xref: /openbmc/linux/drivers/net/macvlan.c (revision 37be287c)
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 	__u8 ip_summed = skb->ip_summed;
267 
268 	if (vlan->mode == MACVLAN_MODE_BRIDGE) {
269 		const struct ethhdr *eth = (void *)skb->data;
270 		skb->ip_summed = CHECKSUM_UNNECESSARY;
271 
272 		/* send to other bridge ports directly */
273 		if (is_multicast_ether_addr(eth->h_dest)) {
274 			macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
275 			goto xmit_world;
276 		}
277 
278 		dest = macvlan_hash_lookup(port, eth->h_dest);
279 		if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
280 			/* send to lowerdev first for its network taps */
281 			dev_forward_skb(vlan->lowerdev, skb);
282 
283 			return NET_XMIT_SUCCESS;
284 		}
285 	}
286 
287 xmit_world:
288 	skb->ip_summed = ip_summed;
289 	skb->dev = vlan->lowerdev;
290 	return dev_queue_xmit(skb);
291 }
292 
293 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
294 				      struct net_device *dev)
295 {
296 	unsigned int len = skb->len;
297 	int ret;
298 	const struct macvlan_dev *vlan = netdev_priv(dev);
299 
300 	if (vlan->fwd_priv) {
301 		skb->dev = vlan->lowerdev;
302 		ret = dev_queue_xmit_accel(skb, vlan->fwd_priv);
303 	} else {
304 		ret = macvlan_queue_xmit(skb, dev);
305 	}
306 
307 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
308 		struct vlan_pcpu_stats *pcpu_stats;
309 
310 		pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
311 		u64_stats_update_begin(&pcpu_stats->syncp);
312 		pcpu_stats->tx_packets++;
313 		pcpu_stats->tx_bytes += len;
314 		u64_stats_update_end(&pcpu_stats->syncp);
315 	} else {
316 		this_cpu_inc(vlan->pcpu_stats->tx_dropped);
317 	}
318 	return ret;
319 }
320 
321 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
322 			       unsigned short type, const void *daddr,
323 			       const void *saddr, unsigned len)
324 {
325 	const struct macvlan_dev *vlan = netdev_priv(dev);
326 	struct net_device *lowerdev = vlan->lowerdev;
327 
328 	return dev_hard_header(skb, lowerdev, type, daddr,
329 			       saddr ? : dev->dev_addr, len);
330 }
331 
332 static const struct header_ops macvlan_hard_header_ops = {
333 	.create  	= macvlan_hard_header,
334 	.rebuild	= eth_rebuild_header,
335 	.parse		= eth_header_parse,
336 	.cache		= eth_header_cache,
337 	.cache_update	= eth_header_cache_update,
338 };
339 
340 static struct rtnl_link_ops macvlan_link_ops;
341 
342 static int macvlan_open(struct net_device *dev)
343 {
344 	struct macvlan_dev *vlan = netdev_priv(dev);
345 	struct net_device *lowerdev = vlan->lowerdev;
346 	int err;
347 
348 	if (vlan->port->passthru) {
349 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
350 			err = dev_set_promiscuity(lowerdev, 1);
351 			if (err < 0)
352 				goto out;
353 		}
354 		goto hash_add;
355 	}
356 
357 	if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD &&
358 	    dev->rtnl_link_ops == &macvlan_link_ops) {
359 		vlan->fwd_priv =
360 		      lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
361 
362 		/* If we get a NULL pointer back, or if we get an error
363 		 * then we should just fall through to the non accelerated path
364 		 */
365 		if (IS_ERR_OR_NULL(vlan->fwd_priv)) {
366 			vlan->fwd_priv = NULL;
367 		} else
368 			return 0;
369 	}
370 
371 	err = -EBUSY;
372 	if (macvlan_addr_busy(vlan->port, dev->dev_addr))
373 		goto out;
374 
375 	err = dev_uc_add(lowerdev, dev->dev_addr);
376 	if (err < 0)
377 		goto out;
378 	if (dev->flags & IFF_ALLMULTI) {
379 		err = dev_set_allmulti(lowerdev, 1);
380 		if (err < 0)
381 			goto del_unicast;
382 	}
383 
384 hash_add:
385 	macvlan_hash_add(vlan);
386 	return 0;
387 
388 del_unicast:
389 	dev_uc_del(lowerdev, dev->dev_addr);
390 out:
391 	if (vlan->fwd_priv) {
392 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
393 							   vlan->fwd_priv);
394 		vlan->fwd_priv = NULL;
395 	}
396 	return err;
397 }
398 
399 static int macvlan_stop(struct net_device *dev)
400 {
401 	struct macvlan_dev *vlan = netdev_priv(dev);
402 	struct net_device *lowerdev = vlan->lowerdev;
403 
404 	if (vlan->fwd_priv) {
405 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
406 							   vlan->fwd_priv);
407 		vlan->fwd_priv = NULL;
408 		return 0;
409 	}
410 
411 	dev_uc_unsync(lowerdev, dev);
412 	dev_mc_unsync(lowerdev, dev);
413 
414 	if (vlan->port->passthru) {
415 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
416 			dev_set_promiscuity(lowerdev, -1);
417 		goto hash_del;
418 	}
419 
420 	if (dev->flags & IFF_ALLMULTI)
421 		dev_set_allmulti(lowerdev, -1);
422 
423 	dev_uc_del(lowerdev, dev->dev_addr);
424 
425 hash_del:
426 	macvlan_hash_del(vlan, !dev->dismantle);
427 	return 0;
428 }
429 
430 static int macvlan_set_mac_address(struct net_device *dev, void *p)
431 {
432 	struct macvlan_dev *vlan = netdev_priv(dev);
433 	struct net_device *lowerdev = vlan->lowerdev;
434 	struct sockaddr *addr = p;
435 	int err;
436 
437 	if (!is_valid_ether_addr(addr->sa_data))
438 		return -EADDRNOTAVAIL;
439 
440 	if (!(dev->flags & IFF_UP)) {
441 		/* Just copy in the new address */
442 		memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
443 	} else {
444 		/* Rehash and update the device filters */
445 		if (macvlan_addr_busy(vlan->port, addr->sa_data))
446 			return -EBUSY;
447 
448 		err = dev_uc_add(lowerdev, addr->sa_data);
449 		if (err)
450 			return err;
451 
452 		dev_uc_del(lowerdev, dev->dev_addr);
453 
454 		macvlan_hash_change_addr(vlan, addr->sa_data);
455 	}
456 	return 0;
457 }
458 
459 static void macvlan_change_rx_flags(struct net_device *dev, int change)
460 {
461 	struct macvlan_dev *vlan = netdev_priv(dev);
462 	struct net_device *lowerdev = vlan->lowerdev;
463 
464 	if (change & IFF_ALLMULTI)
465 		dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
466 }
467 
468 static void macvlan_set_mac_lists(struct net_device *dev)
469 {
470 	struct macvlan_dev *vlan = netdev_priv(dev);
471 
472 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
473 		bitmap_fill(vlan->mc_filter, MACVLAN_MC_FILTER_SZ);
474 	} else {
475 		struct netdev_hw_addr *ha;
476 		DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
477 
478 		bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
479 		netdev_for_each_mc_addr(ha, dev) {
480 			__set_bit(mc_hash(vlan, ha->addr), filter);
481 		}
482 
483 		__set_bit(mc_hash(vlan, dev->broadcast), filter);
484 
485 		bitmap_copy(vlan->mc_filter, filter, MACVLAN_MC_FILTER_SZ);
486 	}
487 	dev_uc_sync(vlan->lowerdev, dev);
488 	dev_mc_sync(vlan->lowerdev, dev);
489 }
490 
491 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
492 {
493 	struct macvlan_dev *vlan = netdev_priv(dev);
494 
495 	if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
496 		return -EINVAL;
497 	dev->mtu = new_mtu;
498 	return 0;
499 }
500 
501 /*
502  * macvlan network devices have devices nesting below it and are a special
503  * "super class" of normal network devices; split their locks off into a
504  * separate class since they always nest.
505  */
506 static struct lock_class_key macvlan_netdev_xmit_lock_key;
507 static struct lock_class_key macvlan_netdev_addr_lock_key;
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 	int i;
538 
539 	dev->state		= (dev->state & ~MACVLAN_STATE_MASK) |
540 				  (lowerdev->state & MACVLAN_STATE_MASK);
541 	dev->features 		= lowerdev->features & MACVLAN_FEATURES;
542 	dev->features		|= NETIF_F_LLTX;
543 	dev->gso_max_size	= lowerdev->gso_max_size;
544 	dev->iflink		= lowerdev->ifindex;
545 	dev->hard_header_len	= lowerdev->hard_header_len;
546 
547 	macvlan_set_lockdep_class(dev);
548 
549 	vlan->pcpu_stats = alloc_percpu(struct vlan_pcpu_stats);
550 	if (!vlan->pcpu_stats)
551 		return -ENOMEM;
552 
553 	for_each_possible_cpu(i) {
554 		struct vlan_pcpu_stats *mvlstats;
555 		mvlstats = per_cpu_ptr(vlan->pcpu_stats, i);
556 		u64_stats_init(&mvlstats->syncp);
557 	}
558 
559 	return 0;
560 }
561 
562 static void macvlan_uninit(struct net_device *dev)
563 {
564 	struct macvlan_dev *vlan = netdev_priv(dev);
565 	struct macvlan_port *port = vlan->port;
566 
567 	free_percpu(vlan->pcpu_stats);
568 
569 	port->count -= 1;
570 	if (!port->count)
571 		macvlan_port_destroy(port->dev);
572 }
573 
574 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
575 							 struct rtnl_link_stats64 *stats)
576 {
577 	struct macvlan_dev *vlan = netdev_priv(dev);
578 
579 	if (vlan->pcpu_stats) {
580 		struct vlan_pcpu_stats *p;
581 		u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
582 		u32 rx_errors = 0, tx_dropped = 0;
583 		unsigned int start;
584 		int i;
585 
586 		for_each_possible_cpu(i) {
587 			p = per_cpu_ptr(vlan->pcpu_stats, i);
588 			do {
589 				start = u64_stats_fetch_begin_bh(&p->syncp);
590 				rx_packets	= p->rx_packets;
591 				rx_bytes	= p->rx_bytes;
592 				rx_multicast	= p->rx_multicast;
593 				tx_packets	= p->tx_packets;
594 				tx_bytes	= p->tx_bytes;
595 			} while (u64_stats_fetch_retry_bh(&p->syncp, start));
596 
597 			stats->rx_packets	+= rx_packets;
598 			stats->rx_bytes		+= rx_bytes;
599 			stats->multicast	+= rx_multicast;
600 			stats->tx_packets	+= tx_packets;
601 			stats->tx_bytes		+= tx_bytes;
602 			/* rx_errors & tx_dropped are u32, updated
603 			 * without syncp protection.
604 			 */
605 			rx_errors	+= p->rx_errors;
606 			tx_dropped	+= p->tx_dropped;
607 		}
608 		stats->rx_errors	= rx_errors;
609 		stats->rx_dropped	= rx_errors;
610 		stats->tx_dropped	= tx_dropped;
611 	}
612 	return stats;
613 }
614 
615 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
616 				   __be16 proto, u16 vid)
617 {
618 	struct macvlan_dev *vlan = netdev_priv(dev);
619 	struct net_device *lowerdev = vlan->lowerdev;
620 
621 	return vlan_vid_add(lowerdev, proto, vid);
622 }
623 
624 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
625 				    __be16 proto, u16 vid)
626 {
627 	struct macvlan_dev *vlan = netdev_priv(dev);
628 	struct net_device *lowerdev = vlan->lowerdev;
629 
630 	vlan_vid_del(lowerdev, proto, vid);
631 	return 0;
632 }
633 
634 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
635 			   struct net_device *dev,
636 			   const unsigned char *addr,
637 			   u16 flags)
638 {
639 	struct macvlan_dev *vlan = netdev_priv(dev);
640 	int err = -EINVAL;
641 
642 	if (!vlan->port->passthru)
643 		return -EOPNOTSUPP;
644 
645 	if (flags & NLM_F_REPLACE)
646 		return -EOPNOTSUPP;
647 
648 	if (is_unicast_ether_addr(addr))
649 		err = dev_uc_add_excl(dev, addr);
650 	else if (is_multicast_ether_addr(addr))
651 		err = dev_mc_add_excl(dev, addr);
652 
653 	return err;
654 }
655 
656 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
657 			   struct net_device *dev,
658 			   const unsigned char *addr)
659 {
660 	struct macvlan_dev *vlan = netdev_priv(dev);
661 	int err = -EINVAL;
662 
663 	if (!vlan->port->passthru)
664 		return -EOPNOTSUPP;
665 
666 	if (is_unicast_ether_addr(addr))
667 		err = dev_uc_del(dev, addr);
668 	else if (is_multicast_ether_addr(addr))
669 		err = dev_mc_del(dev, addr);
670 
671 	return err;
672 }
673 
674 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
675 					struct ethtool_drvinfo *drvinfo)
676 {
677 	strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
678 	strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
679 }
680 
681 static int macvlan_ethtool_get_settings(struct net_device *dev,
682 					struct ethtool_cmd *cmd)
683 {
684 	const struct macvlan_dev *vlan = netdev_priv(dev);
685 
686 	return __ethtool_get_settings(vlan->lowerdev, cmd);
687 }
688 
689 static netdev_features_t macvlan_fix_features(struct net_device *dev,
690 					      netdev_features_t features)
691 {
692 	struct macvlan_dev *vlan = netdev_priv(dev);
693 	netdev_features_t mask;
694 
695 	features |= NETIF_F_ALL_FOR_ALL;
696 	features &= (vlan->set_features | ~MACVLAN_FEATURES);
697 	mask = features;
698 
699 	features = netdev_increment_features(vlan->lowerdev->features,
700 					     features,
701 					     mask);
702 	features |= NETIF_F_LLTX;
703 
704 	return features;
705 }
706 
707 static const struct ethtool_ops macvlan_ethtool_ops = {
708 	.get_link		= ethtool_op_get_link,
709 	.get_settings		= macvlan_ethtool_get_settings,
710 	.get_drvinfo		= macvlan_ethtool_get_drvinfo,
711 };
712 
713 static const struct net_device_ops macvlan_netdev_ops = {
714 	.ndo_init		= macvlan_init,
715 	.ndo_uninit		= macvlan_uninit,
716 	.ndo_open		= macvlan_open,
717 	.ndo_stop		= macvlan_stop,
718 	.ndo_start_xmit		= macvlan_start_xmit,
719 	.ndo_change_mtu		= macvlan_change_mtu,
720 	.ndo_fix_features	= macvlan_fix_features,
721 	.ndo_change_rx_flags	= macvlan_change_rx_flags,
722 	.ndo_set_mac_address	= macvlan_set_mac_address,
723 	.ndo_set_rx_mode	= macvlan_set_mac_lists,
724 	.ndo_get_stats64	= macvlan_dev_get_stats64,
725 	.ndo_validate_addr	= eth_validate_addr,
726 	.ndo_vlan_rx_add_vid	= macvlan_vlan_rx_add_vid,
727 	.ndo_vlan_rx_kill_vid	= macvlan_vlan_rx_kill_vid,
728 	.ndo_fdb_add		= macvlan_fdb_add,
729 	.ndo_fdb_del		= macvlan_fdb_del,
730 	.ndo_fdb_dump		= ndo_dflt_fdb_dump,
731 };
732 
733 void macvlan_common_setup(struct net_device *dev)
734 {
735 	ether_setup(dev);
736 
737 	dev->priv_flags	       &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
738 	dev->priv_flags	       |= IFF_UNICAST_FLT;
739 	dev->netdev_ops		= &macvlan_netdev_ops;
740 	dev->destructor		= free_netdev;
741 	dev->header_ops		= &macvlan_hard_header_ops;
742 	dev->ethtool_ops	= &macvlan_ethtool_ops;
743 }
744 EXPORT_SYMBOL_GPL(macvlan_common_setup);
745 
746 static void macvlan_setup(struct net_device *dev)
747 {
748 	macvlan_common_setup(dev);
749 	dev->tx_queue_len	= 0;
750 }
751 
752 static int macvlan_port_create(struct net_device *dev)
753 {
754 	struct macvlan_port *port;
755 	unsigned int i;
756 	int err;
757 
758 	if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
759 		return -EINVAL;
760 
761 	port = kzalloc(sizeof(*port), GFP_KERNEL);
762 	if (port == NULL)
763 		return -ENOMEM;
764 
765 	port->passthru = false;
766 	port->dev = dev;
767 	INIT_LIST_HEAD(&port->vlans);
768 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
769 		INIT_HLIST_HEAD(&port->vlan_hash[i]);
770 
771 	err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
772 	if (err)
773 		kfree(port);
774 	else
775 		dev->priv_flags |= IFF_MACVLAN_PORT;
776 	return err;
777 }
778 
779 static void macvlan_port_destroy(struct net_device *dev)
780 {
781 	struct macvlan_port *port = macvlan_port_get_rtnl(dev);
782 
783 	dev->priv_flags &= ~IFF_MACVLAN_PORT;
784 	netdev_rx_handler_unregister(dev);
785 	kfree_rcu(port, rcu);
786 }
787 
788 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
789 {
790 	if (tb[IFLA_ADDRESS]) {
791 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
792 			return -EINVAL;
793 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
794 			return -EADDRNOTAVAIL;
795 	}
796 
797 	if (data && data[IFLA_MACVLAN_FLAGS] &&
798 	    nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
799 		return -EINVAL;
800 
801 	if (data && data[IFLA_MACVLAN_MODE]) {
802 		switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
803 		case MACVLAN_MODE_PRIVATE:
804 		case MACVLAN_MODE_VEPA:
805 		case MACVLAN_MODE_BRIDGE:
806 		case MACVLAN_MODE_PASSTHRU:
807 			break;
808 		default:
809 			return -EINVAL;
810 		}
811 	}
812 	return 0;
813 }
814 
815 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
816 			   struct nlattr *tb[], struct nlattr *data[])
817 {
818 	struct macvlan_dev *vlan = netdev_priv(dev);
819 	struct macvlan_port *port;
820 	struct net_device *lowerdev;
821 	int err;
822 
823 	if (!tb[IFLA_LINK])
824 		return -EINVAL;
825 
826 	lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
827 	if (lowerdev == NULL)
828 		return -ENODEV;
829 
830 	/* When creating macvlans or macvtaps on top of other macvlans - use
831 	 * the real device as the lowerdev.
832 	 */
833 	if (netif_is_macvlan(lowerdev))
834 		lowerdev = macvlan_dev_real_dev(lowerdev);
835 
836 	if (!tb[IFLA_MTU])
837 		dev->mtu = lowerdev->mtu;
838 	else if (dev->mtu > lowerdev->mtu)
839 		return -EINVAL;
840 
841 	if (!tb[IFLA_ADDRESS])
842 		eth_hw_addr_random(dev);
843 
844 	if (!macvlan_port_exists(lowerdev)) {
845 		err = macvlan_port_create(lowerdev);
846 		if (err < 0)
847 			return err;
848 	}
849 	port = macvlan_port_get_rtnl(lowerdev);
850 
851 	/* Only 1 macvlan device can be created in passthru mode */
852 	if (port->passthru)
853 		return -EINVAL;
854 
855 	vlan->lowerdev = lowerdev;
856 	vlan->dev      = dev;
857 	vlan->port     = port;
858 	vlan->set_features = MACVLAN_FEATURES;
859 
860 	vlan->mode     = MACVLAN_MODE_VEPA;
861 	if (data && data[IFLA_MACVLAN_MODE])
862 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
863 
864 	if (data && data[IFLA_MACVLAN_FLAGS])
865 		vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
866 
867 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
868 		if (port->count)
869 			return -EINVAL;
870 		port->passthru = true;
871 		eth_hw_addr_inherit(dev, lowerdev);
872 	}
873 
874 	port->count += 1;
875 	err = register_netdevice(dev);
876 	if (err < 0)
877 		goto destroy_port;
878 
879 	dev->priv_flags |= IFF_MACVLAN;
880 	err = netdev_upper_dev_link(lowerdev, dev);
881 	if (err)
882 		goto destroy_port;
883 
884 
885 	list_add_tail_rcu(&vlan->list, &port->vlans);
886 	netif_stacked_transfer_operstate(lowerdev, dev);
887 
888 	return 0;
889 
890 destroy_port:
891 	port->count -= 1;
892 	if (!port->count)
893 		macvlan_port_destroy(lowerdev);
894 
895 	return err;
896 }
897 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
898 
899 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
900 			   struct nlattr *tb[], struct nlattr *data[])
901 {
902 	return macvlan_common_newlink(src_net, dev, tb, data);
903 }
904 
905 void macvlan_dellink(struct net_device *dev, struct list_head *head)
906 {
907 	struct macvlan_dev *vlan = netdev_priv(dev);
908 
909 	list_del_rcu(&vlan->list);
910 	unregister_netdevice_queue(dev, head);
911 	netdev_upper_dev_unlink(vlan->lowerdev, dev);
912 }
913 EXPORT_SYMBOL_GPL(macvlan_dellink);
914 
915 static int macvlan_changelink(struct net_device *dev,
916 		struct nlattr *tb[], struct nlattr *data[])
917 {
918 	struct macvlan_dev *vlan = netdev_priv(dev);
919 	enum macvlan_mode mode;
920 	bool set_mode = false;
921 
922 	/* Validate mode, but don't set yet: setting flags may fail. */
923 	if (data && data[IFLA_MACVLAN_MODE]) {
924 		set_mode = true;
925 		mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
926 		/* Passthrough mode can't be set or cleared dynamically */
927 		if ((mode == MACVLAN_MODE_PASSTHRU) !=
928 		    (vlan->mode == MACVLAN_MODE_PASSTHRU))
929 			return -EINVAL;
930 	}
931 
932 	if (data && data[IFLA_MACVLAN_FLAGS]) {
933 		__u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
934 		bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
935 		if (vlan->port->passthru && promisc) {
936 			int err;
937 
938 			if (flags & MACVLAN_FLAG_NOPROMISC)
939 				err = dev_set_promiscuity(vlan->lowerdev, -1);
940 			else
941 				err = dev_set_promiscuity(vlan->lowerdev, 1);
942 			if (err < 0)
943 				return err;
944 		}
945 		vlan->flags = flags;
946 	}
947 	if (set_mode)
948 		vlan->mode = mode;
949 	return 0;
950 }
951 
952 static size_t macvlan_get_size(const struct net_device *dev)
953 {
954 	return (0
955 		+ nla_total_size(4) /* IFLA_MACVLAN_MODE */
956 		+ nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
957 		);
958 }
959 
960 static int macvlan_fill_info(struct sk_buff *skb,
961 				const struct net_device *dev)
962 {
963 	struct macvlan_dev *vlan = netdev_priv(dev);
964 
965 	if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
966 		goto nla_put_failure;
967 	if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
968 		goto nla_put_failure;
969 	return 0;
970 
971 nla_put_failure:
972 	return -EMSGSIZE;
973 }
974 
975 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
976 	[IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
977 	[IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
978 };
979 
980 int macvlan_link_register(struct rtnl_link_ops *ops)
981 {
982 	/* common fields */
983 	ops->priv_size		= sizeof(struct macvlan_dev);
984 	ops->validate		= macvlan_validate;
985 	ops->maxtype		= IFLA_MACVLAN_MAX;
986 	ops->policy		= macvlan_policy;
987 	ops->changelink		= macvlan_changelink;
988 	ops->get_size		= macvlan_get_size;
989 	ops->fill_info		= macvlan_fill_info;
990 
991 	return rtnl_link_register(ops);
992 };
993 EXPORT_SYMBOL_GPL(macvlan_link_register);
994 
995 static struct rtnl_link_ops macvlan_link_ops = {
996 	.kind		= "macvlan",
997 	.setup		= macvlan_setup,
998 	.newlink	= macvlan_newlink,
999 	.dellink	= macvlan_dellink,
1000 };
1001 
1002 static int macvlan_device_event(struct notifier_block *unused,
1003 				unsigned long event, void *ptr)
1004 {
1005 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1006 	struct macvlan_dev *vlan, *next;
1007 	struct macvlan_port *port;
1008 	LIST_HEAD(list_kill);
1009 
1010 	if (!macvlan_port_exists(dev))
1011 		return NOTIFY_DONE;
1012 
1013 	port = macvlan_port_get_rtnl(dev);
1014 
1015 	switch (event) {
1016 	case NETDEV_CHANGE:
1017 		list_for_each_entry(vlan, &port->vlans, list)
1018 			netif_stacked_transfer_operstate(vlan->lowerdev,
1019 							 vlan->dev);
1020 		break;
1021 	case NETDEV_FEAT_CHANGE:
1022 		list_for_each_entry(vlan, &port->vlans, list) {
1023 			vlan->dev->gso_max_size = dev->gso_max_size;
1024 			netdev_update_features(vlan->dev);
1025 		}
1026 		break;
1027 	case NETDEV_UNREGISTER:
1028 		/* twiddle thumbs on netns device moves */
1029 		if (dev->reg_state != NETREG_UNREGISTERING)
1030 			break;
1031 
1032 		list_for_each_entry_safe(vlan, next, &port->vlans, list)
1033 			vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1034 		unregister_netdevice_many(&list_kill);
1035 		list_del(&list_kill);
1036 		break;
1037 	case NETDEV_PRE_TYPE_CHANGE:
1038 		/* Forbid underlaying device to change its type. */
1039 		return NOTIFY_BAD;
1040 	}
1041 	return NOTIFY_DONE;
1042 }
1043 
1044 static struct notifier_block macvlan_notifier_block __read_mostly = {
1045 	.notifier_call	= macvlan_device_event,
1046 };
1047 
1048 static int __init macvlan_init_module(void)
1049 {
1050 	int err;
1051 
1052 	register_netdevice_notifier(&macvlan_notifier_block);
1053 
1054 	err = macvlan_link_register(&macvlan_link_ops);
1055 	if (err < 0)
1056 		goto err1;
1057 	return 0;
1058 err1:
1059 	unregister_netdevice_notifier(&macvlan_notifier_block);
1060 	return err;
1061 }
1062 
1063 static void __exit macvlan_cleanup_module(void)
1064 {
1065 	rtnl_link_unregister(&macvlan_link_ops);
1066 	unregister_netdevice_notifier(&macvlan_notifier_block);
1067 }
1068 
1069 module_init(macvlan_init_module);
1070 module_exit(macvlan_cleanup_module);
1071 
1072 MODULE_LICENSE("GPL");
1073 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1074 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1075 MODULE_ALIAS_RTNL_LINK("macvlan");
1076