xref: /linux/drivers/net/macvlan.c (revision c5aec4c76af1a2d89ee2f2d4d5463b2ad2d85de5)
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 (dev->flags & IFF_UP) {
462 		if (change & IFF_ALLMULTI)
463 			dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
464 	}
465 }
466 
467 static void macvlan_set_mac_lists(struct net_device *dev)
468 {
469 	struct macvlan_dev *vlan = netdev_priv(dev);
470 
471 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
472 		bitmap_fill(vlan->mc_filter, MACVLAN_MC_FILTER_SZ);
473 	} else {
474 		struct netdev_hw_addr *ha;
475 		DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
476 
477 		bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
478 		netdev_for_each_mc_addr(ha, dev) {
479 			__set_bit(mc_hash(vlan, ha->addr), filter);
480 		}
481 
482 		__set_bit(mc_hash(vlan, dev->broadcast), filter);
483 
484 		bitmap_copy(vlan->mc_filter, filter, MACVLAN_MC_FILTER_SZ);
485 	}
486 	dev_uc_sync(vlan->lowerdev, dev);
487 	dev_mc_sync(vlan->lowerdev, dev);
488 }
489 
490 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
491 {
492 	struct macvlan_dev *vlan = netdev_priv(dev);
493 
494 	if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
495 		return -EINVAL;
496 	dev->mtu = new_mtu;
497 	return 0;
498 }
499 
500 /*
501  * macvlan network devices have devices nesting below it and are a special
502  * "super class" of normal network devices; split their locks off into a
503  * separate class since they always nest.
504  */
505 static struct lock_class_key macvlan_netdev_xmit_lock_key;
506 static struct lock_class_key macvlan_netdev_addr_lock_key;
507 
508 #define ALWAYS_ON_FEATURES \
509 	(NETIF_F_SG | NETIF_F_GEN_CSUM | NETIF_F_GSO_SOFTWARE | NETIF_F_LLTX)
510 
511 #define MACVLAN_FEATURES \
512 	(NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
513 	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
514 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
515 	 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
516 
517 #define MACVLAN_STATE_MASK \
518 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
519 
520 static int macvlan_get_nest_level(struct net_device *dev)
521 {
522 	return ((struct macvlan_dev *)netdev_priv(dev))->nest_level;
523 }
524 
525 static void macvlan_set_lockdep_class_one(struct net_device *dev,
526 					  struct netdev_queue *txq,
527 					  void *_unused)
528 {
529 	lockdep_set_class(&txq->_xmit_lock,
530 			  &macvlan_netdev_xmit_lock_key);
531 }
532 
533 static void macvlan_set_lockdep_class(struct net_device *dev)
534 {
535 	lockdep_set_class_and_subclass(&dev->addr_list_lock,
536 				       &macvlan_netdev_addr_lock_key,
537 				       macvlan_get_nest_level(dev));
538 	netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
539 }
540 
541 static int macvlan_init(struct net_device *dev)
542 {
543 	struct macvlan_dev *vlan = netdev_priv(dev);
544 	const struct net_device *lowerdev = vlan->lowerdev;
545 
546 	dev->state		= (dev->state & ~MACVLAN_STATE_MASK) |
547 				  (lowerdev->state & MACVLAN_STATE_MASK);
548 	dev->features 		= lowerdev->features & MACVLAN_FEATURES;
549 	dev->features		|= ALWAYS_ON_FEATURES;
550 	dev->gso_max_size	= lowerdev->gso_max_size;
551 	dev->iflink		= lowerdev->ifindex;
552 	dev->hard_header_len	= lowerdev->hard_header_len;
553 
554 	macvlan_set_lockdep_class(dev);
555 
556 	vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
557 	if (!vlan->pcpu_stats)
558 		return -ENOMEM;
559 
560 	return 0;
561 }
562 
563 static void macvlan_uninit(struct net_device *dev)
564 {
565 	struct macvlan_dev *vlan = netdev_priv(dev);
566 	struct macvlan_port *port = vlan->port;
567 
568 	free_percpu(vlan->pcpu_stats);
569 
570 	port->count -= 1;
571 	if (!port->count)
572 		macvlan_port_destroy(port->dev);
573 }
574 
575 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
576 							 struct rtnl_link_stats64 *stats)
577 {
578 	struct macvlan_dev *vlan = netdev_priv(dev);
579 
580 	if (vlan->pcpu_stats) {
581 		struct vlan_pcpu_stats *p;
582 		u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
583 		u32 rx_errors = 0, tx_dropped = 0;
584 		unsigned int start;
585 		int i;
586 
587 		for_each_possible_cpu(i) {
588 			p = per_cpu_ptr(vlan->pcpu_stats, i);
589 			do {
590 				start = u64_stats_fetch_begin_irq(&p->syncp);
591 				rx_packets	= p->rx_packets;
592 				rx_bytes	= p->rx_bytes;
593 				rx_multicast	= p->rx_multicast;
594 				tx_packets	= p->tx_packets;
595 				tx_bytes	= p->tx_bytes;
596 			} while (u64_stats_fetch_retry_irq(&p->syncp, start));
597 
598 			stats->rx_packets	+= rx_packets;
599 			stats->rx_bytes		+= rx_bytes;
600 			stats->multicast	+= rx_multicast;
601 			stats->tx_packets	+= tx_packets;
602 			stats->tx_bytes		+= tx_bytes;
603 			/* rx_errors & tx_dropped are u32, updated
604 			 * without syncp protection.
605 			 */
606 			rx_errors	+= p->rx_errors;
607 			tx_dropped	+= p->tx_dropped;
608 		}
609 		stats->rx_errors	= rx_errors;
610 		stats->rx_dropped	= rx_errors;
611 		stats->tx_dropped	= tx_dropped;
612 	}
613 	return stats;
614 }
615 
616 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
617 				   __be16 proto, u16 vid)
618 {
619 	struct macvlan_dev *vlan = netdev_priv(dev);
620 	struct net_device *lowerdev = vlan->lowerdev;
621 
622 	return vlan_vid_add(lowerdev, proto, vid);
623 }
624 
625 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
626 				    __be16 proto, u16 vid)
627 {
628 	struct macvlan_dev *vlan = netdev_priv(dev);
629 	struct net_device *lowerdev = vlan->lowerdev;
630 
631 	vlan_vid_del(lowerdev, proto, vid);
632 	return 0;
633 }
634 
635 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
636 			   struct net_device *dev,
637 			   const unsigned char *addr,
638 			   u16 flags)
639 {
640 	struct macvlan_dev *vlan = netdev_priv(dev);
641 	int err = -EINVAL;
642 
643 	if (!vlan->port->passthru)
644 		return -EOPNOTSUPP;
645 
646 	if (flags & NLM_F_REPLACE)
647 		return -EOPNOTSUPP;
648 
649 	if (is_unicast_ether_addr(addr))
650 		err = dev_uc_add_excl(dev, addr);
651 	else if (is_multicast_ether_addr(addr))
652 		err = dev_mc_add_excl(dev, addr);
653 
654 	return err;
655 }
656 
657 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
658 			   struct net_device *dev,
659 			   const unsigned char *addr)
660 {
661 	struct macvlan_dev *vlan = netdev_priv(dev);
662 	int err = -EINVAL;
663 
664 	if (!vlan->port->passthru)
665 		return -EOPNOTSUPP;
666 
667 	if (is_unicast_ether_addr(addr))
668 		err = dev_uc_del(dev, addr);
669 	else if (is_multicast_ether_addr(addr))
670 		err = dev_mc_del(dev, addr);
671 
672 	return err;
673 }
674 
675 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
676 					struct ethtool_drvinfo *drvinfo)
677 {
678 	strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
679 	strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
680 }
681 
682 static int macvlan_ethtool_get_settings(struct net_device *dev,
683 					struct ethtool_cmd *cmd)
684 {
685 	const struct macvlan_dev *vlan = netdev_priv(dev);
686 
687 	return __ethtool_get_settings(vlan->lowerdev, cmd);
688 }
689 
690 static netdev_features_t macvlan_fix_features(struct net_device *dev,
691 					      netdev_features_t features)
692 {
693 	struct macvlan_dev *vlan = netdev_priv(dev);
694 	netdev_features_t mask;
695 
696 	features |= NETIF_F_ALL_FOR_ALL;
697 	features &= (vlan->set_features | ~MACVLAN_FEATURES);
698 	mask = features;
699 
700 	features = netdev_increment_features(vlan->lowerdev->features,
701 					     features,
702 					     mask);
703 	features |= ALWAYS_ON_FEATURES;
704 
705 	return features;
706 }
707 
708 static const struct ethtool_ops macvlan_ethtool_ops = {
709 	.get_link		= ethtool_op_get_link,
710 	.get_settings		= macvlan_ethtool_get_settings,
711 	.get_drvinfo		= macvlan_ethtool_get_drvinfo,
712 };
713 
714 static const struct net_device_ops macvlan_netdev_ops = {
715 	.ndo_init		= macvlan_init,
716 	.ndo_uninit		= macvlan_uninit,
717 	.ndo_open		= macvlan_open,
718 	.ndo_stop		= macvlan_stop,
719 	.ndo_start_xmit		= macvlan_start_xmit,
720 	.ndo_change_mtu		= macvlan_change_mtu,
721 	.ndo_fix_features	= macvlan_fix_features,
722 	.ndo_change_rx_flags	= macvlan_change_rx_flags,
723 	.ndo_set_mac_address	= macvlan_set_mac_address,
724 	.ndo_set_rx_mode	= macvlan_set_mac_lists,
725 	.ndo_get_stats64	= macvlan_dev_get_stats64,
726 	.ndo_validate_addr	= eth_validate_addr,
727 	.ndo_vlan_rx_add_vid	= macvlan_vlan_rx_add_vid,
728 	.ndo_vlan_rx_kill_vid	= macvlan_vlan_rx_kill_vid,
729 	.ndo_fdb_add		= macvlan_fdb_add,
730 	.ndo_fdb_del		= macvlan_fdb_del,
731 	.ndo_fdb_dump		= ndo_dflt_fdb_dump,
732 	.ndo_get_lock_subclass  = macvlan_get_nest_level,
733 };
734 
735 void macvlan_common_setup(struct net_device *dev)
736 {
737 	ether_setup(dev);
738 
739 	dev->priv_flags	       &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
740 	dev->priv_flags	       |= IFF_UNICAST_FLT;
741 	dev->netdev_ops		= &macvlan_netdev_ops;
742 	dev->destructor		= free_netdev;
743 	dev->header_ops		= &macvlan_hard_header_ops;
744 	dev->ethtool_ops	= &macvlan_ethtool_ops;
745 }
746 EXPORT_SYMBOL_GPL(macvlan_common_setup);
747 
748 static void macvlan_setup(struct net_device *dev)
749 {
750 	macvlan_common_setup(dev);
751 	dev->tx_queue_len	= 0;
752 }
753 
754 static int macvlan_port_create(struct net_device *dev)
755 {
756 	struct macvlan_port *port;
757 	unsigned int i;
758 	int err;
759 
760 	if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
761 		return -EINVAL;
762 
763 	port = kzalloc(sizeof(*port), GFP_KERNEL);
764 	if (port == NULL)
765 		return -ENOMEM;
766 
767 	port->passthru = false;
768 	port->dev = dev;
769 	INIT_LIST_HEAD(&port->vlans);
770 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
771 		INIT_HLIST_HEAD(&port->vlan_hash[i]);
772 
773 	err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
774 	if (err)
775 		kfree(port);
776 	else
777 		dev->priv_flags |= IFF_MACVLAN_PORT;
778 	return err;
779 }
780 
781 static void macvlan_port_destroy(struct net_device *dev)
782 {
783 	struct macvlan_port *port = macvlan_port_get_rtnl(dev);
784 
785 	dev->priv_flags &= ~IFF_MACVLAN_PORT;
786 	netdev_rx_handler_unregister(dev);
787 	kfree_rcu(port, rcu);
788 }
789 
790 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
791 {
792 	if (tb[IFLA_ADDRESS]) {
793 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
794 			return -EINVAL;
795 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
796 			return -EADDRNOTAVAIL;
797 	}
798 
799 	if (data && data[IFLA_MACVLAN_FLAGS] &&
800 	    nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
801 		return -EINVAL;
802 
803 	if (data && data[IFLA_MACVLAN_MODE]) {
804 		switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
805 		case MACVLAN_MODE_PRIVATE:
806 		case MACVLAN_MODE_VEPA:
807 		case MACVLAN_MODE_BRIDGE:
808 		case MACVLAN_MODE_PASSTHRU:
809 			break;
810 		default:
811 			return -EINVAL;
812 		}
813 	}
814 	return 0;
815 }
816 
817 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
818 			   struct nlattr *tb[], struct nlattr *data[])
819 {
820 	struct macvlan_dev *vlan = netdev_priv(dev);
821 	struct macvlan_port *port;
822 	struct net_device *lowerdev;
823 	int err;
824 
825 	if (!tb[IFLA_LINK])
826 		return -EINVAL;
827 
828 	lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
829 	if (lowerdev == NULL)
830 		return -ENODEV;
831 
832 	/* When creating macvlans or macvtaps on top of other macvlans - use
833 	 * the real device as the lowerdev.
834 	 */
835 	if (netif_is_macvlan(lowerdev))
836 		lowerdev = macvlan_dev_real_dev(lowerdev);
837 
838 	if (!tb[IFLA_MTU])
839 		dev->mtu = lowerdev->mtu;
840 	else if (dev->mtu > lowerdev->mtu)
841 		return -EINVAL;
842 
843 	if (!tb[IFLA_ADDRESS])
844 		eth_hw_addr_random(dev);
845 
846 	if (!macvlan_port_exists(lowerdev)) {
847 		err = macvlan_port_create(lowerdev);
848 		if (err < 0)
849 			return err;
850 	}
851 	port = macvlan_port_get_rtnl(lowerdev);
852 
853 	/* Only 1 macvlan device can be created in passthru mode */
854 	if (port->passthru)
855 		return -EINVAL;
856 
857 	vlan->lowerdev = lowerdev;
858 	vlan->dev      = dev;
859 	vlan->port     = port;
860 	vlan->set_features = MACVLAN_FEATURES;
861 	vlan->nest_level = dev_get_nest_level(lowerdev, netif_is_macvlan) + 1;
862 
863 	vlan->mode     = MACVLAN_MODE_VEPA;
864 	if (data && data[IFLA_MACVLAN_MODE])
865 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
866 
867 	if (data && data[IFLA_MACVLAN_FLAGS])
868 		vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
869 
870 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
871 		if (port->count)
872 			return -EINVAL;
873 		port->passthru = true;
874 		eth_hw_addr_inherit(dev, lowerdev);
875 	}
876 
877 	port->count += 1;
878 	err = register_netdevice(dev);
879 	if (err < 0)
880 		goto destroy_port;
881 
882 	dev->priv_flags |= IFF_MACVLAN;
883 	err = netdev_upper_dev_link(lowerdev, dev);
884 	if (err)
885 		goto unregister_netdev;
886 
887 	list_add_tail_rcu(&vlan->list, &port->vlans);
888 	netif_stacked_transfer_operstate(lowerdev, dev);
889 
890 	return 0;
891 
892 unregister_netdev:
893 	unregister_netdevice(dev);
894 destroy_port:
895 	port->count -= 1;
896 	if (!port->count)
897 		macvlan_port_destroy(lowerdev);
898 
899 	return err;
900 }
901 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
902 
903 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
904 			   struct nlattr *tb[], struct nlattr *data[])
905 {
906 	return macvlan_common_newlink(src_net, dev, tb, data);
907 }
908 
909 void macvlan_dellink(struct net_device *dev, struct list_head *head)
910 {
911 	struct macvlan_dev *vlan = netdev_priv(dev);
912 
913 	list_del_rcu(&vlan->list);
914 	unregister_netdevice_queue(dev, head);
915 	netdev_upper_dev_unlink(vlan->lowerdev, dev);
916 }
917 EXPORT_SYMBOL_GPL(macvlan_dellink);
918 
919 static int macvlan_changelink(struct net_device *dev,
920 		struct nlattr *tb[], struct nlattr *data[])
921 {
922 	struct macvlan_dev *vlan = netdev_priv(dev);
923 	enum macvlan_mode mode;
924 	bool set_mode = false;
925 
926 	/* Validate mode, but don't set yet: setting flags may fail. */
927 	if (data && data[IFLA_MACVLAN_MODE]) {
928 		set_mode = true;
929 		mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
930 		/* Passthrough mode can't be set or cleared dynamically */
931 		if ((mode == MACVLAN_MODE_PASSTHRU) !=
932 		    (vlan->mode == MACVLAN_MODE_PASSTHRU))
933 			return -EINVAL;
934 	}
935 
936 	if (data && data[IFLA_MACVLAN_FLAGS]) {
937 		__u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
938 		bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
939 		if (vlan->port->passthru && promisc) {
940 			int err;
941 
942 			if (flags & MACVLAN_FLAG_NOPROMISC)
943 				err = dev_set_promiscuity(vlan->lowerdev, -1);
944 			else
945 				err = dev_set_promiscuity(vlan->lowerdev, 1);
946 			if (err < 0)
947 				return err;
948 		}
949 		vlan->flags = flags;
950 	}
951 	if (set_mode)
952 		vlan->mode = mode;
953 	return 0;
954 }
955 
956 static size_t macvlan_get_size(const struct net_device *dev)
957 {
958 	return (0
959 		+ nla_total_size(4) /* IFLA_MACVLAN_MODE */
960 		+ nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
961 		);
962 }
963 
964 static int macvlan_fill_info(struct sk_buff *skb,
965 				const struct net_device *dev)
966 {
967 	struct macvlan_dev *vlan = netdev_priv(dev);
968 
969 	if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
970 		goto nla_put_failure;
971 	if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
972 		goto nla_put_failure;
973 	return 0;
974 
975 nla_put_failure:
976 	return -EMSGSIZE;
977 }
978 
979 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
980 	[IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
981 	[IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
982 };
983 
984 int macvlan_link_register(struct rtnl_link_ops *ops)
985 {
986 	/* common fields */
987 	ops->priv_size		= sizeof(struct macvlan_dev);
988 	ops->validate		= macvlan_validate;
989 	ops->maxtype		= IFLA_MACVLAN_MAX;
990 	ops->policy		= macvlan_policy;
991 	ops->changelink		= macvlan_changelink;
992 	ops->get_size		= macvlan_get_size;
993 	ops->fill_info		= macvlan_fill_info;
994 
995 	return rtnl_link_register(ops);
996 };
997 EXPORT_SYMBOL_GPL(macvlan_link_register);
998 
999 static struct rtnl_link_ops macvlan_link_ops = {
1000 	.kind		= "macvlan",
1001 	.setup		= macvlan_setup,
1002 	.newlink	= macvlan_newlink,
1003 	.dellink	= macvlan_dellink,
1004 };
1005 
1006 static int macvlan_device_event(struct notifier_block *unused,
1007 				unsigned long event, void *ptr)
1008 {
1009 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1010 	struct macvlan_dev *vlan, *next;
1011 	struct macvlan_port *port;
1012 	LIST_HEAD(list_kill);
1013 
1014 	if (!macvlan_port_exists(dev))
1015 		return NOTIFY_DONE;
1016 
1017 	port = macvlan_port_get_rtnl(dev);
1018 
1019 	switch (event) {
1020 	case NETDEV_CHANGE:
1021 		list_for_each_entry(vlan, &port->vlans, list)
1022 			netif_stacked_transfer_operstate(vlan->lowerdev,
1023 							 vlan->dev);
1024 		break;
1025 	case NETDEV_FEAT_CHANGE:
1026 		list_for_each_entry(vlan, &port->vlans, list) {
1027 			vlan->dev->gso_max_size = dev->gso_max_size;
1028 			netdev_update_features(vlan->dev);
1029 		}
1030 		break;
1031 	case NETDEV_UNREGISTER:
1032 		/* twiddle thumbs on netns device moves */
1033 		if (dev->reg_state != NETREG_UNREGISTERING)
1034 			break;
1035 
1036 		list_for_each_entry_safe(vlan, next, &port->vlans, list)
1037 			vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1038 		unregister_netdevice_many(&list_kill);
1039 		list_del(&list_kill);
1040 		break;
1041 	case NETDEV_PRE_TYPE_CHANGE:
1042 		/* Forbid underlaying device to change its type. */
1043 		return NOTIFY_BAD;
1044 	}
1045 	return NOTIFY_DONE;
1046 }
1047 
1048 static struct notifier_block macvlan_notifier_block __read_mostly = {
1049 	.notifier_call	= macvlan_device_event,
1050 };
1051 
1052 static int __init macvlan_init_module(void)
1053 {
1054 	int err;
1055 
1056 	register_netdevice_notifier(&macvlan_notifier_block);
1057 
1058 	err = macvlan_link_register(&macvlan_link_ops);
1059 	if (err < 0)
1060 		goto err1;
1061 	return 0;
1062 err1:
1063 	unregister_netdevice_notifier(&macvlan_notifier_block);
1064 	return err;
1065 }
1066 
1067 static void __exit macvlan_cleanup_module(void)
1068 {
1069 	rtnl_link_unregister(&macvlan_link_ops);
1070 	unregister_netdevice_notifier(&macvlan_notifier_block);
1071 }
1072 
1073 module_init(macvlan_init_module);
1074 module_exit(macvlan_cleanup_module);
1075 
1076 MODULE_LICENSE("GPL");
1077 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1078 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1079 MODULE_ALIAS_RTNL_LINK("macvlan");
1080