xref: /linux/drivers/net/macvlan.c (revision e676f197a7a9aae9c75b0d9acc97e07de07dd1f0)
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 <linux/workqueue.h>
34 #include <net/rtnetlink.h>
35 #include <net/xfrm.h>
36 
37 #define MACVLAN_HASH_SIZE	(1 << BITS_PER_BYTE)
38 
39 struct macvlan_port {
40 	struct net_device	*dev;
41 	struct hlist_head	vlan_hash[MACVLAN_HASH_SIZE];
42 	struct list_head	vlans;
43 	struct rcu_head		rcu;
44 	struct sk_buff_head	bc_queue;
45 	struct work_struct	bc_work;
46 	bool 			passthru;
47 	int			count;
48 };
49 
50 struct macvlan_skb_cb {
51 	const struct macvlan_dev *src;
52 };
53 
54 #define MACVLAN_SKB_CB(__skb) ((struct macvlan_skb_cb *)&((__skb)->cb[0]))
55 
56 static void macvlan_port_destroy(struct net_device *dev);
57 
58 static struct macvlan_port *macvlan_port_get_rcu(const struct net_device *dev)
59 {
60 	return rcu_dereference(dev->rx_handler_data);
61 }
62 
63 static struct macvlan_port *macvlan_port_get_rtnl(const struct net_device *dev)
64 {
65 	return rtnl_dereference(dev->rx_handler_data);
66 }
67 
68 #define macvlan_port_exists(dev) (dev->priv_flags & IFF_MACVLAN_PORT)
69 
70 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
71 					       const unsigned char *addr)
72 {
73 	struct macvlan_dev *vlan;
74 
75 	hlist_for_each_entry_rcu(vlan, &port->vlan_hash[addr[5]], hlist) {
76 		if (ether_addr_equal_64bits(vlan->dev->dev_addr, addr))
77 			return vlan;
78 	}
79 	return NULL;
80 }
81 
82 static void macvlan_hash_add(struct macvlan_dev *vlan)
83 {
84 	struct macvlan_port *port = vlan->port;
85 	const unsigned char *addr = vlan->dev->dev_addr;
86 
87 	hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[addr[5]]);
88 }
89 
90 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync)
91 {
92 	hlist_del_rcu(&vlan->hlist);
93 	if (sync)
94 		synchronize_rcu();
95 }
96 
97 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
98 					const unsigned char *addr)
99 {
100 	macvlan_hash_del(vlan, true);
101 	/* Now that we are unhashed it is safe to change the device
102 	 * address without confusing packet delivery.
103 	 */
104 	memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
105 	macvlan_hash_add(vlan);
106 }
107 
108 static int macvlan_addr_busy(const struct macvlan_port *port,
109 				const unsigned char *addr)
110 {
111 	/* Test to see if the specified multicast address is
112 	 * currently in use by the underlying device or
113 	 * another macvlan.
114 	 */
115 	if (ether_addr_equal_64bits(port->dev->dev_addr, addr))
116 		return 1;
117 
118 	if (macvlan_hash_lookup(port, addr))
119 		return 1;
120 
121 	return 0;
122 }
123 
124 
125 static int macvlan_broadcast_one(struct sk_buff *skb,
126 				 const struct macvlan_dev *vlan,
127 				 const struct ethhdr *eth, bool local)
128 {
129 	struct net_device *dev = vlan->dev;
130 
131 	if (local)
132 		return __dev_forward_skb(dev, skb);
133 
134 	skb->dev = dev;
135 	if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
136 		skb->pkt_type = PACKET_BROADCAST;
137 	else
138 		skb->pkt_type = PACKET_MULTICAST;
139 
140 	return 0;
141 }
142 
143 static u32 macvlan_hash_mix(const struct macvlan_dev *vlan)
144 {
145 	return (u32)(((unsigned long)vlan) >> L1_CACHE_SHIFT);
146 }
147 
148 
149 static unsigned int mc_hash(const struct macvlan_dev *vlan,
150 			    const unsigned char *addr)
151 {
152 	u32 val = __get_unaligned_cpu32(addr + 2);
153 
154 	val ^= macvlan_hash_mix(vlan);
155 	return hash_32(val, MACVLAN_MC_FILTER_BITS);
156 }
157 
158 static void macvlan_broadcast(struct sk_buff *skb,
159 			      const struct macvlan_port *port,
160 			      struct net_device *src,
161 			      enum macvlan_mode mode)
162 {
163 	const struct ethhdr *eth = eth_hdr(skb);
164 	const struct macvlan_dev *vlan;
165 	struct sk_buff *nskb;
166 	unsigned int i;
167 	int err;
168 	unsigned int hash;
169 
170 	if (skb->protocol == htons(ETH_P_PAUSE))
171 		return;
172 
173 	for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
174 		hlist_for_each_entry_rcu(vlan, &port->vlan_hash[i], hlist) {
175 			if (vlan->dev == src || !(vlan->mode & mode))
176 				continue;
177 
178 			hash = mc_hash(vlan, eth->h_dest);
179 			if (!test_bit(hash, vlan->mc_filter))
180 				continue;
181 
182 			err = NET_RX_DROP;
183 			nskb = skb_clone(skb, GFP_ATOMIC);
184 			if (likely(nskb))
185 				err = macvlan_broadcast_one(
186 					nskb, vlan, eth,
187 					mode == MACVLAN_MODE_BRIDGE) ?:
188 				      netif_rx_ni(nskb);
189 			macvlan_count_rx(vlan, skb->len + ETH_HLEN,
190 					 err == NET_RX_SUCCESS, 1);
191 		}
192 	}
193 }
194 
195 static void macvlan_process_broadcast(struct work_struct *w)
196 {
197 	struct macvlan_port *port = container_of(w, struct macvlan_port,
198 						 bc_work);
199 	struct sk_buff *skb;
200 	struct sk_buff_head list;
201 
202 	skb_queue_head_init(&list);
203 
204 	spin_lock_bh(&port->bc_queue.lock);
205 	skb_queue_splice_tail_init(&port->bc_queue, &list);
206 	spin_unlock_bh(&port->bc_queue.lock);
207 
208 	while ((skb = __skb_dequeue(&list))) {
209 		const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
210 
211 		rcu_read_lock();
212 
213 		if (!src)
214 			/* frame comes from an external address */
215 			macvlan_broadcast(skb, port, NULL,
216 					  MACVLAN_MODE_PRIVATE |
217 					  MACVLAN_MODE_VEPA    |
218 					  MACVLAN_MODE_PASSTHRU|
219 					  MACVLAN_MODE_BRIDGE);
220 		else if (src->mode == MACVLAN_MODE_VEPA)
221 			/* flood to everyone except source */
222 			macvlan_broadcast(skb, port, src->dev,
223 					  MACVLAN_MODE_VEPA |
224 					  MACVLAN_MODE_BRIDGE);
225 		else
226 			/*
227 			 * flood only to VEPA ports, bridge ports
228 			 * already saw the frame on the way out.
229 			 */
230 			macvlan_broadcast(skb, port, src->dev,
231 					  MACVLAN_MODE_VEPA);
232 
233 		rcu_read_unlock();
234 
235 		kfree_skb(skb);
236 	}
237 }
238 
239 static void macvlan_broadcast_enqueue(struct macvlan_port *port,
240 				      struct sk_buff *skb)
241 {
242 	struct sk_buff *nskb;
243 	int err = -ENOMEM;
244 
245 	nskb = skb_clone(skb, GFP_ATOMIC);
246 	if (!nskb)
247 		goto err;
248 
249 	spin_lock(&port->bc_queue.lock);
250 	if (skb_queue_len(&port->bc_queue) < skb->dev->tx_queue_len) {
251 		__skb_queue_tail(&port->bc_queue, nskb);
252 		err = 0;
253 	}
254 	spin_unlock(&port->bc_queue.lock);
255 
256 	if (err)
257 		goto free_nskb;
258 
259 	schedule_work(&port->bc_work);
260 	return;
261 
262 free_nskb:
263 	kfree_skb(nskb);
264 err:
265 	atomic_long_inc(&skb->dev->rx_dropped);
266 }
267 
268 /* called under rcu_read_lock() from netif_receive_skb */
269 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
270 {
271 	struct macvlan_port *port;
272 	struct sk_buff *skb = *pskb;
273 	const struct ethhdr *eth = eth_hdr(skb);
274 	const struct macvlan_dev *vlan;
275 	const struct macvlan_dev *src;
276 	struct net_device *dev;
277 	unsigned int len = 0;
278 	int ret = NET_RX_DROP;
279 
280 	port = macvlan_port_get_rcu(skb->dev);
281 	if (is_multicast_ether_addr(eth->h_dest)) {
282 		skb = ip_check_defrag(skb, IP_DEFRAG_MACVLAN);
283 		if (!skb)
284 			return RX_HANDLER_CONSUMED;
285 		eth = eth_hdr(skb);
286 		src = macvlan_hash_lookup(port, eth->h_source);
287 		if (src && src->mode != MACVLAN_MODE_VEPA &&
288 		    src->mode != MACVLAN_MODE_BRIDGE) {
289 			/* forward to original port. */
290 			vlan = src;
291 			ret = macvlan_broadcast_one(skb, vlan, eth, 0) ?:
292 			      netif_rx(skb);
293 			goto out;
294 		}
295 
296 		MACVLAN_SKB_CB(skb)->src = src;
297 		macvlan_broadcast_enqueue(port, skb);
298 
299 		return RX_HANDLER_PASS;
300 	}
301 
302 	if (port->passthru)
303 		vlan = list_first_or_null_rcu(&port->vlans,
304 					      struct macvlan_dev, list);
305 	else
306 		vlan = macvlan_hash_lookup(port, eth->h_dest);
307 	if (vlan == NULL)
308 		return RX_HANDLER_PASS;
309 
310 	dev = vlan->dev;
311 	if (unlikely(!(dev->flags & IFF_UP))) {
312 		kfree_skb(skb);
313 		return RX_HANDLER_CONSUMED;
314 	}
315 	len = skb->len + ETH_HLEN;
316 	skb = skb_share_check(skb, GFP_ATOMIC);
317 	if (!skb)
318 		goto out;
319 
320 	skb->dev = dev;
321 	skb->pkt_type = PACKET_HOST;
322 
323 	ret = netif_rx(skb);
324 
325 out:
326 	macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, 0);
327 	return RX_HANDLER_CONSUMED;
328 }
329 
330 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
331 {
332 	const struct macvlan_dev *vlan = netdev_priv(dev);
333 	const struct macvlan_port *port = vlan->port;
334 	const struct macvlan_dev *dest;
335 	__u8 ip_summed = skb->ip_summed;
336 
337 	if (vlan->mode == MACVLAN_MODE_BRIDGE) {
338 		const struct ethhdr *eth = (void *)skb->data;
339 		skb->ip_summed = CHECKSUM_UNNECESSARY;
340 
341 		/* send to other bridge ports directly */
342 		if (is_multicast_ether_addr(eth->h_dest)) {
343 			macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
344 			goto xmit_world;
345 		}
346 
347 		dest = macvlan_hash_lookup(port, eth->h_dest);
348 		if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
349 			/* send to lowerdev first for its network taps */
350 			dev_forward_skb(vlan->lowerdev, skb);
351 
352 			return NET_XMIT_SUCCESS;
353 		}
354 	}
355 
356 xmit_world:
357 	skb->ip_summed = ip_summed;
358 	skb->dev = vlan->lowerdev;
359 	return dev_queue_xmit(skb);
360 }
361 
362 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
363 				      struct net_device *dev)
364 {
365 	unsigned int len = skb->len;
366 	int ret;
367 	const struct macvlan_dev *vlan = netdev_priv(dev);
368 
369 	if (vlan->fwd_priv) {
370 		skb->dev = vlan->lowerdev;
371 		ret = dev_queue_xmit_accel(skb, vlan->fwd_priv);
372 	} else {
373 		ret = macvlan_queue_xmit(skb, dev);
374 	}
375 
376 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
377 		struct vlan_pcpu_stats *pcpu_stats;
378 
379 		pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
380 		u64_stats_update_begin(&pcpu_stats->syncp);
381 		pcpu_stats->tx_packets++;
382 		pcpu_stats->tx_bytes += len;
383 		u64_stats_update_end(&pcpu_stats->syncp);
384 	} else {
385 		this_cpu_inc(vlan->pcpu_stats->tx_dropped);
386 	}
387 	return ret;
388 }
389 
390 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
391 			       unsigned short type, const void *daddr,
392 			       const void *saddr, unsigned len)
393 {
394 	const struct macvlan_dev *vlan = netdev_priv(dev);
395 	struct net_device *lowerdev = vlan->lowerdev;
396 
397 	return dev_hard_header(skb, lowerdev, type, daddr,
398 			       saddr ? : dev->dev_addr, len);
399 }
400 
401 static const struct header_ops macvlan_hard_header_ops = {
402 	.create  	= macvlan_hard_header,
403 	.rebuild	= eth_rebuild_header,
404 	.parse		= eth_header_parse,
405 	.cache		= eth_header_cache,
406 	.cache_update	= eth_header_cache_update,
407 };
408 
409 static struct rtnl_link_ops macvlan_link_ops;
410 
411 static int macvlan_open(struct net_device *dev)
412 {
413 	struct macvlan_dev *vlan = netdev_priv(dev);
414 	struct net_device *lowerdev = vlan->lowerdev;
415 	int err;
416 
417 	if (vlan->port->passthru) {
418 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
419 			err = dev_set_promiscuity(lowerdev, 1);
420 			if (err < 0)
421 				goto out;
422 		}
423 		goto hash_add;
424 	}
425 
426 	if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD &&
427 	    dev->rtnl_link_ops == &macvlan_link_ops) {
428 		vlan->fwd_priv =
429 		      lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
430 
431 		/* If we get a NULL pointer back, or if we get an error
432 		 * then we should just fall through to the non accelerated path
433 		 */
434 		if (IS_ERR_OR_NULL(vlan->fwd_priv)) {
435 			vlan->fwd_priv = NULL;
436 		} else
437 			return 0;
438 	}
439 
440 	err = -EBUSY;
441 	if (macvlan_addr_busy(vlan->port, dev->dev_addr))
442 		goto out;
443 
444 	err = dev_uc_add(lowerdev, dev->dev_addr);
445 	if (err < 0)
446 		goto out;
447 	if (dev->flags & IFF_ALLMULTI) {
448 		err = dev_set_allmulti(lowerdev, 1);
449 		if (err < 0)
450 			goto del_unicast;
451 	}
452 
453 hash_add:
454 	macvlan_hash_add(vlan);
455 	return 0;
456 
457 del_unicast:
458 	dev_uc_del(lowerdev, dev->dev_addr);
459 out:
460 	if (vlan->fwd_priv) {
461 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
462 							   vlan->fwd_priv);
463 		vlan->fwd_priv = NULL;
464 	}
465 	return err;
466 }
467 
468 static int macvlan_stop(struct net_device *dev)
469 {
470 	struct macvlan_dev *vlan = netdev_priv(dev);
471 	struct net_device *lowerdev = vlan->lowerdev;
472 
473 	if (vlan->fwd_priv) {
474 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
475 							   vlan->fwd_priv);
476 		vlan->fwd_priv = NULL;
477 		return 0;
478 	}
479 
480 	dev_uc_unsync(lowerdev, dev);
481 	dev_mc_unsync(lowerdev, dev);
482 
483 	if (vlan->port->passthru) {
484 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
485 			dev_set_promiscuity(lowerdev, -1);
486 		goto hash_del;
487 	}
488 
489 	if (dev->flags & IFF_ALLMULTI)
490 		dev_set_allmulti(lowerdev, -1);
491 
492 	dev_uc_del(lowerdev, dev->dev_addr);
493 
494 hash_del:
495 	macvlan_hash_del(vlan, !dev->dismantle);
496 	return 0;
497 }
498 
499 static int macvlan_set_mac_address(struct net_device *dev, void *p)
500 {
501 	struct macvlan_dev *vlan = netdev_priv(dev);
502 	struct net_device *lowerdev = vlan->lowerdev;
503 	struct sockaddr *addr = p;
504 	int err;
505 
506 	if (!is_valid_ether_addr(addr->sa_data))
507 		return -EADDRNOTAVAIL;
508 
509 	if (!(dev->flags & IFF_UP)) {
510 		/* Just copy in the new address */
511 		memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
512 	} else {
513 		/* Rehash and update the device filters */
514 		if (macvlan_addr_busy(vlan->port, addr->sa_data))
515 			return -EBUSY;
516 
517 		err = dev_uc_add(lowerdev, addr->sa_data);
518 		if (err)
519 			return err;
520 
521 		dev_uc_del(lowerdev, dev->dev_addr);
522 
523 		macvlan_hash_change_addr(vlan, addr->sa_data);
524 	}
525 	return 0;
526 }
527 
528 static void macvlan_change_rx_flags(struct net_device *dev, int change)
529 {
530 	struct macvlan_dev *vlan = netdev_priv(dev);
531 	struct net_device *lowerdev = vlan->lowerdev;
532 
533 	if (change & IFF_ALLMULTI)
534 		dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
535 }
536 
537 static void macvlan_set_mac_lists(struct net_device *dev)
538 {
539 	struct macvlan_dev *vlan = netdev_priv(dev);
540 
541 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
542 		bitmap_fill(vlan->mc_filter, MACVLAN_MC_FILTER_SZ);
543 	} else {
544 		struct netdev_hw_addr *ha;
545 		DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
546 
547 		bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
548 		netdev_for_each_mc_addr(ha, dev) {
549 			__set_bit(mc_hash(vlan, ha->addr), filter);
550 		}
551 
552 		__set_bit(mc_hash(vlan, dev->broadcast), filter);
553 
554 		bitmap_copy(vlan->mc_filter, filter, MACVLAN_MC_FILTER_SZ);
555 	}
556 	dev_uc_sync(vlan->lowerdev, dev);
557 	dev_mc_sync(vlan->lowerdev, dev);
558 }
559 
560 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
561 {
562 	struct macvlan_dev *vlan = netdev_priv(dev);
563 
564 	if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
565 		return -EINVAL;
566 	dev->mtu = new_mtu;
567 	return 0;
568 }
569 
570 /*
571  * macvlan network devices have devices nesting below it and are a special
572  * "super class" of normal network devices; split their locks off into a
573  * separate class since they always nest.
574  */
575 static struct lock_class_key macvlan_netdev_xmit_lock_key;
576 static struct lock_class_key macvlan_netdev_addr_lock_key;
577 
578 #define ALWAYS_ON_FEATURES \
579 	(NETIF_F_SG | NETIF_F_GEN_CSUM | NETIF_F_GSO_SOFTWARE | NETIF_F_LLTX)
580 
581 #define MACVLAN_FEATURES \
582 	(NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
583 	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
584 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
585 	 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
586 
587 #define MACVLAN_STATE_MASK \
588 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
589 
590 static void macvlan_set_lockdep_class_one(struct net_device *dev,
591 					  struct netdev_queue *txq,
592 					  void *_unused)
593 {
594 	lockdep_set_class(&txq->_xmit_lock,
595 			  &macvlan_netdev_xmit_lock_key);
596 }
597 
598 static void macvlan_set_lockdep_class(struct net_device *dev)
599 {
600 	lockdep_set_class(&dev->addr_list_lock,
601 			  &macvlan_netdev_addr_lock_key);
602 	netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
603 }
604 
605 static int macvlan_init(struct net_device *dev)
606 {
607 	struct macvlan_dev *vlan = netdev_priv(dev);
608 	const struct net_device *lowerdev = vlan->lowerdev;
609 
610 	dev->state		= (dev->state & ~MACVLAN_STATE_MASK) |
611 				  (lowerdev->state & MACVLAN_STATE_MASK);
612 	dev->features 		= lowerdev->features & MACVLAN_FEATURES;
613 	dev->features		|= ALWAYS_ON_FEATURES;
614 	dev->gso_max_size	= lowerdev->gso_max_size;
615 	dev->iflink		= lowerdev->ifindex;
616 	dev->hard_header_len	= lowerdev->hard_header_len;
617 
618 	macvlan_set_lockdep_class(dev);
619 
620 	vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
621 	if (!vlan->pcpu_stats)
622 		return -ENOMEM;
623 
624 	return 0;
625 }
626 
627 static void macvlan_uninit(struct net_device *dev)
628 {
629 	struct macvlan_dev *vlan = netdev_priv(dev);
630 	struct macvlan_port *port = vlan->port;
631 
632 	free_percpu(vlan->pcpu_stats);
633 
634 	port->count -= 1;
635 	if (!port->count)
636 		macvlan_port_destroy(port->dev);
637 }
638 
639 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
640 							 struct rtnl_link_stats64 *stats)
641 {
642 	struct macvlan_dev *vlan = netdev_priv(dev);
643 
644 	if (vlan->pcpu_stats) {
645 		struct vlan_pcpu_stats *p;
646 		u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
647 		u32 rx_errors = 0, tx_dropped = 0;
648 		unsigned int start;
649 		int i;
650 
651 		for_each_possible_cpu(i) {
652 			p = per_cpu_ptr(vlan->pcpu_stats, i);
653 			do {
654 				start = u64_stats_fetch_begin_irq(&p->syncp);
655 				rx_packets	= p->rx_packets;
656 				rx_bytes	= p->rx_bytes;
657 				rx_multicast	= p->rx_multicast;
658 				tx_packets	= p->tx_packets;
659 				tx_bytes	= p->tx_bytes;
660 			} while (u64_stats_fetch_retry_irq(&p->syncp, start));
661 
662 			stats->rx_packets	+= rx_packets;
663 			stats->rx_bytes		+= rx_bytes;
664 			stats->multicast	+= rx_multicast;
665 			stats->tx_packets	+= tx_packets;
666 			stats->tx_bytes		+= tx_bytes;
667 			/* rx_errors & tx_dropped are u32, updated
668 			 * without syncp protection.
669 			 */
670 			rx_errors	+= p->rx_errors;
671 			tx_dropped	+= p->tx_dropped;
672 		}
673 		stats->rx_errors	= rx_errors;
674 		stats->rx_dropped	= rx_errors;
675 		stats->tx_dropped	= tx_dropped;
676 	}
677 	return stats;
678 }
679 
680 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
681 				   __be16 proto, u16 vid)
682 {
683 	struct macvlan_dev *vlan = netdev_priv(dev);
684 	struct net_device *lowerdev = vlan->lowerdev;
685 
686 	return vlan_vid_add(lowerdev, proto, vid);
687 }
688 
689 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
690 				    __be16 proto, u16 vid)
691 {
692 	struct macvlan_dev *vlan = netdev_priv(dev);
693 	struct net_device *lowerdev = vlan->lowerdev;
694 
695 	vlan_vid_del(lowerdev, proto, vid);
696 	return 0;
697 }
698 
699 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
700 			   struct net_device *dev,
701 			   const unsigned char *addr,
702 			   u16 flags)
703 {
704 	struct macvlan_dev *vlan = netdev_priv(dev);
705 	int err = -EINVAL;
706 
707 	if (!vlan->port->passthru)
708 		return -EOPNOTSUPP;
709 
710 	if (flags & NLM_F_REPLACE)
711 		return -EOPNOTSUPP;
712 
713 	if (is_unicast_ether_addr(addr))
714 		err = dev_uc_add_excl(dev, addr);
715 	else if (is_multicast_ether_addr(addr))
716 		err = dev_mc_add_excl(dev, addr);
717 
718 	return err;
719 }
720 
721 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
722 			   struct net_device *dev,
723 			   const unsigned char *addr)
724 {
725 	struct macvlan_dev *vlan = netdev_priv(dev);
726 	int err = -EINVAL;
727 
728 	if (!vlan->port->passthru)
729 		return -EOPNOTSUPP;
730 
731 	if (is_unicast_ether_addr(addr))
732 		err = dev_uc_del(dev, addr);
733 	else if (is_multicast_ether_addr(addr))
734 		err = dev_mc_del(dev, addr);
735 
736 	return err;
737 }
738 
739 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
740 					struct ethtool_drvinfo *drvinfo)
741 {
742 	strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
743 	strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
744 }
745 
746 static int macvlan_ethtool_get_settings(struct net_device *dev,
747 					struct ethtool_cmd *cmd)
748 {
749 	const struct macvlan_dev *vlan = netdev_priv(dev);
750 
751 	return __ethtool_get_settings(vlan->lowerdev, cmd);
752 }
753 
754 static netdev_features_t macvlan_fix_features(struct net_device *dev,
755 					      netdev_features_t features)
756 {
757 	struct macvlan_dev *vlan = netdev_priv(dev);
758 	netdev_features_t mask;
759 
760 	features |= NETIF_F_ALL_FOR_ALL;
761 	features &= (vlan->set_features | ~MACVLAN_FEATURES);
762 	mask = features;
763 
764 	features = netdev_increment_features(vlan->lowerdev->features,
765 					     features,
766 					     mask);
767 	features |= ALWAYS_ON_FEATURES;
768 
769 	return features;
770 }
771 
772 static const struct ethtool_ops macvlan_ethtool_ops = {
773 	.get_link		= ethtool_op_get_link,
774 	.get_settings		= macvlan_ethtool_get_settings,
775 	.get_drvinfo		= macvlan_ethtool_get_drvinfo,
776 };
777 
778 static const struct net_device_ops macvlan_netdev_ops = {
779 	.ndo_init		= macvlan_init,
780 	.ndo_uninit		= macvlan_uninit,
781 	.ndo_open		= macvlan_open,
782 	.ndo_stop		= macvlan_stop,
783 	.ndo_start_xmit		= macvlan_start_xmit,
784 	.ndo_change_mtu		= macvlan_change_mtu,
785 	.ndo_fix_features	= macvlan_fix_features,
786 	.ndo_change_rx_flags	= macvlan_change_rx_flags,
787 	.ndo_set_mac_address	= macvlan_set_mac_address,
788 	.ndo_set_rx_mode	= macvlan_set_mac_lists,
789 	.ndo_get_stats64	= macvlan_dev_get_stats64,
790 	.ndo_validate_addr	= eth_validate_addr,
791 	.ndo_vlan_rx_add_vid	= macvlan_vlan_rx_add_vid,
792 	.ndo_vlan_rx_kill_vid	= macvlan_vlan_rx_kill_vid,
793 	.ndo_fdb_add		= macvlan_fdb_add,
794 	.ndo_fdb_del		= macvlan_fdb_del,
795 	.ndo_fdb_dump		= ndo_dflt_fdb_dump,
796 };
797 
798 void macvlan_common_setup(struct net_device *dev)
799 {
800 	ether_setup(dev);
801 
802 	dev->priv_flags	       &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
803 	dev->priv_flags	       |= IFF_UNICAST_FLT;
804 	dev->netdev_ops		= &macvlan_netdev_ops;
805 	dev->destructor		= free_netdev;
806 	dev->header_ops		= &macvlan_hard_header_ops;
807 	dev->ethtool_ops	= &macvlan_ethtool_ops;
808 }
809 EXPORT_SYMBOL_GPL(macvlan_common_setup);
810 
811 static void macvlan_setup(struct net_device *dev)
812 {
813 	macvlan_common_setup(dev);
814 	dev->tx_queue_len	= 0;
815 }
816 
817 static int macvlan_port_create(struct net_device *dev)
818 {
819 	struct macvlan_port *port;
820 	unsigned int i;
821 	int err;
822 
823 	if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
824 		return -EINVAL;
825 
826 	port = kzalloc(sizeof(*port), GFP_KERNEL);
827 	if (port == NULL)
828 		return -ENOMEM;
829 
830 	port->passthru = false;
831 	port->dev = dev;
832 	INIT_LIST_HEAD(&port->vlans);
833 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
834 		INIT_HLIST_HEAD(&port->vlan_hash[i]);
835 
836 	skb_queue_head_init(&port->bc_queue);
837 	INIT_WORK(&port->bc_work, macvlan_process_broadcast);
838 
839 	err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
840 	if (err)
841 		kfree(port);
842 	else
843 		dev->priv_flags |= IFF_MACVLAN_PORT;
844 	return err;
845 }
846 
847 static void macvlan_port_destroy(struct net_device *dev)
848 {
849 	struct macvlan_port *port = macvlan_port_get_rtnl(dev);
850 
851 	cancel_work_sync(&port->bc_work);
852 	dev->priv_flags &= ~IFF_MACVLAN_PORT;
853 	netdev_rx_handler_unregister(dev);
854 	kfree_rcu(port, rcu);
855 }
856 
857 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
858 {
859 	if (tb[IFLA_ADDRESS]) {
860 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
861 			return -EINVAL;
862 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
863 			return -EADDRNOTAVAIL;
864 	}
865 
866 	if (data && data[IFLA_MACVLAN_FLAGS] &&
867 	    nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
868 		return -EINVAL;
869 
870 	if (data && data[IFLA_MACVLAN_MODE]) {
871 		switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
872 		case MACVLAN_MODE_PRIVATE:
873 		case MACVLAN_MODE_VEPA:
874 		case MACVLAN_MODE_BRIDGE:
875 		case MACVLAN_MODE_PASSTHRU:
876 			break;
877 		default:
878 			return -EINVAL;
879 		}
880 	}
881 	return 0;
882 }
883 
884 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
885 			   struct nlattr *tb[], struct nlattr *data[])
886 {
887 	struct macvlan_dev *vlan = netdev_priv(dev);
888 	struct macvlan_port *port;
889 	struct net_device *lowerdev;
890 	int err;
891 
892 	if (!tb[IFLA_LINK])
893 		return -EINVAL;
894 
895 	lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
896 	if (lowerdev == NULL)
897 		return -ENODEV;
898 
899 	/* When creating macvlans or macvtaps on top of other macvlans - use
900 	 * the real device as the lowerdev.
901 	 */
902 	if (netif_is_macvlan(lowerdev))
903 		lowerdev = macvlan_dev_real_dev(lowerdev);
904 
905 	if (!tb[IFLA_MTU])
906 		dev->mtu = lowerdev->mtu;
907 	else if (dev->mtu > lowerdev->mtu)
908 		return -EINVAL;
909 
910 	if (!tb[IFLA_ADDRESS])
911 		eth_hw_addr_random(dev);
912 
913 	if (!macvlan_port_exists(lowerdev)) {
914 		err = macvlan_port_create(lowerdev);
915 		if (err < 0)
916 			return err;
917 	}
918 	port = macvlan_port_get_rtnl(lowerdev);
919 
920 	/* Only 1 macvlan device can be created in passthru mode */
921 	if (port->passthru)
922 		return -EINVAL;
923 
924 	vlan->lowerdev = lowerdev;
925 	vlan->dev      = dev;
926 	vlan->port     = port;
927 	vlan->set_features = MACVLAN_FEATURES;
928 
929 	vlan->mode     = MACVLAN_MODE_VEPA;
930 	if (data && data[IFLA_MACVLAN_MODE])
931 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
932 
933 	if (data && data[IFLA_MACVLAN_FLAGS])
934 		vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
935 
936 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
937 		if (port->count)
938 			return -EINVAL;
939 		port->passthru = true;
940 		eth_hw_addr_inherit(dev, lowerdev);
941 	}
942 
943 	port->count += 1;
944 	err = register_netdevice(dev);
945 	if (err < 0)
946 		goto destroy_port;
947 
948 	dev->priv_flags |= IFF_MACVLAN;
949 	err = netdev_upper_dev_link(lowerdev, dev);
950 	if (err)
951 		goto unregister_netdev;
952 
953 	list_add_tail_rcu(&vlan->list, &port->vlans);
954 	netif_stacked_transfer_operstate(lowerdev, dev);
955 
956 	return 0;
957 
958 unregister_netdev:
959 	unregister_netdevice(dev);
960 destroy_port:
961 	port->count -= 1;
962 	if (!port->count)
963 		macvlan_port_destroy(lowerdev);
964 
965 	return err;
966 }
967 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
968 
969 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
970 			   struct nlattr *tb[], struct nlattr *data[])
971 {
972 	return macvlan_common_newlink(src_net, dev, tb, data);
973 }
974 
975 void macvlan_dellink(struct net_device *dev, struct list_head *head)
976 {
977 	struct macvlan_dev *vlan = netdev_priv(dev);
978 
979 	list_del_rcu(&vlan->list);
980 	unregister_netdevice_queue(dev, head);
981 	netdev_upper_dev_unlink(vlan->lowerdev, dev);
982 }
983 EXPORT_SYMBOL_GPL(macvlan_dellink);
984 
985 static int macvlan_changelink(struct net_device *dev,
986 		struct nlattr *tb[], struct nlattr *data[])
987 {
988 	struct macvlan_dev *vlan = netdev_priv(dev);
989 	enum macvlan_mode mode;
990 	bool set_mode = false;
991 
992 	/* Validate mode, but don't set yet: setting flags may fail. */
993 	if (data && data[IFLA_MACVLAN_MODE]) {
994 		set_mode = true;
995 		mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
996 		/* Passthrough mode can't be set or cleared dynamically */
997 		if ((mode == MACVLAN_MODE_PASSTHRU) !=
998 		    (vlan->mode == MACVLAN_MODE_PASSTHRU))
999 			return -EINVAL;
1000 	}
1001 
1002 	if (data && data[IFLA_MACVLAN_FLAGS]) {
1003 		__u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1004 		bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
1005 		if (vlan->port->passthru && promisc) {
1006 			int err;
1007 
1008 			if (flags & MACVLAN_FLAG_NOPROMISC)
1009 				err = dev_set_promiscuity(vlan->lowerdev, -1);
1010 			else
1011 				err = dev_set_promiscuity(vlan->lowerdev, 1);
1012 			if (err < 0)
1013 				return err;
1014 		}
1015 		vlan->flags = flags;
1016 	}
1017 	if (set_mode)
1018 		vlan->mode = mode;
1019 	return 0;
1020 }
1021 
1022 static size_t macvlan_get_size(const struct net_device *dev)
1023 {
1024 	return (0
1025 		+ nla_total_size(4) /* IFLA_MACVLAN_MODE */
1026 		+ nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
1027 		);
1028 }
1029 
1030 static int macvlan_fill_info(struct sk_buff *skb,
1031 				const struct net_device *dev)
1032 {
1033 	struct macvlan_dev *vlan = netdev_priv(dev);
1034 
1035 	if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
1036 		goto nla_put_failure;
1037 	if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
1038 		goto nla_put_failure;
1039 	return 0;
1040 
1041 nla_put_failure:
1042 	return -EMSGSIZE;
1043 }
1044 
1045 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
1046 	[IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
1047 	[IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
1048 };
1049 
1050 int macvlan_link_register(struct rtnl_link_ops *ops)
1051 {
1052 	/* common fields */
1053 	ops->priv_size		= sizeof(struct macvlan_dev);
1054 	ops->validate		= macvlan_validate;
1055 	ops->maxtype		= IFLA_MACVLAN_MAX;
1056 	ops->policy		= macvlan_policy;
1057 	ops->changelink		= macvlan_changelink;
1058 	ops->get_size		= macvlan_get_size;
1059 	ops->fill_info		= macvlan_fill_info;
1060 
1061 	return rtnl_link_register(ops);
1062 };
1063 EXPORT_SYMBOL_GPL(macvlan_link_register);
1064 
1065 static struct rtnl_link_ops macvlan_link_ops = {
1066 	.kind		= "macvlan",
1067 	.setup		= macvlan_setup,
1068 	.newlink	= macvlan_newlink,
1069 	.dellink	= macvlan_dellink,
1070 };
1071 
1072 static int macvlan_device_event(struct notifier_block *unused,
1073 				unsigned long event, void *ptr)
1074 {
1075 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1076 	struct macvlan_dev *vlan, *next;
1077 	struct macvlan_port *port;
1078 	LIST_HEAD(list_kill);
1079 
1080 	if (!macvlan_port_exists(dev))
1081 		return NOTIFY_DONE;
1082 
1083 	port = macvlan_port_get_rtnl(dev);
1084 
1085 	switch (event) {
1086 	case NETDEV_CHANGE:
1087 		list_for_each_entry(vlan, &port->vlans, list)
1088 			netif_stacked_transfer_operstate(vlan->lowerdev,
1089 							 vlan->dev);
1090 		break;
1091 	case NETDEV_FEAT_CHANGE:
1092 		list_for_each_entry(vlan, &port->vlans, list) {
1093 			vlan->dev->gso_max_size = dev->gso_max_size;
1094 			netdev_update_features(vlan->dev);
1095 		}
1096 		break;
1097 	case NETDEV_UNREGISTER:
1098 		/* twiddle thumbs on netns device moves */
1099 		if (dev->reg_state != NETREG_UNREGISTERING)
1100 			break;
1101 
1102 		list_for_each_entry_safe(vlan, next, &port->vlans, list)
1103 			vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1104 		unregister_netdevice_many(&list_kill);
1105 		list_del(&list_kill);
1106 		break;
1107 	case NETDEV_PRE_TYPE_CHANGE:
1108 		/* Forbid underlaying device to change its type. */
1109 		return NOTIFY_BAD;
1110 	}
1111 	return NOTIFY_DONE;
1112 }
1113 
1114 static struct notifier_block macvlan_notifier_block __read_mostly = {
1115 	.notifier_call	= macvlan_device_event,
1116 };
1117 
1118 static int __init macvlan_init_module(void)
1119 {
1120 	int err;
1121 
1122 	register_netdevice_notifier(&macvlan_notifier_block);
1123 
1124 	err = macvlan_link_register(&macvlan_link_ops);
1125 	if (err < 0)
1126 		goto err1;
1127 	return 0;
1128 err1:
1129 	unregister_netdevice_notifier(&macvlan_notifier_block);
1130 	return err;
1131 }
1132 
1133 static void __exit macvlan_cleanup_module(void)
1134 {
1135 	rtnl_link_unregister(&macvlan_link_ops);
1136 	unregister_netdevice_notifier(&macvlan_notifier_block);
1137 }
1138 
1139 module_init(macvlan_init_module);
1140 module_exit(macvlan_cleanup_module);
1141 
1142 MODULE_LICENSE("GPL");
1143 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1144 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1145 MODULE_ALIAS_RTNL_LINK("macvlan");
1146