xref: /openbmc/linux/drivers/net/ipvlan/ipvlan_main.c (revision 2cf1c348)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Copyright (c) 2014 Mahesh Bandewar <maheshb@google.com>
3  */
4 
5 #include <linux/ethtool.h>
6 
7 #include "ipvlan.h"
8 
9 static int ipvlan_set_port_mode(struct ipvl_port *port, u16 nval,
10 				struct netlink_ext_ack *extack)
11 {
12 	struct ipvl_dev *ipvlan;
13 	unsigned int flags;
14 	int err;
15 
16 	ASSERT_RTNL();
17 	if (port->mode != nval) {
18 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
19 			flags = ipvlan->dev->flags;
20 			if (nval == IPVLAN_MODE_L3 || nval == IPVLAN_MODE_L3S) {
21 				err = dev_change_flags(ipvlan->dev,
22 						       flags | IFF_NOARP,
23 						       extack);
24 			} else {
25 				err = dev_change_flags(ipvlan->dev,
26 						       flags & ~IFF_NOARP,
27 						       extack);
28 			}
29 			if (unlikely(err))
30 				goto fail;
31 		}
32 		if (nval == IPVLAN_MODE_L3S) {
33 			/* New mode is L3S */
34 			err = ipvlan_l3s_register(port);
35 			if (err)
36 				goto fail;
37 		} else if (port->mode == IPVLAN_MODE_L3S) {
38 			/* Old mode was L3S */
39 			ipvlan_l3s_unregister(port);
40 		}
41 		port->mode = nval;
42 	}
43 	return 0;
44 
45 fail:
46 	/* Undo the flags changes that have been done so far. */
47 	list_for_each_entry_continue_reverse(ipvlan, &port->ipvlans, pnode) {
48 		flags = ipvlan->dev->flags;
49 		if (port->mode == IPVLAN_MODE_L3 ||
50 		    port->mode == IPVLAN_MODE_L3S)
51 			dev_change_flags(ipvlan->dev, flags | IFF_NOARP,
52 					 NULL);
53 		else
54 			dev_change_flags(ipvlan->dev, flags & ~IFF_NOARP,
55 					 NULL);
56 	}
57 
58 	return err;
59 }
60 
61 static int ipvlan_port_create(struct net_device *dev)
62 {
63 	struct ipvl_port *port;
64 	int err, idx;
65 
66 	port = kzalloc(sizeof(struct ipvl_port), GFP_KERNEL);
67 	if (!port)
68 		return -ENOMEM;
69 
70 	write_pnet(&port->pnet, dev_net(dev));
71 	port->dev = dev;
72 	port->mode = IPVLAN_MODE_L3;
73 	INIT_LIST_HEAD(&port->ipvlans);
74 	for (idx = 0; idx < IPVLAN_HASH_SIZE; idx++)
75 		INIT_HLIST_HEAD(&port->hlhead[idx]);
76 
77 	skb_queue_head_init(&port->backlog);
78 	INIT_WORK(&port->wq, ipvlan_process_multicast);
79 	ida_init(&port->ida);
80 	port->dev_id_start = 1;
81 
82 	err = netdev_rx_handler_register(dev, ipvlan_handle_frame, port);
83 	if (err)
84 		goto err;
85 
86 	return 0;
87 
88 err:
89 	kfree(port);
90 	return err;
91 }
92 
93 static void ipvlan_port_destroy(struct net_device *dev)
94 {
95 	struct ipvl_port *port = ipvlan_port_get_rtnl(dev);
96 	struct sk_buff *skb;
97 
98 	if (port->mode == IPVLAN_MODE_L3S)
99 		ipvlan_l3s_unregister(port);
100 	netdev_rx_handler_unregister(dev);
101 	cancel_work_sync(&port->wq);
102 	while ((skb = __skb_dequeue(&port->backlog)) != NULL) {
103 		dev_put(skb->dev);
104 		kfree_skb(skb);
105 	}
106 	ida_destroy(&port->ida);
107 	kfree(port);
108 }
109 
110 #define IPVLAN_ALWAYS_ON_OFLOADS \
111 	(NETIF_F_SG | NETIF_F_HW_CSUM | \
112 	 NETIF_F_GSO_ROBUST | NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL)
113 
114 #define IPVLAN_ALWAYS_ON \
115 	(IPVLAN_ALWAYS_ON_OFLOADS | NETIF_F_LLTX | NETIF_F_VLAN_CHALLENGED)
116 
117 #define IPVLAN_FEATURES \
118 	(NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
119 	 NETIF_F_GSO | NETIF_F_ALL_TSO | NETIF_F_GSO_ROBUST | \
120 	 NETIF_F_GRO | NETIF_F_RXCSUM | \
121 	 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
122 
123 	/* NETIF_F_GSO_ENCAP_ALL NETIF_F_GSO_SOFTWARE Newly added */
124 
125 #define IPVLAN_STATE_MASK \
126 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
127 
128 static int ipvlan_init(struct net_device *dev)
129 {
130 	struct ipvl_dev *ipvlan = netdev_priv(dev);
131 	struct net_device *phy_dev = ipvlan->phy_dev;
132 	struct ipvl_port *port;
133 	int err;
134 
135 	dev->state = (dev->state & ~IPVLAN_STATE_MASK) |
136 		     (phy_dev->state & IPVLAN_STATE_MASK);
137 	dev->features = phy_dev->features & IPVLAN_FEATURES;
138 	dev->features |= IPVLAN_ALWAYS_ON;
139 	dev->vlan_features = phy_dev->vlan_features & IPVLAN_FEATURES;
140 	dev->vlan_features |= IPVLAN_ALWAYS_ON_OFLOADS;
141 	dev->hw_enc_features |= dev->features;
142 	netif_set_gso_max_size(dev, phy_dev->gso_max_size);
143 	netif_set_gso_max_segs(dev, phy_dev->gso_max_segs);
144 	dev->hard_header_len = phy_dev->hard_header_len;
145 
146 	netdev_lockdep_set_classes(dev);
147 
148 	ipvlan->pcpu_stats = netdev_alloc_pcpu_stats(struct ipvl_pcpu_stats);
149 	if (!ipvlan->pcpu_stats)
150 		return -ENOMEM;
151 
152 	if (!netif_is_ipvlan_port(phy_dev)) {
153 		err = ipvlan_port_create(phy_dev);
154 		if (err < 0) {
155 			free_percpu(ipvlan->pcpu_stats);
156 			return err;
157 		}
158 	}
159 	port = ipvlan_port_get_rtnl(phy_dev);
160 	port->count += 1;
161 	return 0;
162 }
163 
164 static void ipvlan_uninit(struct net_device *dev)
165 {
166 	struct ipvl_dev *ipvlan = netdev_priv(dev);
167 	struct net_device *phy_dev = ipvlan->phy_dev;
168 	struct ipvl_port *port;
169 
170 	free_percpu(ipvlan->pcpu_stats);
171 
172 	port = ipvlan_port_get_rtnl(phy_dev);
173 	port->count -= 1;
174 	if (!port->count)
175 		ipvlan_port_destroy(port->dev);
176 }
177 
178 static int ipvlan_open(struct net_device *dev)
179 {
180 	struct ipvl_dev *ipvlan = netdev_priv(dev);
181 	struct ipvl_addr *addr;
182 
183 	if (ipvlan->port->mode == IPVLAN_MODE_L3 ||
184 	    ipvlan->port->mode == IPVLAN_MODE_L3S)
185 		dev->flags |= IFF_NOARP;
186 	else
187 		dev->flags &= ~IFF_NOARP;
188 
189 	rcu_read_lock();
190 	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
191 		ipvlan_ht_addr_add(ipvlan, addr);
192 	rcu_read_unlock();
193 
194 	return 0;
195 }
196 
197 static int ipvlan_stop(struct net_device *dev)
198 {
199 	struct ipvl_dev *ipvlan = netdev_priv(dev);
200 	struct net_device *phy_dev = ipvlan->phy_dev;
201 	struct ipvl_addr *addr;
202 
203 	dev_uc_unsync(phy_dev, dev);
204 	dev_mc_unsync(phy_dev, dev);
205 
206 	rcu_read_lock();
207 	list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
208 		ipvlan_ht_addr_del(addr);
209 	rcu_read_unlock();
210 
211 	return 0;
212 }
213 
214 static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
215 				     struct net_device *dev)
216 {
217 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
218 	int skblen = skb->len;
219 	int ret;
220 
221 	ret = ipvlan_queue_xmit(skb, dev);
222 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
223 		struct ipvl_pcpu_stats *pcptr;
224 
225 		pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
226 
227 		u64_stats_update_begin(&pcptr->syncp);
228 		pcptr->tx_pkts++;
229 		pcptr->tx_bytes += skblen;
230 		u64_stats_update_end(&pcptr->syncp);
231 	} else {
232 		this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
233 	}
234 	return ret;
235 }
236 
237 static netdev_features_t ipvlan_fix_features(struct net_device *dev,
238 					     netdev_features_t features)
239 {
240 	struct ipvl_dev *ipvlan = netdev_priv(dev);
241 
242 	features |= NETIF_F_ALL_FOR_ALL;
243 	features &= (ipvlan->sfeatures | ~IPVLAN_FEATURES);
244 	features = netdev_increment_features(ipvlan->phy_dev->features,
245 					     features, features);
246 	features |= IPVLAN_ALWAYS_ON;
247 	features &= (IPVLAN_FEATURES | IPVLAN_ALWAYS_ON);
248 
249 	return features;
250 }
251 
252 static void ipvlan_change_rx_flags(struct net_device *dev, int change)
253 {
254 	struct ipvl_dev *ipvlan = netdev_priv(dev);
255 	struct net_device *phy_dev = ipvlan->phy_dev;
256 
257 	if (change & IFF_ALLMULTI)
258 		dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
259 }
260 
261 static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
262 {
263 	struct ipvl_dev *ipvlan = netdev_priv(dev);
264 
265 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
266 		bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
267 	} else {
268 		struct netdev_hw_addr *ha;
269 		DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
270 
271 		bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
272 		netdev_for_each_mc_addr(ha, dev)
273 			__set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
274 
275 		/* Turn-on broadcast bit irrespective of address family,
276 		 * since broadcast is deferred to a work-queue, hence no
277 		 * impact on fast-path processing.
278 		 */
279 		__set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
280 
281 		bitmap_copy(ipvlan->mac_filters, mc_filters,
282 			    IPVLAN_MAC_FILTER_SIZE);
283 	}
284 	dev_uc_sync(ipvlan->phy_dev, dev);
285 	dev_mc_sync(ipvlan->phy_dev, dev);
286 }
287 
288 static void ipvlan_get_stats64(struct net_device *dev,
289 			       struct rtnl_link_stats64 *s)
290 {
291 	struct ipvl_dev *ipvlan = netdev_priv(dev);
292 
293 	if (ipvlan->pcpu_stats) {
294 		struct ipvl_pcpu_stats *pcptr;
295 		u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
296 		u32 rx_errs = 0, tx_drps = 0;
297 		u32 strt;
298 		int idx;
299 
300 		for_each_possible_cpu(idx) {
301 			pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
302 			do {
303 				strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
304 				rx_pkts = pcptr->rx_pkts;
305 				rx_bytes = pcptr->rx_bytes;
306 				rx_mcast = pcptr->rx_mcast;
307 				tx_pkts = pcptr->tx_pkts;
308 				tx_bytes = pcptr->tx_bytes;
309 			} while (u64_stats_fetch_retry_irq(&pcptr->syncp,
310 							   strt));
311 
312 			s->rx_packets += rx_pkts;
313 			s->rx_bytes += rx_bytes;
314 			s->multicast += rx_mcast;
315 			s->tx_packets += tx_pkts;
316 			s->tx_bytes += tx_bytes;
317 
318 			/* u32 values are updated without syncp protection. */
319 			rx_errs += pcptr->rx_errs;
320 			tx_drps += pcptr->tx_drps;
321 		}
322 		s->rx_errors = rx_errs;
323 		s->rx_dropped = rx_errs;
324 		s->tx_dropped = tx_drps;
325 	}
326 }
327 
328 static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
329 {
330 	struct ipvl_dev *ipvlan = netdev_priv(dev);
331 	struct net_device *phy_dev = ipvlan->phy_dev;
332 
333 	return vlan_vid_add(phy_dev, proto, vid);
334 }
335 
336 static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
337 				   u16 vid)
338 {
339 	struct ipvl_dev *ipvlan = netdev_priv(dev);
340 	struct net_device *phy_dev = ipvlan->phy_dev;
341 
342 	vlan_vid_del(phy_dev, proto, vid);
343 	return 0;
344 }
345 
346 static int ipvlan_get_iflink(const struct net_device *dev)
347 {
348 	struct ipvl_dev *ipvlan = netdev_priv(dev);
349 
350 	return ipvlan->phy_dev->ifindex;
351 }
352 
353 static const struct net_device_ops ipvlan_netdev_ops = {
354 	.ndo_init		= ipvlan_init,
355 	.ndo_uninit		= ipvlan_uninit,
356 	.ndo_open		= ipvlan_open,
357 	.ndo_stop		= ipvlan_stop,
358 	.ndo_start_xmit		= ipvlan_start_xmit,
359 	.ndo_fix_features	= ipvlan_fix_features,
360 	.ndo_change_rx_flags	= ipvlan_change_rx_flags,
361 	.ndo_set_rx_mode	= ipvlan_set_multicast_mac_filter,
362 	.ndo_get_stats64	= ipvlan_get_stats64,
363 	.ndo_vlan_rx_add_vid	= ipvlan_vlan_rx_add_vid,
364 	.ndo_vlan_rx_kill_vid	= ipvlan_vlan_rx_kill_vid,
365 	.ndo_get_iflink		= ipvlan_get_iflink,
366 };
367 
368 static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
369 			      unsigned short type, const void *daddr,
370 			      const void *saddr, unsigned len)
371 {
372 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
373 	struct net_device *phy_dev = ipvlan->phy_dev;
374 
375 	/* TODO Probably use a different field than dev_addr so that the
376 	 * mac-address on the virtual device is portable and can be carried
377 	 * while the packets use the mac-addr on the physical device.
378 	 */
379 	return dev_hard_header(skb, phy_dev, type, daddr,
380 			       saddr ? : phy_dev->dev_addr, len);
381 }
382 
383 static const struct header_ops ipvlan_header_ops = {
384 	.create  	= ipvlan_hard_header,
385 	.parse		= eth_header_parse,
386 	.cache		= eth_header_cache,
387 	.cache_update	= eth_header_cache_update,
388 };
389 
390 static void ipvlan_adjust_mtu(struct ipvl_dev *ipvlan, struct net_device *dev)
391 {
392 	ipvlan->dev->mtu = dev->mtu;
393 }
394 
395 static bool netif_is_ipvlan(const struct net_device *dev)
396 {
397 	/* both ipvlan and ipvtap devices use the same netdev_ops */
398 	return dev->netdev_ops == &ipvlan_netdev_ops;
399 }
400 
401 static int ipvlan_ethtool_get_link_ksettings(struct net_device *dev,
402 					     struct ethtool_link_ksettings *cmd)
403 {
404 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
405 
406 	return __ethtool_get_link_ksettings(ipvlan->phy_dev, cmd);
407 }
408 
409 static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
410 				       struct ethtool_drvinfo *drvinfo)
411 {
412 	strlcpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
413 	strlcpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
414 }
415 
416 static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
417 {
418 	const struct ipvl_dev *ipvlan = netdev_priv(dev);
419 
420 	return ipvlan->msg_enable;
421 }
422 
423 static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
424 {
425 	struct ipvl_dev *ipvlan = netdev_priv(dev);
426 
427 	ipvlan->msg_enable = value;
428 }
429 
430 static const struct ethtool_ops ipvlan_ethtool_ops = {
431 	.get_link	= ethtool_op_get_link,
432 	.get_link_ksettings	= ipvlan_ethtool_get_link_ksettings,
433 	.get_drvinfo	= ipvlan_ethtool_get_drvinfo,
434 	.get_msglevel	= ipvlan_ethtool_get_msglevel,
435 	.set_msglevel	= ipvlan_ethtool_set_msglevel,
436 };
437 
438 static int ipvlan_nl_changelink(struct net_device *dev,
439 				struct nlattr *tb[], struct nlattr *data[],
440 				struct netlink_ext_ack *extack)
441 {
442 	struct ipvl_dev *ipvlan = netdev_priv(dev);
443 	struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
444 	int err = 0;
445 
446 	if (!data)
447 		return 0;
448 	if (!ns_capable(dev_net(ipvlan->phy_dev)->user_ns, CAP_NET_ADMIN))
449 		return -EPERM;
450 
451 	if (data[IFLA_IPVLAN_MODE]) {
452 		u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
453 
454 		err = ipvlan_set_port_mode(port, nmode, extack);
455 	}
456 
457 	if (!err && data[IFLA_IPVLAN_FLAGS]) {
458 		u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
459 
460 		if (flags & IPVLAN_F_PRIVATE)
461 			ipvlan_mark_private(port);
462 		else
463 			ipvlan_clear_private(port);
464 
465 		if (flags & IPVLAN_F_VEPA)
466 			ipvlan_mark_vepa(port);
467 		else
468 			ipvlan_clear_vepa(port);
469 	}
470 
471 	return err;
472 }
473 
474 static size_t ipvlan_nl_getsize(const struct net_device *dev)
475 {
476 	return (0
477 		+ nla_total_size(2) /* IFLA_IPVLAN_MODE */
478 		+ nla_total_size(2) /* IFLA_IPVLAN_FLAGS */
479 		);
480 }
481 
482 static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[],
483 			      struct netlink_ext_ack *extack)
484 {
485 	if (!data)
486 		return 0;
487 
488 	if (data[IFLA_IPVLAN_MODE]) {
489 		u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
490 
491 		if (mode >= IPVLAN_MODE_MAX)
492 			return -EINVAL;
493 	}
494 	if (data[IFLA_IPVLAN_FLAGS]) {
495 		u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
496 
497 		/* Only two bits are used at this moment. */
498 		if (flags & ~(IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
499 			return -EINVAL;
500 		/* Also both flags can't be active at the same time. */
501 		if ((flags & (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA)) ==
502 		    (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
503 			return -EINVAL;
504 	}
505 
506 	return 0;
507 }
508 
509 static int ipvlan_nl_fillinfo(struct sk_buff *skb,
510 			      const struct net_device *dev)
511 {
512 	struct ipvl_dev *ipvlan = netdev_priv(dev);
513 	struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
514 	int ret = -EINVAL;
515 
516 	if (!port)
517 		goto err;
518 
519 	ret = -EMSGSIZE;
520 	if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
521 		goto err;
522 	if (nla_put_u16(skb, IFLA_IPVLAN_FLAGS, port->flags))
523 		goto err;
524 
525 	return 0;
526 
527 err:
528 	return ret;
529 }
530 
531 int ipvlan_link_new(struct net *src_net, struct net_device *dev,
532 		    struct nlattr *tb[], struct nlattr *data[],
533 		    struct netlink_ext_ack *extack)
534 {
535 	struct ipvl_dev *ipvlan = netdev_priv(dev);
536 	struct ipvl_port *port;
537 	struct net_device *phy_dev;
538 	int err;
539 	u16 mode = IPVLAN_MODE_L3;
540 
541 	if (!tb[IFLA_LINK])
542 		return -EINVAL;
543 
544 	phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
545 	if (!phy_dev)
546 		return -ENODEV;
547 
548 	if (netif_is_ipvlan(phy_dev)) {
549 		struct ipvl_dev *tmp = netdev_priv(phy_dev);
550 
551 		phy_dev = tmp->phy_dev;
552 		if (!ns_capable(dev_net(phy_dev)->user_ns, CAP_NET_ADMIN))
553 			return -EPERM;
554 	} else if (!netif_is_ipvlan_port(phy_dev)) {
555 		/* Exit early if the underlying link is invalid or busy */
556 		if (phy_dev->type != ARPHRD_ETHER ||
557 		    phy_dev->flags & IFF_LOOPBACK) {
558 			netdev_err(phy_dev,
559 				   "Master is either lo or non-ether device\n");
560 			return -EINVAL;
561 		}
562 
563 		if (netdev_is_rx_handler_busy(phy_dev)) {
564 			netdev_err(phy_dev, "Device is already in use.\n");
565 			return -EBUSY;
566 		}
567 	}
568 
569 	ipvlan->phy_dev = phy_dev;
570 	ipvlan->dev = dev;
571 	ipvlan->sfeatures = IPVLAN_FEATURES;
572 	if (!tb[IFLA_MTU])
573 		ipvlan_adjust_mtu(ipvlan, phy_dev);
574 	INIT_LIST_HEAD(&ipvlan->addrs);
575 	spin_lock_init(&ipvlan->addrs_lock);
576 
577 	/* TODO Probably put random address here to be presented to the
578 	 * world but keep using the physical-dev address for the outgoing
579 	 * packets.
580 	 */
581 	eth_hw_addr_set(dev, phy_dev->dev_addr);
582 
583 	dev->priv_flags |= IFF_NO_RX_HANDLER;
584 
585 	err = register_netdevice(dev);
586 	if (err < 0)
587 		return err;
588 
589 	/* ipvlan_init() would have created the port, if required */
590 	port = ipvlan_port_get_rtnl(phy_dev);
591 	ipvlan->port = port;
592 
593 	/* If the port-id base is at the MAX value, then wrap it around and
594 	 * begin from 0x1 again. This may be due to a busy system where lots
595 	 * of slaves are getting created and deleted.
596 	 */
597 	if (port->dev_id_start == 0xFFFE)
598 		port->dev_id_start = 0x1;
599 
600 	/* Since L2 address is shared among all IPvlan slaves including
601 	 * master, use unique 16 bit dev-ids to diffentiate among them.
602 	 * Assign IDs between 0x1 and 0xFFFE (used by the master) to each
603 	 * slave link [see addrconf_ifid_eui48()].
604 	 */
605 	err = ida_simple_get(&port->ida, port->dev_id_start, 0xFFFE,
606 			     GFP_KERNEL);
607 	if (err < 0)
608 		err = ida_simple_get(&port->ida, 0x1, port->dev_id_start,
609 				     GFP_KERNEL);
610 	if (err < 0)
611 		goto unregister_netdev;
612 	dev->dev_id = err;
613 
614 	/* Increment id-base to the next slot for the future assignment */
615 	port->dev_id_start = err + 1;
616 
617 	err = netdev_upper_dev_link(phy_dev, dev, extack);
618 	if (err)
619 		goto remove_ida;
620 
621 	/* Flags are per port and latest update overrides. User has
622 	 * to be consistent in setting it just like the mode attribute.
623 	 */
624 	if (data && data[IFLA_IPVLAN_FLAGS])
625 		port->flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
626 
627 	if (data && data[IFLA_IPVLAN_MODE])
628 		mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
629 
630 	err = ipvlan_set_port_mode(port, mode, extack);
631 	if (err)
632 		goto unlink_netdev;
633 
634 	list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
635 	netif_stacked_transfer_operstate(phy_dev, dev);
636 	return 0;
637 
638 unlink_netdev:
639 	netdev_upper_dev_unlink(phy_dev, dev);
640 remove_ida:
641 	ida_simple_remove(&port->ida, dev->dev_id);
642 unregister_netdev:
643 	unregister_netdevice(dev);
644 	return err;
645 }
646 EXPORT_SYMBOL_GPL(ipvlan_link_new);
647 
648 void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
649 {
650 	struct ipvl_dev *ipvlan = netdev_priv(dev);
651 	struct ipvl_addr *addr, *next;
652 
653 	spin_lock_bh(&ipvlan->addrs_lock);
654 	list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
655 		ipvlan_ht_addr_del(addr);
656 		list_del_rcu(&addr->anode);
657 		kfree_rcu(addr, rcu);
658 	}
659 	spin_unlock_bh(&ipvlan->addrs_lock);
660 
661 	ida_simple_remove(&ipvlan->port->ida, dev->dev_id);
662 	list_del_rcu(&ipvlan->pnode);
663 	unregister_netdevice_queue(dev, head);
664 	netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
665 }
666 EXPORT_SYMBOL_GPL(ipvlan_link_delete);
667 
668 void ipvlan_link_setup(struct net_device *dev)
669 {
670 	ether_setup(dev);
671 
672 	dev->max_mtu = ETH_MAX_MTU;
673 	dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
674 	dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
675 	dev->netdev_ops = &ipvlan_netdev_ops;
676 	dev->needs_free_netdev = true;
677 	dev->header_ops = &ipvlan_header_ops;
678 	dev->ethtool_ops = &ipvlan_ethtool_ops;
679 }
680 EXPORT_SYMBOL_GPL(ipvlan_link_setup);
681 
682 static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
683 {
684 	[IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
685 	[IFLA_IPVLAN_FLAGS] = { .type = NLA_U16 },
686 };
687 
688 static struct net *ipvlan_get_link_net(const struct net_device *dev)
689 {
690 	struct ipvl_dev *ipvlan = netdev_priv(dev);
691 
692 	return dev_net(ipvlan->phy_dev);
693 }
694 
695 static struct rtnl_link_ops ipvlan_link_ops = {
696 	.kind		= "ipvlan",
697 	.priv_size	= sizeof(struct ipvl_dev),
698 
699 	.setup		= ipvlan_link_setup,
700 	.newlink	= ipvlan_link_new,
701 	.dellink	= ipvlan_link_delete,
702 	.get_link_net   = ipvlan_get_link_net,
703 };
704 
705 int ipvlan_link_register(struct rtnl_link_ops *ops)
706 {
707 	ops->get_size	= ipvlan_nl_getsize;
708 	ops->policy	= ipvlan_nl_policy;
709 	ops->validate	= ipvlan_nl_validate;
710 	ops->fill_info	= ipvlan_nl_fillinfo;
711 	ops->changelink = ipvlan_nl_changelink;
712 	ops->maxtype	= IFLA_IPVLAN_MAX;
713 	return rtnl_link_register(ops);
714 }
715 EXPORT_SYMBOL_GPL(ipvlan_link_register);
716 
717 static int ipvlan_device_event(struct notifier_block *unused,
718 			       unsigned long event, void *ptr)
719 {
720 	struct netlink_ext_ack *extack = netdev_notifier_info_to_extack(ptr);
721 	struct netdev_notifier_pre_changeaddr_info *prechaddr_info;
722 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
723 	struct ipvl_dev *ipvlan, *next;
724 	struct ipvl_port *port;
725 	LIST_HEAD(lst_kill);
726 	int err;
727 
728 	if (!netif_is_ipvlan_port(dev))
729 		return NOTIFY_DONE;
730 
731 	port = ipvlan_port_get_rtnl(dev);
732 
733 	switch (event) {
734 	case NETDEV_UP:
735 	case NETDEV_CHANGE:
736 		list_for_each_entry(ipvlan, &port->ipvlans, pnode)
737 			netif_stacked_transfer_operstate(ipvlan->phy_dev,
738 							 ipvlan->dev);
739 		break;
740 
741 	case NETDEV_REGISTER: {
742 		struct net *oldnet, *newnet = dev_net(dev);
743 
744 		oldnet = read_pnet(&port->pnet);
745 		if (net_eq(newnet, oldnet))
746 			break;
747 
748 		write_pnet(&port->pnet, newnet);
749 
750 		ipvlan_migrate_l3s_hook(oldnet, newnet);
751 		break;
752 	}
753 	case NETDEV_UNREGISTER:
754 		if (dev->reg_state != NETREG_UNREGISTERING)
755 			break;
756 
757 		list_for_each_entry_safe(ipvlan, next, &port->ipvlans, pnode)
758 			ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
759 							    &lst_kill);
760 		unregister_netdevice_many(&lst_kill);
761 		break;
762 
763 	case NETDEV_FEAT_CHANGE:
764 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
765 			netif_set_gso_max_size(ipvlan->dev, dev->gso_max_size);
766 			netif_set_gso_max_segs(ipvlan->dev, dev->gso_max_segs);
767 			netdev_update_features(ipvlan->dev);
768 		}
769 		break;
770 
771 	case NETDEV_CHANGEMTU:
772 		list_for_each_entry(ipvlan, &port->ipvlans, pnode)
773 			ipvlan_adjust_mtu(ipvlan, dev);
774 		break;
775 
776 	case NETDEV_PRE_CHANGEADDR:
777 		prechaddr_info = ptr;
778 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
779 			err = dev_pre_changeaddr_notify(ipvlan->dev,
780 						    prechaddr_info->dev_addr,
781 						    extack);
782 			if (err)
783 				return notifier_from_errno(err);
784 		}
785 		break;
786 
787 	case NETDEV_CHANGEADDR:
788 		list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
789 			eth_hw_addr_set(ipvlan->dev, dev->dev_addr);
790 			call_netdevice_notifiers(NETDEV_CHANGEADDR, ipvlan->dev);
791 		}
792 		break;
793 
794 	case NETDEV_PRE_TYPE_CHANGE:
795 		/* Forbid underlying device to change its type. */
796 		return NOTIFY_BAD;
797 	}
798 	return NOTIFY_DONE;
799 }
800 
801 /* the caller must held the addrs lock */
802 static int ipvlan_add_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
803 {
804 	struct ipvl_addr *addr;
805 
806 	addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
807 	if (!addr)
808 		return -ENOMEM;
809 
810 	addr->master = ipvlan;
811 	if (!is_v6) {
812 		memcpy(&addr->ip4addr, iaddr, sizeof(struct in_addr));
813 		addr->atype = IPVL_IPV4;
814 #if IS_ENABLED(CONFIG_IPV6)
815 	} else {
816 		memcpy(&addr->ip6addr, iaddr, sizeof(struct in6_addr));
817 		addr->atype = IPVL_IPV6;
818 #endif
819 	}
820 
821 	list_add_tail_rcu(&addr->anode, &ipvlan->addrs);
822 
823 	/* If the interface is not up, the address will be added to the hash
824 	 * list by ipvlan_open.
825 	 */
826 	if (netif_running(ipvlan->dev))
827 		ipvlan_ht_addr_add(ipvlan, addr);
828 
829 	return 0;
830 }
831 
832 static void ipvlan_del_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
833 {
834 	struct ipvl_addr *addr;
835 
836 	spin_lock_bh(&ipvlan->addrs_lock);
837 	addr = ipvlan_find_addr(ipvlan, iaddr, is_v6);
838 	if (!addr) {
839 		spin_unlock_bh(&ipvlan->addrs_lock);
840 		return;
841 	}
842 
843 	ipvlan_ht_addr_del(addr);
844 	list_del_rcu(&addr->anode);
845 	spin_unlock_bh(&ipvlan->addrs_lock);
846 	kfree_rcu(addr, rcu);
847 }
848 
849 static bool ipvlan_is_valid_dev(const struct net_device *dev)
850 {
851 	struct ipvl_dev *ipvlan = netdev_priv(dev);
852 
853 	if (!netif_is_ipvlan(dev))
854 		return false;
855 
856 	if (!ipvlan || !ipvlan->port)
857 		return false;
858 
859 	return true;
860 }
861 
862 #if IS_ENABLED(CONFIG_IPV6)
863 static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
864 {
865 	int ret = -EINVAL;
866 
867 	spin_lock_bh(&ipvlan->addrs_lock);
868 	if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true))
869 		netif_err(ipvlan, ifup, ipvlan->dev,
870 			  "Failed to add IPv6=%pI6c addr for %s intf\n",
871 			  ip6_addr, ipvlan->dev->name);
872 	else
873 		ret = ipvlan_add_addr(ipvlan, ip6_addr, true);
874 	spin_unlock_bh(&ipvlan->addrs_lock);
875 	return ret;
876 }
877 
878 static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
879 {
880 	return ipvlan_del_addr(ipvlan, ip6_addr, true);
881 }
882 
883 static int ipvlan_addr6_event(struct notifier_block *unused,
884 			      unsigned long event, void *ptr)
885 {
886 	struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
887 	struct net_device *dev = (struct net_device *)if6->idev->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_add_addr6(ipvlan, &if6->addr))
896 			return NOTIFY_BAD;
897 		break;
898 
899 	case NETDEV_DOWN:
900 		ipvlan_del_addr6(ipvlan, &if6->addr);
901 		break;
902 	}
903 
904 	return NOTIFY_OK;
905 }
906 
907 static int ipvlan_addr6_validator_event(struct notifier_block *unused,
908 					unsigned long event, void *ptr)
909 {
910 	struct in6_validator_info *i6vi = (struct in6_validator_info *)ptr;
911 	struct net_device *dev = (struct net_device *)i6vi->i6vi_dev->dev;
912 	struct ipvl_dev *ipvlan = netdev_priv(dev);
913 
914 	if (!ipvlan_is_valid_dev(dev))
915 		return NOTIFY_DONE;
916 
917 	switch (event) {
918 	case NETDEV_UP:
919 		if (ipvlan_addr_busy(ipvlan->port, &i6vi->i6vi_addr, true)) {
920 			NL_SET_ERR_MSG(i6vi->extack,
921 				       "Address already assigned to an ipvlan device");
922 			return notifier_from_errno(-EADDRINUSE);
923 		}
924 		break;
925 	}
926 
927 	return NOTIFY_OK;
928 }
929 #endif
930 
931 static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
932 {
933 	int ret = -EINVAL;
934 
935 	spin_lock_bh(&ipvlan->addrs_lock);
936 	if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false))
937 		netif_err(ipvlan, ifup, ipvlan->dev,
938 			  "Failed to add IPv4=%pI4 on %s intf.\n",
939 			  ip4_addr, ipvlan->dev->name);
940 	else
941 		ret = ipvlan_add_addr(ipvlan, ip4_addr, false);
942 	spin_unlock_bh(&ipvlan->addrs_lock);
943 	return ret;
944 }
945 
946 static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
947 {
948 	return ipvlan_del_addr(ipvlan, ip4_addr, false);
949 }
950 
951 static int ipvlan_addr4_event(struct notifier_block *unused,
952 			      unsigned long event, void *ptr)
953 {
954 	struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
955 	struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
956 	struct ipvl_dev *ipvlan = netdev_priv(dev);
957 	struct in_addr ip4_addr;
958 
959 	if (!ipvlan_is_valid_dev(dev))
960 		return NOTIFY_DONE;
961 
962 	switch (event) {
963 	case NETDEV_UP:
964 		ip4_addr.s_addr = if4->ifa_address;
965 		if (ipvlan_add_addr4(ipvlan, &ip4_addr))
966 			return NOTIFY_BAD;
967 		break;
968 
969 	case NETDEV_DOWN:
970 		ip4_addr.s_addr = if4->ifa_address;
971 		ipvlan_del_addr4(ipvlan, &ip4_addr);
972 		break;
973 	}
974 
975 	return NOTIFY_OK;
976 }
977 
978 static int ipvlan_addr4_validator_event(struct notifier_block *unused,
979 					unsigned long event, void *ptr)
980 {
981 	struct in_validator_info *ivi = (struct in_validator_info *)ptr;
982 	struct net_device *dev = (struct net_device *)ivi->ivi_dev->dev;
983 	struct ipvl_dev *ipvlan = netdev_priv(dev);
984 
985 	if (!ipvlan_is_valid_dev(dev))
986 		return NOTIFY_DONE;
987 
988 	switch (event) {
989 	case NETDEV_UP:
990 		if (ipvlan_addr_busy(ipvlan->port, &ivi->ivi_addr, false)) {
991 			NL_SET_ERR_MSG(ivi->extack,
992 				       "Address already assigned to an ipvlan device");
993 			return notifier_from_errno(-EADDRINUSE);
994 		}
995 		break;
996 	}
997 
998 	return NOTIFY_OK;
999 }
1000 
1001 static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
1002 	.notifier_call = ipvlan_addr4_event,
1003 };
1004 
1005 static struct notifier_block ipvlan_addr4_vtor_notifier_block __read_mostly = {
1006 	.notifier_call = ipvlan_addr4_validator_event,
1007 };
1008 
1009 static struct notifier_block ipvlan_notifier_block __read_mostly = {
1010 	.notifier_call = ipvlan_device_event,
1011 };
1012 
1013 #if IS_ENABLED(CONFIG_IPV6)
1014 static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
1015 	.notifier_call = ipvlan_addr6_event,
1016 };
1017 
1018 static struct notifier_block ipvlan_addr6_vtor_notifier_block __read_mostly = {
1019 	.notifier_call = ipvlan_addr6_validator_event,
1020 };
1021 #endif
1022 
1023 static int __init ipvlan_init_module(void)
1024 {
1025 	int err;
1026 
1027 	ipvlan_init_secret();
1028 	register_netdevice_notifier(&ipvlan_notifier_block);
1029 #if IS_ENABLED(CONFIG_IPV6)
1030 	register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1031 	register_inet6addr_validator_notifier(
1032 	    &ipvlan_addr6_vtor_notifier_block);
1033 #endif
1034 	register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1035 	register_inetaddr_validator_notifier(&ipvlan_addr4_vtor_notifier_block);
1036 
1037 	err = ipvlan_l3s_init();
1038 	if (err < 0)
1039 		goto error;
1040 
1041 	err = ipvlan_link_register(&ipvlan_link_ops);
1042 	if (err < 0) {
1043 		ipvlan_l3s_cleanup();
1044 		goto error;
1045 	}
1046 
1047 	return 0;
1048 error:
1049 	unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1050 	unregister_inetaddr_validator_notifier(
1051 	    &ipvlan_addr4_vtor_notifier_block);
1052 #if IS_ENABLED(CONFIG_IPV6)
1053 	unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1054 	unregister_inet6addr_validator_notifier(
1055 	    &ipvlan_addr6_vtor_notifier_block);
1056 #endif
1057 	unregister_netdevice_notifier(&ipvlan_notifier_block);
1058 	return err;
1059 }
1060 
1061 static void __exit ipvlan_cleanup_module(void)
1062 {
1063 	rtnl_link_unregister(&ipvlan_link_ops);
1064 	ipvlan_l3s_cleanup();
1065 	unregister_netdevice_notifier(&ipvlan_notifier_block);
1066 	unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1067 	unregister_inetaddr_validator_notifier(
1068 	    &ipvlan_addr4_vtor_notifier_block);
1069 #if IS_ENABLED(CONFIG_IPV6)
1070 	unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1071 	unregister_inet6addr_validator_notifier(
1072 	    &ipvlan_addr6_vtor_notifier_block);
1073 #endif
1074 }
1075 
1076 module_init(ipvlan_init_module);
1077 module_exit(ipvlan_cleanup_module);
1078 
1079 MODULE_LICENSE("GPL");
1080 MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
1081 MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
1082 MODULE_ALIAS_RTNL_LINK("ipvlan");
1083