xref: /linux/drivers/net/macvlan.c (revision e58e871becec2d3b04ed91c0c16fe8deac9c9dfa)
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 #include <linux/netpoll.h>
37 
38 #define MACVLAN_HASH_BITS	8
39 #define MACVLAN_HASH_SIZE	(1<<MACVLAN_HASH_BITS)
40 #define MACVLAN_BC_QUEUE_LEN	1000
41 
42 struct macvlan_port {
43 	struct net_device	*dev;
44 	struct hlist_head	vlan_hash[MACVLAN_HASH_SIZE];
45 	struct list_head	vlans;
46 	struct sk_buff_head	bc_queue;
47 	struct work_struct	bc_work;
48 	bool 			passthru;
49 	int			count;
50 	struct hlist_head	vlan_source_hash[MACVLAN_HASH_SIZE];
51 	DECLARE_BITMAP(mc_filter, MACVLAN_MC_FILTER_SZ);
52 };
53 
54 struct macvlan_source_entry {
55 	struct hlist_node	hlist;
56 	struct macvlan_dev	*vlan;
57 	unsigned char		addr[6+2] __aligned(sizeof(u16));
58 	struct rcu_head		rcu;
59 };
60 
61 struct macvlan_skb_cb {
62 	const struct macvlan_dev *src;
63 };
64 
65 #define MACVLAN_SKB_CB(__skb) ((struct macvlan_skb_cb *)&((__skb)->cb[0]))
66 
67 static void macvlan_port_destroy(struct net_device *dev);
68 
69 /* Hash Ethernet address */
70 static u32 macvlan_eth_hash(const unsigned char *addr)
71 {
72 	u64 value = get_unaligned((u64 *)addr);
73 
74 	/* only want 6 bytes */
75 #ifdef __BIG_ENDIAN
76 	value >>= 16;
77 #else
78 	value <<= 16;
79 #endif
80 	return hash_64(value, MACVLAN_HASH_BITS);
81 }
82 
83 static struct macvlan_port *macvlan_port_get_rcu(const struct net_device *dev)
84 {
85 	return rcu_dereference(dev->rx_handler_data);
86 }
87 
88 static struct macvlan_port *macvlan_port_get_rtnl(const struct net_device *dev)
89 {
90 	return rtnl_dereference(dev->rx_handler_data);
91 }
92 
93 #define macvlan_port_exists(dev) (dev->priv_flags & IFF_MACVLAN_PORT)
94 
95 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
96 					       const unsigned char *addr)
97 {
98 	struct macvlan_dev *vlan;
99 	u32 idx = macvlan_eth_hash(addr);
100 
101 	hlist_for_each_entry_rcu(vlan, &port->vlan_hash[idx], hlist) {
102 		if (ether_addr_equal_64bits(vlan->dev->dev_addr, addr))
103 			return vlan;
104 	}
105 	return NULL;
106 }
107 
108 static struct macvlan_source_entry *macvlan_hash_lookup_source(
109 	const struct macvlan_dev *vlan,
110 	const unsigned char *addr)
111 {
112 	struct macvlan_source_entry *entry;
113 	u32 idx = macvlan_eth_hash(addr);
114 	struct hlist_head *h = &vlan->port->vlan_source_hash[idx];
115 
116 	hlist_for_each_entry_rcu(entry, h, hlist) {
117 		if (ether_addr_equal_64bits(entry->addr, addr) &&
118 		    entry->vlan == vlan)
119 			return entry;
120 	}
121 	return NULL;
122 }
123 
124 static int macvlan_hash_add_source(struct macvlan_dev *vlan,
125 				   const unsigned char *addr)
126 {
127 	struct macvlan_port *port = vlan->port;
128 	struct macvlan_source_entry *entry;
129 	struct hlist_head *h;
130 
131 	entry = macvlan_hash_lookup_source(vlan, addr);
132 	if (entry)
133 		return 0;
134 
135 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
136 	if (!entry)
137 		return -ENOMEM;
138 
139 	ether_addr_copy(entry->addr, addr);
140 	entry->vlan = vlan;
141 	h = &port->vlan_source_hash[macvlan_eth_hash(addr)];
142 	hlist_add_head_rcu(&entry->hlist, h);
143 	vlan->macaddr_count++;
144 
145 	return 0;
146 }
147 
148 static void macvlan_hash_add(struct macvlan_dev *vlan)
149 {
150 	struct macvlan_port *port = vlan->port;
151 	const unsigned char *addr = vlan->dev->dev_addr;
152 	u32 idx = macvlan_eth_hash(addr);
153 
154 	hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[idx]);
155 }
156 
157 static void macvlan_hash_del_source(struct macvlan_source_entry *entry)
158 {
159 	hlist_del_rcu(&entry->hlist);
160 	kfree_rcu(entry, rcu);
161 }
162 
163 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync)
164 {
165 	hlist_del_rcu(&vlan->hlist);
166 	if (sync)
167 		synchronize_rcu();
168 }
169 
170 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
171 					const unsigned char *addr)
172 {
173 	macvlan_hash_del(vlan, true);
174 	/* Now that we are unhashed it is safe to change the device
175 	 * address without confusing packet delivery.
176 	 */
177 	memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
178 	macvlan_hash_add(vlan);
179 }
180 
181 static bool macvlan_addr_busy(const struct macvlan_port *port,
182 			      const unsigned char *addr)
183 {
184 	/* Test to see if the specified multicast address is
185 	 * currently in use by the underlying device or
186 	 * another macvlan.
187 	 */
188 	if (ether_addr_equal_64bits(port->dev->dev_addr, addr))
189 		return true;
190 
191 	if (macvlan_hash_lookup(port, addr))
192 		return true;
193 
194 	return false;
195 }
196 
197 
198 static int macvlan_broadcast_one(struct sk_buff *skb,
199 				 const struct macvlan_dev *vlan,
200 				 const struct ethhdr *eth, bool local)
201 {
202 	struct net_device *dev = vlan->dev;
203 
204 	if (local)
205 		return __dev_forward_skb(dev, skb);
206 
207 	skb->dev = dev;
208 	if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
209 		skb->pkt_type = PACKET_BROADCAST;
210 	else
211 		skb->pkt_type = PACKET_MULTICAST;
212 
213 	return 0;
214 }
215 
216 static u32 macvlan_hash_mix(const struct macvlan_dev *vlan)
217 {
218 	return (u32)(((unsigned long)vlan) >> L1_CACHE_SHIFT);
219 }
220 
221 
222 static unsigned int mc_hash(const struct macvlan_dev *vlan,
223 			    const unsigned char *addr)
224 {
225 	u32 val = __get_unaligned_cpu32(addr + 2);
226 
227 	val ^= macvlan_hash_mix(vlan);
228 	return hash_32(val, MACVLAN_MC_FILTER_BITS);
229 }
230 
231 static void macvlan_broadcast(struct sk_buff *skb,
232 			      const struct macvlan_port *port,
233 			      struct net_device *src,
234 			      enum macvlan_mode mode)
235 {
236 	const struct ethhdr *eth = eth_hdr(skb);
237 	const struct macvlan_dev *vlan;
238 	struct sk_buff *nskb;
239 	unsigned int i;
240 	int err;
241 	unsigned int hash;
242 
243 	if (skb->protocol == htons(ETH_P_PAUSE))
244 		return;
245 
246 	for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
247 		hlist_for_each_entry_rcu(vlan, &port->vlan_hash[i], hlist) {
248 			if (vlan->dev == src || !(vlan->mode & mode))
249 				continue;
250 
251 			hash = mc_hash(vlan, eth->h_dest);
252 			if (!test_bit(hash, vlan->mc_filter))
253 				continue;
254 
255 			err = NET_RX_DROP;
256 			nskb = skb_clone(skb, GFP_ATOMIC);
257 			if (likely(nskb))
258 				err = macvlan_broadcast_one(
259 					nskb, vlan, eth,
260 					mode == MACVLAN_MODE_BRIDGE) ?:
261 				      netif_rx_ni(nskb);
262 			macvlan_count_rx(vlan, skb->len + ETH_HLEN,
263 					 err == NET_RX_SUCCESS, true);
264 		}
265 	}
266 }
267 
268 static void macvlan_process_broadcast(struct work_struct *w)
269 {
270 	struct macvlan_port *port = container_of(w, struct macvlan_port,
271 						 bc_work);
272 	struct sk_buff *skb;
273 	struct sk_buff_head list;
274 
275 	__skb_queue_head_init(&list);
276 
277 	spin_lock_bh(&port->bc_queue.lock);
278 	skb_queue_splice_tail_init(&port->bc_queue, &list);
279 	spin_unlock_bh(&port->bc_queue.lock);
280 
281 	while ((skb = __skb_dequeue(&list))) {
282 		const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
283 
284 		rcu_read_lock();
285 
286 		if (!src)
287 			/* frame comes from an external address */
288 			macvlan_broadcast(skb, port, NULL,
289 					  MACVLAN_MODE_PRIVATE |
290 					  MACVLAN_MODE_VEPA    |
291 					  MACVLAN_MODE_PASSTHRU|
292 					  MACVLAN_MODE_BRIDGE);
293 		else if (src->mode == MACVLAN_MODE_VEPA)
294 			/* flood to everyone except source */
295 			macvlan_broadcast(skb, port, src->dev,
296 					  MACVLAN_MODE_VEPA |
297 					  MACVLAN_MODE_BRIDGE);
298 		else
299 			/*
300 			 * flood only to VEPA ports, bridge ports
301 			 * already saw the frame on the way out.
302 			 */
303 			macvlan_broadcast(skb, port, src->dev,
304 					  MACVLAN_MODE_VEPA);
305 
306 		rcu_read_unlock();
307 
308 		if (src)
309 			dev_put(src->dev);
310 		kfree_skb(skb);
311 	}
312 }
313 
314 static void macvlan_broadcast_enqueue(struct macvlan_port *port,
315 				      const struct macvlan_dev *src,
316 				      struct sk_buff *skb)
317 {
318 	struct sk_buff *nskb;
319 	int err = -ENOMEM;
320 
321 	nskb = skb_clone(skb, GFP_ATOMIC);
322 	if (!nskb)
323 		goto err;
324 
325 	MACVLAN_SKB_CB(nskb)->src = src;
326 
327 	spin_lock(&port->bc_queue.lock);
328 	if (skb_queue_len(&port->bc_queue) < MACVLAN_BC_QUEUE_LEN) {
329 		if (src)
330 			dev_hold(src->dev);
331 		__skb_queue_tail(&port->bc_queue, nskb);
332 		err = 0;
333 	}
334 	spin_unlock(&port->bc_queue.lock);
335 
336 	if (err)
337 		goto free_nskb;
338 
339 	schedule_work(&port->bc_work);
340 	return;
341 
342 free_nskb:
343 	kfree_skb(nskb);
344 err:
345 	atomic_long_inc(&skb->dev->rx_dropped);
346 }
347 
348 static void macvlan_flush_sources(struct macvlan_port *port,
349 				  struct macvlan_dev *vlan)
350 {
351 	int i;
352 
353 	for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
354 		struct hlist_node *h, *n;
355 
356 		hlist_for_each_safe(h, n, &port->vlan_source_hash[i]) {
357 			struct macvlan_source_entry *entry;
358 
359 			entry = hlist_entry(h, struct macvlan_source_entry,
360 					    hlist);
361 			if (entry->vlan == vlan)
362 				macvlan_hash_del_source(entry);
363 		}
364 	}
365 	vlan->macaddr_count = 0;
366 }
367 
368 static void macvlan_forward_source_one(struct sk_buff *skb,
369 				       struct macvlan_dev *vlan)
370 {
371 	struct sk_buff *nskb;
372 	struct net_device *dev;
373 	int len;
374 	int ret;
375 
376 	dev = vlan->dev;
377 	if (unlikely(!(dev->flags & IFF_UP)))
378 		return;
379 
380 	nskb = skb_clone(skb, GFP_ATOMIC);
381 	if (!nskb)
382 		return;
383 
384 	len = nskb->len + ETH_HLEN;
385 	nskb->dev = dev;
386 	nskb->pkt_type = PACKET_HOST;
387 
388 	ret = netif_rx(nskb);
389 	macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
390 }
391 
392 static void macvlan_forward_source(struct sk_buff *skb,
393 				   struct macvlan_port *port,
394 				   const unsigned char *addr)
395 {
396 	struct macvlan_source_entry *entry;
397 	u32 idx = macvlan_eth_hash(addr);
398 	struct hlist_head *h = &port->vlan_source_hash[idx];
399 
400 	hlist_for_each_entry_rcu(entry, h, hlist) {
401 		if (ether_addr_equal_64bits(entry->addr, addr))
402 			macvlan_forward_source_one(skb, entry->vlan);
403 	}
404 }
405 
406 /* called under rcu_read_lock() from netif_receive_skb */
407 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
408 {
409 	struct macvlan_port *port;
410 	struct sk_buff *skb = *pskb;
411 	const struct ethhdr *eth = eth_hdr(skb);
412 	const struct macvlan_dev *vlan;
413 	const struct macvlan_dev *src;
414 	struct net_device *dev;
415 	unsigned int len = 0;
416 	int ret;
417 	rx_handler_result_t handle_res;
418 
419 	port = macvlan_port_get_rcu(skb->dev);
420 	if (is_multicast_ether_addr(eth->h_dest)) {
421 		unsigned int hash;
422 
423 		skb = ip_check_defrag(dev_net(skb->dev), skb, IP_DEFRAG_MACVLAN);
424 		if (!skb)
425 			return RX_HANDLER_CONSUMED;
426 		*pskb = skb;
427 		eth = eth_hdr(skb);
428 		macvlan_forward_source(skb, port, eth->h_source);
429 		src = macvlan_hash_lookup(port, eth->h_source);
430 		if (src && src->mode != MACVLAN_MODE_VEPA &&
431 		    src->mode != MACVLAN_MODE_BRIDGE) {
432 			/* forward to original port. */
433 			vlan = src;
434 			ret = macvlan_broadcast_one(skb, vlan, eth, 0) ?:
435 			      netif_rx(skb);
436 			handle_res = RX_HANDLER_CONSUMED;
437 			goto out;
438 		}
439 
440 		hash = mc_hash(NULL, eth->h_dest);
441 		if (test_bit(hash, port->mc_filter))
442 			macvlan_broadcast_enqueue(port, src, skb);
443 
444 		return RX_HANDLER_PASS;
445 	}
446 
447 	macvlan_forward_source(skb, port, eth->h_source);
448 	if (port->passthru)
449 		vlan = list_first_or_null_rcu(&port->vlans,
450 					      struct macvlan_dev, list);
451 	else
452 		vlan = macvlan_hash_lookup(port, eth->h_dest);
453 	if (vlan == NULL)
454 		return RX_HANDLER_PASS;
455 
456 	dev = vlan->dev;
457 	if (unlikely(!(dev->flags & IFF_UP))) {
458 		kfree_skb(skb);
459 		return RX_HANDLER_CONSUMED;
460 	}
461 	len = skb->len + ETH_HLEN;
462 	skb = skb_share_check(skb, GFP_ATOMIC);
463 	if (!skb) {
464 		ret = NET_RX_DROP;
465 		handle_res = RX_HANDLER_CONSUMED;
466 		goto out;
467 	}
468 
469 	*pskb = skb;
470 	skb->dev = dev;
471 	skb->pkt_type = PACKET_HOST;
472 
473 	ret = NET_RX_SUCCESS;
474 	handle_res = RX_HANDLER_ANOTHER;
475 out:
476 	macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
477 	return handle_res;
478 }
479 
480 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
481 {
482 	const struct macvlan_dev *vlan = netdev_priv(dev);
483 	const struct macvlan_port *port = vlan->port;
484 	const struct macvlan_dev *dest;
485 
486 	if (vlan->mode == MACVLAN_MODE_BRIDGE) {
487 		const struct ethhdr *eth = (void *)skb->data;
488 
489 		/* send to other bridge ports directly */
490 		if (is_multicast_ether_addr(eth->h_dest)) {
491 			macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
492 			goto xmit_world;
493 		}
494 
495 		dest = macvlan_hash_lookup(port, eth->h_dest);
496 		if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
497 			/* send to lowerdev first for its network taps */
498 			dev_forward_skb(vlan->lowerdev, skb);
499 
500 			return NET_XMIT_SUCCESS;
501 		}
502 	}
503 
504 xmit_world:
505 	skb->dev = vlan->lowerdev;
506 	return dev_queue_xmit(skb);
507 }
508 
509 static inline netdev_tx_t macvlan_netpoll_send_skb(struct macvlan_dev *vlan, struct sk_buff *skb)
510 {
511 #ifdef CONFIG_NET_POLL_CONTROLLER
512 	if (vlan->netpoll)
513 		netpoll_send_skb(vlan->netpoll, skb);
514 #else
515 	BUG();
516 #endif
517 	return NETDEV_TX_OK;
518 }
519 
520 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
521 				      struct net_device *dev)
522 {
523 	unsigned int len = skb->len;
524 	int ret;
525 	struct macvlan_dev *vlan = netdev_priv(dev);
526 
527 	if (unlikely(netpoll_tx_running(dev)))
528 		return macvlan_netpoll_send_skb(vlan, skb);
529 
530 	if (vlan->fwd_priv) {
531 		skb->dev = vlan->lowerdev;
532 		ret = dev_queue_xmit_accel(skb, vlan->fwd_priv);
533 	} else {
534 		ret = macvlan_queue_xmit(skb, dev);
535 	}
536 
537 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
538 		struct vlan_pcpu_stats *pcpu_stats;
539 
540 		pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
541 		u64_stats_update_begin(&pcpu_stats->syncp);
542 		pcpu_stats->tx_packets++;
543 		pcpu_stats->tx_bytes += len;
544 		u64_stats_update_end(&pcpu_stats->syncp);
545 	} else {
546 		this_cpu_inc(vlan->pcpu_stats->tx_dropped);
547 	}
548 	return ret;
549 }
550 
551 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
552 			       unsigned short type, const void *daddr,
553 			       const void *saddr, unsigned len)
554 {
555 	const struct macvlan_dev *vlan = netdev_priv(dev);
556 	struct net_device *lowerdev = vlan->lowerdev;
557 
558 	return dev_hard_header(skb, lowerdev, type, daddr,
559 			       saddr ? : dev->dev_addr, len);
560 }
561 
562 static const struct header_ops macvlan_hard_header_ops = {
563 	.create  	= macvlan_hard_header,
564 	.parse		= eth_header_parse,
565 	.cache		= eth_header_cache,
566 	.cache_update	= eth_header_cache_update,
567 };
568 
569 static struct rtnl_link_ops macvlan_link_ops;
570 
571 static int macvlan_open(struct net_device *dev)
572 {
573 	struct macvlan_dev *vlan = netdev_priv(dev);
574 	struct net_device *lowerdev = vlan->lowerdev;
575 	int err;
576 
577 	if (vlan->port->passthru) {
578 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
579 			err = dev_set_promiscuity(lowerdev, 1);
580 			if (err < 0)
581 				goto out;
582 		}
583 		goto hash_add;
584 	}
585 
586 	if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD &&
587 	    dev->rtnl_link_ops == &macvlan_link_ops) {
588 		vlan->fwd_priv =
589 		      lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
590 
591 		/* If we get a NULL pointer back, or if we get an error
592 		 * then we should just fall through to the non accelerated path
593 		 */
594 		if (IS_ERR_OR_NULL(vlan->fwd_priv)) {
595 			vlan->fwd_priv = NULL;
596 		} else
597 			return 0;
598 	}
599 
600 	err = -EBUSY;
601 	if (macvlan_addr_busy(vlan->port, dev->dev_addr))
602 		goto out;
603 
604 	err = dev_uc_add(lowerdev, dev->dev_addr);
605 	if (err < 0)
606 		goto out;
607 	if (dev->flags & IFF_ALLMULTI) {
608 		err = dev_set_allmulti(lowerdev, 1);
609 		if (err < 0)
610 			goto del_unicast;
611 	}
612 
613 	if (dev->flags & IFF_PROMISC) {
614 		err = dev_set_promiscuity(lowerdev, 1);
615 		if (err < 0)
616 			goto clear_multi;
617 	}
618 
619 hash_add:
620 	macvlan_hash_add(vlan);
621 	return 0;
622 
623 clear_multi:
624 	if (dev->flags & IFF_ALLMULTI)
625 		dev_set_allmulti(lowerdev, -1);
626 del_unicast:
627 	dev_uc_del(lowerdev, dev->dev_addr);
628 out:
629 	if (vlan->fwd_priv) {
630 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
631 							   vlan->fwd_priv);
632 		vlan->fwd_priv = NULL;
633 	}
634 	return err;
635 }
636 
637 static int macvlan_stop(struct net_device *dev)
638 {
639 	struct macvlan_dev *vlan = netdev_priv(dev);
640 	struct net_device *lowerdev = vlan->lowerdev;
641 
642 	if (vlan->fwd_priv) {
643 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
644 							   vlan->fwd_priv);
645 		vlan->fwd_priv = NULL;
646 		return 0;
647 	}
648 
649 	dev_uc_unsync(lowerdev, dev);
650 	dev_mc_unsync(lowerdev, dev);
651 
652 	if (vlan->port->passthru) {
653 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
654 			dev_set_promiscuity(lowerdev, -1);
655 		goto hash_del;
656 	}
657 
658 	if (dev->flags & IFF_ALLMULTI)
659 		dev_set_allmulti(lowerdev, -1);
660 
661 	if (dev->flags & IFF_PROMISC)
662 		dev_set_promiscuity(lowerdev, -1);
663 
664 	dev_uc_del(lowerdev, dev->dev_addr);
665 
666 hash_del:
667 	macvlan_hash_del(vlan, !dev->dismantle);
668 	return 0;
669 }
670 
671 static int macvlan_sync_address(struct net_device *dev, unsigned char *addr)
672 {
673 	struct macvlan_dev *vlan = netdev_priv(dev);
674 	struct net_device *lowerdev = vlan->lowerdev;
675 	int err;
676 
677 	if (!(dev->flags & IFF_UP)) {
678 		/* Just copy in the new address */
679 		ether_addr_copy(dev->dev_addr, addr);
680 	} else {
681 		/* Rehash and update the device filters */
682 		if (macvlan_addr_busy(vlan->port, addr))
683 			return -EBUSY;
684 
685 		if (!vlan->port->passthru) {
686 			err = dev_uc_add(lowerdev, addr);
687 			if (err)
688 				return err;
689 
690 			dev_uc_del(lowerdev, dev->dev_addr);
691 		}
692 
693 		macvlan_hash_change_addr(vlan, addr);
694 	}
695 	return 0;
696 }
697 
698 static int macvlan_set_mac_address(struct net_device *dev, void *p)
699 {
700 	struct macvlan_dev *vlan = netdev_priv(dev);
701 	struct sockaddr *addr = p;
702 
703 	if (!is_valid_ether_addr(addr->sa_data))
704 		return -EADDRNOTAVAIL;
705 
706 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
707 		dev_set_mac_address(vlan->lowerdev, addr);
708 		return 0;
709 	}
710 
711 	return macvlan_sync_address(dev, addr->sa_data);
712 }
713 
714 static void macvlan_change_rx_flags(struct net_device *dev, int change)
715 {
716 	struct macvlan_dev *vlan = netdev_priv(dev);
717 	struct net_device *lowerdev = vlan->lowerdev;
718 
719 	if (dev->flags & IFF_UP) {
720 		if (change & IFF_ALLMULTI)
721 			dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
722 		if (change & IFF_PROMISC)
723 			dev_set_promiscuity(lowerdev,
724 					    dev->flags & IFF_PROMISC ? 1 : -1);
725 
726 	}
727 }
728 
729 static void macvlan_compute_filter(unsigned long *mc_filter,
730 				   struct net_device *dev,
731 				   struct macvlan_dev *vlan)
732 {
733 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
734 		bitmap_fill(mc_filter, MACVLAN_MC_FILTER_SZ);
735 	} else {
736 		struct netdev_hw_addr *ha;
737 		DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
738 
739 		bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
740 		netdev_for_each_mc_addr(ha, dev) {
741 			__set_bit(mc_hash(vlan, ha->addr), filter);
742 		}
743 
744 		__set_bit(mc_hash(vlan, dev->broadcast), filter);
745 
746 		bitmap_copy(mc_filter, filter, MACVLAN_MC_FILTER_SZ);
747 	}
748 }
749 
750 static void macvlan_set_mac_lists(struct net_device *dev)
751 {
752 	struct macvlan_dev *vlan = netdev_priv(dev);
753 
754 	macvlan_compute_filter(vlan->mc_filter, dev, vlan);
755 
756 	dev_uc_sync(vlan->lowerdev, dev);
757 	dev_mc_sync(vlan->lowerdev, dev);
758 
759 	/* This is slightly inaccurate as we're including the subscription
760 	 * list of vlan->lowerdev too.
761 	 *
762 	 * Bug alert: This only works if everyone has the same broadcast
763 	 * address as lowerdev.  As soon as someone changes theirs this
764 	 * will break.
765 	 *
766 	 * However, this is already broken as when you change your broadcast
767 	 * address we don't get called.
768 	 *
769 	 * The solution is to maintain a list of broadcast addresses like
770 	 * we do for uc/mc, if you care.
771 	 */
772 	macvlan_compute_filter(vlan->port->mc_filter, vlan->lowerdev, NULL);
773 }
774 
775 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
776 {
777 	struct macvlan_dev *vlan = netdev_priv(dev);
778 
779 	if (vlan->lowerdev->mtu < new_mtu)
780 		return -EINVAL;
781 	dev->mtu = new_mtu;
782 	return 0;
783 }
784 
785 /*
786  * macvlan network devices have devices nesting below it and are a special
787  * "super class" of normal network devices; split their locks off into a
788  * separate class since they always nest.
789  */
790 static struct lock_class_key macvlan_netdev_addr_lock_key;
791 
792 #define ALWAYS_ON_OFFLOADS \
793 	(NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE | \
794 	 NETIF_F_GSO_ROBUST)
795 
796 #define ALWAYS_ON_FEATURES (ALWAYS_ON_OFFLOADS | NETIF_F_LLTX)
797 
798 #define MACVLAN_FEATURES \
799 	(NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
800 	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_LRO | \
801 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
802 	 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
803 
804 #define MACVLAN_STATE_MASK \
805 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
806 
807 static int macvlan_get_nest_level(struct net_device *dev)
808 {
809 	return ((struct macvlan_dev *)netdev_priv(dev))->nest_level;
810 }
811 
812 static void macvlan_set_lockdep_class(struct net_device *dev)
813 {
814 	netdev_lockdep_set_classes(dev);
815 	lockdep_set_class_and_subclass(&dev->addr_list_lock,
816 				       &macvlan_netdev_addr_lock_key,
817 				       macvlan_get_nest_level(dev));
818 }
819 
820 static int macvlan_init(struct net_device *dev)
821 {
822 	struct macvlan_dev *vlan = netdev_priv(dev);
823 	const struct net_device *lowerdev = vlan->lowerdev;
824 	struct macvlan_port *port = vlan->port;
825 
826 	dev->state		= (dev->state & ~MACVLAN_STATE_MASK) |
827 				  (lowerdev->state & MACVLAN_STATE_MASK);
828 	dev->features 		= lowerdev->features & MACVLAN_FEATURES;
829 	dev->features		|= ALWAYS_ON_FEATURES;
830 	dev->hw_features	|= NETIF_F_LRO;
831 	dev->vlan_features	= lowerdev->vlan_features & MACVLAN_FEATURES;
832 	dev->vlan_features	|= ALWAYS_ON_OFFLOADS;
833 	dev->gso_max_size	= lowerdev->gso_max_size;
834 	dev->gso_max_segs	= lowerdev->gso_max_segs;
835 	dev->hard_header_len	= lowerdev->hard_header_len;
836 
837 	macvlan_set_lockdep_class(dev);
838 
839 	vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
840 	if (!vlan->pcpu_stats)
841 		return -ENOMEM;
842 
843 	port->count += 1;
844 
845 	return 0;
846 }
847 
848 static void macvlan_uninit(struct net_device *dev)
849 {
850 	struct macvlan_dev *vlan = netdev_priv(dev);
851 	struct macvlan_port *port = vlan->port;
852 
853 	free_percpu(vlan->pcpu_stats);
854 
855 	macvlan_flush_sources(port, vlan);
856 	port->count -= 1;
857 	if (!port->count)
858 		macvlan_port_destroy(port->dev);
859 }
860 
861 static void macvlan_dev_get_stats64(struct net_device *dev,
862 				    struct rtnl_link_stats64 *stats)
863 {
864 	struct macvlan_dev *vlan = netdev_priv(dev);
865 
866 	if (vlan->pcpu_stats) {
867 		struct vlan_pcpu_stats *p;
868 		u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
869 		u32 rx_errors = 0, tx_dropped = 0;
870 		unsigned int start;
871 		int i;
872 
873 		for_each_possible_cpu(i) {
874 			p = per_cpu_ptr(vlan->pcpu_stats, i);
875 			do {
876 				start = u64_stats_fetch_begin_irq(&p->syncp);
877 				rx_packets	= p->rx_packets;
878 				rx_bytes	= p->rx_bytes;
879 				rx_multicast	= p->rx_multicast;
880 				tx_packets	= p->tx_packets;
881 				tx_bytes	= p->tx_bytes;
882 			} while (u64_stats_fetch_retry_irq(&p->syncp, start));
883 
884 			stats->rx_packets	+= rx_packets;
885 			stats->rx_bytes		+= rx_bytes;
886 			stats->multicast	+= rx_multicast;
887 			stats->tx_packets	+= tx_packets;
888 			stats->tx_bytes		+= tx_bytes;
889 			/* rx_errors & tx_dropped are u32, updated
890 			 * without syncp protection.
891 			 */
892 			rx_errors	+= p->rx_errors;
893 			tx_dropped	+= p->tx_dropped;
894 		}
895 		stats->rx_errors	= rx_errors;
896 		stats->rx_dropped	= rx_errors;
897 		stats->tx_dropped	= tx_dropped;
898 	}
899 }
900 
901 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
902 				   __be16 proto, u16 vid)
903 {
904 	struct macvlan_dev *vlan = netdev_priv(dev);
905 	struct net_device *lowerdev = vlan->lowerdev;
906 
907 	return vlan_vid_add(lowerdev, proto, vid);
908 }
909 
910 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
911 				    __be16 proto, u16 vid)
912 {
913 	struct macvlan_dev *vlan = netdev_priv(dev);
914 	struct net_device *lowerdev = vlan->lowerdev;
915 
916 	vlan_vid_del(lowerdev, proto, vid);
917 	return 0;
918 }
919 
920 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
921 			   struct net_device *dev,
922 			   const unsigned char *addr, u16 vid,
923 			   u16 flags)
924 {
925 	struct macvlan_dev *vlan = netdev_priv(dev);
926 	int err = -EINVAL;
927 
928 	/* Support unicast filter only on passthru devices.
929 	 * Multicast filter should be allowed on all devices.
930 	 */
931 	if (!vlan->port->passthru && is_unicast_ether_addr(addr))
932 		return -EOPNOTSUPP;
933 
934 	if (flags & NLM_F_REPLACE)
935 		return -EOPNOTSUPP;
936 
937 	if (is_unicast_ether_addr(addr))
938 		err = dev_uc_add_excl(dev, addr);
939 	else if (is_multicast_ether_addr(addr))
940 		err = dev_mc_add_excl(dev, addr);
941 
942 	return err;
943 }
944 
945 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
946 			   struct net_device *dev,
947 			   const unsigned char *addr, u16 vid)
948 {
949 	struct macvlan_dev *vlan = netdev_priv(dev);
950 	int err = -EINVAL;
951 
952 	/* Support unicast filter only on passthru devices.
953 	 * Multicast filter should be allowed on all devices.
954 	 */
955 	if (!vlan->port->passthru && is_unicast_ether_addr(addr))
956 		return -EOPNOTSUPP;
957 
958 	if (is_unicast_ether_addr(addr))
959 		err = dev_uc_del(dev, addr);
960 	else if (is_multicast_ether_addr(addr))
961 		err = dev_mc_del(dev, addr);
962 
963 	return err;
964 }
965 
966 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
967 					struct ethtool_drvinfo *drvinfo)
968 {
969 	strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
970 	strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
971 }
972 
973 static int macvlan_ethtool_get_link_ksettings(struct net_device *dev,
974 					      struct ethtool_link_ksettings *cmd)
975 {
976 	const struct macvlan_dev *vlan = netdev_priv(dev);
977 
978 	return __ethtool_get_link_ksettings(vlan->lowerdev, cmd);
979 }
980 
981 static netdev_features_t macvlan_fix_features(struct net_device *dev,
982 					      netdev_features_t features)
983 {
984 	struct macvlan_dev *vlan = netdev_priv(dev);
985 	netdev_features_t lowerdev_features = vlan->lowerdev->features;
986 	netdev_features_t mask;
987 
988 	features |= NETIF_F_ALL_FOR_ALL;
989 	features &= (vlan->set_features | ~MACVLAN_FEATURES);
990 	mask = features;
991 
992 	lowerdev_features &= (features | ~NETIF_F_LRO);
993 	features = netdev_increment_features(lowerdev_features, features, mask);
994 	features |= ALWAYS_ON_FEATURES;
995 	features &= ~NETIF_F_NETNS_LOCAL;
996 
997 	return features;
998 }
999 
1000 #ifdef CONFIG_NET_POLL_CONTROLLER
1001 static void macvlan_dev_poll_controller(struct net_device *dev)
1002 {
1003 	return;
1004 }
1005 
1006 static int macvlan_dev_netpoll_setup(struct net_device *dev, struct netpoll_info *npinfo)
1007 {
1008 	struct macvlan_dev *vlan = netdev_priv(dev);
1009 	struct net_device *real_dev = vlan->lowerdev;
1010 	struct netpoll *netpoll;
1011 	int err = 0;
1012 
1013 	netpoll = kzalloc(sizeof(*netpoll), GFP_KERNEL);
1014 	err = -ENOMEM;
1015 	if (!netpoll)
1016 		goto out;
1017 
1018 	err = __netpoll_setup(netpoll, real_dev);
1019 	if (err) {
1020 		kfree(netpoll);
1021 		goto out;
1022 	}
1023 
1024 	vlan->netpoll = netpoll;
1025 
1026 out:
1027 	return err;
1028 }
1029 
1030 static void macvlan_dev_netpoll_cleanup(struct net_device *dev)
1031 {
1032 	struct macvlan_dev *vlan = netdev_priv(dev);
1033 	struct netpoll *netpoll = vlan->netpoll;
1034 
1035 	if (!netpoll)
1036 		return;
1037 
1038 	vlan->netpoll = NULL;
1039 
1040 	__netpoll_free_async(netpoll);
1041 }
1042 #endif	/* CONFIG_NET_POLL_CONTROLLER */
1043 
1044 static int macvlan_dev_get_iflink(const struct net_device *dev)
1045 {
1046 	struct macvlan_dev *vlan = netdev_priv(dev);
1047 
1048 	return vlan->lowerdev->ifindex;
1049 }
1050 
1051 static const struct ethtool_ops macvlan_ethtool_ops = {
1052 	.get_link		= ethtool_op_get_link,
1053 	.get_link_ksettings	= macvlan_ethtool_get_link_ksettings,
1054 	.get_drvinfo		= macvlan_ethtool_get_drvinfo,
1055 };
1056 
1057 static const struct net_device_ops macvlan_netdev_ops = {
1058 	.ndo_init		= macvlan_init,
1059 	.ndo_uninit		= macvlan_uninit,
1060 	.ndo_open		= macvlan_open,
1061 	.ndo_stop		= macvlan_stop,
1062 	.ndo_start_xmit		= macvlan_start_xmit,
1063 	.ndo_change_mtu		= macvlan_change_mtu,
1064 	.ndo_fix_features	= macvlan_fix_features,
1065 	.ndo_change_rx_flags	= macvlan_change_rx_flags,
1066 	.ndo_set_mac_address	= macvlan_set_mac_address,
1067 	.ndo_set_rx_mode	= macvlan_set_mac_lists,
1068 	.ndo_get_stats64	= macvlan_dev_get_stats64,
1069 	.ndo_validate_addr	= eth_validate_addr,
1070 	.ndo_vlan_rx_add_vid	= macvlan_vlan_rx_add_vid,
1071 	.ndo_vlan_rx_kill_vid	= macvlan_vlan_rx_kill_vid,
1072 	.ndo_fdb_add		= macvlan_fdb_add,
1073 	.ndo_fdb_del		= macvlan_fdb_del,
1074 	.ndo_fdb_dump		= ndo_dflt_fdb_dump,
1075 	.ndo_get_lock_subclass  = macvlan_get_nest_level,
1076 #ifdef CONFIG_NET_POLL_CONTROLLER
1077 	.ndo_poll_controller	= macvlan_dev_poll_controller,
1078 	.ndo_netpoll_setup	= macvlan_dev_netpoll_setup,
1079 	.ndo_netpoll_cleanup	= macvlan_dev_netpoll_cleanup,
1080 #endif
1081 	.ndo_get_iflink		= macvlan_dev_get_iflink,
1082 	.ndo_features_check	= passthru_features_check,
1083 };
1084 
1085 void macvlan_common_setup(struct net_device *dev)
1086 {
1087 	ether_setup(dev);
1088 
1089 	dev->min_mtu		= 0;
1090 	dev->max_mtu		= ETH_MAX_MTU;
1091 	dev->priv_flags	       &= ~IFF_TX_SKB_SHARING;
1092 	netif_keep_dst(dev);
1093 	dev->priv_flags	       |= IFF_UNICAST_FLT;
1094 	dev->netdev_ops		= &macvlan_netdev_ops;
1095 	dev->destructor		= free_netdev;
1096 	dev->header_ops		= &macvlan_hard_header_ops;
1097 	dev->ethtool_ops	= &macvlan_ethtool_ops;
1098 }
1099 EXPORT_SYMBOL_GPL(macvlan_common_setup);
1100 
1101 static void macvlan_setup(struct net_device *dev)
1102 {
1103 	macvlan_common_setup(dev);
1104 	dev->priv_flags |= IFF_NO_QUEUE;
1105 }
1106 
1107 static int macvlan_port_create(struct net_device *dev)
1108 {
1109 	struct macvlan_port *port;
1110 	unsigned int i;
1111 	int err;
1112 
1113 	if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
1114 		return -EINVAL;
1115 
1116 	if (netdev_is_rx_handler_busy(dev))
1117 		return -EBUSY;
1118 
1119 	port = kzalloc(sizeof(*port), GFP_KERNEL);
1120 	if (port == NULL)
1121 		return -ENOMEM;
1122 
1123 	port->passthru = false;
1124 	port->dev = dev;
1125 	INIT_LIST_HEAD(&port->vlans);
1126 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1127 		INIT_HLIST_HEAD(&port->vlan_hash[i]);
1128 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1129 		INIT_HLIST_HEAD(&port->vlan_source_hash[i]);
1130 
1131 	skb_queue_head_init(&port->bc_queue);
1132 	INIT_WORK(&port->bc_work, macvlan_process_broadcast);
1133 
1134 	err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
1135 	if (err)
1136 		kfree(port);
1137 	else
1138 		dev->priv_flags |= IFF_MACVLAN_PORT;
1139 	return err;
1140 }
1141 
1142 static void macvlan_port_destroy(struct net_device *dev)
1143 {
1144 	struct macvlan_port *port = macvlan_port_get_rtnl(dev);
1145 	struct sk_buff *skb;
1146 
1147 	dev->priv_flags &= ~IFF_MACVLAN_PORT;
1148 	netdev_rx_handler_unregister(dev);
1149 
1150 	/* After this point, no packet can schedule bc_work anymore,
1151 	 * but we need to cancel it and purge left skbs if any.
1152 	 */
1153 	cancel_work_sync(&port->bc_work);
1154 
1155 	while ((skb = __skb_dequeue(&port->bc_queue))) {
1156 		const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
1157 
1158 		if (src)
1159 			dev_put(src->dev);
1160 
1161 		kfree_skb(skb);
1162 	}
1163 
1164 	kfree(port);
1165 }
1166 
1167 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
1168 {
1169 	if (tb[IFLA_ADDRESS]) {
1170 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1171 			return -EINVAL;
1172 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1173 			return -EADDRNOTAVAIL;
1174 	}
1175 
1176 	if (data && data[IFLA_MACVLAN_FLAGS] &&
1177 	    nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
1178 		return -EINVAL;
1179 
1180 	if (data && data[IFLA_MACVLAN_MODE]) {
1181 		switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
1182 		case MACVLAN_MODE_PRIVATE:
1183 		case MACVLAN_MODE_VEPA:
1184 		case MACVLAN_MODE_BRIDGE:
1185 		case MACVLAN_MODE_PASSTHRU:
1186 		case MACVLAN_MODE_SOURCE:
1187 			break;
1188 		default:
1189 			return -EINVAL;
1190 		}
1191 	}
1192 
1193 	if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1194 		switch (nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE])) {
1195 		case MACVLAN_MACADDR_ADD:
1196 		case MACVLAN_MACADDR_DEL:
1197 		case MACVLAN_MACADDR_FLUSH:
1198 		case MACVLAN_MACADDR_SET:
1199 			break;
1200 		default:
1201 			return -EINVAL;
1202 		}
1203 	}
1204 
1205 	if (data && data[IFLA_MACVLAN_MACADDR]) {
1206 		if (nla_len(data[IFLA_MACVLAN_MACADDR]) != ETH_ALEN)
1207 			return -EINVAL;
1208 
1209 		if (!is_valid_ether_addr(nla_data(data[IFLA_MACVLAN_MACADDR])))
1210 			return -EADDRNOTAVAIL;
1211 	}
1212 
1213 	if (data && data[IFLA_MACVLAN_MACADDR_COUNT])
1214 		return -EINVAL;
1215 
1216 	return 0;
1217 }
1218 
1219 /**
1220  * reconfigure list of remote source mac address
1221  * (only for macvlan devices in source mode)
1222  * Note regarding alignment: all netlink data is aligned to 4 Byte, which
1223  * suffices for both ether_addr_copy and ether_addr_equal_64bits usage.
1224  */
1225 static int macvlan_changelink_sources(struct macvlan_dev *vlan, u32 mode,
1226 				      struct nlattr *data[])
1227 {
1228 	char *addr = NULL;
1229 	int ret, rem, len;
1230 	struct nlattr *nla, *head;
1231 	struct macvlan_source_entry *entry;
1232 
1233 	if (data[IFLA_MACVLAN_MACADDR])
1234 		addr = nla_data(data[IFLA_MACVLAN_MACADDR]);
1235 
1236 	if (mode == MACVLAN_MACADDR_ADD) {
1237 		if (!addr)
1238 			return -EINVAL;
1239 
1240 		return macvlan_hash_add_source(vlan, addr);
1241 
1242 	} else if (mode == MACVLAN_MACADDR_DEL) {
1243 		if (!addr)
1244 			return -EINVAL;
1245 
1246 		entry = macvlan_hash_lookup_source(vlan, addr);
1247 		if (entry) {
1248 			macvlan_hash_del_source(entry);
1249 			vlan->macaddr_count--;
1250 		}
1251 	} else if (mode == MACVLAN_MACADDR_FLUSH) {
1252 		macvlan_flush_sources(vlan->port, vlan);
1253 	} else if (mode == MACVLAN_MACADDR_SET) {
1254 		macvlan_flush_sources(vlan->port, vlan);
1255 
1256 		if (addr) {
1257 			ret = macvlan_hash_add_source(vlan, addr);
1258 			if (ret)
1259 				return ret;
1260 		}
1261 
1262 		if (!data || !data[IFLA_MACVLAN_MACADDR_DATA])
1263 			return 0;
1264 
1265 		head = nla_data(data[IFLA_MACVLAN_MACADDR_DATA]);
1266 		len = nla_len(data[IFLA_MACVLAN_MACADDR_DATA]);
1267 
1268 		nla_for_each_attr(nla, head, len, rem) {
1269 			if (nla_type(nla) != IFLA_MACVLAN_MACADDR ||
1270 			    nla_len(nla) != ETH_ALEN)
1271 				continue;
1272 
1273 			addr = nla_data(nla);
1274 			ret = macvlan_hash_add_source(vlan, addr);
1275 			if (ret)
1276 				return ret;
1277 		}
1278 	} else {
1279 		return -EINVAL;
1280 	}
1281 
1282 	return 0;
1283 }
1284 
1285 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
1286 			   struct nlattr *tb[], struct nlattr *data[])
1287 {
1288 	struct macvlan_dev *vlan = netdev_priv(dev);
1289 	struct macvlan_port *port;
1290 	struct net_device *lowerdev;
1291 	int err;
1292 	int macmode;
1293 	bool create = false;
1294 
1295 	if (!tb[IFLA_LINK])
1296 		return -EINVAL;
1297 
1298 	lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
1299 	if (lowerdev == NULL)
1300 		return -ENODEV;
1301 
1302 	/* When creating macvlans or macvtaps on top of other macvlans - use
1303 	 * the real device as the lowerdev.
1304 	 */
1305 	if (netif_is_macvlan(lowerdev))
1306 		lowerdev = macvlan_dev_real_dev(lowerdev);
1307 
1308 	if (!tb[IFLA_MTU])
1309 		dev->mtu = lowerdev->mtu;
1310 	else if (dev->mtu > lowerdev->mtu)
1311 		return -EINVAL;
1312 
1313 	/* MTU range: 68 - lowerdev->max_mtu */
1314 	dev->min_mtu = ETH_MIN_MTU;
1315 	dev->max_mtu = lowerdev->max_mtu;
1316 
1317 	if (!tb[IFLA_ADDRESS])
1318 		eth_hw_addr_random(dev);
1319 
1320 	if (!macvlan_port_exists(lowerdev)) {
1321 		err = macvlan_port_create(lowerdev);
1322 		if (err < 0)
1323 			return err;
1324 		create = true;
1325 	}
1326 	port = macvlan_port_get_rtnl(lowerdev);
1327 
1328 	/* Only 1 macvlan device can be created in passthru mode */
1329 	if (port->passthru) {
1330 		/* The macvlan port must be not created this time,
1331 		 * still goto destroy_macvlan_port for readability.
1332 		 */
1333 		err = -EINVAL;
1334 		goto destroy_macvlan_port;
1335 	}
1336 
1337 	vlan->lowerdev = lowerdev;
1338 	vlan->dev      = dev;
1339 	vlan->port     = port;
1340 	vlan->set_features = MACVLAN_FEATURES;
1341 	vlan->nest_level = dev_get_nest_level(lowerdev) + 1;
1342 
1343 	vlan->mode     = MACVLAN_MODE_VEPA;
1344 	if (data && data[IFLA_MACVLAN_MODE])
1345 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1346 
1347 	if (data && data[IFLA_MACVLAN_FLAGS])
1348 		vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1349 
1350 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
1351 		if (port->count) {
1352 			err = -EINVAL;
1353 			goto destroy_macvlan_port;
1354 		}
1355 		port->passthru = true;
1356 		eth_hw_addr_inherit(dev, lowerdev);
1357 	}
1358 
1359 	if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1360 		if (vlan->mode != MACVLAN_MODE_SOURCE) {
1361 			err = -EINVAL;
1362 			goto destroy_macvlan_port;
1363 		}
1364 		macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1365 		err = macvlan_changelink_sources(vlan, macmode, data);
1366 		if (err)
1367 			goto destroy_macvlan_port;
1368 	}
1369 
1370 	err = register_netdevice(dev);
1371 	if (err < 0)
1372 		goto destroy_macvlan_port;
1373 
1374 	dev->priv_flags |= IFF_MACVLAN;
1375 	err = netdev_upper_dev_link(lowerdev, dev);
1376 	if (err)
1377 		goto unregister_netdev;
1378 
1379 	list_add_tail_rcu(&vlan->list, &port->vlans);
1380 	netif_stacked_transfer_operstate(lowerdev, dev);
1381 	linkwatch_fire_event(dev);
1382 
1383 	return 0;
1384 
1385 unregister_netdev:
1386 	unregister_netdevice(dev);
1387 destroy_macvlan_port:
1388 	if (create)
1389 		macvlan_port_destroy(port->dev);
1390 	return err;
1391 }
1392 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
1393 
1394 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
1395 			   struct nlattr *tb[], struct nlattr *data[])
1396 {
1397 	return macvlan_common_newlink(src_net, dev, tb, data);
1398 }
1399 
1400 void macvlan_dellink(struct net_device *dev, struct list_head *head)
1401 {
1402 	struct macvlan_dev *vlan = netdev_priv(dev);
1403 
1404 	if (vlan->mode == MACVLAN_MODE_SOURCE)
1405 		macvlan_flush_sources(vlan->port, vlan);
1406 	list_del_rcu(&vlan->list);
1407 	unregister_netdevice_queue(dev, head);
1408 	netdev_upper_dev_unlink(vlan->lowerdev, dev);
1409 }
1410 EXPORT_SYMBOL_GPL(macvlan_dellink);
1411 
1412 static int macvlan_changelink(struct net_device *dev,
1413 		struct nlattr *tb[], struct nlattr *data[])
1414 {
1415 	struct macvlan_dev *vlan = netdev_priv(dev);
1416 	enum macvlan_mode mode;
1417 	bool set_mode = false;
1418 	enum macvlan_macaddr_mode macmode;
1419 	int ret;
1420 
1421 	/* Validate mode, but don't set yet: setting flags may fail. */
1422 	if (data && data[IFLA_MACVLAN_MODE]) {
1423 		set_mode = true;
1424 		mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1425 		/* Passthrough mode can't be set or cleared dynamically */
1426 		if ((mode == MACVLAN_MODE_PASSTHRU) !=
1427 		    (vlan->mode == MACVLAN_MODE_PASSTHRU))
1428 			return -EINVAL;
1429 		if (vlan->mode == MACVLAN_MODE_SOURCE &&
1430 		    vlan->mode != mode)
1431 			macvlan_flush_sources(vlan->port, vlan);
1432 	}
1433 
1434 	if (data && data[IFLA_MACVLAN_FLAGS]) {
1435 		__u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1436 		bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
1437 		if (vlan->port->passthru && promisc) {
1438 			int err;
1439 
1440 			if (flags & MACVLAN_FLAG_NOPROMISC)
1441 				err = dev_set_promiscuity(vlan->lowerdev, -1);
1442 			else
1443 				err = dev_set_promiscuity(vlan->lowerdev, 1);
1444 			if (err < 0)
1445 				return err;
1446 		}
1447 		vlan->flags = flags;
1448 	}
1449 	if (set_mode)
1450 		vlan->mode = mode;
1451 	if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1452 		if (vlan->mode != MACVLAN_MODE_SOURCE)
1453 			return -EINVAL;
1454 		macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1455 		ret = macvlan_changelink_sources(vlan, macmode, data);
1456 		if (ret)
1457 			return ret;
1458 	}
1459 	return 0;
1460 }
1461 
1462 static size_t macvlan_get_size_mac(const struct macvlan_dev *vlan)
1463 {
1464 	if (vlan->macaddr_count == 0)
1465 		return 0;
1466 	return nla_total_size(0) /* IFLA_MACVLAN_MACADDR_DATA */
1467 		+ vlan->macaddr_count * nla_total_size(sizeof(u8) * ETH_ALEN);
1468 }
1469 
1470 static size_t macvlan_get_size(const struct net_device *dev)
1471 {
1472 	struct macvlan_dev *vlan = netdev_priv(dev);
1473 
1474 	return (0
1475 		+ nla_total_size(4) /* IFLA_MACVLAN_MODE */
1476 		+ nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
1477 		+ nla_total_size(4) /* IFLA_MACVLAN_MACADDR_COUNT */
1478 		+ macvlan_get_size_mac(vlan) /* IFLA_MACVLAN_MACADDR */
1479 		);
1480 }
1481 
1482 static int macvlan_fill_info_macaddr(struct sk_buff *skb,
1483 				     const struct macvlan_dev *vlan,
1484 				     const int i)
1485 {
1486 	struct hlist_head *h = &vlan->port->vlan_source_hash[i];
1487 	struct macvlan_source_entry *entry;
1488 
1489 	hlist_for_each_entry_rcu(entry, h, hlist) {
1490 		if (entry->vlan != vlan)
1491 			continue;
1492 		if (nla_put(skb, IFLA_MACVLAN_MACADDR, ETH_ALEN, entry->addr))
1493 			return 1;
1494 	}
1495 	return 0;
1496 }
1497 
1498 static int macvlan_fill_info(struct sk_buff *skb,
1499 				const struct net_device *dev)
1500 {
1501 	struct macvlan_dev *vlan = netdev_priv(dev);
1502 	int i;
1503 	struct nlattr *nest;
1504 
1505 	if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
1506 		goto nla_put_failure;
1507 	if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
1508 		goto nla_put_failure;
1509 	if (nla_put_u32(skb, IFLA_MACVLAN_MACADDR_COUNT, vlan->macaddr_count))
1510 		goto nla_put_failure;
1511 	if (vlan->macaddr_count > 0) {
1512 		nest = nla_nest_start(skb, IFLA_MACVLAN_MACADDR_DATA);
1513 		if (nest == NULL)
1514 			goto nla_put_failure;
1515 
1516 		for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
1517 			if (macvlan_fill_info_macaddr(skb, vlan, i))
1518 				goto nla_put_failure;
1519 		}
1520 		nla_nest_end(skb, nest);
1521 	}
1522 	return 0;
1523 
1524 nla_put_failure:
1525 	return -EMSGSIZE;
1526 }
1527 
1528 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
1529 	[IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
1530 	[IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
1531 	[IFLA_MACVLAN_MACADDR_MODE] = { .type = NLA_U32 },
1532 	[IFLA_MACVLAN_MACADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1533 	[IFLA_MACVLAN_MACADDR_DATA] = { .type = NLA_NESTED },
1534 	[IFLA_MACVLAN_MACADDR_COUNT] = { .type = NLA_U32 },
1535 };
1536 
1537 int macvlan_link_register(struct rtnl_link_ops *ops)
1538 {
1539 	/* common fields */
1540 	ops->validate		= macvlan_validate;
1541 	ops->maxtype		= IFLA_MACVLAN_MAX;
1542 	ops->policy		= macvlan_policy;
1543 	ops->changelink		= macvlan_changelink;
1544 	ops->get_size		= macvlan_get_size;
1545 	ops->fill_info		= macvlan_fill_info;
1546 
1547 	return rtnl_link_register(ops);
1548 };
1549 EXPORT_SYMBOL_GPL(macvlan_link_register);
1550 
1551 static struct net *macvlan_get_link_net(const struct net_device *dev)
1552 {
1553 	return dev_net(macvlan_dev_real_dev(dev));
1554 }
1555 
1556 static struct rtnl_link_ops macvlan_link_ops = {
1557 	.kind		= "macvlan",
1558 	.setup		= macvlan_setup,
1559 	.newlink	= macvlan_newlink,
1560 	.dellink	= macvlan_dellink,
1561 	.get_link_net	= macvlan_get_link_net,
1562 	.priv_size      = sizeof(struct macvlan_dev),
1563 };
1564 
1565 static int macvlan_device_event(struct notifier_block *unused,
1566 				unsigned long event, void *ptr)
1567 {
1568 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1569 	struct macvlan_dev *vlan, *next;
1570 	struct macvlan_port *port;
1571 	LIST_HEAD(list_kill);
1572 
1573 	if (!macvlan_port_exists(dev))
1574 		return NOTIFY_DONE;
1575 
1576 	port = macvlan_port_get_rtnl(dev);
1577 
1578 	switch (event) {
1579 	case NETDEV_UP:
1580 	case NETDEV_CHANGE:
1581 		list_for_each_entry(vlan, &port->vlans, list)
1582 			netif_stacked_transfer_operstate(vlan->lowerdev,
1583 							 vlan->dev);
1584 		break;
1585 	case NETDEV_FEAT_CHANGE:
1586 		list_for_each_entry(vlan, &port->vlans, list) {
1587 			vlan->dev->gso_max_size = dev->gso_max_size;
1588 			vlan->dev->gso_max_segs = dev->gso_max_segs;
1589 			netdev_update_features(vlan->dev);
1590 		}
1591 		break;
1592 	case NETDEV_CHANGEMTU:
1593 		list_for_each_entry(vlan, &port->vlans, list) {
1594 			if (vlan->dev->mtu <= dev->mtu)
1595 				continue;
1596 			dev_set_mtu(vlan->dev, dev->mtu);
1597 		}
1598 		break;
1599 	case NETDEV_CHANGEADDR:
1600 		if (!port->passthru)
1601 			return NOTIFY_DONE;
1602 
1603 		vlan = list_first_entry_or_null(&port->vlans,
1604 						struct macvlan_dev,
1605 						list);
1606 
1607 		if (macvlan_sync_address(vlan->dev, dev->dev_addr))
1608 			return NOTIFY_BAD;
1609 
1610 		break;
1611 	case NETDEV_UNREGISTER:
1612 		/* twiddle thumbs on netns device moves */
1613 		if (dev->reg_state != NETREG_UNREGISTERING)
1614 			break;
1615 
1616 		list_for_each_entry_safe(vlan, next, &port->vlans, list)
1617 			vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1618 		unregister_netdevice_many(&list_kill);
1619 		break;
1620 	case NETDEV_PRE_TYPE_CHANGE:
1621 		/* Forbid underlaying device to change its type. */
1622 		return NOTIFY_BAD;
1623 
1624 	case NETDEV_NOTIFY_PEERS:
1625 	case NETDEV_BONDING_FAILOVER:
1626 	case NETDEV_RESEND_IGMP:
1627 		/* Propagate to all vlans */
1628 		list_for_each_entry(vlan, &port->vlans, list)
1629 			call_netdevice_notifiers(event, vlan->dev);
1630 	}
1631 	return NOTIFY_DONE;
1632 }
1633 
1634 static struct notifier_block macvlan_notifier_block __read_mostly = {
1635 	.notifier_call	= macvlan_device_event,
1636 };
1637 
1638 static int __init macvlan_init_module(void)
1639 {
1640 	int err;
1641 
1642 	register_netdevice_notifier(&macvlan_notifier_block);
1643 
1644 	err = macvlan_link_register(&macvlan_link_ops);
1645 	if (err < 0)
1646 		goto err1;
1647 	return 0;
1648 err1:
1649 	unregister_netdevice_notifier(&macvlan_notifier_block);
1650 	return err;
1651 }
1652 
1653 static void __exit macvlan_cleanup_module(void)
1654 {
1655 	rtnl_link_unregister(&macvlan_link_ops);
1656 	unregister_netdevice_notifier(&macvlan_notifier_block);
1657 }
1658 
1659 module_init(macvlan_init_module);
1660 module_exit(macvlan_cleanup_module);
1661 
1662 MODULE_LICENSE("GPL");
1663 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1664 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1665 MODULE_ALIAS_RTNL_LINK("macvlan");
1666