xref: /openbmc/linux/drivers/net/macvlan.c (revision b13ba1b83f524732523db1079e56478b32c85c96)
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 vlan->forward(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 vlan->receive(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 = vlan->receive(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 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_hard_start_xmit(skb, skb->dev, NULL, 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 macvlan_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 EXPORT_SYMBOL_GPL(macvlan_start_xmit);
321 
322 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
323 			       unsigned short type, const void *daddr,
324 			       const void *saddr, unsigned len)
325 {
326 	const struct macvlan_dev *vlan = netdev_priv(dev);
327 	struct net_device *lowerdev = vlan->lowerdev;
328 
329 	return dev_hard_header(skb, lowerdev, type, daddr,
330 			       saddr ? : dev->dev_addr, len);
331 }
332 
333 static const struct header_ops macvlan_hard_header_ops = {
334 	.create  	= macvlan_hard_header,
335 	.rebuild	= eth_rebuild_header,
336 	.parse		= eth_header_parse,
337 	.cache		= eth_header_cache,
338 	.cache_update	= eth_header_cache_update,
339 };
340 
341 static struct rtnl_link_ops macvlan_link_ops;
342 
343 static int macvlan_open(struct net_device *dev)
344 {
345 	struct macvlan_dev *vlan = netdev_priv(dev);
346 	struct net_device *lowerdev = vlan->lowerdev;
347 	int err;
348 
349 	if (vlan->port->passthru) {
350 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
351 			err = dev_set_promiscuity(lowerdev, 1);
352 			if (err < 0)
353 				goto out;
354 		}
355 		goto hash_add;
356 	}
357 
358 	if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD &&
359 	    dev->rtnl_link_ops == &macvlan_link_ops) {
360 		vlan->fwd_priv =
361 		      lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
362 
363 		/* If we get a NULL pointer back, or if we get an error
364 		 * then we should just fall through to the non accelerated path
365 		 */
366 		if (IS_ERR_OR_NULL(vlan->fwd_priv)) {
367 			vlan->fwd_priv = NULL;
368 		} else {
369 			dev->features &= ~NETIF_F_LLTX;
370 			return 0;
371 		}
372 	}
373 
374 	err = -EBUSY;
375 	if (macvlan_addr_busy(vlan->port, dev->dev_addr))
376 		goto out;
377 
378 	err = dev_uc_add(lowerdev, dev->dev_addr);
379 	if (err < 0)
380 		goto out;
381 	if (dev->flags & IFF_ALLMULTI) {
382 		err = dev_set_allmulti(lowerdev, 1);
383 		if (err < 0)
384 			goto del_unicast;
385 	}
386 
387 hash_add:
388 	macvlan_hash_add(vlan);
389 	return 0;
390 
391 del_unicast:
392 	dev_uc_del(lowerdev, dev->dev_addr);
393 out:
394 	if (vlan->fwd_priv) {
395 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
396 							   vlan->fwd_priv);
397 		vlan->fwd_priv = NULL;
398 	}
399 	return err;
400 }
401 
402 static int macvlan_stop(struct net_device *dev)
403 {
404 	struct macvlan_dev *vlan = netdev_priv(dev);
405 	struct net_device *lowerdev = vlan->lowerdev;
406 
407 	if (vlan->fwd_priv) {
408 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
409 							   vlan->fwd_priv);
410 		vlan->fwd_priv = NULL;
411 		return 0;
412 	}
413 
414 	dev_uc_unsync(lowerdev, dev);
415 	dev_mc_unsync(lowerdev, dev);
416 
417 	if (vlan->port->passthru) {
418 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
419 			dev_set_promiscuity(lowerdev, -1);
420 		goto hash_del;
421 	}
422 
423 	if (dev->flags & IFF_ALLMULTI)
424 		dev_set_allmulti(lowerdev, -1);
425 
426 	dev_uc_del(lowerdev, dev->dev_addr);
427 
428 hash_del:
429 	macvlan_hash_del(vlan, !dev->dismantle);
430 	return 0;
431 }
432 
433 static int macvlan_set_mac_address(struct net_device *dev, void *p)
434 {
435 	struct macvlan_dev *vlan = netdev_priv(dev);
436 	struct net_device *lowerdev = vlan->lowerdev;
437 	struct sockaddr *addr = p;
438 	int err;
439 
440 	if (!is_valid_ether_addr(addr->sa_data))
441 		return -EADDRNOTAVAIL;
442 
443 	if (!(dev->flags & IFF_UP)) {
444 		/* Just copy in the new address */
445 		memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
446 	} else {
447 		/* Rehash and update the device filters */
448 		if (macvlan_addr_busy(vlan->port, addr->sa_data))
449 			return -EBUSY;
450 
451 		err = dev_uc_add(lowerdev, addr->sa_data);
452 		if (err)
453 			return err;
454 
455 		dev_uc_del(lowerdev, dev->dev_addr);
456 
457 		macvlan_hash_change_addr(vlan, addr->sa_data);
458 	}
459 	return 0;
460 }
461 
462 static void macvlan_change_rx_flags(struct net_device *dev, int change)
463 {
464 	struct macvlan_dev *vlan = netdev_priv(dev);
465 	struct net_device *lowerdev = vlan->lowerdev;
466 
467 	if (change & IFF_ALLMULTI)
468 		dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
469 }
470 
471 static void macvlan_set_mac_lists(struct net_device *dev)
472 {
473 	struct macvlan_dev *vlan = netdev_priv(dev);
474 
475 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
476 		bitmap_fill(vlan->mc_filter, MACVLAN_MC_FILTER_SZ);
477 	} else {
478 		struct netdev_hw_addr *ha;
479 		DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
480 
481 		bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
482 		netdev_for_each_mc_addr(ha, dev) {
483 			__set_bit(mc_hash(vlan, ha->addr), filter);
484 		}
485 
486 		__set_bit(mc_hash(vlan, dev->broadcast), filter);
487 
488 		bitmap_copy(vlan->mc_filter, filter, MACVLAN_MC_FILTER_SZ);
489 	}
490 	dev_uc_sync(vlan->lowerdev, dev);
491 	dev_mc_sync(vlan->lowerdev, dev);
492 }
493 
494 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
495 {
496 	struct macvlan_dev *vlan = netdev_priv(dev);
497 
498 	if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
499 		return -EINVAL;
500 	dev->mtu = new_mtu;
501 	return 0;
502 }
503 
504 /*
505  * macvlan network devices have devices nesting below it and are a special
506  * "super class" of normal network devices; split their locks off into a
507  * separate class since they always nest.
508  */
509 static struct lock_class_key macvlan_netdev_xmit_lock_key;
510 static struct lock_class_key macvlan_netdev_addr_lock_key;
511 
512 #define MACVLAN_FEATURES \
513 	(NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
514 	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
515 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
516 	 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
517 
518 #define MACVLAN_STATE_MASK \
519 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
520 
521 static void macvlan_set_lockdep_class_one(struct net_device *dev,
522 					  struct netdev_queue *txq,
523 					  void *_unused)
524 {
525 	lockdep_set_class(&txq->_xmit_lock,
526 			  &macvlan_netdev_xmit_lock_key);
527 }
528 
529 static void macvlan_set_lockdep_class(struct net_device *dev)
530 {
531 	lockdep_set_class(&dev->addr_list_lock,
532 			  &macvlan_netdev_addr_lock_key);
533 	netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
534 }
535 
536 static int macvlan_init(struct net_device *dev)
537 {
538 	struct macvlan_dev *vlan = netdev_priv(dev);
539 	const struct net_device *lowerdev = vlan->lowerdev;
540 	int i;
541 
542 	dev->state		= (dev->state & ~MACVLAN_STATE_MASK) |
543 				  (lowerdev->state & MACVLAN_STATE_MASK);
544 	dev->features 		= lowerdev->features & MACVLAN_FEATURES;
545 	dev->features		|= NETIF_F_LLTX;
546 	dev->gso_max_size	= lowerdev->gso_max_size;
547 	dev->iflink		= lowerdev->ifindex;
548 	dev->hard_header_len	= lowerdev->hard_header_len;
549 
550 	macvlan_set_lockdep_class(dev);
551 
552 	vlan->pcpu_stats = alloc_percpu(struct macvlan_pcpu_stats);
553 	if (!vlan->pcpu_stats)
554 		return -ENOMEM;
555 
556 	for_each_possible_cpu(i) {
557 		struct macvlan_pcpu_stats *mvlstats;
558 		mvlstats = per_cpu_ptr(vlan->pcpu_stats, i);
559 		u64_stats_init(&mvlstats->syncp);
560 	}
561 
562 	return 0;
563 }
564 
565 static void macvlan_uninit(struct net_device *dev)
566 {
567 	struct macvlan_dev *vlan = netdev_priv(dev);
568 	struct macvlan_port *port = vlan->port;
569 
570 	free_percpu(vlan->pcpu_stats);
571 
572 	port->count -= 1;
573 	if (!port->count)
574 		macvlan_port_destroy(port->dev);
575 }
576 
577 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
578 							 struct rtnl_link_stats64 *stats)
579 {
580 	struct macvlan_dev *vlan = netdev_priv(dev);
581 
582 	if (vlan->pcpu_stats) {
583 		struct macvlan_pcpu_stats *p;
584 		u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
585 		u32 rx_errors = 0, tx_dropped = 0;
586 		unsigned int start;
587 		int i;
588 
589 		for_each_possible_cpu(i) {
590 			p = per_cpu_ptr(vlan->pcpu_stats, i);
591 			do {
592 				start = u64_stats_fetch_begin_bh(&p->syncp);
593 				rx_packets	= p->rx_packets;
594 				rx_bytes	= p->rx_bytes;
595 				rx_multicast	= p->rx_multicast;
596 				tx_packets	= p->tx_packets;
597 				tx_bytes	= p->tx_bytes;
598 			} while (u64_stats_fetch_retry_bh(&p->syncp, start));
599 
600 			stats->rx_packets	+= rx_packets;
601 			stats->rx_bytes		+= rx_bytes;
602 			stats->multicast	+= rx_multicast;
603 			stats->tx_packets	+= tx_packets;
604 			stats->tx_bytes		+= tx_bytes;
605 			/* rx_errors & tx_dropped are u32, updated
606 			 * without syncp protection.
607 			 */
608 			rx_errors	+= p->rx_errors;
609 			tx_dropped	+= p->tx_dropped;
610 		}
611 		stats->rx_errors	= rx_errors;
612 		stats->rx_dropped	= rx_errors;
613 		stats->tx_dropped	= tx_dropped;
614 	}
615 	return stats;
616 }
617 
618 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
619 				   __be16 proto, u16 vid)
620 {
621 	struct macvlan_dev *vlan = netdev_priv(dev);
622 	struct net_device *lowerdev = vlan->lowerdev;
623 
624 	return vlan_vid_add(lowerdev, proto, vid);
625 }
626 
627 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
628 				    __be16 proto, u16 vid)
629 {
630 	struct macvlan_dev *vlan = netdev_priv(dev);
631 	struct net_device *lowerdev = vlan->lowerdev;
632 
633 	vlan_vid_del(lowerdev, proto, vid);
634 	return 0;
635 }
636 
637 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
638 			   struct net_device *dev,
639 			   const unsigned char *addr,
640 			   u16 flags)
641 {
642 	struct macvlan_dev *vlan = netdev_priv(dev);
643 	int err = -EINVAL;
644 
645 	if (!vlan->port->passthru)
646 		return -EOPNOTSUPP;
647 
648 	if (flags & NLM_F_REPLACE)
649 		return -EOPNOTSUPP;
650 
651 	if (is_unicast_ether_addr(addr))
652 		err = dev_uc_add_excl(dev, addr);
653 	else if (is_multicast_ether_addr(addr))
654 		err = dev_mc_add_excl(dev, addr);
655 
656 	return err;
657 }
658 
659 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
660 			   struct net_device *dev,
661 			   const unsigned char *addr)
662 {
663 	struct macvlan_dev *vlan = netdev_priv(dev);
664 	int err = -EINVAL;
665 
666 	if (!vlan->port->passthru)
667 		return -EOPNOTSUPP;
668 
669 	if (is_unicast_ether_addr(addr))
670 		err = dev_uc_del(dev, addr);
671 	else if (is_multicast_ether_addr(addr))
672 		err = dev_mc_del(dev, addr);
673 
674 	return err;
675 }
676 
677 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
678 					struct ethtool_drvinfo *drvinfo)
679 {
680 	strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
681 	strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
682 }
683 
684 static int macvlan_ethtool_get_settings(struct net_device *dev,
685 					struct ethtool_cmd *cmd)
686 {
687 	const struct macvlan_dev *vlan = netdev_priv(dev);
688 
689 	return __ethtool_get_settings(vlan->lowerdev, cmd);
690 }
691 
692 static netdev_features_t macvlan_fix_features(struct net_device *dev,
693 					      netdev_features_t features)
694 {
695 	struct macvlan_dev *vlan = netdev_priv(dev);
696 	netdev_features_t mask;
697 
698 	features |= NETIF_F_ALL_FOR_ALL;
699 	features &= (vlan->set_features | ~MACVLAN_FEATURES);
700 	mask = features;
701 
702 	features = netdev_increment_features(vlan->lowerdev->features,
703 					     features,
704 					     mask);
705 	if (!vlan->fwd_priv)
706 		features |= NETIF_F_LLTX;
707 
708 	return features;
709 }
710 
711 static const struct ethtool_ops macvlan_ethtool_ops = {
712 	.get_link		= ethtool_op_get_link,
713 	.get_settings		= macvlan_ethtool_get_settings,
714 	.get_drvinfo		= macvlan_ethtool_get_drvinfo,
715 };
716 
717 static const struct net_device_ops macvlan_netdev_ops = {
718 	.ndo_init		= macvlan_init,
719 	.ndo_uninit		= macvlan_uninit,
720 	.ndo_open		= macvlan_open,
721 	.ndo_stop		= macvlan_stop,
722 	.ndo_start_xmit		= macvlan_start_xmit,
723 	.ndo_change_mtu		= macvlan_change_mtu,
724 	.ndo_fix_features	= macvlan_fix_features,
725 	.ndo_change_rx_flags	= macvlan_change_rx_flags,
726 	.ndo_set_mac_address	= macvlan_set_mac_address,
727 	.ndo_set_rx_mode	= macvlan_set_mac_lists,
728 	.ndo_get_stats64	= macvlan_dev_get_stats64,
729 	.ndo_validate_addr	= eth_validate_addr,
730 	.ndo_vlan_rx_add_vid	= macvlan_vlan_rx_add_vid,
731 	.ndo_vlan_rx_kill_vid	= macvlan_vlan_rx_kill_vid,
732 	.ndo_fdb_add		= macvlan_fdb_add,
733 	.ndo_fdb_del		= macvlan_fdb_del,
734 	.ndo_fdb_dump		= ndo_dflt_fdb_dump,
735 };
736 
737 void macvlan_common_setup(struct net_device *dev)
738 {
739 	ether_setup(dev);
740 
741 	dev->priv_flags	       &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
742 	dev->priv_flags	       |= IFF_UNICAST_FLT;
743 	dev->netdev_ops		= &macvlan_netdev_ops;
744 	dev->destructor		= free_netdev;
745 	dev->header_ops		= &macvlan_hard_header_ops;
746 	dev->ethtool_ops	= &macvlan_ethtool_ops;
747 }
748 EXPORT_SYMBOL_GPL(macvlan_common_setup);
749 
750 static void macvlan_setup(struct net_device *dev)
751 {
752 	macvlan_common_setup(dev);
753 	dev->tx_queue_len	= 0;
754 }
755 
756 static int macvlan_port_create(struct net_device *dev)
757 {
758 	struct macvlan_port *port;
759 	unsigned int i;
760 	int err;
761 
762 	if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
763 		return -EINVAL;
764 
765 	port = kzalloc(sizeof(*port), GFP_KERNEL);
766 	if (port == NULL)
767 		return -ENOMEM;
768 
769 	port->passthru = false;
770 	port->dev = dev;
771 	INIT_LIST_HEAD(&port->vlans);
772 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
773 		INIT_HLIST_HEAD(&port->vlan_hash[i]);
774 
775 	err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
776 	if (err)
777 		kfree(port);
778 	else
779 		dev->priv_flags |= IFF_MACVLAN_PORT;
780 	return err;
781 }
782 
783 static void macvlan_port_destroy(struct net_device *dev)
784 {
785 	struct macvlan_port *port = macvlan_port_get_rtnl(dev);
786 
787 	dev->priv_flags &= ~IFF_MACVLAN_PORT;
788 	netdev_rx_handler_unregister(dev);
789 	kfree_rcu(port, rcu);
790 }
791 
792 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
793 {
794 	if (tb[IFLA_ADDRESS]) {
795 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
796 			return -EINVAL;
797 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
798 			return -EADDRNOTAVAIL;
799 	}
800 
801 	if (data && data[IFLA_MACVLAN_FLAGS] &&
802 	    nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
803 		return -EINVAL;
804 
805 	if (data && data[IFLA_MACVLAN_MODE]) {
806 		switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
807 		case MACVLAN_MODE_PRIVATE:
808 		case MACVLAN_MODE_VEPA:
809 		case MACVLAN_MODE_BRIDGE:
810 		case MACVLAN_MODE_PASSTHRU:
811 			break;
812 		default:
813 			return -EINVAL;
814 		}
815 	}
816 	return 0;
817 }
818 
819 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
820 			   struct nlattr *tb[], struct nlattr *data[],
821 			   int (*receive)(struct sk_buff *skb),
822 			   int (*forward)(struct net_device *dev,
823 					  struct sk_buff *skb))
824 {
825 	struct macvlan_dev *vlan = netdev_priv(dev);
826 	struct macvlan_port *port;
827 	struct net_device *lowerdev;
828 	int err;
829 
830 	if (!tb[IFLA_LINK])
831 		return -EINVAL;
832 
833 	lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
834 	if (lowerdev == NULL)
835 		return -ENODEV;
836 
837 	/* When creating macvlans on top of other macvlans - use
838 	 * the real device as the lowerdev.
839 	 */
840 	if (lowerdev->rtnl_link_ops == dev->rtnl_link_ops) {
841 		struct macvlan_dev *lowervlan = netdev_priv(lowerdev);
842 		lowerdev = lowervlan->lowerdev;
843 	}
844 
845 	if (!tb[IFLA_MTU])
846 		dev->mtu = lowerdev->mtu;
847 	else if (dev->mtu > lowerdev->mtu)
848 		return -EINVAL;
849 
850 	if (!tb[IFLA_ADDRESS])
851 		eth_hw_addr_random(dev);
852 
853 	if (!macvlan_port_exists(lowerdev)) {
854 		err = macvlan_port_create(lowerdev);
855 		if (err < 0)
856 			return err;
857 	}
858 	port = macvlan_port_get_rtnl(lowerdev);
859 
860 	/* Only 1 macvlan device can be created in passthru mode */
861 	if (port->passthru)
862 		return -EINVAL;
863 
864 	vlan->lowerdev = lowerdev;
865 	vlan->dev      = dev;
866 	vlan->port     = port;
867 	vlan->receive  = receive;
868 	vlan->forward  = forward;
869 	vlan->set_features = MACVLAN_FEATURES;
870 
871 	vlan->mode     = MACVLAN_MODE_VEPA;
872 	if (data && data[IFLA_MACVLAN_MODE])
873 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
874 
875 	if (data && data[IFLA_MACVLAN_FLAGS])
876 		vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
877 
878 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
879 		if (port->count)
880 			return -EINVAL;
881 		port->passthru = true;
882 		eth_hw_addr_inherit(dev, lowerdev);
883 	}
884 
885 	port->count += 1;
886 	err = register_netdevice(dev);
887 	if (err < 0)
888 		goto destroy_port;
889 
890 	dev->priv_flags |= IFF_MACVLAN;
891 	err = netdev_upper_dev_link(lowerdev, dev);
892 	if (err)
893 		goto destroy_port;
894 
895 
896 	list_add_tail_rcu(&vlan->list, &port->vlans);
897 	netif_stacked_transfer_operstate(lowerdev, dev);
898 
899 	return 0;
900 
901 destroy_port:
902 	port->count -= 1;
903 	if (!port->count)
904 		macvlan_port_destroy(lowerdev);
905 
906 	return err;
907 }
908 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
909 
910 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
911 			   struct nlattr *tb[], struct nlattr *data[])
912 {
913 	return macvlan_common_newlink(src_net, dev, tb, data,
914 				      netif_rx,
915 				      dev_forward_skb);
916 }
917 
918 void macvlan_dellink(struct net_device *dev, struct list_head *head)
919 {
920 	struct macvlan_dev *vlan = netdev_priv(dev);
921 
922 	list_del_rcu(&vlan->list);
923 	unregister_netdevice_queue(dev, head);
924 	netdev_upper_dev_unlink(vlan->lowerdev, dev);
925 }
926 EXPORT_SYMBOL_GPL(macvlan_dellink);
927 
928 static int macvlan_changelink(struct net_device *dev,
929 		struct nlattr *tb[], struct nlattr *data[])
930 {
931 	struct macvlan_dev *vlan = netdev_priv(dev);
932 	enum macvlan_mode mode;
933 	bool set_mode = false;
934 
935 	/* Validate mode, but don't set yet: setting flags may fail. */
936 	if (data && data[IFLA_MACVLAN_MODE]) {
937 		set_mode = true;
938 		mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
939 		/* Passthrough mode can't be set or cleared dynamically */
940 		if ((mode == MACVLAN_MODE_PASSTHRU) !=
941 		    (vlan->mode == MACVLAN_MODE_PASSTHRU))
942 			return -EINVAL;
943 	}
944 
945 	if (data && data[IFLA_MACVLAN_FLAGS]) {
946 		__u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
947 		bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
948 		if (vlan->port->passthru && promisc) {
949 			int err;
950 
951 			if (flags & MACVLAN_FLAG_NOPROMISC)
952 				err = dev_set_promiscuity(vlan->lowerdev, -1);
953 			else
954 				err = dev_set_promiscuity(vlan->lowerdev, 1);
955 			if (err < 0)
956 				return err;
957 		}
958 		vlan->flags = flags;
959 	}
960 	if (set_mode)
961 		vlan->mode = mode;
962 	return 0;
963 }
964 
965 static size_t macvlan_get_size(const struct net_device *dev)
966 {
967 	return (0
968 		+ nla_total_size(4) /* IFLA_MACVLAN_MODE */
969 		+ nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
970 		);
971 }
972 
973 static int macvlan_fill_info(struct sk_buff *skb,
974 				const struct net_device *dev)
975 {
976 	struct macvlan_dev *vlan = netdev_priv(dev);
977 
978 	if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
979 		goto nla_put_failure;
980 	if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
981 		goto nla_put_failure;
982 	return 0;
983 
984 nla_put_failure:
985 	return -EMSGSIZE;
986 }
987 
988 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
989 	[IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
990 	[IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
991 };
992 
993 int macvlan_link_register(struct rtnl_link_ops *ops)
994 {
995 	/* common fields */
996 	ops->priv_size		= sizeof(struct macvlan_dev);
997 	ops->validate		= macvlan_validate;
998 	ops->maxtype		= IFLA_MACVLAN_MAX;
999 	ops->policy		= macvlan_policy;
1000 	ops->changelink		= macvlan_changelink;
1001 	ops->get_size		= macvlan_get_size;
1002 	ops->fill_info		= macvlan_fill_info;
1003 
1004 	return rtnl_link_register(ops);
1005 };
1006 EXPORT_SYMBOL_GPL(macvlan_link_register);
1007 
1008 static struct rtnl_link_ops macvlan_link_ops = {
1009 	.kind		= "macvlan",
1010 	.setup		= macvlan_setup,
1011 	.newlink	= macvlan_newlink,
1012 	.dellink	= macvlan_dellink,
1013 };
1014 
1015 static int macvlan_device_event(struct notifier_block *unused,
1016 				unsigned long event, void *ptr)
1017 {
1018 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1019 	struct macvlan_dev *vlan, *next;
1020 	struct macvlan_port *port;
1021 	LIST_HEAD(list_kill);
1022 
1023 	if (!macvlan_port_exists(dev))
1024 		return NOTIFY_DONE;
1025 
1026 	port = macvlan_port_get_rtnl(dev);
1027 
1028 	switch (event) {
1029 	case NETDEV_CHANGE:
1030 		list_for_each_entry(vlan, &port->vlans, list)
1031 			netif_stacked_transfer_operstate(vlan->lowerdev,
1032 							 vlan->dev);
1033 		break;
1034 	case NETDEV_FEAT_CHANGE:
1035 		list_for_each_entry(vlan, &port->vlans, list) {
1036 			vlan->dev->gso_max_size = dev->gso_max_size;
1037 			netdev_update_features(vlan->dev);
1038 		}
1039 		break;
1040 	case NETDEV_UNREGISTER:
1041 		/* twiddle thumbs on netns device moves */
1042 		if (dev->reg_state != NETREG_UNREGISTERING)
1043 			break;
1044 
1045 		list_for_each_entry_safe(vlan, next, &port->vlans, list)
1046 			vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1047 		unregister_netdevice_many(&list_kill);
1048 		list_del(&list_kill);
1049 		break;
1050 	case NETDEV_PRE_TYPE_CHANGE:
1051 		/* Forbid underlaying device to change its type. */
1052 		return NOTIFY_BAD;
1053 	}
1054 	return NOTIFY_DONE;
1055 }
1056 
1057 static struct notifier_block macvlan_notifier_block __read_mostly = {
1058 	.notifier_call	= macvlan_device_event,
1059 };
1060 
1061 static int __init macvlan_init_module(void)
1062 {
1063 	int err;
1064 
1065 	register_netdevice_notifier(&macvlan_notifier_block);
1066 
1067 	err = macvlan_link_register(&macvlan_link_ops);
1068 	if (err < 0)
1069 		goto err1;
1070 	return 0;
1071 err1:
1072 	unregister_netdevice_notifier(&macvlan_notifier_block);
1073 	return err;
1074 }
1075 
1076 static void __exit macvlan_cleanup_module(void)
1077 {
1078 	rtnl_link_unregister(&macvlan_link_ops);
1079 	unregister_netdevice_notifier(&macvlan_notifier_block);
1080 }
1081 
1082 module_init(macvlan_init_module);
1083 module_exit(macvlan_cleanup_module);
1084 
1085 MODULE_LICENSE("GPL");
1086 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1087 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1088 MODULE_ALIAS_RTNL_LINK("macvlan");
1089