xref: /linux/drivers/net/macvlan.c (revision 2fe05e1139a555ae91f00a812cb9520e7d3022ab)
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 #define MACVLAN_F_PASSTHRU	1
43 #define MACVLAN_F_ADDRCHANGE	2
44 
45 struct macvlan_port {
46 	struct net_device	*dev;
47 	struct hlist_head	vlan_hash[MACVLAN_HASH_SIZE];
48 	struct list_head	vlans;
49 	struct sk_buff_head	bc_queue;
50 	struct work_struct	bc_work;
51 	u32			flags;
52 	int			count;
53 	struct hlist_head	vlan_source_hash[MACVLAN_HASH_SIZE];
54 	DECLARE_BITMAP(mc_filter, MACVLAN_MC_FILTER_SZ);
55 	unsigned char           perm_addr[ETH_ALEN];
56 };
57 
58 struct macvlan_source_entry {
59 	struct hlist_node	hlist;
60 	struct macvlan_dev	*vlan;
61 	unsigned char		addr[6+2] __aligned(sizeof(u16));
62 	struct rcu_head		rcu;
63 };
64 
65 struct macvlan_skb_cb {
66 	const struct macvlan_dev *src;
67 };
68 
69 #define MACVLAN_SKB_CB(__skb) ((struct macvlan_skb_cb *)&((__skb)->cb[0]))
70 
71 static void macvlan_port_destroy(struct net_device *dev);
72 
73 static inline bool macvlan_passthru(const struct macvlan_port *port)
74 {
75 	return port->flags & MACVLAN_F_PASSTHRU;
76 }
77 
78 static inline void macvlan_set_passthru(struct macvlan_port *port)
79 {
80 	port->flags |= MACVLAN_F_PASSTHRU;
81 }
82 
83 static inline bool macvlan_addr_change(const struct macvlan_port *port)
84 {
85 	return port->flags & MACVLAN_F_ADDRCHANGE;
86 }
87 
88 static inline void macvlan_set_addr_change(struct macvlan_port *port)
89 {
90 	port->flags |= MACVLAN_F_ADDRCHANGE;
91 }
92 
93 static inline void macvlan_clear_addr_change(struct macvlan_port *port)
94 {
95 	port->flags &= ~MACVLAN_F_ADDRCHANGE;
96 }
97 
98 /* Hash Ethernet address */
99 static u32 macvlan_eth_hash(const unsigned char *addr)
100 {
101 	u64 value = get_unaligned((u64 *)addr);
102 
103 	/* only want 6 bytes */
104 #ifdef __BIG_ENDIAN
105 	value >>= 16;
106 #else
107 	value <<= 16;
108 #endif
109 	return hash_64(value, MACVLAN_HASH_BITS);
110 }
111 
112 static struct macvlan_port *macvlan_port_get_rcu(const struct net_device *dev)
113 {
114 	return rcu_dereference(dev->rx_handler_data);
115 }
116 
117 static struct macvlan_port *macvlan_port_get_rtnl(const struct net_device *dev)
118 {
119 	return rtnl_dereference(dev->rx_handler_data);
120 }
121 
122 #define macvlan_port_exists(dev) (dev->priv_flags & IFF_MACVLAN_PORT)
123 
124 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
125 					       const unsigned char *addr)
126 {
127 	struct macvlan_dev *vlan;
128 	u32 idx = macvlan_eth_hash(addr);
129 
130 	hlist_for_each_entry_rcu(vlan, &port->vlan_hash[idx], hlist) {
131 		if (ether_addr_equal_64bits(vlan->dev->dev_addr, addr))
132 			return vlan;
133 	}
134 	return NULL;
135 }
136 
137 static struct macvlan_source_entry *macvlan_hash_lookup_source(
138 	const struct macvlan_dev *vlan,
139 	const unsigned char *addr)
140 {
141 	struct macvlan_source_entry *entry;
142 	u32 idx = macvlan_eth_hash(addr);
143 	struct hlist_head *h = &vlan->port->vlan_source_hash[idx];
144 
145 	hlist_for_each_entry_rcu(entry, h, hlist) {
146 		if (ether_addr_equal_64bits(entry->addr, addr) &&
147 		    entry->vlan == vlan)
148 			return entry;
149 	}
150 	return NULL;
151 }
152 
153 static int macvlan_hash_add_source(struct macvlan_dev *vlan,
154 				   const unsigned char *addr)
155 {
156 	struct macvlan_port *port = vlan->port;
157 	struct macvlan_source_entry *entry;
158 	struct hlist_head *h;
159 
160 	entry = macvlan_hash_lookup_source(vlan, addr);
161 	if (entry)
162 		return 0;
163 
164 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
165 	if (!entry)
166 		return -ENOMEM;
167 
168 	ether_addr_copy(entry->addr, addr);
169 	entry->vlan = vlan;
170 	h = &port->vlan_source_hash[macvlan_eth_hash(addr)];
171 	hlist_add_head_rcu(&entry->hlist, h);
172 	vlan->macaddr_count++;
173 
174 	return 0;
175 }
176 
177 static void macvlan_hash_add(struct macvlan_dev *vlan)
178 {
179 	struct macvlan_port *port = vlan->port;
180 	const unsigned char *addr = vlan->dev->dev_addr;
181 	u32 idx = macvlan_eth_hash(addr);
182 
183 	hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[idx]);
184 }
185 
186 static void macvlan_hash_del_source(struct macvlan_source_entry *entry)
187 {
188 	hlist_del_rcu(&entry->hlist);
189 	kfree_rcu(entry, rcu);
190 }
191 
192 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync)
193 {
194 	hlist_del_rcu(&vlan->hlist);
195 	if (sync)
196 		synchronize_rcu();
197 }
198 
199 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
200 					const unsigned char *addr)
201 {
202 	macvlan_hash_del(vlan, true);
203 	/* Now that we are unhashed it is safe to change the device
204 	 * address without confusing packet delivery.
205 	 */
206 	memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
207 	macvlan_hash_add(vlan);
208 }
209 
210 static bool macvlan_addr_busy(const struct macvlan_port *port,
211 			      const unsigned char *addr)
212 {
213 	/* Test to see if the specified address is
214 	 * currently in use by the underlying device or
215 	 * another macvlan.
216 	 */
217 	if (!macvlan_passthru(port) && !macvlan_addr_change(port) &&
218 	    ether_addr_equal_64bits(port->dev->dev_addr, addr))
219 		return true;
220 
221 	if (macvlan_hash_lookup(port, addr))
222 		return true;
223 
224 	return false;
225 }
226 
227 
228 static int macvlan_broadcast_one(struct sk_buff *skb,
229 				 const struct macvlan_dev *vlan,
230 				 const struct ethhdr *eth, bool local)
231 {
232 	struct net_device *dev = vlan->dev;
233 
234 	if (local)
235 		return __dev_forward_skb(dev, skb);
236 
237 	skb->dev = dev;
238 	if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
239 		skb->pkt_type = PACKET_BROADCAST;
240 	else
241 		skb->pkt_type = PACKET_MULTICAST;
242 
243 	return 0;
244 }
245 
246 static u32 macvlan_hash_mix(const struct macvlan_dev *vlan)
247 {
248 	return (u32)(((unsigned long)vlan) >> L1_CACHE_SHIFT);
249 }
250 
251 
252 static unsigned int mc_hash(const struct macvlan_dev *vlan,
253 			    const unsigned char *addr)
254 {
255 	u32 val = __get_unaligned_cpu32(addr + 2);
256 
257 	val ^= macvlan_hash_mix(vlan);
258 	return hash_32(val, MACVLAN_MC_FILTER_BITS);
259 }
260 
261 static void macvlan_broadcast(struct sk_buff *skb,
262 			      const struct macvlan_port *port,
263 			      struct net_device *src,
264 			      enum macvlan_mode mode)
265 {
266 	const struct ethhdr *eth = eth_hdr(skb);
267 	const struct macvlan_dev *vlan;
268 	struct sk_buff *nskb;
269 	unsigned int i;
270 	int err;
271 	unsigned int hash;
272 
273 	if (skb->protocol == htons(ETH_P_PAUSE))
274 		return;
275 
276 	for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
277 		hlist_for_each_entry_rcu(vlan, &port->vlan_hash[i], hlist) {
278 			if (vlan->dev == src || !(vlan->mode & mode))
279 				continue;
280 
281 			hash = mc_hash(vlan, eth->h_dest);
282 			if (!test_bit(hash, vlan->mc_filter))
283 				continue;
284 
285 			err = NET_RX_DROP;
286 			nskb = skb_clone(skb, GFP_ATOMIC);
287 			if (likely(nskb))
288 				err = macvlan_broadcast_one(
289 					nskb, vlan, eth,
290 					mode == MACVLAN_MODE_BRIDGE) ?:
291 				      netif_rx_ni(nskb);
292 			macvlan_count_rx(vlan, skb->len + ETH_HLEN,
293 					 err == NET_RX_SUCCESS, true);
294 		}
295 	}
296 }
297 
298 static void macvlan_process_broadcast(struct work_struct *w)
299 {
300 	struct macvlan_port *port = container_of(w, struct macvlan_port,
301 						 bc_work);
302 	struct sk_buff *skb;
303 	struct sk_buff_head list;
304 
305 	__skb_queue_head_init(&list);
306 
307 	spin_lock_bh(&port->bc_queue.lock);
308 	skb_queue_splice_tail_init(&port->bc_queue, &list);
309 	spin_unlock_bh(&port->bc_queue.lock);
310 
311 	while ((skb = __skb_dequeue(&list))) {
312 		const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
313 
314 		rcu_read_lock();
315 
316 		if (!src)
317 			/* frame comes from an external address */
318 			macvlan_broadcast(skb, port, NULL,
319 					  MACVLAN_MODE_PRIVATE |
320 					  MACVLAN_MODE_VEPA    |
321 					  MACVLAN_MODE_PASSTHRU|
322 					  MACVLAN_MODE_BRIDGE);
323 		else if (src->mode == MACVLAN_MODE_VEPA)
324 			/* flood to everyone except source */
325 			macvlan_broadcast(skb, port, src->dev,
326 					  MACVLAN_MODE_VEPA |
327 					  MACVLAN_MODE_BRIDGE);
328 		else
329 			/*
330 			 * flood only to VEPA ports, bridge ports
331 			 * already saw the frame on the way out.
332 			 */
333 			macvlan_broadcast(skb, port, src->dev,
334 					  MACVLAN_MODE_VEPA);
335 
336 		rcu_read_unlock();
337 
338 		if (src)
339 			dev_put(src->dev);
340 		kfree_skb(skb);
341 	}
342 }
343 
344 static void macvlan_broadcast_enqueue(struct macvlan_port *port,
345 				      const struct macvlan_dev *src,
346 				      struct sk_buff *skb)
347 {
348 	struct sk_buff *nskb;
349 	int err = -ENOMEM;
350 
351 	nskb = skb_clone(skb, GFP_ATOMIC);
352 	if (!nskb)
353 		goto err;
354 
355 	MACVLAN_SKB_CB(nskb)->src = src;
356 
357 	spin_lock(&port->bc_queue.lock);
358 	if (skb_queue_len(&port->bc_queue) < MACVLAN_BC_QUEUE_LEN) {
359 		if (src)
360 			dev_hold(src->dev);
361 		__skb_queue_tail(&port->bc_queue, nskb);
362 		err = 0;
363 	}
364 	spin_unlock(&port->bc_queue.lock);
365 
366 	if (err)
367 		goto free_nskb;
368 
369 	schedule_work(&port->bc_work);
370 	return;
371 
372 free_nskb:
373 	kfree_skb(nskb);
374 err:
375 	atomic_long_inc(&skb->dev->rx_dropped);
376 }
377 
378 static void macvlan_flush_sources(struct macvlan_port *port,
379 				  struct macvlan_dev *vlan)
380 {
381 	int i;
382 
383 	for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
384 		struct hlist_node *h, *n;
385 
386 		hlist_for_each_safe(h, n, &port->vlan_source_hash[i]) {
387 			struct macvlan_source_entry *entry;
388 
389 			entry = hlist_entry(h, struct macvlan_source_entry,
390 					    hlist);
391 			if (entry->vlan == vlan)
392 				macvlan_hash_del_source(entry);
393 		}
394 	}
395 	vlan->macaddr_count = 0;
396 }
397 
398 static void macvlan_forward_source_one(struct sk_buff *skb,
399 				       struct macvlan_dev *vlan)
400 {
401 	struct sk_buff *nskb;
402 	struct net_device *dev;
403 	int len;
404 	int ret;
405 
406 	dev = vlan->dev;
407 	if (unlikely(!(dev->flags & IFF_UP)))
408 		return;
409 
410 	nskb = skb_clone(skb, GFP_ATOMIC);
411 	if (!nskb)
412 		return;
413 
414 	len = nskb->len + ETH_HLEN;
415 	nskb->dev = dev;
416 	nskb->pkt_type = PACKET_HOST;
417 
418 	ret = netif_rx(nskb);
419 	macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
420 }
421 
422 static void macvlan_forward_source(struct sk_buff *skb,
423 				   struct macvlan_port *port,
424 				   const unsigned char *addr)
425 {
426 	struct macvlan_source_entry *entry;
427 	u32 idx = macvlan_eth_hash(addr);
428 	struct hlist_head *h = &port->vlan_source_hash[idx];
429 
430 	hlist_for_each_entry_rcu(entry, h, hlist) {
431 		if (ether_addr_equal_64bits(entry->addr, addr))
432 			macvlan_forward_source_one(skb, entry->vlan);
433 	}
434 }
435 
436 /* called under rcu_read_lock() from netif_receive_skb */
437 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
438 {
439 	struct macvlan_port *port;
440 	struct sk_buff *skb = *pskb;
441 	const struct ethhdr *eth = eth_hdr(skb);
442 	const struct macvlan_dev *vlan;
443 	const struct macvlan_dev *src;
444 	struct net_device *dev;
445 	unsigned int len = 0;
446 	int ret;
447 	rx_handler_result_t handle_res;
448 
449 	port = macvlan_port_get_rcu(skb->dev);
450 	if (is_multicast_ether_addr(eth->h_dest)) {
451 		unsigned int hash;
452 
453 		skb = ip_check_defrag(dev_net(skb->dev), skb, IP_DEFRAG_MACVLAN);
454 		if (!skb)
455 			return RX_HANDLER_CONSUMED;
456 		*pskb = skb;
457 		eth = eth_hdr(skb);
458 		macvlan_forward_source(skb, port, eth->h_source);
459 		src = macvlan_hash_lookup(port, eth->h_source);
460 		if (src && src->mode != MACVLAN_MODE_VEPA &&
461 		    src->mode != MACVLAN_MODE_BRIDGE) {
462 			/* forward to original port. */
463 			vlan = src;
464 			ret = macvlan_broadcast_one(skb, vlan, eth, 0) ?:
465 			      netif_rx(skb);
466 			handle_res = RX_HANDLER_CONSUMED;
467 			goto out;
468 		}
469 
470 		hash = mc_hash(NULL, eth->h_dest);
471 		if (test_bit(hash, port->mc_filter))
472 			macvlan_broadcast_enqueue(port, src, skb);
473 
474 		return RX_HANDLER_PASS;
475 	}
476 
477 	macvlan_forward_source(skb, port, eth->h_source);
478 	if (macvlan_passthru(port))
479 		vlan = list_first_or_null_rcu(&port->vlans,
480 					      struct macvlan_dev, list);
481 	else
482 		vlan = macvlan_hash_lookup(port, eth->h_dest);
483 	if (vlan == NULL)
484 		return RX_HANDLER_PASS;
485 
486 	dev = vlan->dev;
487 	if (unlikely(!(dev->flags & IFF_UP))) {
488 		kfree_skb(skb);
489 		return RX_HANDLER_CONSUMED;
490 	}
491 	len = skb->len + ETH_HLEN;
492 	skb = skb_share_check(skb, GFP_ATOMIC);
493 	if (!skb) {
494 		ret = NET_RX_DROP;
495 		handle_res = RX_HANDLER_CONSUMED;
496 		goto out;
497 	}
498 
499 	*pskb = skb;
500 	skb->dev = dev;
501 	skb->pkt_type = PACKET_HOST;
502 
503 	ret = NET_RX_SUCCESS;
504 	handle_res = RX_HANDLER_ANOTHER;
505 out:
506 	macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
507 	return handle_res;
508 }
509 
510 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
511 {
512 	const struct macvlan_dev *vlan = netdev_priv(dev);
513 	const struct macvlan_port *port = vlan->port;
514 	const struct macvlan_dev *dest;
515 
516 	if (vlan->mode == MACVLAN_MODE_BRIDGE) {
517 		const struct ethhdr *eth = (void *)skb->data;
518 
519 		/* send to other bridge ports directly */
520 		if (is_multicast_ether_addr(eth->h_dest)) {
521 			macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
522 			goto xmit_world;
523 		}
524 
525 		dest = macvlan_hash_lookup(port, eth->h_dest);
526 		if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
527 			/* send to lowerdev first for its network taps */
528 			dev_forward_skb(vlan->lowerdev, skb);
529 
530 			return NET_XMIT_SUCCESS;
531 		}
532 	}
533 
534 xmit_world:
535 	skb->dev = vlan->lowerdev;
536 	return dev_queue_xmit(skb);
537 }
538 
539 static inline netdev_tx_t macvlan_netpoll_send_skb(struct macvlan_dev *vlan, struct sk_buff *skb)
540 {
541 #ifdef CONFIG_NET_POLL_CONTROLLER
542 	if (vlan->netpoll)
543 		netpoll_send_skb(vlan->netpoll, skb);
544 #else
545 	BUG();
546 #endif
547 	return NETDEV_TX_OK;
548 }
549 
550 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
551 				      struct net_device *dev)
552 {
553 	unsigned int len = skb->len;
554 	int ret;
555 	struct macvlan_dev *vlan = netdev_priv(dev);
556 
557 	if (unlikely(netpoll_tx_running(dev)))
558 		return macvlan_netpoll_send_skb(vlan, skb);
559 
560 	if (vlan->fwd_priv) {
561 		skb->dev = vlan->lowerdev;
562 		ret = dev_queue_xmit_accel(skb, vlan->fwd_priv);
563 	} else {
564 		ret = macvlan_queue_xmit(skb, dev);
565 	}
566 
567 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
568 		struct vlan_pcpu_stats *pcpu_stats;
569 
570 		pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
571 		u64_stats_update_begin(&pcpu_stats->syncp);
572 		pcpu_stats->tx_packets++;
573 		pcpu_stats->tx_bytes += len;
574 		u64_stats_update_end(&pcpu_stats->syncp);
575 	} else {
576 		this_cpu_inc(vlan->pcpu_stats->tx_dropped);
577 	}
578 	return ret;
579 }
580 
581 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
582 			       unsigned short type, const void *daddr,
583 			       const void *saddr, unsigned len)
584 {
585 	const struct macvlan_dev *vlan = netdev_priv(dev);
586 	struct net_device *lowerdev = vlan->lowerdev;
587 
588 	return dev_hard_header(skb, lowerdev, type, daddr,
589 			       saddr ? : dev->dev_addr, len);
590 }
591 
592 static const struct header_ops macvlan_hard_header_ops = {
593 	.create  	= macvlan_hard_header,
594 	.parse		= eth_header_parse,
595 	.cache		= eth_header_cache,
596 	.cache_update	= eth_header_cache_update,
597 };
598 
599 static struct rtnl_link_ops macvlan_link_ops;
600 
601 static int macvlan_open(struct net_device *dev)
602 {
603 	struct macvlan_dev *vlan = netdev_priv(dev);
604 	struct net_device *lowerdev = vlan->lowerdev;
605 	int err;
606 
607 	if (macvlan_passthru(vlan->port)) {
608 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
609 			err = dev_set_promiscuity(lowerdev, 1);
610 			if (err < 0)
611 				goto out;
612 		}
613 		goto hash_add;
614 	}
615 
616 	if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD &&
617 	    dev->rtnl_link_ops == &macvlan_link_ops) {
618 		vlan->fwd_priv =
619 		      lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
620 
621 		/* If we get a NULL pointer back, or if we get an error
622 		 * then we should just fall through to the non accelerated path
623 		 */
624 		if (IS_ERR_OR_NULL(vlan->fwd_priv)) {
625 			vlan->fwd_priv = NULL;
626 		} else
627 			return 0;
628 	}
629 
630 	err = -EBUSY;
631 	if (macvlan_addr_busy(vlan->port, dev->dev_addr))
632 		goto out;
633 
634 	err = dev_uc_add(lowerdev, dev->dev_addr);
635 	if (err < 0)
636 		goto out;
637 	if (dev->flags & IFF_ALLMULTI) {
638 		err = dev_set_allmulti(lowerdev, 1);
639 		if (err < 0)
640 			goto del_unicast;
641 	}
642 
643 	if (dev->flags & IFF_PROMISC) {
644 		err = dev_set_promiscuity(lowerdev, 1);
645 		if (err < 0)
646 			goto clear_multi;
647 	}
648 
649 hash_add:
650 	macvlan_hash_add(vlan);
651 	return 0;
652 
653 clear_multi:
654 	if (dev->flags & IFF_ALLMULTI)
655 		dev_set_allmulti(lowerdev, -1);
656 del_unicast:
657 	dev_uc_del(lowerdev, dev->dev_addr);
658 out:
659 	if (vlan->fwd_priv) {
660 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
661 							   vlan->fwd_priv);
662 		vlan->fwd_priv = NULL;
663 	}
664 	return err;
665 }
666 
667 static int macvlan_stop(struct net_device *dev)
668 {
669 	struct macvlan_dev *vlan = netdev_priv(dev);
670 	struct net_device *lowerdev = vlan->lowerdev;
671 
672 	if (vlan->fwd_priv) {
673 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
674 							   vlan->fwd_priv);
675 		vlan->fwd_priv = NULL;
676 		return 0;
677 	}
678 
679 	dev_uc_unsync(lowerdev, dev);
680 	dev_mc_unsync(lowerdev, dev);
681 
682 	if (macvlan_passthru(vlan->port)) {
683 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
684 			dev_set_promiscuity(lowerdev, -1);
685 		goto hash_del;
686 	}
687 
688 	if (dev->flags & IFF_ALLMULTI)
689 		dev_set_allmulti(lowerdev, -1);
690 
691 	if (dev->flags & IFF_PROMISC)
692 		dev_set_promiscuity(lowerdev, -1);
693 
694 	dev_uc_del(lowerdev, dev->dev_addr);
695 
696 hash_del:
697 	macvlan_hash_del(vlan, !dev->dismantle);
698 	return 0;
699 }
700 
701 static int macvlan_sync_address(struct net_device *dev, unsigned char *addr)
702 {
703 	struct macvlan_dev *vlan = netdev_priv(dev);
704 	struct net_device *lowerdev = vlan->lowerdev;
705 	struct macvlan_port *port = vlan->port;
706 	int err;
707 
708 	if (!(dev->flags & IFF_UP)) {
709 		/* Just copy in the new address */
710 		ether_addr_copy(dev->dev_addr, addr);
711 	} else {
712 		/* Rehash and update the device filters */
713 		if (macvlan_addr_busy(vlan->port, addr))
714 			return -EBUSY;
715 
716 		if (!macvlan_passthru(port)) {
717 			err = dev_uc_add(lowerdev, addr);
718 			if (err)
719 				return err;
720 
721 			dev_uc_del(lowerdev, dev->dev_addr);
722 		}
723 
724 		macvlan_hash_change_addr(vlan, addr);
725 	}
726 	if (macvlan_passthru(port) && !macvlan_addr_change(port)) {
727 		/* Since addr_change isn't set, we are here due to lower
728 		 * device change.  Save the lower-dev address so we can
729 		 * restore it later.
730 		 */
731 		ether_addr_copy(vlan->port->perm_addr,
732 				lowerdev->dev_addr);
733 	}
734 	macvlan_clear_addr_change(port);
735 	return 0;
736 }
737 
738 static int macvlan_set_mac_address(struct net_device *dev, void *p)
739 {
740 	struct macvlan_dev *vlan = netdev_priv(dev);
741 	struct sockaddr *addr = p;
742 
743 	if (!is_valid_ether_addr(addr->sa_data))
744 		return -EADDRNOTAVAIL;
745 
746 	/* If the addresses are the same, this is a no-op */
747 	if (ether_addr_equal(dev->dev_addr, addr->sa_data))
748 		return 0;
749 
750 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
751 		macvlan_set_addr_change(vlan->port);
752 		return dev_set_mac_address(vlan->lowerdev, addr);
753 	}
754 
755 	return macvlan_sync_address(dev, addr->sa_data);
756 }
757 
758 static void macvlan_change_rx_flags(struct net_device *dev, int change)
759 {
760 	struct macvlan_dev *vlan = netdev_priv(dev);
761 	struct net_device *lowerdev = vlan->lowerdev;
762 
763 	if (dev->flags & IFF_UP) {
764 		if (change & IFF_ALLMULTI)
765 			dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
766 		if (change & IFF_PROMISC)
767 			dev_set_promiscuity(lowerdev,
768 					    dev->flags & IFF_PROMISC ? 1 : -1);
769 
770 	}
771 }
772 
773 static void macvlan_compute_filter(unsigned long *mc_filter,
774 				   struct net_device *dev,
775 				   struct macvlan_dev *vlan)
776 {
777 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
778 		bitmap_fill(mc_filter, MACVLAN_MC_FILTER_SZ);
779 	} else {
780 		struct netdev_hw_addr *ha;
781 		DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
782 
783 		bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
784 		netdev_for_each_mc_addr(ha, dev) {
785 			__set_bit(mc_hash(vlan, ha->addr), filter);
786 		}
787 
788 		__set_bit(mc_hash(vlan, dev->broadcast), filter);
789 
790 		bitmap_copy(mc_filter, filter, MACVLAN_MC_FILTER_SZ);
791 	}
792 }
793 
794 static void macvlan_set_mac_lists(struct net_device *dev)
795 {
796 	struct macvlan_dev *vlan = netdev_priv(dev);
797 
798 	macvlan_compute_filter(vlan->mc_filter, dev, vlan);
799 
800 	dev_uc_sync(vlan->lowerdev, dev);
801 	dev_mc_sync(vlan->lowerdev, dev);
802 
803 	/* This is slightly inaccurate as we're including the subscription
804 	 * list of vlan->lowerdev too.
805 	 *
806 	 * Bug alert: This only works if everyone has the same broadcast
807 	 * address as lowerdev.  As soon as someone changes theirs this
808 	 * will break.
809 	 *
810 	 * However, this is already broken as when you change your broadcast
811 	 * address we don't get called.
812 	 *
813 	 * The solution is to maintain a list of broadcast addresses like
814 	 * we do for uc/mc, if you care.
815 	 */
816 	macvlan_compute_filter(vlan->port->mc_filter, vlan->lowerdev, NULL);
817 }
818 
819 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
820 {
821 	struct macvlan_dev *vlan = netdev_priv(dev);
822 
823 	if (vlan->lowerdev->mtu < new_mtu)
824 		return -EINVAL;
825 	dev->mtu = new_mtu;
826 	return 0;
827 }
828 
829 /*
830  * macvlan network devices have devices nesting below it and are a special
831  * "super class" of normal network devices; split their locks off into a
832  * separate class since they always nest.
833  */
834 static struct lock_class_key macvlan_netdev_addr_lock_key;
835 
836 #define ALWAYS_ON_OFFLOADS \
837 	(NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE | \
838 	 NETIF_F_GSO_ROBUST)
839 
840 #define ALWAYS_ON_FEATURES (ALWAYS_ON_OFFLOADS | NETIF_F_LLTX)
841 
842 #define MACVLAN_FEATURES \
843 	(NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
844 	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_LRO | \
845 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
846 	 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
847 
848 #define MACVLAN_STATE_MASK \
849 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
850 
851 static int macvlan_get_nest_level(struct net_device *dev)
852 {
853 	return ((struct macvlan_dev *)netdev_priv(dev))->nest_level;
854 }
855 
856 static void macvlan_set_lockdep_class(struct net_device *dev)
857 {
858 	netdev_lockdep_set_classes(dev);
859 	lockdep_set_class_and_subclass(&dev->addr_list_lock,
860 				       &macvlan_netdev_addr_lock_key,
861 				       macvlan_get_nest_level(dev));
862 }
863 
864 static int macvlan_init(struct net_device *dev)
865 {
866 	struct macvlan_dev *vlan = netdev_priv(dev);
867 	const struct net_device *lowerdev = vlan->lowerdev;
868 	struct macvlan_port *port = vlan->port;
869 
870 	dev->state		= (dev->state & ~MACVLAN_STATE_MASK) |
871 				  (lowerdev->state & MACVLAN_STATE_MASK);
872 	dev->features 		= lowerdev->features & MACVLAN_FEATURES;
873 	dev->features		|= ALWAYS_ON_FEATURES;
874 	dev->hw_features	|= NETIF_F_LRO;
875 	dev->vlan_features	= lowerdev->vlan_features & MACVLAN_FEATURES;
876 	dev->vlan_features	|= ALWAYS_ON_OFFLOADS;
877 	dev->gso_max_size	= lowerdev->gso_max_size;
878 	dev->gso_max_segs	= lowerdev->gso_max_segs;
879 	dev->hard_header_len	= lowerdev->hard_header_len;
880 
881 	macvlan_set_lockdep_class(dev);
882 
883 	vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
884 	if (!vlan->pcpu_stats)
885 		return -ENOMEM;
886 
887 	port->count += 1;
888 
889 	return 0;
890 }
891 
892 static void macvlan_uninit(struct net_device *dev)
893 {
894 	struct macvlan_dev *vlan = netdev_priv(dev);
895 	struct macvlan_port *port = vlan->port;
896 
897 	free_percpu(vlan->pcpu_stats);
898 
899 	macvlan_flush_sources(port, vlan);
900 	port->count -= 1;
901 	if (!port->count)
902 		macvlan_port_destroy(port->dev);
903 }
904 
905 static void macvlan_dev_get_stats64(struct net_device *dev,
906 				    struct rtnl_link_stats64 *stats)
907 {
908 	struct macvlan_dev *vlan = netdev_priv(dev);
909 
910 	if (vlan->pcpu_stats) {
911 		struct vlan_pcpu_stats *p;
912 		u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
913 		u32 rx_errors = 0, tx_dropped = 0;
914 		unsigned int start;
915 		int i;
916 
917 		for_each_possible_cpu(i) {
918 			p = per_cpu_ptr(vlan->pcpu_stats, i);
919 			do {
920 				start = u64_stats_fetch_begin_irq(&p->syncp);
921 				rx_packets	= p->rx_packets;
922 				rx_bytes	= p->rx_bytes;
923 				rx_multicast	= p->rx_multicast;
924 				tx_packets	= p->tx_packets;
925 				tx_bytes	= p->tx_bytes;
926 			} while (u64_stats_fetch_retry_irq(&p->syncp, start));
927 
928 			stats->rx_packets	+= rx_packets;
929 			stats->rx_bytes		+= rx_bytes;
930 			stats->multicast	+= rx_multicast;
931 			stats->tx_packets	+= tx_packets;
932 			stats->tx_bytes		+= tx_bytes;
933 			/* rx_errors & tx_dropped are u32, updated
934 			 * without syncp protection.
935 			 */
936 			rx_errors	+= p->rx_errors;
937 			tx_dropped	+= p->tx_dropped;
938 		}
939 		stats->rx_errors	= rx_errors;
940 		stats->rx_dropped	= rx_errors;
941 		stats->tx_dropped	= tx_dropped;
942 	}
943 }
944 
945 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
946 				   __be16 proto, u16 vid)
947 {
948 	struct macvlan_dev *vlan = netdev_priv(dev);
949 	struct net_device *lowerdev = vlan->lowerdev;
950 
951 	return vlan_vid_add(lowerdev, proto, vid);
952 }
953 
954 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
955 				    __be16 proto, u16 vid)
956 {
957 	struct macvlan_dev *vlan = netdev_priv(dev);
958 	struct net_device *lowerdev = vlan->lowerdev;
959 
960 	vlan_vid_del(lowerdev, proto, vid);
961 	return 0;
962 }
963 
964 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
965 			   struct net_device *dev,
966 			   const unsigned char *addr, u16 vid,
967 			   u16 flags)
968 {
969 	struct macvlan_dev *vlan = netdev_priv(dev);
970 	int err = -EINVAL;
971 
972 	/* Support unicast filter only on passthru devices.
973 	 * Multicast filter should be allowed on all devices.
974 	 */
975 	if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr))
976 		return -EOPNOTSUPP;
977 
978 	if (flags & NLM_F_REPLACE)
979 		return -EOPNOTSUPP;
980 
981 	if (is_unicast_ether_addr(addr))
982 		err = dev_uc_add_excl(dev, addr);
983 	else if (is_multicast_ether_addr(addr))
984 		err = dev_mc_add_excl(dev, addr);
985 
986 	return err;
987 }
988 
989 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
990 			   struct net_device *dev,
991 			   const unsigned char *addr, u16 vid)
992 {
993 	struct macvlan_dev *vlan = netdev_priv(dev);
994 	int err = -EINVAL;
995 
996 	/* Support unicast filter only on passthru devices.
997 	 * Multicast filter should be allowed on all devices.
998 	 */
999 	if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr))
1000 		return -EOPNOTSUPP;
1001 
1002 	if (is_unicast_ether_addr(addr))
1003 		err = dev_uc_del(dev, addr);
1004 	else if (is_multicast_ether_addr(addr))
1005 		err = dev_mc_del(dev, addr);
1006 
1007 	return err;
1008 }
1009 
1010 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
1011 					struct ethtool_drvinfo *drvinfo)
1012 {
1013 	strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
1014 	strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
1015 }
1016 
1017 static int macvlan_ethtool_get_link_ksettings(struct net_device *dev,
1018 					      struct ethtool_link_ksettings *cmd)
1019 {
1020 	const struct macvlan_dev *vlan = netdev_priv(dev);
1021 
1022 	return __ethtool_get_link_ksettings(vlan->lowerdev, cmd);
1023 }
1024 
1025 static netdev_features_t macvlan_fix_features(struct net_device *dev,
1026 					      netdev_features_t features)
1027 {
1028 	struct macvlan_dev *vlan = netdev_priv(dev);
1029 	netdev_features_t lowerdev_features = vlan->lowerdev->features;
1030 	netdev_features_t mask;
1031 
1032 	features |= NETIF_F_ALL_FOR_ALL;
1033 	features &= (vlan->set_features | ~MACVLAN_FEATURES);
1034 	mask = features;
1035 
1036 	lowerdev_features &= (features | ~NETIF_F_LRO);
1037 	features = netdev_increment_features(lowerdev_features, features, mask);
1038 	features |= ALWAYS_ON_FEATURES;
1039 	features &= ~NETIF_F_NETNS_LOCAL;
1040 
1041 	return features;
1042 }
1043 
1044 #ifdef CONFIG_NET_POLL_CONTROLLER
1045 static void macvlan_dev_poll_controller(struct net_device *dev)
1046 {
1047 	return;
1048 }
1049 
1050 static int macvlan_dev_netpoll_setup(struct net_device *dev, struct netpoll_info *npinfo)
1051 {
1052 	struct macvlan_dev *vlan = netdev_priv(dev);
1053 	struct net_device *real_dev = vlan->lowerdev;
1054 	struct netpoll *netpoll;
1055 	int err = 0;
1056 
1057 	netpoll = kzalloc(sizeof(*netpoll), GFP_KERNEL);
1058 	err = -ENOMEM;
1059 	if (!netpoll)
1060 		goto out;
1061 
1062 	err = __netpoll_setup(netpoll, real_dev);
1063 	if (err) {
1064 		kfree(netpoll);
1065 		goto out;
1066 	}
1067 
1068 	vlan->netpoll = netpoll;
1069 
1070 out:
1071 	return err;
1072 }
1073 
1074 static void macvlan_dev_netpoll_cleanup(struct net_device *dev)
1075 {
1076 	struct macvlan_dev *vlan = netdev_priv(dev);
1077 	struct netpoll *netpoll = vlan->netpoll;
1078 
1079 	if (!netpoll)
1080 		return;
1081 
1082 	vlan->netpoll = NULL;
1083 
1084 	__netpoll_free_async(netpoll);
1085 }
1086 #endif	/* CONFIG_NET_POLL_CONTROLLER */
1087 
1088 static int macvlan_dev_get_iflink(const struct net_device *dev)
1089 {
1090 	struct macvlan_dev *vlan = netdev_priv(dev);
1091 
1092 	return vlan->lowerdev->ifindex;
1093 }
1094 
1095 static const struct ethtool_ops macvlan_ethtool_ops = {
1096 	.get_link		= ethtool_op_get_link,
1097 	.get_link_ksettings	= macvlan_ethtool_get_link_ksettings,
1098 	.get_drvinfo		= macvlan_ethtool_get_drvinfo,
1099 };
1100 
1101 static const struct net_device_ops macvlan_netdev_ops = {
1102 	.ndo_init		= macvlan_init,
1103 	.ndo_uninit		= macvlan_uninit,
1104 	.ndo_open		= macvlan_open,
1105 	.ndo_stop		= macvlan_stop,
1106 	.ndo_start_xmit		= macvlan_start_xmit,
1107 	.ndo_change_mtu		= macvlan_change_mtu,
1108 	.ndo_fix_features	= macvlan_fix_features,
1109 	.ndo_change_rx_flags	= macvlan_change_rx_flags,
1110 	.ndo_set_mac_address	= macvlan_set_mac_address,
1111 	.ndo_set_rx_mode	= macvlan_set_mac_lists,
1112 	.ndo_get_stats64	= macvlan_dev_get_stats64,
1113 	.ndo_validate_addr	= eth_validate_addr,
1114 	.ndo_vlan_rx_add_vid	= macvlan_vlan_rx_add_vid,
1115 	.ndo_vlan_rx_kill_vid	= macvlan_vlan_rx_kill_vid,
1116 	.ndo_fdb_add		= macvlan_fdb_add,
1117 	.ndo_fdb_del		= macvlan_fdb_del,
1118 	.ndo_fdb_dump		= ndo_dflt_fdb_dump,
1119 	.ndo_get_lock_subclass  = macvlan_get_nest_level,
1120 #ifdef CONFIG_NET_POLL_CONTROLLER
1121 	.ndo_poll_controller	= macvlan_dev_poll_controller,
1122 	.ndo_netpoll_setup	= macvlan_dev_netpoll_setup,
1123 	.ndo_netpoll_cleanup	= macvlan_dev_netpoll_cleanup,
1124 #endif
1125 	.ndo_get_iflink		= macvlan_dev_get_iflink,
1126 	.ndo_features_check	= passthru_features_check,
1127 };
1128 
1129 void macvlan_common_setup(struct net_device *dev)
1130 {
1131 	ether_setup(dev);
1132 
1133 	dev->min_mtu		= 0;
1134 	dev->max_mtu		= ETH_MAX_MTU;
1135 	dev->priv_flags	       &= ~IFF_TX_SKB_SHARING;
1136 	netif_keep_dst(dev);
1137 	dev->priv_flags	       |= IFF_UNICAST_FLT;
1138 	dev->netdev_ops		= &macvlan_netdev_ops;
1139 	dev->needs_free_netdev	= true;
1140 	dev->header_ops		= &macvlan_hard_header_ops;
1141 	dev->ethtool_ops	= &macvlan_ethtool_ops;
1142 }
1143 EXPORT_SYMBOL_GPL(macvlan_common_setup);
1144 
1145 static void macvlan_setup(struct net_device *dev)
1146 {
1147 	macvlan_common_setup(dev);
1148 	dev->priv_flags |= IFF_NO_QUEUE;
1149 }
1150 
1151 static int macvlan_port_create(struct net_device *dev)
1152 {
1153 	struct macvlan_port *port;
1154 	unsigned int i;
1155 	int err;
1156 
1157 	if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
1158 		return -EINVAL;
1159 
1160 	if (netdev_is_rx_handler_busy(dev))
1161 		return -EBUSY;
1162 
1163 	port = kzalloc(sizeof(*port), GFP_KERNEL);
1164 	if (port == NULL)
1165 		return -ENOMEM;
1166 
1167 	port->dev = dev;
1168 	ether_addr_copy(port->perm_addr, dev->dev_addr);
1169 	INIT_LIST_HEAD(&port->vlans);
1170 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1171 		INIT_HLIST_HEAD(&port->vlan_hash[i]);
1172 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1173 		INIT_HLIST_HEAD(&port->vlan_source_hash[i]);
1174 
1175 	skb_queue_head_init(&port->bc_queue);
1176 	INIT_WORK(&port->bc_work, macvlan_process_broadcast);
1177 
1178 	err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
1179 	if (err)
1180 		kfree(port);
1181 	else
1182 		dev->priv_flags |= IFF_MACVLAN_PORT;
1183 	return err;
1184 }
1185 
1186 static void macvlan_port_destroy(struct net_device *dev)
1187 {
1188 	struct macvlan_port *port = macvlan_port_get_rtnl(dev);
1189 	struct sk_buff *skb;
1190 
1191 	dev->priv_flags &= ~IFF_MACVLAN_PORT;
1192 	netdev_rx_handler_unregister(dev);
1193 
1194 	/* After this point, no packet can schedule bc_work anymore,
1195 	 * but we need to cancel it and purge left skbs if any.
1196 	 */
1197 	cancel_work_sync(&port->bc_work);
1198 
1199 	while ((skb = __skb_dequeue(&port->bc_queue))) {
1200 		const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
1201 
1202 		if (src)
1203 			dev_put(src->dev);
1204 
1205 		kfree_skb(skb);
1206 	}
1207 
1208 	/* If the lower device address has been changed by passthru
1209 	 * macvlan, put it back.
1210 	 */
1211 	if (macvlan_passthru(port) &&
1212 	    !ether_addr_equal(port->dev->dev_addr, port->perm_addr)) {
1213 		struct sockaddr sa;
1214 
1215 		sa.sa_family = port->dev->type;
1216 		memcpy(&sa.sa_data, port->perm_addr, port->dev->addr_len);
1217 		dev_set_mac_address(port->dev, &sa);
1218 	}
1219 
1220 	kfree(port);
1221 }
1222 
1223 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[],
1224 			    struct netlink_ext_ack *extack)
1225 {
1226 	if (tb[IFLA_ADDRESS]) {
1227 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1228 			return -EINVAL;
1229 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1230 			return -EADDRNOTAVAIL;
1231 	}
1232 
1233 	if (data && data[IFLA_MACVLAN_FLAGS] &&
1234 	    nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
1235 		return -EINVAL;
1236 
1237 	if (data && data[IFLA_MACVLAN_MODE]) {
1238 		switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
1239 		case MACVLAN_MODE_PRIVATE:
1240 		case MACVLAN_MODE_VEPA:
1241 		case MACVLAN_MODE_BRIDGE:
1242 		case MACVLAN_MODE_PASSTHRU:
1243 		case MACVLAN_MODE_SOURCE:
1244 			break;
1245 		default:
1246 			return -EINVAL;
1247 		}
1248 	}
1249 
1250 	if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1251 		switch (nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE])) {
1252 		case MACVLAN_MACADDR_ADD:
1253 		case MACVLAN_MACADDR_DEL:
1254 		case MACVLAN_MACADDR_FLUSH:
1255 		case MACVLAN_MACADDR_SET:
1256 			break;
1257 		default:
1258 			return -EINVAL;
1259 		}
1260 	}
1261 
1262 	if (data && data[IFLA_MACVLAN_MACADDR]) {
1263 		if (nla_len(data[IFLA_MACVLAN_MACADDR]) != ETH_ALEN)
1264 			return -EINVAL;
1265 
1266 		if (!is_valid_ether_addr(nla_data(data[IFLA_MACVLAN_MACADDR])))
1267 			return -EADDRNOTAVAIL;
1268 	}
1269 
1270 	if (data && data[IFLA_MACVLAN_MACADDR_COUNT])
1271 		return -EINVAL;
1272 
1273 	return 0;
1274 }
1275 
1276 /**
1277  * reconfigure list of remote source mac address
1278  * (only for macvlan devices in source mode)
1279  * Note regarding alignment: all netlink data is aligned to 4 Byte, which
1280  * suffices for both ether_addr_copy and ether_addr_equal_64bits usage.
1281  */
1282 static int macvlan_changelink_sources(struct macvlan_dev *vlan, u32 mode,
1283 				      struct nlattr *data[])
1284 {
1285 	char *addr = NULL;
1286 	int ret, rem, len;
1287 	struct nlattr *nla, *head;
1288 	struct macvlan_source_entry *entry;
1289 
1290 	if (data[IFLA_MACVLAN_MACADDR])
1291 		addr = nla_data(data[IFLA_MACVLAN_MACADDR]);
1292 
1293 	if (mode == MACVLAN_MACADDR_ADD) {
1294 		if (!addr)
1295 			return -EINVAL;
1296 
1297 		return macvlan_hash_add_source(vlan, addr);
1298 
1299 	} else if (mode == MACVLAN_MACADDR_DEL) {
1300 		if (!addr)
1301 			return -EINVAL;
1302 
1303 		entry = macvlan_hash_lookup_source(vlan, addr);
1304 		if (entry) {
1305 			macvlan_hash_del_source(entry);
1306 			vlan->macaddr_count--;
1307 		}
1308 	} else if (mode == MACVLAN_MACADDR_FLUSH) {
1309 		macvlan_flush_sources(vlan->port, vlan);
1310 	} else if (mode == MACVLAN_MACADDR_SET) {
1311 		macvlan_flush_sources(vlan->port, vlan);
1312 
1313 		if (addr) {
1314 			ret = macvlan_hash_add_source(vlan, addr);
1315 			if (ret)
1316 				return ret;
1317 		}
1318 
1319 		if (!data || !data[IFLA_MACVLAN_MACADDR_DATA])
1320 			return 0;
1321 
1322 		head = nla_data(data[IFLA_MACVLAN_MACADDR_DATA]);
1323 		len = nla_len(data[IFLA_MACVLAN_MACADDR_DATA]);
1324 
1325 		nla_for_each_attr(nla, head, len, rem) {
1326 			if (nla_type(nla) != IFLA_MACVLAN_MACADDR ||
1327 			    nla_len(nla) != ETH_ALEN)
1328 				continue;
1329 
1330 			addr = nla_data(nla);
1331 			ret = macvlan_hash_add_source(vlan, addr);
1332 			if (ret)
1333 				return ret;
1334 		}
1335 	} else {
1336 		return -EINVAL;
1337 	}
1338 
1339 	return 0;
1340 }
1341 
1342 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
1343 			   struct nlattr *tb[], struct nlattr *data[])
1344 {
1345 	struct macvlan_dev *vlan = netdev_priv(dev);
1346 	struct macvlan_port *port;
1347 	struct net_device *lowerdev;
1348 	int err;
1349 	int macmode;
1350 	bool create = false;
1351 
1352 	if (!tb[IFLA_LINK])
1353 		return -EINVAL;
1354 
1355 	lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
1356 	if (lowerdev == NULL)
1357 		return -ENODEV;
1358 
1359 	/* When creating macvlans or macvtaps on top of other macvlans - use
1360 	 * the real device as the lowerdev.
1361 	 */
1362 	if (netif_is_macvlan(lowerdev))
1363 		lowerdev = macvlan_dev_real_dev(lowerdev);
1364 
1365 	if (!tb[IFLA_MTU])
1366 		dev->mtu = lowerdev->mtu;
1367 	else if (dev->mtu > lowerdev->mtu)
1368 		return -EINVAL;
1369 
1370 	/* MTU range: 68 - lowerdev->max_mtu */
1371 	dev->min_mtu = ETH_MIN_MTU;
1372 	dev->max_mtu = lowerdev->max_mtu;
1373 
1374 	if (!tb[IFLA_ADDRESS])
1375 		eth_hw_addr_random(dev);
1376 
1377 	if (!macvlan_port_exists(lowerdev)) {
1378 		err = macvlan_port_create(lowerdev);
1379 		if (err < 0)
1380 			return err;
1381 		create = true;
1382 	}
1383 	port = macvlan_port_get_rtnl(lowerdev);
1384 
1385 	/* Only 1 macvlan device can be created in passthru mode */
1386 	if (macvlan_passthru(port)) {
1387 		/* The macvlan port must be not created this time,
1388 		 * still goto destroy_macvlan_port for readability.
1389 		 */
1390 		err = -EINVAL;
1391 		goto destroy_macvlan_port;
1392 	}
1393 
1394 	vlan->lowerdev = lowerdev;
1395 	vlan->dev      = dev;
1396 	vlan->port     = port;
1397 	vlan->set_features = MACVLAN_FEATURES;
1398 	vlan->nest_level = dev_get_nest_level(lowerdev) + 1;
1399 
1400 	vlan->mode     = MACVLAN_MODE_VEPA;
1401 	if (data && data[IFLA_MACVLAN_MODE])
1402 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1403 
1404 	if (data && data[IFLA_MACVLAN_FLAGS])
1405 		vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1406 
1407 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
1408 		if (port->count) {
1409 			err = -EINVAL;
1410 			goto destroy_macvlan_port;
1411 		}
1412 		macvlan_set_passthru(port);
1413 		eth_hw_addr_inherit(dev, lowerdev);
1414 	}
1415 
1416 	if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1417 		if (vlan->mode != MACVLAN_MODE_SOURCE) {
1418 			err = -EINVAL;
1419 			goto destroy_macvlan_port;
1420 		}
1421 		macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1422 		err = macvlan_changelink_sources(vlan, macmode, data);
1423 		if (err)
1424 			goto destroy_macvlan_port;
1425 	}
1426 
1427 	err = register_netdevice(dev);
1428 	if (err < 0)
1429 		goto destroy_macvlan_port;
1430 
1431 	dev->priv_flags |= IFF_MACVLAN;
1432 	err = netdev_upper_dev_link(lowerdev, dev);
1433 	if (err)
1434 		goto unregister_netdev;
1435 
1436 	list_add_tail_rcu(&vlan->list, &port->vlans);
1437 	netif_stacked_transfer_operstate(lowerdev, dev);
1438 	linkwatch_fire_event(dev);
1439 
1440 	return 0;
1441 
1442 unregister_netdev:
1443 	unregister_netdevice(dev);
1444 destroy_macvlan_port:
1445 	if (create)
1446 		macvlan_port_destroy(port->dev);
1447 	return err;
1448 }
1449 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
1450 
1451 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
1452 			   struct nlattr *tb[], struct nlattr *data[],
1453 			   struct netlink_ext_ack *extack)
1454 {
1455 	return macvlan_common_newlink(src_net, dev, tb, data);
1456 }
1457 
1458 void macvlan_dellink(struct net_device *dev, struct list_head *head)
1459 {
1460 	struct macvlan_dev *vlan = netdev_priv(dev);
1461 
1462 	if (vlan->mode == MACVLAN_MODE_SOURCE)
1463 		macvlan_flush_sources(vlan->port, vlan);
1464 	list_del_rcu(&vlan->list);
1465 	unregister_netdevice_queue(dev, head);
1466 	netdev_upper_dev_unlink(vlan->lowerdev, dev);
1467 }
1468 EXPORT_SYMBOL_GPL(macvlan_dellink);
1469 
1470 static int macvlan_changelink(struct net_device *dev,
1471 			      struct nlattr *tb[], struct nlattr *data[],
1472 			      struct netlink_ext_ack *extack)
1473 {
1474 	struct macvlan_dev *vlan = netdev_priv(dev);
1475 	enum macvlan_mode mode;
1476 	bool set_mode = false;
1477 	enum macvlan_macaddr_mode macmode;
1478 	int ret;
1479 
1480 	/* Validate mode, but don't set yet: setting flags may fail. */
1481 	if (data && data[IFLA_MACVLAN_MODE]) {
1482 		set_mode = true;
1483 		mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1484 		/* Passthrough mode can't be set or cleared dynamically */
1485 		if ((mode == MACVLAN_MODE_PASSTHRU) !=
1486 		    (vlan->mode == MACVLAN_MODE_PASSTHRU))
1487 			return -EINVAL;
1488 		if (vlan->mode == MACVLAN_MODE_SOURCE &&
1489 		    vlan->mode != mode)
1490 			macvlan_flush_sources(vlan->port, vlan);
1491 	}
1492 
1493 	if (data && data[IFLA_MACVLAN_FLAGS]) {
1494 		__u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1495 		bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
1496 		if (macvlan_passthru(vlan->port) && promisc) {
1497 			int err;
1498 
1499 			if (flags & MACVLAN_FLAG_NOPROMISC)
1500 				err = dev_set_promiscuity(vlan->lowerdev, -1);
1501 			else
1502 				err = dev_set_promiscuity(vlan->lowerdev, 1);
1503 			if (err < 0)
1504 				return err;
1505 		}
1506 		vlan->flags = flags;
1507 	}
1508 	if (set_mode)
1509 		vlan->mode = mode;
1510 	if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1511 		if (vlan->mode != MACVLAN_MODE_SOURCE)
1512 			return -EINVAL;
1513 		macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1514 		ret = macvlan_changelink_sources(vlan, macmode, data);
1515 		if (ret)
1516 			return ret;
1517 	}
1518 	return 0;
1519 }
1520 
1521 static size_t macvlan_get_size_mac(const struct macvlan_dev *vlan)
1522 {
1523 	if (vlan->macaddr_count == 0)
1524 		return 0;
1525 	return nla_total_size(0) /* IFLA_MACVLAN_MACADDR_DATA */
1526 		+ vlan->macaddr_count * nla_total_size(sizeof(u8) * ETH_ALEN);
1527 }
1528 
1529 static size_t macvlan_get_size(const struct net_device *dev)
1530 {
1531 	struct macvlan_dev *vlan = netdev_priv(dev);
1532 
1533 	return (0
1534 		+ nla_total_size(4) /* IFLA_MACVLAN_MODE */
1535 		+ nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
1536 		+ nla_total_size(4) /* IFLA_MACVLAN_MACADDR_COUNT */
1537 		+ macvlan_get_size_mac(vlan) /* IFLA_MACVLAN_MACADDR */
1538 		);
1539 }
1540 
1541 static int macvlan_fill_info_macaddr(struct sk_buff *skb,
1542 				     const struct macvlan_dev *vlan,
1543 				     const int i)
1544 {
1545 	struct hlist_head *h = &vlan->port->vlan_source_hash[i];
1546 	struct macvlan_source_entry *entry;
1547 
1548 	hlist_for_each_entry_rcu(entry, h, hlist) {
1549 		if (entry->vlan != vlan)
1550 			continue;
1551 		if (nla_put(skb, IFLA_MACVLAN_MACADDR, ETH_ALEN, entry->addr))
1552 			return 1;
1553 	}
1554 	return 0;
1555 }
1556 
1557 static int macvlan_fill_info(struct sk_buff *skb,
1558 				const struct net_device *dev)
1559 {
1560 	struct macvlan_dev *vlan = netdev_priv(dev);
1561 	int i;
1562 	struct nlattr *nest;
1563 
1564 	if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
1565 		goto nla_put_failure;
1566 	if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
1567 		goto nla_put_failure;
1568 	if (nla_put_u32(skb, IFLA_MACVLAN_MACADDR_COUNT, vlan->macaddr_count))
1569 		goto nla_put_failure;
1570 	if (vlan->macaddr_count > 0) {
1571 		nest = nla_nest_start(skb, IFLA_MACVLAN_MACADDR_DATA);
1572 		if (nest == NULL)
1573 			goto nla_put_failure;
1574 
1575 		for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
1576 			if (macvlan_fill_info_macaddr(skb, vlan, i))
1577 				goto nla_put_failure;
1578 		}
1579 		nla_nest_end(skb, nest);
1580 	}
1581 	return 0;
1582 
1583 nla_put_failure:
1584 	return -EMSGSIZE;
1585 }
1586 
1587 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
1588 	[IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
1589 	[IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
1590 	[IFLA_MACVLAN_MACADDR_MODE] = { .type = NLA_U32 },
1591 	[IFLA_MACVLAN_MACADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1592 	[IFLA_MACVLAN_MACADDR_DATA] = { .type = NLA_NESTED },
1593 	[IFLA_MACVLAN_MACADDR_COUNT] = { .type = NLA_U32 },
1594 };
1595 
1596 int macvlan_link_register(struct rtnl_link_ops *ops)
1597 {
1598 	/* common fields */
1599 	ops->validate		= macvlan_validate;
1600 	ops->maxtype		= IFLA_MACVLAN_MAX;
1601 	ops->policy		= macvlan_policy;
1602 	ops->changelink		= macvlan_changelink;
1603 	ops->get_size		= macvlan_get_size;
1604 	ops->fill_info		= macvlan_fill_info;
1605 
1606 	return rtnl_link_register(ops);
1607 };
1608 EXPORT_SYMBOL_GPL(macvlan_link_register);
1609 
1610 static struct net *macvlan_get_link_net(const struct net_device *dev)
1611 {
1612 	return dev_net(macvlan_dev_real_dev(dev));
1613 }
1614 
1615 static struct rtnl_link_ops macvlan_link_ops = {
1616 	.kind		= "macvlan",
1617 	.setup		= macvlan_setup,
1618 	.newlink	= macvlan_newlink,
1619 	.dellink	= macvlan_dellink,
1620 	.get_link_net	= macvlan_get_link_net,
1621 	.priv_size      = sizeof(struct macvlan_dev),
1622 };
1623 
1624 static int macvlan_device_event(struct notifier_block *unused,
1625 				unsigned long event, void *ptr)
1626 {
1627 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1628 	struct macvlan_dev *vlan, *next;
1629 	struct macvlan_port *port;
1630 	LIST_HEAD(list_kill);
1631 
1632 	if (!macvlan_port_exists(dev))
1633 		return NOTIFY_DONE;
1634 
1635 	port = macvlan_port_get_rtnl(dev);
1636 
1637 	switch (event) {
1638 	case NETDEV_UP:
1639 	case NETDEV_CHANGE:
1640 		list_for_each_entry(vlan, &port->vlans, list)
1641 			netif_stacked_transfer_operstate(vlan->lowerdev,
1642 							 vlan->dev);
1643 		break;
1644 	case NETDEV_FEAT_CHANGE:
1645 		list_for_each_entry(vlan, &port->vlans, list) {
1646 			vlan->dev->gso_max_size = dev->gso_max_size;
1647 			vlan->dev->gso_max_segs = dev->gso_max_segs;
1648 			netdev_update_features(vlan->dev);
1649 		}
1650 		break;
1651 	case NETDEV_CHANGEMTU:
1652 		list_for_each_entry(vlan, &port->vlans, list) {
1653 			if (vlan->dev->mtu <= dev->mtu)
1654 				continue;
1655 			dev_set_mtu(vlan->dev, dev->mtu);
1656 		}
1657 		break;
1658 	case NETDEV_CHANGEADDR:
1659 		if (!macvlan_passthru(port))
1660 			return NOTIFY_DONE;
1661 
1662 		vlan = list_first_entry_or_null(&port->vlans,
1663 						struct macvlan_dev,
1664 						list);
1665 
1666 		if (macvlan_sync_address(vlan->dev, dev->dev_addr))
1667 			return NOTIFY_BAD;
1668 
1669 		break;
1670 	case NETDEV_UNREGISTER:
1671 		/* twiddle thumbs on netns device moves */
1672 		if (dev->reg_state != NETREG_UNREGISTERING)
1673 			break;
1674 
1675 		list_for_each_entry_safe(vlan, next, &port->vlans, list)
1676 			vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1677 		unregister_netdevice_many(&list_kill);
1678 		break;
1679 	case NETDEV_PRE_TYPE_CHANGE:
1680 		/* Forbid underlaying device to change its type. */
1681 		return NOTIFY_BAD;
1682 
1683 	case NETDEV_NOTIFY_PEERS:
1684 	case NETDEV_BONDING_FAILOVER:
1685 	case NETDEV_RESEND_IGMP:
1686 		/* Propagate to all vlans */
1687 		list_for_each_entry(vlan, &port->vlans, list)
1688 			call_netdevice_notifiers(event, vlan->dev);
1689 	}
1690 	return NOTIFY_DONE;
1691 }
1692 
1693 static struct notifier_block macvlan_notifier_block __read_mostly = {
1694 	.notifier_call	= macvlan_device_event,
1695 };
1696 
1697 static int __init macvlan_init_module(void)
1698 {
1699 	int err;
1700 
1701 	register_netdevice_notifier(&macvlan_notifier_block);
1702 
1703 	err = macvlan_link_register(&macvlan_link_ops);
1704 	if (err < 0)
1705 		goto err1;
1706 	return 0;
1707 err1:
1708 	unregister_netdevice_notifier(&macvlan_notifier_block);
1709 	return err;
1710 }
1711 
1712 static void __exit macvlan_cleanup_module(void)
1713 {
1714 	rtnl_link_unregister(&macvlan_link_ops);
1715 	unregister_netdevice_notifier(&macvlan_notifier_block);
1716 }
1717 
1718 module_init(macvlan_init_module);
1719 module_exit(macvlan_cleanup_module);
1720 
1721 MODULE_LICENSE("GPL");
1722 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1723 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1724 MODULE_ALIAS_RTNL_LINK("macvlan");
1725