan_get_stats64(struct net_device *dev,
290 			       struct rtnl_link_stats64 *s)
291 {
292 	struct ipvl_dev *ipvlan = netdev_priv(dev);
293 
294 	if (ipvlan->pcpu_stats) {
295 		struct ipvl_pcpu_stats *pcptr;
296 		u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
297 		u32 rx_errs = 0, tx_drps = 0;
298 		u32 strt;
299 		int idx;
300 
301 		for_each_possible_cpu(idx) {
302 			pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
303 			do {
304 				strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
305 				rx_pkts = pcptr->rx_pkts;
306 				rx_bytes = pcptr->rx_bytes;
307 				rx_mcast = pcptr->rx_mcast;
308 				tx_pkts = pcptr->tx_pkts;
309 				tx_bytes = pcptr->tx_bytes;
310 			} while (u64_stats_fetch_retry_irq(&pcptr->syncp,
311 							   strt));
312 
313 			s->rx_packets += rx_pkts;
314 			s->rx_bytes += rx_bytes;
315 			s->multicast += rx_mcast;
316 			s->tx_packets += tx_pkts;
317 			s->tx_bytes += tx_bytes;
318 
319 			/* u32 values are updated without syncp protection. */
320 			rx_errs += pcptr->rx_errs;
321 			tx_drps += pcptr->tx_drps;
322 		}
323 		s->rx_errors = rx_errs;
324 		s->rx_dropped = rx_errs;
325 		s->tx_dropped = tx_drps;
326 	}
327 }
328 
329 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
330 {
331 	struct ipvl_dev *ipvlan = netdev_priv(dev);
332 	struct net_device *phy_dev = ipvlan->phy_dev;
333 
334 	return vlan_vid_add(phy_dev, proto, vid);
335 }
336 
337 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
338 				   u16 vid)
339 {
340 	struct ipvl_dev *ipvlan = netdev_priv(dev);
341 	struct net_device *phy_dev = ipvlan->phy_dev;
342 
343 	vlan_vid_del(phy_dev, proto, vid);
344 	return 0;
345 }
346 
347 static int ipvlan_get_iflink(const struct net_device *dev)
348 {
349 	struct ipvl_dev *ipvlan = netdev_priv(dev);
350 
351 	return ipvlan->phy_dev->ifindex;
352 }
353 
354 static const struct net_device_ops ipvlan_netdev_ops = {
355 	.ndo_init		= ipvlan_init,
356 	.ndo_uninit		= ipvlan_uninit,
357 	.ndo_open		= ipvlan_open,
358 	.ndo_stop		= ipvlan_stop,
359 	.ndo_start_xmit		= ipvlan_start_xmit,
360 	.ndo_fix_features	= ipvlan_fix_features,
361 	.ndo_change_rx_flags	= ipvlan_change_rx_flags,
362 	.ndo_set_rx_mode	= ipvlan_set_multicast_mac_filter,
363 	.ndo_get_stats64	= ipvlan_get_stats64,
364 	.ndo_vlan_rx_add_vid	= ipvlan_vlan_rx_add_vid,
365 	.ndo_vlan_rx_kill_vid	= ipvlan_vlan_rx_kill_vid,
366 	.ndo_get_iflink		= ipvlan_get_iflink,
367 };
368 
369 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
370 			      unsigned short type, const void *daddr,
371 			      const void *saddr, unsigned len)
372 {
373 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
374 	struct net_device *phy_dev = ipvlan->phy_dev;
375 
376 	/* TODO Probably use a different field than dev_addr so that the
377 	 * mac-address on the virtual device is portable and can be carried
378 	 * while the packets use the mac-addr on the physical device.
379 	 */
380 	return dev_hard_header(skb, phy_dev, type, daddr,
381 			       saddr ? : phy_dev->dev_addr, len);
382 }
383 
384 static const struct header_ops ipvlan_header_ops = {
385 	.create  	= ipvlan_hard_header,
386 	.parse		= eth_header_parse,
387 	.cache		= eth_header_cache,
388 	.cache_update	= eth_header_cache_update,
389 };
390 
391 static void ipvlan_adjust_mtu(struct ipvl_dev *ipvlan, struct net_device *dev)
392 {
393 	ipvlan->dev->mtu = dev->mtu;
394 }
395 
396 static bool netif_is_ipvlan(const struct net_device *dev)
397 {
398 	/* both ipvlan and ipvtap devices use the same netdev_ops */
399 	return dev->netdev_ops == &ipvlan_netdev_ops;
400 }
401 
402 static int ipvlan_ethtool_get_link_ksettings(struct net_device *dev,
403 					     struct ethtool_link_ksettings *cmd)
404 {
405 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
406 
407 	return __ethtool_get_link_ksettings(ipvlan->phy_dev, cmd);
408 }
409 
410 static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
411 				       struct ethtool_drvinfo *drvinfo)
412 {
413 	strlcpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
414 	strlcpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
415 }
416 
417 static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
418 {
419 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
420 
421 	return ipvlan->msg_enable;
422 }
423 
424 static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
425 {
426 	struct ipvl_dev *ipvlan = netdev_priv(dev);
427 
428 	ipvlan->msg_enable = value;
429 }
430 
431 static const struct ethtool_ops ipvlan_ethtool_ops = {
432 	.get_link	= ethtool_op_get_link,
433 	.get_link_ksettings	= ipvlan_ethtool_get_link_ksettings,
434 	.get_drvinfo	= ipvlan_ethtool_get_drvinfo,
435 	.get_msglevel	= ipvlan_ethtool_get_msglevel,
436 	.set_msglevel	= ipvlan_ethtool_set_msglevel,
437 };
438 
439 static int ipvlan_nl_changelink(struct net_device *dev,
440 				struct nlattr *tb[], struct nlattr *data[],
441 				struct netlink_ext_ack *extack)
442 {
443 	struct ipvl_dev *ipvlan = netdev_priv(dev);
444 	struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
445 	int err = 0;
446 
447 	if (!data)
448 		return 0;
449 	if (!ns_capable(dev_net(ipvlan->phy_dev)->user_ns, CAP_NET_ADMIN))
450 		return -EPERM;
451 
452 	if (data[IFLA_IPVLAN_MODE]) {
453 		u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
454 
455 		err = ipvlan_set_port_mode(port, nmode, extack);
456 	}
457 
458 	if (!err && data[IFLA_IPVLAN_FLAGS]) {
459 		u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
460 
461 		if (flags & IPVLAN_F_PRIVATE)
462 			ipvlan_mark_private(port);
463 		else
464 			ipvlan_clear_private(port);
465 
466 		if (flags & IPVLAN_F_VEPA)
467 			ipvlan_mark_vepa(port);
468 		else
469 			ipvlan_clear_vepa(port);
470 	}
471 
472 	return err;
473 }
474 
475 static size_t ipvlan_nl_getsize(const struct net_device *dev)
476 {
477 	return (0
478 		+ nla_total_size(2) /* IFLA_IPVLAN_MODE */
479 		+ nla_total_size(2) /* IFLA_IPVLAN_FLAGS */
480 		);
481 }
482 
483 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[],
484 			      struct netlink_ext_ack *extack)
485 {
486 	if (!data)
487 		return 0;
488 
489 	if (data[IFLA_IPVLAN_MODE]) {
490 		u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
491 
492 		if (mode >= IPVLAN_MODE_MAX)
493 			return -EINVAL;
494 	}
495 	if (data[IFLA_IPVLAN_FLAGS]) {
496 		u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
497 
498 		/* Only two bits are used at this moment. */
499 		if (flags & ~(IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
500 			return -EINVAL;
501 		/* Also both flags can't be active at the same time. */
502 		if ((flags & (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA)) ==
503 		    (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
504 			return -EINVAL;
505 	}
506 
507 	return 0;
508 }
509 
510 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
511 			      const struct net_device *dev)
512 {
513 	struct ipvl_dev *ipvlan = netdev_priv(dev);
514 	struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
515 	int ret = -EINVAL;
516 
517 	if (!port)
518 		goto err;
519 
520 	ret = -EMSGSIZE;
521 	if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
522 		goto err;
523 	if (nla_put_u16(skb, IFLA_IPVLAN_FLAGS, port->flags))
524 		goto err;
525 
526 	return 0;
527 
528 err:
529 	return ret;
530 }
531 
532 int ipvlan_link_new(struct net *src_net, struct net_device *dev,
533 		    struct nlattr *tb[], struct nlattr *data[],
534 		    struct netlink_ext_ack *extack)
535 {
536 	struct ipvl_dev *ipvlan = netdev_priv(dev);
537 	struct ipvl_port *port;
538 	struct net_device *phy_dev;
539 	int err;
540 	u16 mode = IPVLAN_MODE_L3;
541 
542 	if (!tb[IFLA_LINK])
543 		return -EINVAL;
544 
545 	phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
546 	if (!phy_dev)
547 		return -ENODEV;
548 
549 	if (netif_is_ipvlan(phy_dev)) {
550 		struct ipvl_dev *tmp = netdev_priv(phy_dev);
551 
552 		phy_dev = tmp->phy_dev;
553 		if (!ns_capable(dev_net(phy_dev)->user_ns, CAP_NET_ADMIN))
554 			return -EPERM;
555 	} else if (!netif_is_ipvlan_port(phy_dev)) {
556 		/* Exit early if the underlying link is invalid or busy */
557 		if (phy_dev->type != ARPHRD_ETHER ||
558 		    phy_dev->flags & IFF_LOOPBACK) {
559 			netdev_err(phy_dev,
560 				   "Master is either lo or non-ether device\n");
561 			return -EINVAL;
562 		}
563 
564 		if (netdev_is_rx_handler_busy(phy_dev)) {
565 			netdev_err(phy_dev, "Device is already in use.\n");
566 			return -EBUSY;
567 		}
568 	}
569 
570 	ipvlan->phy_dev = phy_dev;
571 	ipvlan->dev = dev;
572 	ipvlan->sfeatures = IPVLAN_FEATURES;
573 	if (!tb[IFLA_MTU])
574 		ipvlan_adjust_mtu(ipvlan, phy_dev);
575 	INIT_LIST_HEAD(&ipvlan->addrs);
576 	spin_lock_init(&ipvlan->addrs_lock);
577 
578 	/* TODO Probably put random address here to be presented to the
579 	 * world but keep using the physical-dev address for the outgoing
580 	 * packets.
581 	 */
582 	eth_hw_addr_set(dev, phy_dev->dev_addr);
583 
584 	dev->priv_flags |= IFF_NO_RX_HANDLER;
585 
586 	err = register_netdevice(dev);
587 	if (err < 0)
588 		return err;
589 
590 	/* ipvlan_init() would have created the port, if required */
591 	port = ipvlan_port_get_rtnl(phy_dev);
592 	ipvlan->port = port;
593 
594 	/* If the port-id base is at the MAX value, then wrap it around and
595 	 * begin from 0x1 again. This may be due to a busy system where lots
596 	 * of slaves are getting created and deleted.
597 	 */
598 	if (port->dev_id_start == 0xFFFE)
599 		port->dev_id_start = 0x1;
600 
601 	/* Since L2 address is shared among all IPvlan slaves including
602 	 * master, use unique 16 bit dev-ids to diffentiate among them.
603 	 * Assign IDs between 0x1 and 0xFFFE (used by the master) to each
604 	 * slave link [see addrconf_ifid_eui48()].
605 	 */
606 	err = ida_simple_get(&port->ida, port->dev_id_start, 0xFFFE,
607 			     GFP_KERNEL);
608 	if (err < 0)
609 		err = ida_simple_get(&port->ida, 0x1, port->dev_id_start,
610 				     GFP_KERNEL);
611 	if (err < 0)
612 		goto unregister_netdev;
613 	dev->dev_id = err;
614 
615 	/* Increment id-base to the next slot for the future assignment */
616 	port->dev_id_start = err + 1;
617 
618 	err = netdev_upper_dev_link(phy_dev, dev, extack);
619 	if (err)
620 		goto remove_ida;
621 
622 	/* Flags are per port and latest update overrides. User has
623 	 * to be consistent in setting it just like the mode attribute.
624 	 */
625 	if (data && data[IFLA_IPVLAN_FLAGS])
626 		port->flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
627 
628 	if (data && data[IFLA_IPVLAN_MODE])
629 		mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
630 
631 	err = ipvlan_set_port_mode(port, mode, extack);
632 	if (err)
633 		goto unlink_netdev;
634 
635 	list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
636 	netif_stacked_transfer_operstate(phy_dev, dev);
637 	return 0;
638 
639 unlink_netdev:
640 	netdev_upper_dev_unlink(phy_dev, dev);
641 remove_ida:
642 	ida_simple_remove(&port->ida, dev->dev_id);
643 unregister_netdev:
644 	unregister_netdevice(dev);
645 	return err;
646 }
647 EXPORT_SYMBOL_GPL(ipvlan_link_new);
648 
649 void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
650 {
651 	struct ipvl_dev *ipvlan = netdev_priv(dev);
652 	struct ipvl_addr *addr, *next;
653 
654 	spin_lock_bh(&ipvlan->addrs_lock);
655 	list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
656 		ipvlan_ht_addr_del(addr);
657 		list_del_rcu(&addr->anode);
658 		kfree_rcu(addr, rcu);
659 	}
660 	spin_unlock_bh(&ipvlan->addrs_lock);
661 
662 	ida_simple_remove(&ipvlan->port->ida, dev->dev_id);
663 	list_del_rcu(&ipvlan->pnode);
664 	unregister_netdevice_queue(dev, head);
665 	netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
666 }
667 EXPORT_SYMBOL_GPL(ipvlan_link_delete);
668 
669 void ipvlan_link_setup(struct net_device *dev)
670 {
671 	ether_setup(dev);
672 
673 	dev->max_mtu = ETH_MAX_MTU;
674 	dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
675 	dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
676 	dev->netdev_ops = &ipvlan_netdev_ops;
677 	dev->needs_free_netdev = true;
678 	dev->header_ops = &ipvlan_header_ops;
679 	dev->ethtool_ops = &ipvlan_ethtool_ops;
680 }
681 EXPORT_SYMBOL_GPL(ipvlan_link_setup);
682 
683 static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
684 {
685 	[IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
686 	[IFLA_IPVLAN_FLAGS] = { .type = NLA_U16 },
687 };
688 
689 static struct net *ipvlan_get_link_net(const struct net_device *dev)
690 {
691 	struct ipvl_dev *ipvlan = netdev_priv(dev);
692 
693 	return dev_net(ipvlan->phy_dev);
694 }
695 
696 static struct rtnl_link_ops ipvlan_link_ops = {
697 	.kind		= "ipvlan",
698 	.priv_size	= sizeof(struct ipvl_dev),
699 
700 	.setup		= ipvlan_link_setup,
701 	.newlink	= ipvlan_link_new,
702 	.dellink	= ipvlan_link_delete,
703 	.get_link_net   = ipvlan_get_link_net,
704 };
705 
706 int ipvlan_link_register(struct rtnl_link_ops *ops)
707 {
708 	ops->get_size	= ipvlan_nl_getsize;
709 	ops->policy	= ipvlan_nl_policy;
710 	ops->validate	= ipvlan_nl_validate;
711 	ops->fill_info	= ipvlan_nl_fillinfo;
712 	ops->changelink = ipvlan_nl_changelink;
713 	ops->maxtype	= IFLA_IPVLAN_MAX;
714 	return rtnl_link_register(ops);
715 }
716 EXPORT_SYMBOL_GPL(ipvlan_link_register);
717 
718 static int ipvlan_device_event(struct notifier_block *unused,
719 			       unsigned long event, void *ptr)
720 {
721 	struct netlink_ext_ack *extack = netdev_notifier_info_to_extack(ptr);
722 	struct netdev_notifier_pre_changeaddr_info *prechaddr_info;
723 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
724 	struct ipvl_dev *ipvlan, *next;
725 	struct ipvl_port *port;
726 	LIST_HEAD(lst_kill);
727 	int err;
728 
729 	if (!netif_is_ipvlan_port(dev))
730 		return NOTIFY_DONE;
731 
732 	port = ipvlan_port_get_rtnl(dev);
733 
734 	switch (event) {
735 	case NETDEV_UP:
736 	case NETDEV_CHANGE:
737 		list_for_each_entry(ipvlan, &port->ipvlans, pnode)
738 			netif_stacked_transfer_operstate(ipvlan->phy_dev,
739 							 ipvlan->dev);
740 		break;
741 
742 	case NETDEV_REGISTER: {
743 		struct net *oldnet, *newnet = dev_net(dev);
744 
745 		oldnet = read_pnet(&port->pnet);
746 		if (net_eq(newnet, oldnet))
747 			break;
748 
749 		write_pnet(&port->pnet, newnet);
750 
751 		ipvlan_migrate_l3s_hook(oldnet, newnet);
752 		break;
753 	}
754 	case NETDEV_UNREGISTER:
755 		if (dev->reg_state != NETREG_UNREGISTERING)
756 			break;
757 
758 		list_for_each_entry_safe(ipvlan, next, &port->ipvlans, pnode)
759 			ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
760 							    &lst_kill);
761 		unregister_netdevice_many(&lst_kill);
762 		break;
763 
764 	case NETDEV_FEAT_CHANGE:
765 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
766 			ipvlan->dev->gso_max_size = dev->gso_max_size;
767 			ipvlan->dev->gso_max_segs = dev->gso_max_segs;
768 			netdev_update_features(ipvlan->dev);
769 		}
770 		break;
771 
772 	case NETDEV_CHANGEMTU:
773 		list_for_each_entry(ipvlan, &port->ipvlans, pnode)
774 			ipvlan_adjust_mtu(ipvlan, dev);
775 		break;
776 
777 	case NETDEV_PRE_CHANGEADDR:
778 		prechaddr_info = ptr;
779 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
780 			err = dev_pre_changeaddr_notify(ipvlan->dev,
781 						    prechaddr_info->dev_addr,
782 						    extack);
783 			if (err)
784 				return notifier_from_errno(err);
785 		}
786 		break;
787 
788 	case NETDEV_CHANGEADDR:
789 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
790 			eth_hw_addr_set(ipvlan->dev, dev->dev_addr);
791 			call_netdevice_notifiers(NETDEV_CHANGEADDR, ipvlan->dev);
792 		}
793 		break;
794 
795 	case NETDEV_PRE_TYPE_CHANGE:
796 		/* Forbid underlying device to change its type. */
797 		return NOTIFY_BAD;
798 	}
799 	return NOTIFY_DONE;
800 }
801 
802 /* the caller must held the addrs lock */
803 static int ipvlan_add_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
804 {
805 	struct ipvl_addr *addr;
806 
807 	addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
808 	if (!addr)
809 		return -ENOMEM;
810 
811 	addr->master = ipvlan;
812 	if (!is_v6) {
813 		memcpy(&addr->ip4addr, iaddr, sizeof(struct in_addr));
814 		addr->atype = IPVL_IPV4;
815 #if IS_ENABLED(CONFIG_IPV6)
816 	} else {
817 		memcpy(&addr->ip6addr, iaddr, sizeof(struct in6_addr));
818 		addr->atype = IPVL_IPV6;
819 #endif
820 	}
821 
822 	list_add_tail_rcu(&addr->anode, &ipvlan->addrs);
823 
824 	/* If the interface is not up, the address will be added to the hash
825 	 * list by ipvlan_open.
826 	 */
827 	if (netif_running(ipvlan->dev))
828 		ipvlan_ht_addr_add(ipvlan, addr);
829 
830 	return 0;
831 }
832 
833 static void ipvlan_del_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
834 {
835 	struct ipvl_addr *addr;
836 
837 	spin_lock_bh(&ipvlan->addrs_lock);
838 	addr = ipvlan_find_addr(ipvlan, iaddr, is_v6);
839 	if (!addr) {
840 		spin_unlock_bh(&ipvlan->addrs_lock);
841 		return;
842 	}
843 
844 	ipvlan_ht_addr_del(addr);
845 	list_del_rcu(&addr->anode);
846 	spin_unlock_bh(&ipvlan->addrs_lock);
847 	kfree_rcu(addr, rcu);
848 }
849 
850 static bool ipvlan_is_valid_dev(const struct net_device *dev)
851 {
852 	struct ipvl_dev *ipvlan = netdev_priv(dev);
853 
854 	if (!netif_is_ipvlan(dev))
855 		return false;
856 
857 	if (!ipvlan || !ipvlan->port)
858 		return false;
859 
860 	return true;
861 }
862 
863 #if IS_ENABLED(CONFIG_IPV6)
864 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
865 {
866 	int ret = -EINVAL;
867 
868 	spin_lock_bh(&ipvlan->addrs_lock);
869 	if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true))
870 		netif_err(ipvlan, ifup, ipvlan->dev,
871 			  "Failed to add IPv6=%pI6c addr for %s intf\n",
872 			  ip6_addr, ipvlan->dev->name);
873 	else
874 		ret = ipvlan_add_addr(ipvlan, ip6_addr, true);
875 	spin_unlock_bh(&ipvlan->addrs_lock);
876 	return ret;
877 }
878 
879 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
880 {
881 	return ipvlan_del_addr(ipvlan, ip6_addr, true);
882 }
883 
884 static int ipvlan_addr6_event(struct notifier_block *unused,
885 			      unsigned long event, void *ptr)
886 {
887 	struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
888 	struct net_device *dev = (struct net_device *)if6->idev->dev;
889 	struct ipvl_dev *ipvlan = netdev_priv(dev);
890 
891 	if (!ipvlan_is_valid_dev(dev))
892 		return NOTIFY_DONE;
893 
894 	switch (event) {
895 	case NETDEV_UP:
896 		if (ipvlan_add_addr6(ipvlan, &if6->addr))
897 			return NOTIFY_BAD;
898 		break;
899 
900 	case NETDEV_DOWN:
901 		ipvlan_del_addr6(ipvlan, &if6->addr);
902 		break;
903 	}
904 
905 	return NOTIFY_OK;
906 }
907 
908 static int ipvlan_addr6_validator_event(struct notifier_block *unused,
909 					unsigned long event, void *ptr)
910 {
911 	struct in6_validator_info *i6vi = (struct in6_validator_info *)ptr;
912 	struct net_device *dev = (struct net_device *)i6vi->i6vi_dev->dev;
913 	struct ipvl_dev *ipvlan = netdev_priv(dev);
914 
915 	if (!ipvlan_is_valid_dev(dev))
916 		return NOTIFY_DONE;
917 
918 	switch (event) {
919 	case NETDEV_UP:
920 		if (ipvlan_addr_busy(ipvlan->port, &i6vi->i6vi_addr, true)) {
921 			NL_SET_ERR_MSG(i6vi->extack,
922 				       "Address already assigned to an ipvlan device");
923 			return notifier_from_errno(-EADDRINUSE);
924 		}
925 		break;
926 	}
927 
928 	return NOTIFY_OK;
929 }
930 #endif
931 
932 static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
933 {
934 	int ret = -EINVAL;
935 
936 	spin_lock_bh(&ipvlan->addrs_lock);
937 	if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false))
938 		netif_err(ipvlan, ifup, ipvlan->dev,
939 			  "Failed to add IPv4=%pI4 on %s intf.\n",
940 			  ip4_addr, ipvlan->dev->name);
941 	else
942 		ret = ipvlan_add_addr(ipvlan, ip4_addr, false);
943 	spin_unlock_bh(&ipvlan->addrs_lock);
944 	return ret;
945 }
946 
947 static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
948 {
949 	return ipvlan_del_addr(ipvlan, ip4_addr, false);
950 }
951 
952 static int ipvlan_addr4_event(struct notifier_block *unused,
953 			      unsigned long event, void *ptr)
954 {
955 	struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
956 	struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
957 	struct ipvl_dev *ipvlan = netdev_priv(dev);
958 	struct in_addr ip4_addr;
959 
960 	if (!ipvlan_is_valid_dev(dev))
961 		return NOTIFY_DONE;
962 
963 	switch (event) {
964 	case NETDEV_UP:
965 		ip4_addr.s_addr = if4->ifa_address;
966 		if (ipvlan_add_addr4(ipvlan, &ip4_addr))
967 			return NOTIFY_BAD;
968 		break;
969 
970 	case NETDEV_DOWN:
971 		ip4_addr.s_addr = if4->ifa_address;
972 		ipvlan_del_addr4(ipvlan, &ip4_addr);
973 		break;
974 	}
975 
976 	return NOTIFY_OK;
977 }
978 
979 static int ipvlan_addr4_validator_event(struct notifier_block *unused,
980 					unsigned long event, void *ptr)
981 {
982 	struct in_validator_info *ivi = (struct in_validator_info *)ptr;
983 	struct net_device *dev = (struct net_device *)ivi->ivi_dev->dev;
984 	struct ipvl_dev *ipvlan = netdev_priv(dev);
985 
986 	if (!ipvlan_is_valid_dev(dev))
987 		return NOTIFY_DONE;
988 
989 	switch (event) {
990 	case NETDEV_UP:
991 		if (ipvlan_addr_busy(ipvlan->port, &ivi->ivi_addr, false)) {
992 			NL_SET_ERR_MSG(ivi->extack,
993 				       "Address already assigned to an ipvlan device");
994 			return notifier_from_errno(-EADDRINUSE);
995 		}
996 		break;
997 	}
998 
999 	return NOTIFY_OK;
1000 }
1001 
1002 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
1003 	.notifier_call = ipvlan_addr4_event,
1004 };
1005 
1006 static struct notifier_block ipvlan_addr4_vtor_notifier_block __read_mostly = {
1007 	.notifier_call = ipvlan_addr4_validator_event,
1008 };
1009 
1010 static struct notifier_block ipvlan_notifier_block __read_mostly = {
1011 	.notifier_call = ipvlan_device_event,
1012 };
1013 
1014 #if IS_ENABLED(CONFIG_IPV6)
1015 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
1016 	.notifier_call = ipvlan_addr6_event,
1017 };
1018 
1019 static struct notifier_block ipvlan_addr6_vtor_notifier_block __read_mostly = {
1020 	.notifier_call = ipvlan_addr6_validator_event,
1021 };
1022 #endif
1023 
1024 static int __init ipvlan_init_module(void)
1025 {
1026 	int err;
1027 
1028 	ipvlan_init_secret();
1029 	register_netdevice_notifier(&ipvlan_notifier_block);
1030 #if IS_ENABLED(CONFIG_IPV6)
1031 	register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1032 	register_inet6addr_validator_notifier(
1033 	    &ipvlan_addr6_vtor_notifier_block);
1034 #endif
1035 	register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1036 	register_inetaddr_validator_notifier(&ipvlan_addr4_vtor_notifier_block);
1037 
1038 	err = ipvlan_l3s_init();
1039 	if (err < 0)
1040 		goto error;
1041 
1042 	err = ipvlan_link_register(&ipvlan_link_ops);
1043 	if (err < 0) {
1044 		ipvlan_l3s_cleanup();
1045 		goto error;
1046 	}
1047 
1048 	return 0;
1049 error:
1050 	unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1051 	unregister_inetaddr_validator_notifier(
1052 	    &ipvlan_addr4_vtor_notifier_block);
1053 #if IS_ENABLED(CONFIG_IPV6)
1054 	unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1055 	unregister_inet6addr_validator_notifier(
1056 	    &ipvlan_addr6_vtor_notifier_block);
1057 #endif
1058 	unregister_netdevice_notifier(&ipvlan_notifier_block);
1059 	return err;
1060 }
1061 
1062 static void __exit ipvlan_cleanup_module(void)
1063 {
1064 	rtnl_link_unregister(&ipvlan_link_ops);
1065 	ipvlan_l3s_cleanup();
1066 	unregister_netdevice_notifier(&ipvlan_notifier_block);
1067 	unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1068 	unregister_inetaddr_validator_notifier(
1069 	    &ipvlan_addr4_vtor_notifier_block);
1070 #if IS_ENABLED(CONFIG_IPV6)
1071 	unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1072 	unregister_inet6addr_validator_notifier(
1073 	    &ipvlan_addr6_vtor_notifier_block);
1074 #endif
1075 }
1076 
1077 module_init(ipvlan_init_module);
1078 module_exit(ipvlan_cleanup_module);
1079 
1080 MODULE_LICENSE("GPL");
1081 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
1082 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
1083 MODULE_ALIAS_RTNL_LINK("ipvlan");
1084