xref: /linux/drivers/net/macvlan.c (revision 81fa7a69c2174ed8de314b9c231ef30a8718e5e1)
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 
417 	if (ether_addr_equal_64bits(eth_hdr(skb)->h_dest, dev->dev_addr))
418 		nskb->pkt_type = PACKET_HOST;
419 
420 	ret = netif_rx(nskb);
421 	macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
422 }
423 
424 static void macvlan_forward_source(struct sk_buff *skb,
425 				   struct macvlan_port *port,
426 				   const unsigned char *addr)
427 {
428 	struct macvlan_source_entry *entry;
429 	u32 idx = macvlan_eth_hash(addr);
430 	struct hlist_head *h = &port->vlan_source_hash[idx];
431 
432 	hlist_for_each_entry_rcu(entry, h, hlist) {
433 		if (ether_addr_equal_64bits(entry->addr, addr))
434 			macvlan_forward_source_one(skb, entry->vlan);
435 	}
436 }
437 
438 /* called under rcu_read_lock() from netif_receive_skb */
439 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
440 {
441 	struct macvlan_port *port;
442 	struct sk_buff *skb = *pskb;
443 	const struct ethhdr *eth = eth_hdr(skb);
444 	const struct macvlan_dev *vlan;
445 	const struct macvlan_dev *src;
446 	struct net_device *dev;
447 	unsigned int len = 0;
448 	int ret;
449 	rx_handler_result_t handle_res;
450 
451 	port = macvlan_port_get_rcu(skb->dev);
452 	if (is_multicast_ether_addr(eth->h_dest)) {
453 		unsigned int hash;
454 
455 		skb = ip_check_defrag(dev_net(skb->dev), skb, IP_DEFRAG_MACVLAN);
456 		if (!skb)
457 			return RX_HANDLER_CONSUMED;
458 		*pskb = skb;
459 		eth = eth_hdr(skb);
460 		macvlan_forward_source(skb, port, eth->h_source);
461 		src = macvlan_hash_lookup(port, eth->h_source);
462 		if (src && src->mode != MACVLAN_MODE_VEPA &&
463 		    src->mode != MACVLAN_MODE_BRIDGE) {
464 			/* forward to original port. */
465 			vlan = src;
466 			ret = macvlan_broadcast_one(skb, vlan, eth, 0) ?:
467 			      netif_rx(skb);
468 			handle_res = RX_HANDLER_CONSUMED;
469 			goto out;
470 		}
471 
472 		hash = mc_hash(NULL, eth->h_dest);
473 		if (test_bit(hash, port->mc_filter))
474 			macvlan_broadcast_enqueue(port, src, skb);
475 
476 		return RX_HANDLER_PASS;
477 	}
478 
479 	macvlan_forward_source(skb, port, eth->h_source);
480 	if (macvlan_passthru(port))
481 		vlan = list_first_or_null_rcu(&port->vlans,
482 					      struct macvlan_dev, list);
483 	else
484 		vlan = macvlan_hash_lookup(port, eth->h_dest);
485 	if (!vlan || vlan->mode == MACVLAN_MODE_SOURCE)
486 		return RX_HANDLER_PASS;
487 
488 	dev = vlan->dev;
489 	if (unlikely(!(dev->flags & IFF_UP))) {
490 		kfree_skb(skb);
491 		return RX_HANDLER_CONSUMED;
492 	}
493 	len = skb->len + ETH_HLEN;
494 	skb = skb_share_check(skb, GFP_ATOMIC);
495 	if (!skb) {
496 		ret = NET_RX_DROP;
497 		handle_res = RX_HANDLER_CONSUMED;
498 		goto out;
499 	}
500 
501 	*pskb = skb;
502 	skb->dev = dev;
503 	skb->pkt_type = PACKET_HOST;
504 
505 	ret = NET_RX_SUCCESS;
506 	handle_res = RX_HANDLER_ANOTHER;
507 out:
508 	macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
509 	return handle_res;
510 }
511 
512 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
513 {
514 	const struct macvlan_dev *vlan = netdev_priv(dev);
515 	const struct macvlan_port *port = vlan->port;
516 	const struct macvlan_dev *dest;
517 
518 	if (vlan->mode == MACVLAN_MODE_BRIDGE) {
519 		const struct ethhdr *eth = (void *)skb->data;
520 
521 		/* send to other bridge ports directly */
522 		if (is_multicast_ether_addr(eth->h_dest)) {
523 			macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
524 			goto xmit_world;
525 		}
526 
527 		dest = macvlan_hash_lookup(port, eth->h_dest);
528 		if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
529 			/* send to lowerdev first for its network taps */
530 			dev_forward_skb(vlan->lowerdev, skb);
531 
532 			return NET_XMIT_SUCCESS;
533 		}
534 	}
535 xmit_world:
536 	skb->dev = vlan->lowerdev;
537 	return dev_queue_xmit_accel(skb,
538 				    netdev_get_sb_channel(dev) ? dev : NULL);
539 }
540 
541 static inline netdev_tx_t macvlan_netpoll_send_skb(struct macvlan_dev *vlan, struct sk_buff *skb)
542 {
543 #ifdef CONFIG_NET_POLL_CONTROLLER
544 	if (vlan->netpoll)
545 		netpoll_send_skb(vlan->netpoll, skb);
546 #else
547 	BUG();
548 #endif
549 	return NETDEV_TX_OK;
550 }
551 
552 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
553 				      struct net_device *dev)
554 {
555 	struct macvlan_dev *vlan = netdev_priv(dev);
556 	unsigned int len = skb->len;
557 	int ret;
558 
559 	if (unlikely(netpoll_tx_running(dev)))
560 		return macvlan_netpoll_send_skb(vlan, skb);
561 
562 	ret = macvlan_queue_xmit(skb, dev);
563 
564 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
565 		struct vlan_pcpu_stats *pcpu_stats;
566 
567 		pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
568 		u64_stats_update_begin(&pcpu_stats->syncp);
569 		pcpu_stats->tx_packets++;
570 		pcpu_stats->tx_bytes += len;
571 		u64_stats_update_end(&pcpu_stats->syncp);
572 	} else {
573 		this_cpu_inc(vlan->pcpu_stats->tx_dropped);
574 	}
575 	return ret;
576 }
577 
578 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
579 			       unsigned short type, const void *daddr,
580 			       const void *saddr, unsigned len)
581 {
582 	const struct macvlan_dev *vlan = netdev_priv(dev);
583 	struct net_device *lowerdev = vlan->lowerdev;
584 
585 	return dev_hard_header(skb, lowerdev, type, daddr,
586 			       saddr ? : dev->dev_addr, len);
587 }
588 
589 static const struct header_ops macvlan_hard_header_ops = {
590 	.create  	= macvlan_hard_header,
591 	.parse		= eth_header_parse,
592 	.cache		= eth_header_cache,
593 	.cache_update	= eth_header_cache_update,
594 };
595 
596 static int macvlan_open(struct net_device *dev)
597 {
598 	struct macvlan_dev *vlan = netdev_priv(dev);
599 	struct net_device *lowerdev = vlan->lowerdev;
600 	int err;
601 
602 	if (macvlan_passthru(vlan->port)) {
603 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
604 			err = dev_set_promiscuity(lowerdev, 1);
605 			if (err < 0)
606 				goto out;
607 		}
608 		goto hash_add;
609 	}
610 
611 	err = -EBUSY;
612 	if (macvlan_addr_busy(vlan->port, dev->dev_addr))
613 		goto out;
614 
615 	/* Attempt to populate accel_priv which is used to offload the L2
616 	 * forwarding requests for unicast packets.
617 	 */
618 	if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD)
619 		vlan->accel_priv =
620 		      lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
621 
622 	/* If earlier attempt to offload failed, or accel_priv is not
623 	 * populated we must add the unicast address to the lower device.
624 	 */
625 	if (IS_ERR_OR_NULL(vlan->accel_priv)) {
626 		vlan->accel_priv = NULL;
627 		err = dev_uc_add(lowerdev, dev->dev_addr);
628 		if (err < 0)
629 			goto out;
630 	}
631 
632 	if (dev->flags & IFF_ALLMULTI) {
633 		err = dev_set_allmulti(lowerdev, 1);
634 		if (err < 0)
635 			goto del_unicast;
636 	}
637 
638 	if (dev->flags & IFF_PROMISC) {
639 		err = dev_set_promiscuity(lowerdev, 1);
640 		if (err < 0)
641 			goto clear_multi;
642 	}
643 
644 hash_add:
645 	macvlan_hash_add(vlan);
646 	return 0;
647 
648 clear_multi:
649 	if (dev->flags & IFF_ALLMULTI)
650 		dev_set_allmulti(lowerdev, -1);
651 del_unicast:
652 	if (vlan->accel_priv) {
653 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
654 							   vlan->accel_priv);
655 		vlan->accel_priv = NULL;
656 	} else {
657 		dev_uc_del(lowerdev, dev->dev_addr);
658 	}
659 out:
660 	return err;
661 }
662 
663 static int macvlan_stop(struct net_device *dev)
664 {
665 	struct macvlan_dev *vlan = netdev_priv(dev);
666 	struct net_device *lowerdev = vlan->lowerdev;
667 
668 	if (vlan->accel_priv) {
669 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
670 							   vlan->accel_priv);
671 		vlan->accel_priv = NULL;
672 	}
673 
674 	dev_uc_unsync(lowerdev, dev);
675 	dev_mc_unsync(lowerdev, dev);
676 
677 	if (macvlan_passthru(vlan->port)) {
678 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
679 			dev_set_promiscuity(lowerdev, -1);
680 		goto hash_del;
681 	}
682 
683 	if (dev->flags & IFF_ALLMULTI)
684 		dev_set_allmulti(lowerdev, -1);
685 
686 	if (dev->flags & IFF_PROMISC)
687 		dev_set_promiscuity(lowerdev, -1);
688 
689 	dev_uc_del(lowerdev, dev->dev_addr);
690 
691 hash_del:
692 	macvlan_hash_del(vlan, !dev->dismantle);
693 	return 0;
694 }
695 
696 static int macvlan_sync_address(struct net_device *dev, unsigned char *addr)
697 {
698 	struct macvlan_dev *vlan = netdev_priv(dev);
699 	struct net_device *lowerdev = vlan->lowerdev;
700 	struct macvlan_port *port = vlan->port;
701 	int err;
702 
703 	if (!(dev->flags & IFF_UP)) {
704 		/* Just copy in the new address */
705 		ether_addr_copy(dev->dev_addr, addr);
706 	} else {
707 		/* Rehash and update the device filters */
708 		if (macvlan_addr_busy(vlan->port, addr))
709 			return -EBUSY;
710 
711 		if (!macvlan_passthru(port)) {
712 			err = dev_uc_add(lowerdev, addr);
713 			if (err)
714 				return err;
715 
716 			dev_uc_del(lowerdev, dev->dev_addr);
717 		}
718 
719 		macvlan_hash_change_addr(vlan, addr);
720 	}
721 	if (macvlan_passthru(port) && !macvlan_addr_change(port)) {
722 		/* Since addr_change isn't set, we are here due to lower
723 		 * device change.  Save the lower-dev address so we can
724 		 * restore it later.
725 		 */
726 		ether_addr_copy(vlan->port->perm_addr,
727 				lowerdev->dev_addr);
728 	}
729 	macvlan_clear_addr_change(port);
730 	return 0;
731 }
732 
733 static int macvlan_set_mac_address(struct net_device *dev, void *p)
734 {
735 	struct macvlan_dev *vlan = netdev_priv(dev);
736 	struct sockaddr *addr = p;
737 
738 	if (!is_valid_ether_addr(addr->sa_data))
739 		return -EADDRNOTAVAIL;
740 
741 	/* If the addresses are the same, this is a no-op */
742 	if (ether_addr_equal(dev->dev_addr, addr->sa_data))
743 		return 0;
744 
745 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
746 		macvlan_set_addr_change(vlan->port);
747 		return dev_set_mac_address(vlan->lowerdev, addr);
748 	}
749 
750 	return macvlan_sync_address(dev, addr->sa_data);
751 }
752 
753 static void macvlan_change_rx_flags(struct net_device *dev, int change)
754 {
755 	struct macvlan_dev *vlan = netdev_priv(dev);
756 	struct net_device *lowerdev = vlan->lowerdev;
757 
758 	if (dev->flags & IFF_UP) {
759 		if (change & IFF_ALLMULTI)
760 			dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
761 		if (change & IFF_PROMISC)
762 			dev_set_promiscuity(lowerdev,
763 					    dev->flags & IFF_PROMISC ? 1 : -1);
764 
765 	}
766 }
767 
768 static void macvlan_compute_filter(unsigned long *mc_filter,
769 				   struct net_device *dev,
770 				   struct macvlan_dev *vlan)
771 {
772 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
773 		bitmap_fill(mc_filter, MACVLAN_MC_FILTER_SZ);
774 	} else {
775 		struct netdev_hw_addr *ha;
776 		DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
777 
778 		bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
779 		netdev_for_each_mc_addr(ha, dev) {
780 			__set_bit(mc_hash(vlan, ha->addr), filter);
781 		}
782 
783 		__set_bit(mc_hash(vlan, dev->broadcast), filter);
784 
785 		bitmap_copy(mc_filter, filter, MACVLAN_MC_FILTER_SZ);
786 	}
787 }
788 
789 static void macvlan_set_mac_lists(struct net_device *dev)
790 {
791 	struct macvlan_dev *vlan = netdev_priv(dev);
792 
793 	macvlan_compute_filter(vlan->mc_filter, dev, vlan);
794 
795 	dev_uc_sync(vlan->lowerdev, dev);
796 	dev_mc_sync(vlan->lowerdev, dev);
797 
798 	/* This is slightly inaccurate as we're including the subscription
799 	 * list of vlan->lowerdev too.
800 	 *
801 	 * Bug alert: This only works if everyone has the same broadcast
802 	 * address as lowerdev.  As soon as someone changes theirs this
803 	 * will break.
804 	 *
805 	 * However, this is already broken as when you change your broadcast
806 	 * address we don't get called.
807 	 *
808 	 * The solution is to maintain a list of broadcast addresses like
809 	 * we do for uc/mc, if you care.
810 	 */
811 	macvlan_compute_filter(vlan->port->mc_filter, vlan->lowerdev, NULL);
812 }
813 
814 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
815 {
816 	struct macvlan_dev *vlan = netdev_priv(dev);
817 
818 	if (vlan->lowerdev->mtu < new_mtu)
819 		return -EINVAL;
820 	dev->mtu = new_mtu;
821 	return 0;
822 }
823 
824 /*
825  * macvlan network devices have devices nesting below it and are a special
826  * "super class" of normal network devices; split their locks off into a
827  * separate class since they always nest.
828  */
829 static struct lock_class_key macvlan_netdev_addr_lock_key;
830 
831 #define ALWAYS_ON_OFFLOADS \
832 	(NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE | \
833 	 NETIF_F_GSO_ROBUST | NETIF_F_GSO_ENCAP_ALL)
834 
835 #define ALWAYS_ON_FEATURES (ALWAYS_ON_OFFLOADS | NETIF_F_LLTX)
836 
837 #define MACVLAN_FEATURES \
838 	(NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
839 	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_LRO | \
840 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
841 	 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
842 
843 #define MACVLAN_STATE_MASK \
844 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
845 
846 static int macvlan_get_nest_level(struct net_device *dev)
847 {
848 	return ((struct macvlan_dev *)netdev_priv(dev))->nest_level;
849 }
850 
851 static void macvlan_set_lockdep_class(struct net_device *dev)
852 {
853 	netdev_lockdep_set_classes(dev);
854 	lockdep_set_class_and_subclass(&dev->addr_list_lock,
855 				       &macvlan_netdev_addr_lock_key,
856 				       macvlan_get_nest_level(dev));
857 }
858 
859 static int macvlan_init(struct net_device *dev)
860 {
861 	struct macvlan_dev *vlan = netdev_priv(dev);
862 	const struct net_device *lowerdev = vlan->lowerdev;
863 	struct macvlan_port *port = vlan->port;
864 
865 	dev->state		= (dev->state & ~MACVLAN_STATE_MASK) |
866 				  (lowerdev->state & MACVLAN_STATE_MASK);
867 	dev->features 		= lowerdev->features & MACVLAN_FEATURES;
868 	dev->features		|= ALWAYS_ON_FEATURES;
869 	dev->hw_features	|= NETIF_F_LRO;
870 	dev->vlan_features	= lowerdev->vlan_features & MACVLAN_FEATURES;
871 	dev->vlan_features	|= ALWAYS_ON_OFFLOADS;
872 	dev->hw_enc_features    |= dev->features;
873 	dev->gso_max_size	= lowerdev->gso_max_size;
874 	dev->gso_max_segs	= lowerdev->gso_max_segs;
875 	dev->hard_header_len	= lowerdev->hard_header_len;
876 
877 	macvlan_set_lockdep_class(dev);
878 
879 	vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
880 	if (!vlan->pcpu_stats)
881 		return -ENOMEM;
882 
883 	port->count += 1;
884 
885 	return 0;
886 }
887 
888 static void macvlan_uninit(struct net_device *dev)
889 {
890 	struct macvlan_dev *vlan = netdev_priv(dev);
891 	struct macvlan_port *port = vlan->port;
892 
893 	free_percpu(vlan->pcpu_stats);
894 
895 	macvlan_flush_sources(port, vlan);
896 	port->count -= 1;
897 	if (!port->count)
898 		macvlan_port_destroy(port->dev);
899 }
900 
901 static void macvlan_dev_get_stats64(struct net_device *dev,
902 				    struct rtnl_link_stats64 *stats)
903 {
904 	struct macvlan_dev *vlan = netdev_priv(dev);
905 
906 	if (vlan->pcpu_stats) {
907 		struct vlan_pcpu_stats *p;
908 		u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
909 		u32 rx_errors = 0, tx_dropped = 0;
910 		unsigned int start;
911 		int i;
912 
913 		for_each_possible_cpu(i) {
914 			p = per_cpu_ptr(vlan->pcpu_stats, i);
915 			do {
916 				start = u64_stats_fetch_begin_irq(&p->syncp);
917 				rx_packets	= p->rx_packets;
918 				rx_bytes	= p->rx_bytes;
919 				rx_multicast	= p->rx_multicast;
920 				tx_packets	= p->tx_packets;
921 				tx_bytes	= p->tx_bytes;
922 			} while (u64_stats_fetch_retry_irq(&p->syncp, start));
923 
924 			stats->rx_packets	+= rx_packets;
925 			stats->rx_bytes		+= rx_bytes;
926 			stats->multicast	+= rx_multicast;
927 			stats->tx_packets	+= tx_packets;
928 			stats->tx_bytes		+= tx_bytes;
929 			/* rx_errors & tx_dropped are u32, updated
930 			 * without syncp protection.
931 			 */
932 			rx_errors	+= p->rx_errors;
933 			tx_dropped	+= p->tx_dropped;
934 		}
935 		stats->rx_errors	= rx_errors;
936 		stats->rx_dropped	= rx_errors;
937 		stats->tx_dropped	= tx_dropped;
938 	}
939 }
940 
941 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
942 				   __be16 proto, u16 vid)
943 {
944 	struct macvlan_dev *vlan = netdev_priv(dev);
945 	struct net_device *lowerdev = vlan->lowerdev;
946 
947 	return vlan_vid_add(lowerdev, proto, vid);
948 }
949 
950 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
951 				    __be16 proto, u16 vid)
952 {
953 	struct macvlan_dev *vlan = netdev_priv(dev);
954 	struct net_device *lowerdev = vlan->lowerdev;
955 
956 	vlan_vid_del(lowerdev, proto, vid);
957 	return 0;
958 }
959 
960 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
961 			   struct net_device *dev,
962 			   const unsigned char *addr, u16 vid,
963 			   u16 flags)
964 {
965 	struct macvlan_dev *vlan = netdev_priv(dev);
966 	int err = -EINVAL;
967 
968 	/* Support unicast filter only on passthru devices.
969 	 * Multicast filter should be allowed on all devices.
970 	 */
971 	if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr))
972 		return -EOPNOTSUPP;
973 
974 	if (flags & NLM_F_REPLACE)
975 		return -EOPNOTSUPP;
976 
977 	if (is_unicast_ether_addr(addr))
978 		err = dev_uc_add_excl(dev, addr);
979 	else if (is_multicast_ether_addr(addr))
980 		err = dev_mc_add_excl(dev, addr);
981 
982 	return err;
983 }
984 
985 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
986 			   struct net_device *dev,
987 			   const unsigned char *addr, u16 vid)
988 {
989 	struct macvlan_dev *vlan = netdev_priv(dev);
990 	int err = -EINVAL;
991 
992 	/* Support unicast filter only on passthru devices.
993 	 * Multicast filter should be allowed on all devices.
994 	 */
995 	if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr))
996 		return -EOPNOTSUPP;
997 
998 	if (is_unicast_ether_addr(addr))
999 		err = dev_uc_del(dev, addr);
1000 	else if (is_multicast_ether_addr(addr))
1001 		err = dev_mc_del(dev, addr);
1002 
1003 	return err;
1004 }
1005 
1006 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
1007 					struct ethtool_drvinfo *drvinfo)
1008 {
1009 	strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
1010 	strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
1011 }
1012 
1013 static int macvlan_ethtool_get_link_ksettings(struct net_device *dev,
1014 					      struct ethtool_link_ksettings *cmd)
1015 {
1016 	const struct macvlan_dev *vlan = netdev_priv(dev);
1017 
1018 	return __ethtool_get_link_ksettings(vlan->lowerdev, cmd);
1019 }
1020 
1021 static netdev_features_t macvlan_fix_features(struct net_device *dev,
1022 					      netdev_features_t features)
1023 {
1024 	struct macvlan_dev *vlan = netdev_priv(dev);
1025 	netdev_features_t lowerdev_features = vlan->lowerdev->features;
1026 	netdev_features_t mask;
1027 
1028 	features |= NETIF_F_ALL_FOR_ALL;
1029 	features &= (vlan->set_features | ~MACVLAN_FEATURES);
1030 	mask = features;
1031 
1032 	lowerdev_features &= (features | ~NETIF_F_LRO);
1033 	features = netdev_increment_features(lowerdev_features, features, mask);
1034 	features |= ALWAYS_ON_FEATURES;
1035 	features &= (ALWAYS_ON_FEATURES | MACVLAN_FEATURES);
1036 
1037 	return features;
1038 }
1039 
1040 #ifdef CONFIG_NET_POLL_CONTROLLER
1041 static void macvlan_dev_poll_controller(struct net_device *dev)
1042 {
1043 	return;
1044 }
1045 
1046 static int macvlan_dev_netpoll_setup(struct net_device *dev, struct netpoll_info *npinfo)
1047 {
1048 	struct macvlan_dev *vlan = netdev_priv(dev);
1049 	struct net_device *real_dev = vlan->lowerdev;
1050 	struct netpoll *netpoll;
1051 	int err = 0;
1052 
1053 	netpoll = kzalloc(sizeof(*netpoll), GFP_KERNEL);
1054 	err = -ENOMEM;
1055 	if (!netpoll)
1056 		goto out;
1057 
1058 	err = __netpoll_setup(netpoll, real_dev);
1059 	if (err) {
1060 		kfree(netpoll);
1061 		goto out;
1062 	}
1063 
1064 	vlan->netpoll = netpoll;
1065 
1066 out:
1067 	return err;
1068 }
1069 
1070 static void macvlan_dev_netpoll_cleanup(struct net_device *dev)
1071 {
1072 	struct macvlan_dev *vlan = netdev_priv(dev);
1073 	struct netpoll *netpoll = vlan->netpoll;
1074 
1075 	if (!netpoll)
1076 		return;
1077 
1078 	vlan->netpoll = NULL;
1079 
1080 	__netpoll_free(netpoll);
1081 }
1082 #endif	/* CONFIG_NET_POLL_CONTROLLER */
1083 
1084 static int macvlan_dev_get_iflink(const struct net_device *dev)
1085 {
1086 	struct macvlan_dev *vlan = netdev_priv(dev);
1087 
1088 	return vlan->lowerdev->ifindex;
1089 }
1090 
1091 static const struct ethtool_ops macvlan_ethtool_ops = {
1092 	.get_link		= ethtool_op_get_link,
1093 	.get_link_ksettings	= macvlan_ethtool_get_link_ksettings,
1094 	.get_drvinfo		= macvlan_ethtool_get_drvinfo,
1095 };
1096 
1097 static const struct net_device_ops macvlan_netdev_ops = {
1098 	.ndo_init		= macvlan_init,
1099 	.ndo_uninit		= macvlan_uninit,
1100 	.ndo_open		= macvlan_open,
1101 	.ndo_stop		= macvlan_stop,
1102 	.ndo_start_xmit		= macvlan_start_xmit,
1103 	.ndo_change_mtu		= macvlan_change_mtu,
1104 	.ndo_fix_features	= macvlan_fix_features,
1105 	.ndo_change_rx_flags	= macvlan_change_rx_flags,
1106 	.ndo_set_mac_address	= macvlan_set_mac_address,
1107 	.ndo_set_rx_mode	= macvlan_set_mac_lists,
1108 	.ndo_get_stats64	= macvlan_dev_get_stats64,
1109 	.ndo_validate_addr	= eth_validate_addr,
1110 	.ndo_vlan_rx_add_vid	= macvlan_vlan_rx_add_vid,
1111 	.ndo_vlan_rx_kill_vid	= macvlan_vlan_rx_kill_vid,
1112 	.ndo_fdb_add		= macvlan_fdb_add,
1113 	.ndo_fdb_del		= macvlan_fdb_del,
1114 	.ndo_fdb_dump		= ndo_dflt_fdb_dump,
1115 	.ndo_get_lock_subclass  = macvlan_get_nest_level,
1116 #ifdef CONFIG_NET_POLL_CONTROLLER
1117 	.ndo_poll_controller	= macvlan_dev_poll_controller,
1118 	.ndo_netpoll_setup	= macvlan_dev_netpoll_setup,
1119 	.ndo_netpoll_cleanup	= macvlan_dev_netpoll_cleanup,
1120 #endif
1121 	.ndo_get_iflink		= macvlan_dev_get_iflink,
1122 	.ndo_features_check	= passthru_features_check,
1123 };
1124 
1125 void macvlan_common_setup(struct net_device *dev)
1126 {
1127 	ether_setup(dev);
1128 
1129 	dev->min_mtu		= 0;
1130 	dev->max_mtu		= ETH_MAX_MTU;
1131 	dev->priv_flags	       &= ~IFF_TX_SKB_SHARING;
1132 	netif_keep_dst(dev);
1133 	dev->priv_flags	       |= IFF_UNICAST_FLT;
1134 	dev->netdev_ops		= &macvlan_netdev_ops;
1135 	dev->needs_free_netdev	= true;
1136 	dev->header_ops		= &macvlan_hard_header_ops;
1137 	dev->ethtool_ops	= &macvlan_ethtool_ops;
1138 }
1139 EXPORT_SYMBOL_GPL(macvlan_common_setup);
1140 
1141 static void macvlan_setup(struct net_device *dev)
1142 {
1143 	macvlan_common_setup(dev);
1144 	dev->priv_flags |= IFF_NO_QUEUE;
1145 }
1146 
1147 static int macvlan_port_create(struct net_device *dev)
1148 {
1149 	struct macvlan_port *port;
1150 	unsigned int i;
1151 	int err;
1152 
1153 	if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
1154 		return -EINVAL;
1155 
1156 	if (netdev_is_rx_handler_busy(dev))
1157 		return -EBUSY;
1158 
1159 	port = kzalloc(sizeof(*port), GFP_KERNEL);
1160 	if (port == NULL)
1161 		return -ENOMEM;
1162 
1163 	port->dev = dev;
1164 	ether_addr_copy(port->perm_addr, dev->dev_addr);
1165 	INIT_LIST_HEAD(&port->vlans);
1166 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1167 		INIT_HLIST_HEAD(&port->vlan_hash[i]);
1168 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1169 		INIT_HLIST_HEAD(&port->vlan_source_hash[i]);
1170 
1171 	skb_queue_head_init(&port->bc_queue);
1172 	INIT_WORK(&port->bc_work, macvlan_process_broadcast);
1173 
1174 	err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
1175 	if (err)
1176 		kfree(port);
1177 	else
1178 		dev->priv_flags |= IFF_MACVLAN_PORT;
1179 	return err;
1180 }
1181 
1182 static void macvlan_port_destroy(struct net_device *dev)
1183 {
1184 	struct macvlan_port *port = macvlan_port_get_rtnl(dev);
1185 	struct sk_buff *skb;
1186 
1187 	dev->priv_flags &= ~IFF_MACVLAN_PORT;
1188 	netdev_rx_handler_unregister(dev);
1189 
1190 	/* After this point, no packet can schedule bc_work anymore,
1191 	 * but we need to cancel it and purge left skbs if any.
1192 	 */
1193 	cancel_work_sync(&port->bc_work);
1194 
1195 	while ((skb = __skb_dequeue(&port->bc_queue))) {
1196 		const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
1197 
1198 		if (src)
1199 			dev_put(src->dev);
1200 
1201 		kfree_skb(skb);
1202 	}
1203 
1204 	/* If the lower device address has been changed by passthru
1205 	 * macvlan, put it back.
1206 	 */
1207 	if (macvlan_passthru(port) &&
1208 	    !ether_addr_equal(port->dev->dev_addr, port->perm_addr)) {
1209 		struct sockaddr sa;
1210 
1211 		sa.sa_family = port->dev->type;
1212 		memcpy(&sa.sa_data, port->perm_addr, port->dev->addr_len);
1213 		dev_set_mac_address(port->dev, &sa);
1214 	}
1215 
1216 	kfree(port);
1217 }
1218 
1219 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[],
1220 			    struct netlink_ext_ack *extack)
1221 {
1222 	if (tb[IFLA_ADDRESS]) {
1223 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1224 			return -EINVAL;
1225 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1226 			return -EADDRNOTAVAIL;
1227 	}
1228 
1229 	if (!data)
1230 		return 0;
1231 
1232 	if (data[IFLA_MACVLAN_FLAGS] &&
1233 	    nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
1234 		return -EINVAL;
1235 
1236 	if (data[IFLA_MACVLAN_MODE]) {
1237 		switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
1238 		case MACVLAN_MODE_PRIVATE:
1239 		case MACVLAN_MODE_VEPA:
1240 		case MACVLAN_MODE_BRIDGE:
1241 		case MACVLAN_MODE_PASSTHRU:
1242 		case MACVLAN_MODE_SOURCE:
1243 			break;
1244 		default:
1245 			return -EINVAL;
1246 		}
1247 	}
1248 
1249 	if (data[IFLA_MACVLAN_MACADDR_MODE]) {
1250 		switch (nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE])) {
1251 		case MACVLAN_MACADDR_ADD:
1252 		case MACVLAN_MACADDR_DEL:
1253 		case MACVLAN_MACADDR_FLUSH:
1254 		case MACVLAN_MACADDR_SET:
1255 			break;
1256 		default:
1257 			return -EINVAL;
1258 		}
1259 	}
1260 
1261 	if (data[IFLA_MACVLAN_MACADDR]) {
1262 		if (nla_len(data[IFLA_MACVLAN_MACADDR]) != ETH_ALEN)
1263 			return -EINVAL;
1264 
1265 		if (!is_valid_ether_addr(nla_data(data[IFLA_MACVLAN_MACADDR])))
1266 			return -EADDRNOTAVAIL;
1267 	}
1268 
1269 	if (data[IFLA_MACVLAN_MACADDR_COUNT])
1270 		return -EINVAL;
1271 
1272 	return 0;
1273 }
1274 
1275 /**
1276  * reconfigure list of remote source mac address
1277  * (only for macvlan devices in source mode)
1278  * Note regarding alignment: all netlink data is aligned to 4 Byte, which
1279  * suffices for both ether_addr_copy and ether_addr_equal_64bits usage.
1280  */
1281 static int macvlan_changelink_sources(struct macvlan_dev *vlan, u32 mode,
1282 				      struct nlattr *data[])
1283 {
1284 	char *addr = NULL;
1285 	int ret, rem, len;
1286 	struct nlattr *nla, *head;
1287 	struct macvlan_source_entry *entry;
1288 
1289 	if (data[IFLA_MACVLAN_MACADDR])
1290 		addr = nla_data(data[IFLA_MACVLAN_MACADDR]);
1291 
1292 	if (mode == MACVLAN_MACADDR_ADD) {
1293 		if (!addr)
1294 			return -EINVAL;
1295 
1296 		return macvlan_hash_add_source(vlan, addr);
1297 
1298 	} else if (mode == MACVLAN_MACADDR_DEL) {
1299 		if (!addr)
1300 			return -EINVAL;
1301 
1302 		entry = macvlan_hash_lookup_source(vlan, addr);
1303 		if (entry) {
1304 			macvlan_hash_del_source(entry);
1305 			vlan->macaddr_count--;
1306 		}
1307 	} else if (mode == MACVLAN_MACADDR_FLUSH) {
1308 		macvlan_flush_sources(vlan->port, vlan);
1309 	} else if (mode == MACVLAN_MACADDR_SET) {
1310 		macvlan_flush_sources(vlan->port, vlan);
1311 
1312 		if (addr) {
1313 			ret = macvlan_hash_add_source(vlan, addr);
1314 			if (ret)
1315 				return ret;
1316 		}
1317 
1318 		if (!data || !data[IFLA_MACVLAN_MACADDR_DATA])
1319 			return 0;
1320 
1321 		head = nla_data(data[IFLA_MACVLAN_MACADDR_DATA]);
1322 		len = nla_len(data[IFLA_MACVLAN_MACADDR_DATA]);
1323 
1324 		nla_for_each_attr(nla, head, len, rem) {
1325 			if (nla_type(nla) != IFLA_MACVLAN_MACADDR ||
1326 			    nla_len(nla) != ETH_ALEN)
1327 				continue;
1328 
1329 			addr = nla_data(nla);
1330 			ret = macvlan_hash_add_source(vlan, addr);
1331 			if (ret)
1332 				return ret;
1333 		}
1334 	} else {
1335 		return -EINVAL;
1336 	}
1337 
1338 	return 0;
1339 }
1340 
1341 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
1342 			   struct nlattr *tb[], struct nlattr *data[],
1343 			   struct netlink_ext_ack *extack)
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, extack);
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 	/* macvlan_uninit would free the macvlan port */
1444 	unregister_netdevice(dev);
1445 	return err;
1446 destroy_macvlan_port:
1447 	/* the macvlan port may be freed by macvlan_uninit when fail to register.
1448 	 * so we destroy the macvlan port only when it's valid.
1449 	 */
1450 	if (create && macvlan_port_get_rtnl(lowerdev))
1451 		macvlan_port_destroy(port->dev);
1452 	return err;
1453 }
1454 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
1455 
1456 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
1457 			   struct nlattr *tb[], struct nlattr *data[],
1458 			   struct netlink_ext_ack *extack)
1459 {
1460 	return macvlan_common_newlink(src_net, dev, tb, data, extack);
1461 }
1462 
1463 void macvlan_dellink(struct net_device *dev, struct list_head *head)
1464 {
1465 	struct macvlan_dev *vlan = netdev_priv(dev);
1466 
1467 	if (vlan->mode == MACVLAN_MODE_SOURCE)
1468 		macvlan_flush_sources(vlan->port, vlan);
1469 	list_del_rcu(&vlan->list);
1470 	unregister_netdevice_queue(dev, head);
1471 	netdev_upper_dev_unlink(vlan->lowerdev, dev);
1472 }
1473 EXPORT_SYMBOL_GPL(macvlan_dellink);
1474 
1475 static int macvlan_changelink(struct net_device *dev,
1476 			      struct nlattr *tb[], struct nlattr *data[],
1477 			      struct netlink_ext_ack *extack)
1478 {
1479 	struct macvlan_dev *vlan = netdev_priv(dev);
1480 	enum macvlan_mode mode;
1481 	bool set_mode = false;
1482 	enum macvlan_macaddr_mode macmode;
1483 	int ret;
1484 
1485 	/* Validate mode, but don't set yet: setting flags may fail. */
1486 	if (data && data[IFLA_MACVLAN_MODE]) {
1487 		set_mode = true;
1488 		mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1489 		/* Passthrough mode can't be set or cleared dynamically */
1490 		if ((mode == MACVLAN_MODE_PASSTHRU) !=
1491 		    (vlan->mode == MACVLAN_MODE_PASSTHRU))
1492 			return -EINVAL;
1493 		if (vlan->mode == MACVLAN_MODE_SOURCE &&
1494 		    vlan->mode != mode)
1495 			macvlan_flush_sources(vlan->port, vlan);
1496 	}
1497 
1498 	if (data && data[IFLA_MACVLAN_FLAGS]) {
1499 		__u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1500 		bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
1501 		if (macvlan_passthru(vlan->port) && promisc) {
1502 			int err;
1503 
1504 			if (flags & MACVLAN_FLAG_NOPROMISC)
1505 				err = dev_set_promiscuity(vlan->lowerdev, -1);
1506 			else
1507 				err = dev_set_promiscuity(vlan->lowerdev, 1);
1508 			if (err < 0)
1509 				return err;
1510 		}
1511 		vlan->flags = flags;
1512 	}
1513 	if (set_mode)
1514 		vlan->mode = mode;
1515 	if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1516 		if (vlan->mode != MACVLAN_MODE_SOURCE)
1517 			return -EINVAL;
1518 		macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1519 		ret = macvlan_changelink_sources(vlan, macmode, data);
1520 		if (ret)
1521 			return ret;
1522 	}
1523 	return 0;
1524 }
1525 
1526 static size_t macvlan_get_size_mac(const struct macvlan_dev *vlan)
1527 {
1528 	if (vlan->macaddr_count == 0)
1529 		return 0;
1530 	return nla_total_size(0) /* IFLA_MACVLAN_MACADDR_DATA */
1531 		+ vlan->macaddr_count * nla_total_size(sizeof(u8) * ETH_ALEN);
1532 }
1533 
1534 static size_t macvlan_get_size(const struct net_device *dev)
1535 {
1536 	struct macvlan_dev *vlan = netdev_priv(dev);
1537 
1538 	return (0
1539 		+ nla_total_size(4) /* IFLA_MACVLAN_MODE */
1540 		+ nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
1541 		+ nla_total_size(4) /* IFLA_MACVLAN_MACADDR_COUNT */
1542 		+ macvlan_get_size_mac(vlan) /* IFLA_MACVLAN_MACADDR */
1543 		);
1544 }
1545 
1546 static int macvlan_fill_info_macaddr(struct sk_buff *skb,
1547 				     const struct macvlan_dev *vlan,
1548 				     const int i)
1549 {
1550 	struct hlist_head *h = &vlan->port->vlan_source_hash[i];
1551 	struct macvlan_source_entry *entry;
1552 
1553 	hlist_for_each_entry_rcu(entry, h, hlist) {
1554 		if (entry->vlan != vlan)
1555 			continue;
1556 		if (nla_put(skb, IFLA_MACVLAN_MACADDR, ETH_ALEN, entry->addr))
1557 			return 1;
1558 	}
1559 	return 0;
1560 }
1561 
1562 static int macvlan_fill_info(struct sk_buff *skb,
1563 				const struct net_device *dev)
1564 {
1565 	struct macvlan_dev *vlan = netdev_priv(dev);
1566 	int i;
1567 	struct nlattr *nest;
1568 
1569 	if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
1570 		goto nla_put_failure;
1571 	if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
1572 		goto nla_put_failure;
1573 	if (nla_put_u32(skb, IFLA_MACVLAN_MACADDR_COUNT, vlan->macaddr_count))
1574 		goto nla_put_failure;
1575 	if (vlan->macaddr_count > 0) {
1576 		nest = nla_nest_start(skb, IFLA_MACVLAN_MACADDR_DATA);
1577 		if (nest == NULL)
1578 			goto nla_put_failure;
1579 
1580 		for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
1581 			if (macvlan_fill_info_macaddr(skb, vlan, i))
1582 				goto nla_put_failure;
1583 		}
1584 		nla_nest_end(skb, nest);
1585 	}
1586 	return 0;
1587 
1588 nla_put_failure:
1589 	return -EMSGSIZE;
1590 }
1591 
1592 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
1593 	[IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
1594 	[IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
1595 	[IFLA_MACVLAN_MACADDR_MODE] = { .type = NLA_U32 },
1596 	[IFLA_MACVLAN_MACADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1597 	[IFLA_MACVLAN_MACADDR_DATA] = { .type = NLA_NESTED },
1598 	[IFLA_MACVLAN_MACADDR_COUNT] = { .type = NLA_U32 },
1599 };
1600 
1601 int macvlan_link_register(struct rtnl_link_ops *ops)
1602 {
1603 	/* common fields */
1604 	ops->validate		= macvlan_validate;
1605 	ops->maxtype		= IFLA_MACVLAN_MAX;
1606 	ops->policy		= macvlan_policy;
1607 	ops->changelink		= macvlan_changelink;
1608 	ops->get_size		= macvlan_get_size;
1609 	ops->fill_info		= macvlan_fill_info;
1610 
1611 	return rtnl_link_register(ops);
1612 };
1613 EXPORT_SYMBOL_GPL(macvlan_link_register);
1614 
1615 static struct net *macvlan_get_link_net(const struct net_device *dev)
1616 {
1617 	return dev_net(macvlan_dev_real_dev(dev));
1618 }
1619 
1620 static struct rtnl_link_ops macvlan_link_ops = {
1621 	.kind		= "macvlan",
1622 	.setup		= macvlan_setup,
1623 	.newlink	= macvlan_newlink,
1624 	.dellink	= macvlan_dellink,
1625 	.get_link_net	= macvlan_get_link_net,
1626 	.priv_size      = sizeof(struct macvlan_dev),
1627 };
1628 
1629 static int macvlan_device_event(struct notifier_block *unused,
1630 				unsigned long event, void *ptr)
1631 {
1632 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1633 	struct macvlan_dev *vlan, *next;
1634 	struct macvlan_port *port;
1635 	LIST_HEAD(list_kill);
1636 
1637 	if (!macvlan_port_exists(dev))
1638 		return NOTIFY_DONE;
1639 
1640 	port = macvlan_port_get_rtnl(dev);
1641 
1642 	switch (event) {
1643 	case NETDEV_UP:
1644 	case NETDEV_DOWN:
1645 	case NETDEV_CHANGE:
1646 		list_for_each_entry(vlan, &port->vlans, list)
1647 			netif_stacked_transfer_operstate(vlan->lowerdev,
1648 							 vlan->dev);
1649 		break;
1650 	case NETDEV_FEAT_CHANGE:
1651 		list_for_each_entry(vlan, &port->vlans, list) {
1652 			vlan->dev->gso_max_size = dev->gso_max_size;
1653 			vlan->dev->gso_max_segs = dev->gso_max_segs;
1654 			netdev_update_features(vlan->dev);
1655 		}
1656 		break;
1657 	case NETDEV_CHANGEMTU:
1658 		list_for_each_entry(vlan, &port->vlans, list) {
1659 			if (vlan->dev->mtu <= dev->mtu)
1660 				continue;
1661 			dev_set_mtu(vlan->dev, dev->mtu);
1662 		}
1663 		break;
1664 	case NETDEV_CHANGEADDR:
1665 		if (!macvlan_passthru(port))
1666 			return NOTIFY_DONE;
1667 
1668 		vlan = list_first_entry_or_null(&port->vlans,
1669 						struct macvlan_dev,
1670 						list);
1671 
1672 		if (macvlan_sync_address(vlan->dev, dev->dev_addr))
1673 			return NOTIFY_BAD;
1674 
1675 		break;
1676 	case NETDEV_UNREGISTER:
1677 		/* twiddle thumbs on netns device moves */
1678 		if (dev->reg_state != NETREG_UNREGISTERING)
1679 			break;
1680 
1681 		list_for_each_entry_safe(vlan, next, &port->vlans, list)
1682 			vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1683 		unregister_netdevice_many(&list_kill);
1684 		break;
1685 	case NETDEV_PRE_TYPE_CHANGE:
1686 		/* Forbid underlaying device to change its type. */
1687 		return NOTIFY_BAD;
1688 
1689 	case NETDEV_NOTIFY_PEERS:
1690 	case NETDEV_BONDING_FAILOVER:
1691 	case NETDEV_RESEND_IGMP:
1692 		/* Propagate to all vlans */
1693 		list_for_each_entry(vlan, &port->vlans, list)
1694 			call_netdevice_notifiers(event, vlan->dev);
1695 	}
1696 	return NOTIFY_DONE;
1697 }
1698 
1699 static struct notifier_block macvlan_notifier_block __read_mostly = {
1700 	.notifier_call	= macvlan_device_event,
1701 };
1702 
1703 static int __init macvlan_init_module(void)
1704 {
1705 	int err;
1706 
1707 	register_netdevice_notifier(&macvlan_notifier_block);
1708 
1709 	err = macvlan_link_register(&macvlan_link_ops);
1710 	if (err < 0)
1711 		goto err1;
1712 	return 0;
1713 err1:
1714 	unregister_netdevice_notifier(&macvlan_notifier_block);
1715 	return err;
1716 }
1717 
1718 static void __exit macvlan_cleanup_module(void)
1719 {
1720 	rtnl_link_unregister(&macvlan_link_ops);
1721 	unregister_netdevice_notifier(&macvlan_notifier_block);
1722 }
1723 
1724 module_init(macvlan_init_module);
1725 module_exit(macvlan_cleanup_module);
1726 
1727 MODULE_LICENSE("GPL");
1728 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1729 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1730 MODULE_ALIAS_RTNL_LINK("macvlan");
1731