xref: /linux/drivers/net/macvlan.c (revision 00a6d7b6762c27d441e9ac8faff36384bc0fc180)
1 /*
2  * Copyright (c) 2007 Patrick McHardy <kaber@trash.net>
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public License as
6  * published by the Free Software Foundation; either version 2 of
7  * the License, or (at your option) any later version.
8  *
9  * The code this is based on carried the following copyright notice:
10  * ---
11  * (C) Copyright 2001-2006
12  * Alex Zeffertt, Cambridge Broadband Ltd, ajz@cambridgebroadband.com
13  * Re-worked by Ben Greear <greearb@candelatech.com>
14  * ---
15  */
16 #include <linux/kernel.h>
17 #include <linux/types.h>
18 #include <linux/module.h>
19 #include <linux/init.h>
20 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/string.h>
23 #include <linux/rculist.h>
24 #include <linux/notifier.h>
25 #include <linux/netdevice.h>
26 #include <linux/etherdevice.h>
27 #include <linux/ethtool.h>
28 #include <linux/if_arp.h>
29 #include <linux/if_vlan.h>
30 #include <linux/if_link.h>
31 #include <linux/if_macvlan.h>
32 #include <linux/hash.h>
33 #include <net/rtnetlink.h>
34 #include <net/xfrm.h>
35 
36 #define MACVLAN_HASH_SIZE	(1 << BITS_PER_BYTE)
37 
38 struct macvlan_port {
39 	struct net_device	*dev;
40 	struct hlist_head	vlan_hash[MACVLAN_HASH_SIZE];
41 	struct list_head	vlans;
42 	struct rcu_head		rcu;
43 	bool 			passthru;
44 	int			count;
45 };
46 
47 static void macvlan_port_destroy(struct net_device *dev);
48 
49 static struct macvlan_port *macvlan_port_get_rcu(const struct net_device *dev)
50 {
51 	return rcu_dereference(dev->rx_handler_data);
52 }
53 
54 static struct macvlan_port *macvlan_port_get_rtnl(const struct net_device *dev)
55 {
56 	return rtnl_dereference(dev->rx_handler_data);
57 }
58 
59 #define macvlan_port_exists(dev) (dev->priv_flags & IFF_MACVLAN_PORT)
60 
61 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
62 					       const unsigned char *addr)
63 {
64 	struct macvlan_dev *vlan;
65 
66 	hlist_for_each_entry_rcu(vlan, &port->vlan_hash[addr[5]], hlist) {
67 		if (ether_addr_equal_64bits(vlan->dev->dev_addr, addr))
68 			return vlan;
69 	}
70 	return NULL;
71 }
72 
73 static void macvlan_hash_add(struct macvlan_dev *vlan)
74 {
75 	struct macvlan_port *port = vlan->port;
76 	const unsigned char *addr = vlan->dev->dev_addr;
77 
78 	hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[addr[5]]);
79 }
80 
81 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync)
82 {
83 	hlist_del_rcu(&vlan->hlist);
84 	if (sync)
85 		synchronize_rcu();
86 }
87 
88 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
89 					const unsigned char *addr)
90 {
91 	macvlan_hash_del(vlan, true);
92 	/* Now that we are unhashed it is safe to change the device
93 	 * address without confusing packet delivery.
94 	 */
95 	memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
96 	macvlan_hash_add(vlan);
97 }
98 
99 static int macvlan_addr_busy(const struct macvlan_port *port,
100 				const unsigned char *addr)
101 {
102 	/* Test to see if the specified multicast address is
103 	 * currently in use by the underlying device or
104 	 * another macvlan.
105 	 */
106 	if (ether_addr_equal_64bits(port->dev->dev_addr, addr))
107 		return 1;
108 
109 	if (macvlan_hash_lookup(port, addr))
110 		return 1;
111 
112 	return 0;
113 }
114 
115 
116 static int macvlan_broadcast_one(struct sk_buff *skb,
117 				 const struct macvlan_dev *vlan,
118 				 const struct ethhdr *eth, bool local)
119 {
120 	struct net_device *dev = vlan->dev;
121 
122 	if (local)
123 		return dev_forward_skb(dev, skb);
124 
125 	skb->dev = dev;
126 	if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
127 		skb->pkt_type = PACKET_BROADCAST;
128 	else
129 		skb->pkt_type = PACKET_MULTICAST;
130 
131 	return netif_rx(skb);
132 }
133 
134 static u32 macvlan_hash_mix(const struct macvlan_dev *vlan)
135 {
136 	return (u32)(((unsigned long)vlan) >> L1_CACHE_SHIFT);
137 }
138 
139 
140 static unsigned int mc_hash(const struct macvlan_dev *vlan,
141 			    const unsigned char *addr)
142 {
143 	u32 val = __get_unaligned_cpu32(addr + 2);
144 
145 	val ^= macvlan_hash_mix(vlan);
146 	return hash_32(val, MACVLAN_MC_FILTER_BITS);
147 }
148 
149 static void macvlan_broadcast(struct sk_buff *skb,
150 			      const struct macvlan_port *port,
151 			      struct net_device *src,
152 			      enum macvlan_mode mode)
153 {
154 	const struct ethhdr *eth = eth_hdr(skb);
155 	const struct macvlan_dev *vlan;
156 	struct sk_buff *nskb;
157 	unsigned int i;
158 	int err;
159 	unsigned int hash;
160 
161 	if (skb->protocol == htons(ETH_P_PAUSE))
162 		return;
163 
164 	for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
165 		hlist_for_each_entry_rcu(vlan, &port->vlan_hash[i], hlist) {
166 			if (vlan->dev == src || !(vlan->mode & mode))
167 				continue;
168 
169 			hash = mc_hash(vlan, eth->h_dest);
170 			if (!test_bit(hash, vlan->mc_filter))
171 				continue;
172 
173 			err = NET_RX_DROP;
174 			nskb = skb_clone(skb, GFP_ATOMIC);
175 			if (likely(nskb))
176 				err = macvlan_broadcast_one(
177 					nskb, vlan, eth,
178 					mode == MACVLAN_MODE_BRIDGE);
179 			macvlan_count_rx(vlan, skb->len + ETH_HLEN,
180 					 err == NET_RX_SUCCESS, 1);
181 		}
182 	}
183 }
184 
185 /* called under rcu_read_lock() from netif_receive_skb */
186 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
187 {
188 	struct macvlan_port *port;
189 	struct sk_buff *skb = *pskb;
190 	const struct ethhdr *eth = eth_hdr(skb);
191 	const struct macvlan_dev *vlan;
192 	const struct macvlan_dev *src;
193 	struct net_device *dev;
194 	unsigned int len = 0;
195 	int ret = NET_RX_DROP;
196 
197 	port = macvlan_port_get_rcu(skb->dev);
198 	if (is_multicast_ether_addr(eth->h_dest)) {
199 		skb = ip_check_defrag(skb, IP_DEFRAG_MACVLAN);
200 		if (!skb)
201 			return RX_HANDLER_CONSUMED;
202 		eth = eth_hdr(skb);
203 		src = macvlan_hash_lookup(port, eth->h_source);
204 		if (!src)
205 			/* frame comes from an external address */
206 			macvlan_broadcast(skb, port, NULL,
207 					  MACVLAN_MODE_PRIVATE |
208 					  MACVLAN_MODE_VEPA    |
209 					  MACVLAN_MODE_PASSTHRU|
210 					  MACVLAN_MODE_BRIDGE);
211 		else if (src->mode == MACVLAN_MODE_VEPA)
212 			/* flood to everyone except source */
213 			macvlan_broadcast(skb, port, src->dev,
214 					  MACVLAN_MODE_VEPA |
215 					  MACVLAN_MODE_BRIDGE);
216 		else if (src->mode == MACVLAN_MODE_BRIDGE)
217 			/*
218 			 * flood only to VEPA ports, bridge ports
219 			 * already saw the frame on the way out.
220 			 */
221 			macvlan_broadcast(skb, port, src->dev,
222 					  MACVLAN_MODE_VEPA);
223 		else {
224 			/* forward to original port. */
225 			vlan = src;
226 			ret = macvlan_broadcast_one(skb, vlan, eth, 0);
227 			goto out;
228 		}
229 
230 		return RX_HANDLER_PASS;
231 	}
232 
233 	if (port->passthru)
234 		vlan = list_first_or_null_rcu(&port->vlans,
235 					      struct macvlan_dev, list);
236 	else
237 		vlan = macvlan_hash_lookup(port, eth->h_dest);
238 	if (vlan == NULL)
239 		return RX_HANDLER_PASS;
240 
241 	dev = vlan->dev;
242 	if (unlikely(!(dev->flags & IFF_UP))) {
243 		kfree_skb(skb);
244 		return RX_HANDLER_CONSUMED;
245 	}
246 	len = skb->len + ETH_HLEN;
247 	skb = skb_share_check(skb, GFP_ATOMIC);
248 	if (!skb)
249 		goto out;
250 
251 	skb->dev = dev;
252 	skb->pkt_type = PACKET_HOST;
253 
254 	ret = netif_rx(skb);
255 
256 out:
257 	macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, 0);
258 	return RX_HANDLER_CONSUMED;
259 }
260 
261 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
262 {
263 	const struct macvlan_dev *vlan = netdev_priv(dev);
264 	const struct macvlan_port *port = vlan->port;
265 	const struct macvlan_dev *dest;
266 
267 	if (vlan->mode == MACVLAN_MODE_BRIDGE) {
268 		const struct ethhdr *eth = (void *)skb->data;
269 
270 		/* send to other bridge ports directly */
271 		if (is_multicast_ether_addr(eth->h_dest)) {
272 			macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
273 			goto xmit_world;
274 		}
275 
276 		dest = macvlan_hash_lookup(port, eth->h_dest);
277 		if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
278 			/* send to lowerdev first for its network taps */
279 			dev_forward_skb(vlan->lowerdev, skb);
280 
281 			return NET_XMIT_SUCCESS;
282 		}
283 	}
284 
285 xmit_world:
286 	skb->dev = vlan->lowerdev;
287 	return dev_queue_xmit(skb);
288 }
289 
290 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
291 				      struct net_device *dev)
292 {
293 	unsigned int len = skb->len;
294 	int ret;
295 	const struct macvlan_dev *vlan = netdev_priv(dev);
296 
297 	if (vlan->fwd_priv) {
298 		skb->dev = vlan->lowerdev;
299 		ret = dev_queue_xmit_accel(skb, vlan->fwd_priv);
300 	} else {
301 		ret = macvlan_queue_xmit(skb, dev);
302 	}
303 
304 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
305 		struct vlan_pcpu_stats *pcpu_stats;
306 
307 		pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
308 		u64_stats_update_begin(&pcpu_stats->syncp);
309 		pcpu_stats->tx_packets++;
310 		pcpu_stats->tx_bytes += len;
311 		u64_stats_update_end(&pcpu_stats->syncp);
312 	} else {
313 		this_cpu_inc(vlan->pcpu_stats->tx_dropped);
314 	}
315 	return ret;
316 }
317 
318 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
319 			       unsigned short type, const void *daddr,
320 			       const void *saddr, unsigned len)
321 {
322 	const struct macvlan_dev *vlan = netdev_priv(dev);
323 	struct net_device *lowerdev = vlan->lowerdev;
324 
325 	return dev_hard_header(skb, lowerdev, type, daddr,
326 			       saddr ? : dev->dev_addr, len);
327 }
328 
329 static const struct header_ops macvlan_hard_header_ops = {
330 	.create  	= macvlan_hard_header,
331 	.rebuild	= eth_rebuild_header,
332 	.parse		= eth_header_parse,
333 	.cache		= eth_header_cache,
334 	.cache_update	= eth_header_cache_update,
335 };
336 
337 static struct rtnl_link_ops macvlan_link_ops;
338 
339 static int macvlan_open(struct net_device *dev)
340 {
341 	struct macvlan_dev *vlan = netdev_priv(dev);
342 	struct net_device *lowerdev = vlan->lowerdev;
343 	int err;
344 
345 	if (vlan->port->passthru) {
346 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
347 			err = dev_set_promiscuity(lowerdev, 1);
348 			if (err < 0)
349 				goto out;
350 		}
351 		goto hash_add;
352 	}
353 
354 	if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD &&
355 	    dev->rtnl_link_ops == &macvlan_link_ops) {
356 		vlan->fwd_priv =
357 		      lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
358 
359 		/* If we get a NULL pointer back, or if we get an error
360 		 * then we should just fall through to the non accelerated path
361 		 */
362 		if (IS_ERR_OR_NULL(vlan->fwd_priv)) {
363 			vlan->fwd_priv = NULL;
364 		} else
365 			return 0;
366 	}
367 
368 	err = -EBUSY;
369 	if (macvlan_addr_busy(vlan->port, dev->dev_addr))
370 		goto out;
371 
372 	err = dev_uc_add(lowerdev, dev->dev_addr);
373 	if (err < 0)
374 		goto out;
375 	if (dev->flags & IFF_ALLMULTI) {
376 		err = dev_set_allmulti(lowerdev, 1);
377 		if (err < 0)
378 			goto del_unicast;
379 	}
380 
381 hash_add:
382 	macvlan_hash_add(vlan);
383 	return 0;
384 
385 del_unicast:
386 	dev_uc_del(lowerdev, dev->dev_addr);
387 out:
388 	if (vlan->fwd_priv) {
389 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
390 							   vlan->fwd_priv);
391 		vlan->fwd_priv = NULL;
392 	}
393 	return err;
394 }
395 
396 static int macvlan_stop(struct net_device *dev)
397 {
398 	struct macvlan_dev *vlan = netdev_priv(dev);
399 	struct net_device *lowerdev = vlan->lowerdev;
400 
401 	if (vlan->fwd_priv) {
402 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
403 							   vlan->fwd_priv);
404 		vlan->fwd_priv = NULL;
405 		return 0;
406 	}
407 
408 	dev_uc_unsync(lowerdev, dev);
409 	dev_mc_unsync(lowerdev, dev);
410 
411 	if (vlan->port->passthru) {
412 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
413 			dev_set_promiscuity(lowerdev, -1);
414 		goto hash_del;
415 	}
416 
417 	if (dev->flags & IFF_ALLMULTI)
418 		dev_set_allmulti(lowerdev, -1);
419 
420 	dev_uc_del(lowerdev, dev->dev_addr);
421 
422 hash_del:
423 	macvlan_hash_del(vlan, !dev->dismantle);
424 	return 0;
425 }
426 
427 static int macvlan_set_mac_address(struct net_device *dev, void *p)
428 {
429 	struct macvlan_dev *vlan = netdev_priv(dev);
430 	struct net_device *lowerdev = vlan->lowerdev;
431 	struct sockaddr *addr = p;
432 	int err;
433 
434 	if (!is_valid_ether_addr(addr->sa_data))
435 		return -EADDRNOTAVAIL;
436 
437 	if (!(dev->flags & IFF_UP)) {
438 		/* Just copy in the new address */
439 		memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
440 	} else {
441 		/* Rehash and update the device filters */
442 		if (macvlan_addr_busy(vlan->port, addr->sa_data))
443 			return -EBUSY;
444 
445 		err = dev_uc_add(lowerdev, addr->sa_data);
446 		if (err)
447 			return err;
448 
449 		dev_uc_del(lowerdev, dev->dev_addr);
450 
451 		macvlan_hash_change_addr(vlan, addr->sa_data);
452 	}
453 	return 0;
454 }
455 
456 static void macvlan_change_rx_flags(struct net_device *dev, int change)
457 {
458 	struct macvlan_dev *vlan = netdev_priv(dev);
459 	struct net_device *lowerdev = vlan->lowerdev;
460 
461 	if (change & IFF_ALLMULTI)
462 		dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
463 }
464 
465 static void macvlan_set_mac_lists(struct net_device *dev)
466 {
467 	struct macvlan_dev *vlan = netdev_priv(dev);
468 
469 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
470 		bitmap_fill(vlan->mc_filter, MACVLAN_MC_FILTER_SZ);
471 	} else {
472 		struct netdev_hw_addr *ha;
473 		DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
474 
475 		bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
476 		netdev_for_each_mc_addr(ha, dev) {
477 			__set_bit(mc_hash(vlan, ha->addr), filter);
478 		}
479 
480 		__set_bit(mc_hash(vlan, dev->broadcast), filter);
481 
482 		bitmap_copy(vlan->mc_filter, filter, MACVLAN_MC_FILTER_SZ);
483 	}
484 	dev_uc_sync(vlan->lowerdev, dev);
485 	dev_mc_sync(vlan->lowerdev, dev);
486 }
487 
488 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
489 {
490 	struct macvlan_dev *vlan = netdev_priv(dev);
491 
492 	if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
493 		return -EINVAL;
494 	dev->mtu = new_mtu;
495 	return 0;
496 }
497 
498 /*
499  * macvlan network devices have devices nesting below it and are a special
500  * "super class" of normal network devices; split their locks off into a
501  * separate class since they always nest.
502  */
503 static struct lock_class_key macvlan_netdev_xmit_lock_key;
504 static struct lock_class_key macvlan_netdev_addr_lock_key;
505 
506 #define ALWAYS_ON_FEATURES \
507 	(NETIF_F_SG | NETIF_F_GEN_CSUM | NETIF_F_GSO_SOFTWARE | NETIF_F_LLTX)
508 
509 #define MACVLAN_FEATURES \
510 	(NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
511 	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
512 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
513 	 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
514 
515 #define MACVLAN_STATE_MASK \
516 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
517 
518 static void macvlan_set_lockdep_class_one(struct net_device *dev,
519 					  struct netdev_queue *txq,
520 					  void *_unused)
521 {
522 	lockdep_set_class(&txq->_xmit_lock,
523 			  &macvlan_netdev_xmit_lock_key);
524 }
525 
526 static void macvlan_set_lockdep_class(struct net_device *dev)
527 {
528 	lockdep_set_class(&dev->addr_list_lock,
529 			  &macvlan_netdev_addr_lock_key);
530 	netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
531 }
532 
533 static int macvlan_init(struct net_device *dev)
534 {
535 	struct macvlan_dev *vlan = netdev_priv(dev);
536 	const struct net_device *lowerdev = vlan->lowerdev;
537 
538 	dev->state		= (dev->state & ~MACVLAN_STATE_MASK) |
539 				  (lowerdev->state & MACVLAN_STATE_MASK);
540 	dev->features 		= lowerdev->features & MACVLAN_FEATURES;
541 	dev->features		|= ALWAYS_ON_FEATURES;
542 	dev->gso_max_size	= lowerdev->gso_max_size;
543 	dev->iflink		= lowerdev->ifindex;
544 	dev->hard_header_len	= lowerdev->hard_header_len;
545 
546 	macvlan_set_lockdep_class(dev);
547 
548 	vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
549 	if (!vlan->pcpu_stats)
550 		return -ENOMEM;
551 
552 	return 0;
553 }
554 
555 static void macvlan_uninit(struct net_device *dev)
556 {
557 	struct macvlan_dev *vlan = netdev_priv(dev);
558 	struct macvlan_port *port = vlan->port;
559 
560 	free_percpu(vlan->pcpu_stats);
561 
562 	port->count -= 1;
563 	if (!port->count)
564 		macvlan_port_destroy(port->dev);
565 }
566 
567 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
568 							 struct rtnl_link_stats64 *stats)
569 {
570 	struct macvlan_dev *vlan = netdev_priv(dev);
571 
572 	if (vlan->pcpu_stats) {
573 		struct vlan_pcpu_stats *p;
574 		u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
575 		u32 rx_errors = 0, tx_dropped = 0;
576 		unsigned int start;
577 		int i;
578 
579 		for_each_possible_cpu(i) {
580 			p = per_cpu_ptr(vlan->pcpu_stats, i);
581 			do {
582 				start = u64_stats_fetch_begin_irq(&p->syncp);
583 				rx_packets	= p->rx_packets;
584 				rx_bytes	= p->rx_bytes;
585 				rx_multicast	= p->rx_multicast;
586 				tx_packets	= p->tx_packets;
587 				tx_bytes	= p->tx_bytes;
588 			} while (u64_stats_fetch_retry_irq(&p->syncp, start));
589 
590 			stats->rx_packets	+= rx_packets;
591 			stats->rx_bytes		+= rx_bytes;
592 			stats->multicast	+= rx_multicast;
593 			stats->tx_packets	+= tx_packets;
594 			stats->tx_bytes		+= tx_bytes;
595 			/* rx_errors & tx_dropped are u32, updated
596 			 * without syncp protection.
597 			 */
598 			rx_errors	+= p->rx_errors;
599 			tx_dropped	+= p->tx_dropped;
600 		}
601 		stats->rx_errors	= rx_errors;
602 		stats->rx_dropped	= rx_errors;
603 		stats->tx_dropped	= tx_dropped;
604 	}
605 	return stats;
606 }
607 
608 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
609 				   __be16 proto, u16 vid)
610 {
611 	struct macvlan_dev *vlan = netdev_priv(dev);
612 	struct net_device *lowerdev = vlan->lowerdev;
613 
614 	return vlan_vid_add(lowerdev, proto, vid);
615 }
616 
617 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
618 				    __be16 proto, u16 vid)
619 {
620 	struct macvlan_dev *vlan = netdev_priv(dev);
621 	struct net_device *lowerdev = vlan->lowerdev;
622 
623 	vlan_vid_del(lowerdev, proto, vid);
624 	return 0;
625 }
626 
627 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
628 			   struct net_device *dev,
629 			   const unsigned char *addr,
630 			   u16 flags)
631 {
632 	struct macvlan_dev *vlan = netdev_priv(dev);
633 	int err = -EINVAL;
634 
635 	if (!vlan->port->passthru)
636 		return -EOPNOTSUPP;
637 
638 	if (flags & NLM_F_REPLACE)
639 		return -EOPNOTSUPP;
640 
641 	if (is_unicast_ether_addr(addr))
642 		err = dev_uc_add_excl(dev, addr);
643 	else if (is_multicast_ether_addr(addr))
644 		err = dev_mc_add_excl(dev, addr);
645 
646 	return err;
647 }
648 
649 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
650 			   struct net_device *dev,
651 			   const unsigned char *addr)
652 {
653 	struct macvlan_dev *vlan = netdev_priv(dev);
654 	int err = -EINVAL;
655 
656 	if (!vlan->port->passthru)
657 		return -EOPNOTSUPP;
658 
659 	if (is_unicast_ether_addr(addr))
660 		err = dev_uc_del(dev, addr);
661 	else if (is_multicast_ether_addr(addr))
662 		err = dev_mc_del(dev, addr);
663 
664 	return err;
665 }
666 
667 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
668 					struct ethtool_drvinfo *drvinfo)
669 {
670 	strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
671 	strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
672 }
673 
674 static int macvlan_ethtool_get_settings(struct net_device *dev,
675 					struct ethtool_cmd *cmd)
676 {
677 	const struct macvlan_dev *vlan = netdev_priv(dev);
678 
679 	return __ethtool_get_settings(vlan->lowerdev, cmd);
680 }
681 
682 static netdev_features_t macvlan_fix_features(struct net_device *dev,
683 					      netdev_features_t features)
684 {
685 	struct macvlan_dev *vlan = netdev_priv(dev);
686 	netdev_features_t mask;
687 
688 	features |= NETIF_F_ALL_FOR_ALL;
689 	features &= (vlan->set_features | ~MACVLAN_FEATURES);
690 	mask = features;
691 
692 	features = netdev_increment_features(vlan->lowerdev->features,
693 					     features,
694 					     mask);
695 	features |= ALWAYS_ON_FEATURES;
696 
697 	return features;
698 }
699 
700 static const struct ethtool_ops macvlan_ethtool_ops = {
701 	.get_link		= ethtool_op_get_link,
702 	.get_settings		= macvlan_ethtool_get_settings,
703 	.get_drvinfo		= macvlan_ethtool_get_drvinfo,
704 };
705 
706 static const struct net_device_ops macvlan_netdev_ops = {
707 	.ndo_init		= macvlan_init,
708 	.ndo_uninit		= macvlan_uninit,
709 	.ndo_open		= macvlan_open,
710 	.ndo_stop		= macvlan_stop,
711 	.ndo_start_xmit		= macvlan_start_xmit,
712 	.ndo_change_mtu		= macvlan_change_mtu,
713 	.ndo_fix_features	= macvlan_fix_features,
714 	.ndo_change_rx_flags	= macvlan_change_rx_flags,
715 	.ndo_set_mac_address	= macvlan_set_mac_address,
716 	.ndo_set_rx_mode	= macvlan_set_mac_lists,
717 	.ndo_get_stats64	= macvlan_dev_get_stats64,
718 	.ndo_validate_addr	= eth_validate_addr,
719 	.ndo_vlan_rx_add_vid	= macvlan_vlan_rx_add_vid,
720 	.ndo_vlan_rx_kill_vid	= macvlan_vlan_rx_kill_vid,
721 	.ndo_fdb_add		= macvlan_fdb_add,
722 	.ndo_fdb_del		= macvlan_fdb_del,
723 	.ndo_fdb_dump		= ndo_dflt_fdb_dump,
724 };
725 
726 void macvlan_common_setup(struct net_device *dev)
727 {
728 	ether_setup(dev);
729 
730 	dev->priv_flags	       &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
731 	dev->priv_flags	       |= IFF_UNICAST_FLT;
732 	dev->netdev_ops		= &macvlan_netdev_ops;
733 	dev->destructor		= free_netdev;
734 	dev->header_ops		= &macvlan_hard_header_ops;
735 	dev->ethtool_ops	= &macvlan_ethtool_ops;
736 }
737 EXPORT_SYMBOL_GPL(macvlan_common_setup);
738 
739 static void macvlan_setup(struct net_device *dev)
740 {
741 	macvlan_common_setup(dev);
742 	dev->tx_queue_len	= 0;
743 }
744 
745 static int macvlan_port_create(struct net_device *dev)
746 {
747 	struct macvlan_port *port;
748 	unsigned int i;
749 	int err;
750 
751 	if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
752 		return -EINVAL;
753 
754 	port = kzalloc(sizeof(*port), GFP_KERNEL);
755 	if (port == NULL)
756 		return -ENOMEM;
757 
758 	port->passthru = false;
759 	port->dev = dev;
760 	INIT_LIST_HEAD(&port->vlans);
761 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
762 		INIT_HLIST_HEAD(&port->vlan_hash[i]);
763 
764 	err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
765 	if (err)
766 		kfree(port);
767 	else
768 		dev->priv_flags |= IFF_MACVLAN_PORT;
769 	return err;
770 }
771 
772 static void macvlan_port_destroy(struct net_device *dev)
773 {
774 	struct macvlan_port *port = macvlan_port_get_rtnl(dev);
775 
776 	dev->priv_flags &= ~IFF_MACVLAN_PORT;
777 	netdev_rx_handler_unregister(dev);
778 	kfree_rcu(port, rcu);
779 }
780 
781 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
782 {
783 	if (tb[IFLA_ADDRESS]) {
784 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
785 			return -EINVAL;
786 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
787 			return -EADDRNOTAVAIL;
788 	}
789 
790 	if (data && data[IFLA_MACVLAN_FLAGS] &&
791 	    nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
792 		return -EINVAL;
793 
794 	if (data && data[IFLA_MACVLAN_MODE]) {
795 		switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
796 		case MACVLAN_MODE_PRIVATE:
797 		case MACVLAN_MODE_VEPA:
798 		case MACVLAN_MODE_BRIDGE:
799 		case MACVLAN_MODE_PASSTHRU:
800 			break;
801 		default:
802 			return -EINVAL;
803 		}
804 	}
805 	return 0;
806 }
807 
808 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
809 			   struct nlattr *tb[], struct nlattr *data[])
810 {
811 	struct macvlan_dev *vlan = netdev_priv(dev);
812 	struct macvlan_port *port;
813 	struct net_device *lowerdev;
814 	int err;
815 
816 	if (!tb[IFLA_LINK])
817 		return -EINVAL;
818 
819 	lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
820 	if (lowerdev == NULL)
821 		return -ENODEV;
822 
823 	/* When creating macvlans or macvtaps on top of other macvlans - use
824 	 * the real device as the lowerdev.
825 	 */
826 	if (netif_is_macvlan(lowerdev))
827 		lowerdev = macvlan_dev_real_dev(lowerdev);
828 
829 	if (!tb[IFLA_MTU])
830 		dev->mtu = lowerdev->mtu;
831 	else if (dev->mtu > lowerdev->mtu)
832 		return -EINVAL;
833 
834 	if (!tb[IFLA_ADDRESS])
835 		eth_hw_addr_random(dev);
836 
837 	if (!macvlan_port_exists(lowerdev)) {
838 		err = macvlan_port_create(lowerdev);
839 		if (err < 0)
840 			return err;
841 	}
842 	port = macvlan_port_get_rtnl(lowerdev);
843 
844 	/* Only 1 macvlan device can be created in passthru mode */
845 	if (port->passthru)
846 		return -EINVAL;
847 
848 	vlan->lowerdev = lowerdev;
849 	vlan->dev      = dev;
850 	vlan->port     = port;
851 	vlan->set_features = MACVLAN_FEATURES;
852 
853 	vlan->mode     = MACVLAN_MODE_VEPA;
854 	if (data && data[IFLA_MACVLAN_MODE])
855 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
856 
857 	if (data && data[IFLA_MACVLAN_FLAGS])
858 		vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
859 
860 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
861 		if (port->count)
862 			return -EINVAL;
863 		port->passthru = true;
864 		eth_hw_addr_inherit(dev, lowerdev);
865 	}
866 
867 	port->count += 1;
868 	err = register_netdevice(dev);
869 	if (err < 0)
870 		goto destroy_port;
871 
872 	dev->priv_flags |= IFF_MACVLAN;
873 	err = netdev_upper_dev_link(lowerdev, dev);
874 	if (err)
875 		goto unregister_netdev;
876 
877 	list_add_tail_rcu(&vlan->list, &port->vlans);
878 	netif_stacked_transfer_operstate(lowerdev, dev);
879 
880 	return 0;
881 
882 unregister_netdev:
883 	unregister_netdevice(dev);
884 destroy_port:
885 	port->count -= 1;
886 	if (!port->count)
887 		macvlan_port_destroy(lowerdev);
888 
889 	return err;
890 }
891 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
892 
893 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
894 			   struct nlattr *tb[], struct nlattr *data[])
895 {
896 	return macvlan_common_newlink(src_net, dev, tb, data);
897 }
898 
899 void macvlan_dellink(struct net_device *dev, struct list_head *head)
900 {
901 	struct macvlan_dev *vlan = netdev_priv(dev);
902 
903 	list_del_rcu(&vlan->list);
904 	unregister_netdevice_queue(dev, head);
905 	netdev_upper_dev_unlink(vlan->lowerdev, dev);
906 }
907 EXPORT_SYMBOL_GPL(macvlan_dellink);
908 
909 static int macvlan_changelink(struct net_device *dev,
910 		struct nlattr *tb[], struct nlattr *data[])
911 {
912 	struct macvlan_dev *vlan = netdev_priv(dev);
913 	enum macvlan_mode mode;
914 	bool set_mode = false;
915 
916 	/* Validate mode, but don't set yet: setting flags may fail. */
917 	if (data && data[IFLA_MACVLAN_MODE]) {
918 		set_mode = true;
919 		mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
920 		/* Passthrough mode can't be set or cleared dynamically */
921 		if ((mode == MACVLAN_MODE_PASSTHRU) !=
922 		    (vlan->mode == MACVLAN_MODE_PASSTHRU))
923 			return -EINVAL;
924 	}
925 
926 	if (data && data[IFLA_MACVLAN_FLAGS]) {
927 		__u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
928 		bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
929 		if (vlan->port->passthru && promisc) {
930 			int err;
931 
932 			if (flags & MACVLAN_FLAG_NOPROMISC)
933 				err = dev_set_promiscuity(vlan->lowerdev, -1);
934 			else
935 				err = dev_set_promiscuity(vlan->lowerdev, 1);
936 			if (err < 0)
937 				return err;
938 		}
939 		vlan->flags = flags;
940 	}
941 	if (set_mode)
942 		vlan->mode = mode;
943 	return 0;
944 }
945 
946 static size_t macvlan_get_size(const struct net_device *dev)
947 {
948 	return (0
949 		+ nla_total_size(4) /* IFLA_MACVLAN_MODE */
950 		+ nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
951 		);
952 }
953 
954 static int macvlan_fill_info(struct sk_buff *skb,
955 				const struct net_device *dev)
956 {
957 	struct macvlan_dev *vlan = netdev_priv(dev);
958 
959 	if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
960 		goto nla_put_failure;
961 	if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
962 		goto nla_put_failure;
963 	return 0;
964 
965 nla_put_failure:
966 	return -EMSGSIZE;
967 }
968 
969 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
970 	[IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
971 	[IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
972 };
973 
974 int macvlan_link_register(struct rtnl_link_ops *ops)
975 {
976 	/* common fields */
977 	ops->priv_size		= sizeof(struct macvlan_dev);
978 	ops->validate		= macvlan_validate;
979 	ops->maxtype		= IFLA_MACVLAN_MAX;
980 	ops->policy		= macvlan_policy;
981 	ops->changelink		= macvlan_changelink;
982 	ops->get_size		= macvlan_get_size;
983 	ops->fill_info		= macvlan_fill_info;
984 
985 	return rtnl_link_register(ops);
986 };
987 EXPORT_SYMBOL_GPL(macvlan_link_register);
988 
989 static struct rtnl_link_ops macvlan_link_ops = {
990 	.kind		= "macvlan",
991 	.setup		= macvlan_setup,
992 	.newlink	= macvlan_newlink,
993 	.dellink	= macvlan_dellink,
994 };
995 
996 static int macvlan_device_event(struct notifier_block *unused,
997 				unsigned long event, void *ptr)
998 {
999 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1000 	struct macvlan_dev *vlan, *next;
1001 	struct macvlan_port *port;
1002 	LIST_HEAD(list_kill);
1003 
1004 	if (!macvlan_port_exists(dev))
1005 		return NOTIFY_DONE;
1006 
1007 	port = macvlan_port_get_rtnl(dev);
1008 
1009 	switch (event) {
1010 	case NETDEV_CHANGE:
1011 		list_for_each_entry(vlan, &port->vlans, list)
1012 			netif_stacked_transfer_operstate(vlan->lowerdev,
1013 							 vlan->dev);
1014 		break;
1015 	case NETDEV_FEAT_CHANGE:
1016 		list_for_each_entry(vlan, &port->vlans, list) {
1017 			vlan->dev->gso_max_size = dev->gso_max_size;
1018 			netdev_update_features(vlan->dev);
1019 		}
1020 		break;
1021 	case NETDEV_UNREGISTER:
1022 		/* twiddle thumbs on netns device moves */
1023 		if (dev->reg_state != NETREG_UNREGISTERING)
1024 			break;
1025 
1026 		list_for_each_entry_safe(vlan, next, &port->vlans, list)
1027 			vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1028 		unregister_netdevice_many(&list_kill);
1029 		list_del(&list_kill);
1030 		break;
1031 	case NETDEV_PRE_TYPE_CHANGE:
1032 		/* Forbid underlaying device to change its type. */
1033 		return NOTIFY_BAD;
1034 	}
1035 	return NOTIFY_DONE;
1036 }
1037 
1038 static struct notifier_block macvlan_notifier_block __read_mostly = {
1039 	.notifier_call	= macvlan_device_event,
1040 };
1041 
1042 static int __init macvlan_init_module(void)
1043 {
1044 	int err;
1045 
1046 	register_netdevice_notifier(&macvlan_notifier_block);
1047 
1048 	err = macvlan_link_register(&macvlan_link_ops);
1049 	if (err < 0)
1050 		goto err1;
1051 	return 0;
1052 err1:
1053 	unregister_netdevice_notifier(&macvlan_notifier_block);
1054 	return err;
1055 }
1056 
1057 static void __exit macvlan_cleanup_module(void)
1058 {
1059 	rtnl_link_unregister(&macvlan_link_ops);
1060 	unregister_netdevice_notifier(&macvlan_notifier_block);
1061 }
1062 
1063 module_init(macvlan_init_module);
1064 module_exit(macvlan_cleanup_module);
1065 
1066 MODULE_LICENSE("GPL");
1067 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1068 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1069 MODULE_ALIAS_RTNL_LINK("macvlan");
1070