xref: /openbmc/linux/drivers/net/macvlan.c (revision df687341)
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 = -EADDRINUSE;
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 -EADDRINUSE;
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, NULL);
748 	}
749 
750 	if (macvlan_addr_busy(vlan->port, addr->sa_data))
751 		return -EADDRINUSE;
752 
753 	return macvlan_sync_address(dev, addr->sa_data);
754 }
755 
756 static void macvlan_change_rx_flags(struct net_device *dev, int change)
757 {
758 	struct macvlan_dev *vlan = netdev_priv(dev);
759 	struct net_device *lowerdev = vlan->lowerdev;
760 
761 	if (dev->flags & IFF_UP) {
762 		if (change & IFF_ALLMULTI)
763 			dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
764 		if (change & IFF_PROMISC)
765 			dev_set_promiscuity(lowerdev,
766 					    dev->flags & IFF_PROMISC ? 1 : -1);
767 
768 	}
769 }
770 
771 static void macvlan_compute_filter(unsigned long *mc_filter,
772 				   struct net_device *dev,
773 				   struct macvlan_dev *vlan)
774 {
775 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
776 		bitmap_fill(mc_filter, MACVLAN_MC_FILTER_SZ);
777 	} else {
778 		struct netdev_hw_addr *ha;
779 		DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
780 
781 		bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
782 		netdev_for_each_mc_addr(ha, dev) {
783 			__set_bit(mc_hash(vlan, ha->addr), filter);
784 		}
785 
786 		__set_bit(mc_hash(vlan, dev->broadcast), filter);
787 
788 		bitmap_copy(mc_filter, filter, MACVLAN_MC_FILTER_SZ);
789 	}
790 }
791 
792 static void macvlan_set_mac_lists(struct net_device *dev)
793 {
794 	struct macvlan_dev *vlan = netdev_priv(dev);
795 
796 	macvlan_compute_filter(vlan->mc_filter, dev, vlan);
797 
798 	dev_uc_sync(vlan->lowerdev, dev);
799 	dev_mc_sync(vlan->lowerdev, dev);
800 
801 	/* This is slightly inaccurate as we're including the subscription
802 	 * list of vlan->lowerdev too.
803 	 *
804 	 * Bug alert: This only works if everyone has the same broadcast
805 	 * address as lowerdev.  As soon as someone changes theirs this
806 	 * will break.
807 	 *
808 	 * However, this is already broken as when you change your broadcast
809 	 * address we don't get called.
810 	 *
811 	 * The solution is to maintain a list of broadcast addresses like
812 	 * we do for uc/mc, if you care.
813 	 */
814 	macvlan_compute_filter(vlan->port->mc_filter, vlan->lowerdev, NULL);
815 }
816 
817 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
818 {
819 	struct macvlan_dev *vlan = netdev_priv(dev);
820 
821 	if (vlan->lowerdev->mtu < new_mtu)
822 		return -EINVAL;
823 	dev->mtu = new_mtu;
824 	return 0;
825 }
826 
827 /*
828  * macvlan network devices have devices nesting below it and are a special
829  * "super class" of normal network devices; split their locks off into a
830  * separate class since they always nest.
831  */
832 static struct lock_class_key macvlan_netdev_addr_lock_key;
833 
834 #define ALWAYS_ON_OFFLOADS \
835 	(NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE | \
836 	 NETIF_F_GSO_ROBUST | NETIF_F_GSO_ENCAP_ALL)
837 
838 #define ALWAYS_ON_FEATURES (ALWAYS_ON_OFFLOADS | NETIF_F_LLTX)
839 
840 #define MACVLAN_FEATURES \
841 	(NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
842 	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_LRO | \
843 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
844 	 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
845 
846 #define MACVLAN_STATE_MASK \
847 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
848 
849 static int macvlan_get_nest_level(struct net_device *dev)
850 {
851 	return ((struct macvlan_dev *)netdev_priv(dev))->nest_level;
852 }
853 
854 static void macvlan_set_lockdep_class(struct net_device *dev)
855 {
856 	netdev_lockdep_set_classes(dev);
857 	lockdep_set_class_and_subclass(&dev->addr_list_lock,
858 				       &macvlan_netdev_addr_lock_key,
859 				       macvlan_get_nest_level(dev));
860 }
861 
862 static int macvlan_init(struct net_device *dev)
863 {
864 	struct macvlan_dev *vlan = netdev_priv(dev);
865 	const struct net_device *lowerdev = vlan->lowerdev;
866 	struct macvlan_port *port = vlan->port;
867 
868 	dev->state		= (dev->state & ~MACVLAN_STATE_MASK) |
869 				  (lowerdev->state & MACVLAN_STATE_MASK);
870 	dev->features 		= lowerdev->features & MACVLAN_FEATURES;
871 	dev->features		|= ALWAYS_ON_FEATURES;
872 	dev->hw_features	|= NETIF_F_LRO;
873 	dev->vlan_features	= lowerdev->vlan_features & MACVLAN_FEATURES;
874 	dev->vlan_features	|= ALWAYS_ON_OFFLOADS;
875 	dev->hw_enc_features    |= dev->features;
876 	dev->gso_max_size	= lowerdev->gso_max_size;
877 	dev->gso_max_segs	= lowerdev->gso_max_segs;
878 	dev->hard_header_len	= lowerdev->hard_header_len;
879 
880 	macvlan_set_lockdep_class(dev);
881 
882 	vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
883 	if (!vlan->pcpu_stats)
884 		return -ENOMEM;
885 
886 	port->count += 1;
887 
888 	return 0;
889 }
890 
891 static void macvlan_uninit(struct net_device *dev)
892 {
893 	struct macvlan_dev *vlan = netdev_priv(dev);
894 	struct macvlan_port *port = vlan->port;
895 
896 	free_percpu(vlan->pcpu_stats);
897 
898 	macvlan_flush_sources(port, vlan);
899 	port->count -= 1;
900 	if (!port->count)
901 		macvlan_port_destroy(port->dev);
902 }
903 
904 static void macvlan_dev_get_stats64(struct net_device *dev,
905 				    struct rtnl_link_stats64 *stats)
906 {
907 	struct macvlan_dev *vlan = netdev_priv(dev);
908 
909 	if (vlan->pcpu_stats) {
910 		struct vlan_pcpu_stats *p;
911 		u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
912 		u32 rx_errors = 0, tx_dropped = 0;
913 		unsigned int start;
914 		int i;
915 
916 		for_each_possible_cpu(i) {
917 			p = per_cpu_ptr(vlan->pcpu_stats, i);
918 			do {
919 				start = u64_stats_fetch_begin_irq(&p->syncp);
920 				rx_packets	= p->rx_packets;
921 				rx_bytes	= p->rx_bytes;
922 				rx_multicast	= p->rx_multicast;
923 				tx_packets	= p->tx_packets;
924 				tx_bytes	= p->tx_bytes;
925 			} while (u64_stats_fetch_retry_irq(&p->syncp, start));
926 
927 			stats->rx_packets	+= rx_packets;
928 			stats->rx_bytes		+= rx_bytes;
929 			stats->multicast	+= rx_multicast;
930 			stats->tx_packets	+= tx_packets;
931 			stats->tx_bytes		+= tx_bytes;
932 			/* rx_errors & tx_dropped are u32, updated
933 			 * without syncp protection.
934 			 */
935 			rx_errors	+= p->rx_errors;
936 			tx_dropped	+= p->tx_dropped;
937 		}
938 		stats->rx_errors	= rx_errors;
939 		stats->rx_dropped	= rx_errors;
940 		stats->tx_dropped	= tx_dropped;
941 	}
942 }
943 
944 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
945 				   __be16 proto, u16 vid)
946 {
947 	struct macvlan_dev *vlan = netdev_priv(dev);
948 	struct net_device *lowerdev = vlan->lowerdev;
949 
950 	return vlan_vid_add(lowerdev, proto, vid);
951 }
952 
953 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
954 				    __be16 proto, u16 vid)
955 {
956 	struct macvlan_dev *vlan = netdev_priv(dev);
957 	struct net_device *lowerdev = vlan->lowerdev;
958 
959 	vlan_vid_del(lowerdev, proto, vid);
960 	return 0;
961 }
962 
963 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
964 			   struct net_device *dev,
965 			   const unsigned char *addr, u16 vid,
966 			   u16 flags)
967 {
968 	struct macvlan_dev *vlan = netdev_priv(dev);
969 	int err = -EINVAL;
970 
971 	/* Support unicast filter only on passthru devices.
972 	 * Multicast filter should be allowed on all devices.
973 	 */
974 	if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr))
975 		return -EOPNOTSUPP;
976 
977 	if (flags & NLM_F_REPLACE)
978 		return -EOPNOTSUPP;
979 
980 	if (is_unicast_ether_addr(addr))
981 		err = dev_uc_add_excl(dev, addr);
982 	else if (is_multicast_ether_addr(addr))
983 		err = dev_mc_add_excl(dev, addr);
984 
985 	return err;
986 }
987 
988 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
989 			   struct net_device *dev,
990 			   const unsigned char *addr, u16 vid)
991 {
992 	struct macvlan_dev *vlan = netdev_priv(dev);
993 	int err = -EINVAL;
994 
995 	/* Support unicast filter only on passthru devices.
996 	 * Multicast filter should be allowed on all devices.
997 	 */
998 	if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr))
999 		return -EOPNOTSUPP;
1000 
1001 	if (is_unicast_ether_addr(addr))
1002 		err = dev_uc_del(dev, addr);
1003 	else if (is_multicast_ether_addr(addr))
1004 		err = dev_mc_del(dev, addr);
1005 
1006 	return err;
1007 }
1008 
1009 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
1010 					struct ethtool_drvinfo *drvinfo)
1011 {
1012 	strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
1013 	strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
1014 }
1015 
1016 static int macvlan_ethtool_get_link_ksettings(struct net_device *dev,
1017 					      struct ethtool_link_ksettings *cmd)
1018 {
1019 	const struct macvlan_dev *vlan = netdev_priv(dev);
1020 
1021 	return __ethtool_get_link_ksettings(vlan->lowerdev, cmd);
1022 }
1023 
1024 static netdev_features_t macvlan_fix_features(struct net_device *dev,
1025 					      netdev_features_t features)
1026 {
1027 	struct macvlan_dev *vlan = netdev_priv(dev);
1028 	netdev_features_t lowerdev_features = vlan->lowerdev->features;
1029 	netdev_features_t mask;
1030 
1031 	features |= NETIF_F_ALL_FOR_ALL;
1032 	features &= (vlan->set_features | ~MACVLAN_FEATURES);
1033 	mask = features;
1034 
1035 	lowerdev_features &= (features | ~NETIF_F_LRO);
1036 	features = netdev_increment_features(lowerdev_features, features, mask);
1037 	features |= ALWAYS_ON_FEATURES;
1038 	features &= (ALWAYS_ON_FEATURES | MACVLAN_FEATURES);
1039 
1040 	return features;
1041 }
1042 
1043 #ifdef CONFIG_NET_POLL_CONTROLLER
1044 static void macvlan_dev_poll_controller(struct net_device *dev)
1045 {
1046 	return;
1047 }
1048 
1049 static int macvlan_dev_netpoll_setup(struct net_device *dev, struct netpoll_info *npinfo)
1050 {
1051 	struct macvlan_dev *vlan = netdev_priv(dev);
1052 	struct net_device *real_dev = vlan->lowerdev;
1053 	struct netpoll *netpoll;
1054 	int err = 0;
1055 
1056 	netpoll = kzalloc(sizeof(*netpoll), GFP_KERNEL);
1057 	err = -ENOMEM;
1058 	if (!netpoll)
1059 		goto out;
1060 
1061 	err = __netpoll_setup(netpoll, real_dev);
1062 	if (err) {
1063 		kfree(netpoll);
1064 		goto out;
1065 	}
1066 
1067 	vlan->netpoll = netpoll;
1068 
1069 out:
1070 	return err;
1071 }
1072 
1073 static void macvlan_dev_netpoll_cleanup(struct net_device *dev)
1074 {
1075 	struct macvlan_dev *vlan = netdev_priv(dev);
1076 	struct netpoll *netpoll = vlan->netpoll;
1077 
1078 	if (!netpoll)
1079 		return;
1080 
1081 	vlan->netpoll = NULL;
1082 
1083 	__netpoll_free(netpoll);
1084 }
1085 #endif	/* CONFIG_NET_POLL_CONTROLLER */
1086 
1087 static int macvlan_dev_get_iflink(const struct net_device *dev)
1088 {
1089 	struct macvlan_dev *vlan = netdev_priv(dev);
1090 
1091 	return vlan->lowerdev->ifindex;
1092 }
1093 
1094 static const struct ethtool_ops macvlan_ethtool_ops = {
1095 	.get_link		= ethtool_op_get_link,
1096 	.get_link_ksettings	= macvlan_ethtool_get_link_ksettings,
1097 	.get_drvinfo		= macvlan_ethtool_get_drvinfo,
1098 };
1099 
1100 static const struct net_device_ops macvlan_netdev_ops = {
1101 	.ndo_init		= macvlan_init,
1102 	.ndo_uninit		= macvlan_uninit,
1103 	.ndo_open		= macvlan_open,
1104 	.ndo_stop		= macvlan_stop,
1105 	.ndo_start_xmit		= macvlan_start_xmit,
1106 	.ndo_change_mtu		= macvlan_change_mtu,
1107 	.ndo_fix_features	= macvlan_fix_features,
1108 	.ndo_change_rx_flags	= macvlan_change_rx_flags,
1109 	.ndo_set_mac_address	= macvlan_set_mac_address,
1110 	.ndo_set_rx_mode	= macvlan_set_mac_lists,
1111 	.ndo_get_stats64	= macvlan_dev_get_stats64,
1112 	.ndo_validate_addr	= eth_validate_addr,
1113 	.ndo_vlan_rx_add_vid	= macvlan_vlan_rx_add_vid,
1114 	.ndo_vlan_rx_kill_vid	= macvlan_vlan_rx_kill_vid,
1115 	.ndo_fdb_add		= macvlan_fdb_add,
1116 	.ndo_fdb_del		= macvlan_fdb_del,
1117 	.ndo_fdb_dump		= ndo_dflt_fdb_dump,
1118 	.ndo_get_lock_subclass  = macvlan_get_nest_level,
1119 #ifdef CONFIG_NET_POLL_CONTROLLER
1120 	.ndo_poll_controller	= macvlan_dev_poll_controller,
1121 	.ndo_netpoll_setup	= macvlan_dev_netpoll_setup,
1122 	.ndo_netpoll_cleanup	= macvlan_dev_netpoll_cleanup,
1123 #endif
1124 	.ndo_get_iflink		= macvlan_dev_get_iflink,
1125 	.ndo_features_check	= passthru_features_check,
1126 };
1127 
1128 void macvlan_common_setup(struct net_device *dev)
1129 {
1130 	ether_setup(dev);
1131 
1132 	dev->min_mtu		= 0;
1133 	dev->max_mtu		= ETH_MAX_MTU;
1134 	dev->priv_flags	       &= ~IFF_TX_SKB_SHARING;
1135 	netif_keep_dst(dev);
1136 	dev->priv_flags	       |= IFF_UNICAST_FLT;
1137 	dev->netdev_ops		= &macvlan_netdev_ops;
1138 	dev->needs_free_netdev	= true;
1139 	dev->header_ops		= &macvlan_hard_header_ops;
1140 	dev->ethtool_ops	= &macvlan_ethtool_ops;
1141 }
1142 EXPORT_SYMBOL_GPL(macvlan_common_setup);
1143 
1144 static void macvlan_setup(struct net_device *dev)
1145 {
1146 	macvlan_common_setup(dev);
1147 	dev->priv_flags |= IFF_NO_QUEUE;
1148 }
1149 
1150 static int macvlan_port_create(struct net_device *dev)
1151 {
1152 	struct macvlan_port *port;
1153 	unsigned int i;
1154 	int err;
1155 
1156 	if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
1157 		return -EINVAL;
1158 
1159 	if (netdev_is_rx_handler_busy(dev))
1160 		return -EBUSY;
1161 
1162 	port = kzalloc(sizeof(*port), GFP_KERNEL);
1163 	if (port == NULL)
1164 		return -ENOMEM;
1165 
1166 	port->dev = dev;
1167 	ether_addr_copy(port->perm_addr, dev->dev_addr);
1168 	INIT_LIST_HEAD(&port->vlans);
1169 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1170 		INIT_HLIST_HEAD(&port->vlan_hash[i]);
1171 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1172 		INIT_HLIST_HEAD(&port->vlan_source_hash[i]);
1173 
1174 	skb_queue_head_init(&port->bc_queue);
1175 	INIT_WORK(&port->bc_work, macvlan_process_broadcast);
1176 
1177 	err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
1178 	if (err)
1179 		kfree(port);
1180 	else
1181 		dev->priv_flags |= IFF_MACVLAN_PORT;
1182 	return err;
1183 }
1184 
1185 static void macvlan_port_destroy(struct net_device *dev)
1186 {
1187 	struct macvlan_port *port = macvlan_port_get_rtnl(dev);
1188 	struct sk_buff *skb;
1189 
1190 	dev->priv_flags &= ~IFF_MACVLAN_PORT;
1191 	netdev_rx_handler_unregister(dev);
1192 
1193 	/* After this point, no packet can schedule bc_work anymore,
1194 	 * but we need to cancel it and purge left skbs if any.
1195 	 */
1196 	cancel_work_sync(&port->bc_work);
1197 
1198 	while ((skb = __skb_dequeue(&port->bc_queue))) {
1199 		const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
1200 
1201 		if (src)
1202 			dev_put(src->dev);
1203 
1204 		kfree_skb(skb);
1205 	}
1206 
1207 	/* If the lower device address has been changed by passthru
1208 	 * macvlan, put it back.
1209 	 */
1210 	if (macvlan_passthru(port) &&
1211 	    !ether_addr_equal(port->dev->dev_addr, port->perm_addr)) {
1212 		struct sockaddr sa;
1213 
1214 		sa.sa_family = port->dev->type;
1215 		memcpy(&sa.sa_data, port->perm_addr, port->dev->addr_len);
1216 		dev_set_mac_address(port->dev, &sa, NULL);
1217 	}
1218 
1219 	kfree(port);
1220 }
1221 
1222 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[],
1223 			    struct netlink_ext_ack *extack)
1224 {
1225 	if (tb[IFLA_ADDRESS]) {
1226 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1227 			return -EINVAL;
1228 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1229 			return -EADDRNOTAVAIL;
1230 	}
1231 
1232 	if (!data)
1233 		return 0;
1234 
1235 	if (data[IFLA_MACVLAN_FLAGS] &&
1236 	    nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
1237 		return -EINVAL;
1238 
1239 	if (data[IFLA_MACVLAN_MODE]) {
1240 		switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
1241 		case MACVLAN_MODE_PRIVATE:
1242 		case MACVLAN_MODE_VEPA:
1243 		case MACVLAN_MODE_BRIDGE:
1244 		case MACVLAN_MODE_PASSTHRU:
1245 		case MACVLAN_MODE_SOURCE:
1246 			break;
1247 		default:
1248 			return -EINVAL;
1249 		}
1250 	}
1251 
1252 	if (data[IFLA_MACVLAN_MACADDR_MODE]) {
1253 		switch (nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE])) {
1254 		case MACVLAN_MACADDR_ADD:
1255 		case MACVLAN_MACADDR_DEL:
1256 		case MACVLAN_MACADDR_FLUSH:
1257 		case MACVLAN_MACADDR_SET:
1258 			break;
1259 		default:
1260 			return -EINVAL;
1261 		}
1262 	}
1263 
1264 	if (data[IFLA_MACVLAN_MACADDR]) {
1265 		if (nla_len(data[IFLA_MACVLAN_MACADDR]) != ETH_ALEN)
1266 			return -EINVAL;
1267 
1268 		if (!is_valid_ether_addr(nla_data(data[IFLA_MACVLAN_MACADDR])))
1269 			return -EADDRNOTAVAIL;
1270 	}
1271 
1272 	if (data[IFLA_MACVLAN_MACADDR_COUNT])
1273 		return -EINVAL;
1274 
1275 	return 0;
1276 }
1277 
1278 /**
1279  * reconfigure list of remote source mac address
1280  * (only for macvlan devices in source mode)
1281  * Note regarding alignment: all netlink data is aligned to 4 Byte, which
1282  * suffices for both ether_addr_copy and ether_addr_equal_64bits usage.
1283  */
1284 static int macvlan_changelink_sources(struct macvlan_dev *vlan, u32 mode,
1285 				      struct nlattr *data[])
1286 {
1287 	char *addr = NULL;
1288 	int ret, rem, len;
1289 	struct nlattr *nla, *head;
1290 	struct macvlan_source_entry *entry;
1291 
1292 	if (data[IFLA_MACVLAN_MACADDR])
1293 		addr = nla_data(data[IFLA_MACVLAN_MACADDR]);
1294 
1295 	if (mode == MACVLAN_MACADDR_ADD) {
1296 		if (!addr)
1297 			return -EINVAL;
1298 
1299 		return macvlan_hash_add_source(vlan, addr);
1300 
1301 	} else if (mode == MACVLAN_MACADDR_DEL) {
1302 		if (!addr)
1303 			return -EINVAL;
1304 
1305 		entry = macvlan_hash_lookup_source(vlan, addr);
1306 		if (entry) {
1307 			macvlan_hash_del_source(entry);
1308 			vlan->macaddr_count--;
1309 		}
1310 	} else if (mode == MACVLAN_MACADDR_FLUSH) {
1311 		macvlan_flush_sources(vlan->port, vlan);
1312 	} else if (mode == MACVLAN_MACADDR_SET) {
1313 		macvlan_flush_sources(vlan->port, vlan);
1314 
1315 		if (addr) {
1316 			ret = macvlan_hash_add_source(vlan, addr);
1317 			if (ret)
1318 				return ret;
1319 		}
1320 
1321 		if (!data || !data[IFLA_MACVLAN_MACADDR_DATA])
1322 			return 0;
1323 
1324 		head = nla_data(data[IFLA_MACVLAN_MACADDR_DATA]);
1325 		len = nla_len(data[IFLA_MACVLAN_MACADDR_DATA]);
1326 
1327 		nla_for_each_attr(nla, head, len, rem) {
1328 			if (nla_type(nla) != IFLA_MACVLAN_MACADDR ||
1329 			    nla_len(nla) != ETH_ALEN)
1330 				continue;
1331 
1332 			addr = nla_data(nla);
1333 			ret = macvlan_hash_add_source(vlan, addr);
1334 			if (ret)
1335 				return ret;
1336 		}
1337 	} else {
1338 		return -EINVAL;
1339 	}
1340 
1341 	return 0;
1342 }
1343 
1344 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
1345 			   struct nlattr *tb[], struct nlattr *data[],
1346 			   struct netlink_ext_ack *extack)
1347 {
1348 	struct macvlan_dev *vlan = netdev_priv(dev);
1349 	struct macvlan_port *port;
1350 	struct net_device *lowerdev;
1351 	int err;
1352 	int macmode;
1353 	bool create = false;
1354 
1355 	if (!tb[IFLA_LINK])
1356 		return -EINVAL;
1357 
1358 	lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
1359 	if (lowerdev == NULL)
1360 		return -ENODEV;
1361 
1362 	/* When creating macvlans or macvtaps on top of other macvlans - use
1363 	 * the real device as the lowerdev.
1364 	 */
1365 	if (netif_is_macvlan(lowerdev))
1366 		lowerdev = macvlan_dev_real_dev(lowerdev);
1367 
1368 	if (!tb[IFLA_MTU])
1369 		dev->mtu = lowerdev->mtu;
1370 	else if (dev->mtu > lowerdev->mtu)
1371 		return -EINVAL;
1372 
1373 	/* MTU range: 68 - lowerdev->max_mtu */
1374 	dev->min_mtu = ETH_MIN_MTU;
1375 	dev->max_mtu = lowerdev->max_mtu;
1376 
1377 	if (!tb[IFLA_ADDRESS])
1378 		eth_hw_addr_random(dev);
1379 
1380 	if (!macvlan_port_exists(lowerdev)) {
1381 		err = macvlan_port_create(lowerdev);
1382 		if (err < 0)
1383 			return err;
1384 		create = true;
1385 	}
1386 	port = macvlan_port_get_rtnl(lowerdev);
1387 
1388 	/* Only 1 macvlan device can be created in passthru mode */
1389 	if (macvlan_passthru(port)) {
1390 		/* The macvlan port must be not created this time,
1391 		 * still goto destroy_macvlan_port for readability.
1392 		 */
1393 		err = -EINVAL;
1394 		goto destroy_macvlan_port;
1395 	}
1396 
1397 	vlan->lowerdev = lowerdev;
1398 	vlan->dev      = dev;
1399 	vlan->port     = port;
1400 	vlan->set_features = MACVLAN_FEATURES;
1401 	vlan->nest_level = dev_get_nest_level(lowerdev) + 1;
1402 
1403 	vlan->mode     = MACVLAN_MODE_VEPA;
1404 	if (data && data[IFLA_MACVLAN_MODE])
1405 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1406 
1407 	if (data && data[IFLA_MACVLAN_FLAGS])
1408 		vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1409 
1410 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
1411 		if (port->count) {
1412 			err = -EINVAL;
1413 			goto destroy_macvlan_port;
1414 		}
1415 		macvlan_set_passthru(port);
1416 		eth_hw_addr_inherit(dev, lowerdev);
1417 	}
1418 
1419 	if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1420 		if (vlan->mode != MACVLAN_MODE_SOURCE) {
1421 			err = -EINVAL;
1422 			goto destroy_macvlan_port;
1423 		}
1424 		macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1425 		err = macvlan_changelink_sources(vlan, macmode, data);
1426 		if (err)
1427 			goto destroy_macvlan_port;
1428 	}
1429 
1430 	err = register_netdevice(dev);
1431 	if (err < 0)
1432 		goto destroy_macvlan_port;
1433 
1434 	dev->priv_flags |= IFF_MACVLAN;
1435 	err = netdev_upper_dev_link(lowerdev, dev, extack);
1436 	if (err)
1437 		goto unregister_netdev;
1438 
1439 	list_add_tail_rcu(&vlan->list, &port->vlans);
1440 	netif_stacked_transfer_operstate(lowerdev, dev);
1441 	linkwatch_fire_event(dev);
1442 
1443 	return 0;
1444 
1445 unregister_netdev:
1446 	/* macvlan_uninit would free the macvlan port */
1447 	unregister_netdevice(dev);
1448 	return err;
1449 destroy_macvlan_port:
1450 	/* the macvlan port may be freed by macvlan_uninit when fail to register.
1451 	 * so we destroy the macvlan port only when it's valid.
1452 	 */
1453 	if (create && macvlan_port_get_rtnl(lowerdev))
1454 		macvlan_port_destroy(port->dev);
1455 	return err;
1456 }
1457 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
1458 
1459 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
1460 			   struct nlattr *tb[], struct nlattr *data[],
1461 			   struct netlink_ext_ack *extack)
1462 {
1463 	return macvlan_common_newlink(src_net, dev, tb, data, extack);
1464 }
1465 
1466 void macvlan_dellink(struct net_device *dev, struct list_head *head)
1467 {
1468 	struct macvlan_dev *vlan = netdev_priv(dev);
1469 
1470 	if (vlan->mode == MACVLAN_MODE_SOURCE)
1471 		macvlan_flush_sources(vlan->port, vlan);
1472 	list_del_rcu(&vlan->list);
1473 	unregister_netdevice_queue(dev, head);
1474 	netdev_upper_dev_unlink(vlan->lowerdev, dev);
1475 }
1476 EXPORT_SYMBOL_GPL(macvlan_dellink);
1477 
1478 static int macvlan_changelink(struct net_device *dev,
1479 			      struct nlattr *tb[], struct nlattr *data[],
1480 			      struct netlink_ext_ack *extack)
1481 {
1482 	struct macvlan_dev *vlan = netdev_priv(dev);
1483 	enum macvlan_mode mode;
1484 	bool set_mode = false;
1485 	enum macvlan_macaddr_mode macmode;
1486 	int ret;
1487 
1488 	/* Validate mode, but don't set yet: setting flags may fail. */
1489 	if (data && data[IFLA_MACVLAN_MODE]) {
1490 		set_mode = true;
1491 		mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1492 		/* Passthrough mode can't be set or cleared dynamically */
1493 		if ((mode == MACVLAN_MODE_PASSTHRU) !=
1494 		    (vlan->mode == MACVLAN_MODE_PASSTHRU))
1495 			return -EINVAL;
1496 		if (vlan->mode == MACVLAN_MODE_SOURCE &&
1497 		    vlan->mode != mode)
1498 			macvlan_flush_sources(vlan->port, vlan);
1499 	}
1500 
1501 	if (data && data[IFLA_MACVLAN_FLAGS]) {
1502 		__u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1503 		bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
1504 		if (macvlan_passthru(vlan->port) && promisc) {
1505 			int err;
1506 
1507 			if (flags & MACVLAN_FLAG_NOPROMISC)
1508 				err = dev_set_promiscuity(vlan->lowerdev, -1);
1509 			else
1510 				err = dev_set_promiscuity(vlan->lowerdev, 1);
1511 			if (err < 0)
1512 				return err;
1513 		}
1514 		vlan->flags = flags;
1515 	}
1516 	if (set_mode)
1517 		vlan->mode = mode;
1518 	if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1519 		if (vlan->mode != MACVLAN_MODE_SOURCE)
1520 			return -EINVAL;
1521 		macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1522 		ret = macvlan_changelink_sources(vlan, macmode, data);
1523 		if (ret)
1524 			return ret;
1525 	}
1526 	return 0;
1527 }
1528 
1529 static size_t macvlan_get_size_mac(const struct macvlan_dev *vlan)
1530 {
1531 	if (vlan->macaddr_count == 0)
1532 		return 0;
1533 	return nla_total_size(0) /* IFLA_MACVLAN_MACADDR_DATA */
1534 		+ vlan->macaddr_count * nla_total_size(sizeof(u8) * ETH_ALEN);
1535 }
1536 
1537 static size_t macvlan_get_size(const struct net_device *dev)
1538 {
1539 	struct macvlan_dev *vlan = netdev_priv(dev);
1540 
1541 	return (0
1542 		+ nla_total_size(4) /* IFLA_MACVLAN_MODE */
1543 		+ nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
1544 		+ nla_total_size(4) /* IFLA_MACVLAN_MACADDR_COUNT */
1545 		+ macvlan_get_size_mac(vlan) /* IFLA_MACVLAN_MACADDR */
1546 		);
1547 }
1548 
1549 static int macvlan_fill_info_macaddr(struct sk_buff *skb,
1550 				     const struct macvlan_dev *vlan,
1551 				     const int i)
1552 {
1553 	struct hlist_head *h = &vlan->port->vlan_source_hash[i];
1554 	struct macvlan_source_entry *entry;
1555 
1556 	hlist_for_each_entry_rcu(entry, h, hlist) {
1557 		if (entry->vlan != vlan)
1558 			continue;
1559 		if (nla_put(skb, IFLA_MACVLAN_MACADDR, ETH_ALEN, entry->addr))
1560 			return 1;
1561 	}
1562 	return 0;
1563 }
1564 
1565 static int macvlan_fill_info(struct sk_buff *skb,
1566 				const struct net_device *dev)
1567 {
1568 	struct macvlan_dev *vlan = netdev_priv(dev);
1569 	int i;
1570 	struct nlattr *nest;
1571 
1572 	if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
1573 		goto nla_put_failure;
1574 	if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
1575 		goto nla_put_failure;
1576 	if (nla_put_u32(skb, IFLA_MACVLAN_MACADDR_COUNT, vlan->macaddr_count))
1577 		goto nla_put_failure;
1578 	if (vlan->macaddr_count > 0) {
1579 		nest = nla_nest_start(skb, IFLA_MACVLAN_MACADDR_DATA);
1580 		if (nest == NULL)
1581 			goto nla_put_failure;
1582 
1583 		for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
1584 			if (macvlan_fill_info_macaddr(skb, vlan, i))
1585 				goto nla_put_failure;
1586 		}
1587 		nla_nest_end(skb, nest);
1588 	}
1589 	return 0;
1590 
1591 nla_put_failure:
1592 	return -EMSGSIZE;
1593 }
1594 
1595 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
1596 	[IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
1597 	[IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
1598 	[IFLA_MACVLAN_MACADDR_MODE] = { .type = NLA_U32 },
1599 	[IFLA_MACVLAN_MACADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1600 	[IFLA_MACVLAN_MACADDR_DATA] = { .type = NLA_NESTED },
1601 	[IFLA_MACVLAN_MACADDR_COUNT] = { .type = NLA_U32 },
1602 };
1603 
1604 int macvlan_link_register(struct rtnl_link_ops *ops)
1605 {
1606 	/* common fields */
1607 	ops->validate		= macvlan_validate;
1608 	ops->maxtype		= IFLA_MACVLAN_MAX;
1609 	ops->policy		= macvlan_policy;
1610 	ops->changelink		= macvlan_changelink;
1611 	ops->get_size		= macvlan_get_size;
1612 	ops->fill_info		= macvlan_fill_info;
1613 
1614 	return rtnl_link_register(ops);
1615 };
1616 EXPORT_SYMBOL_GPL(macvlan_link_register);
1617 
1618 static struct net *macvlan_get_link_net(const struct net_device *dev)
1619 {
1620 	return dev_net(macvlan_dev_real_dev(dev));
1621 }
1622 
1623 static struct rtnl_link_ops macvlan_link_ops = {
1624 	.kind		= "macvlan",
1625 	.setup		= macvlan_setup,
1626 	.newlink	= macvlan_newlink,
1627 	.dellink	= macvlan_dellink,
1628 	.get_link_net	= macvlan_get_link_net,
1629 	.priv_size      = sizeof(struct macvlan_dev),
1630 };
1631 
1632 static int macvlan_device_event(struct notifier_block *unused,
1633 				unsigned long event, void *ptr)
1634 {
1635 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1636 	struct macvlan_dev *vlan, *next;
1637 	struct macvlan_port *port;
1638 	LIST_HEAD(list_kill);
1639 
1640 	if (!macvlan_port_exists(dev))
1641 		return NOTIFY_DONE;
1642 
1643 	port = macvlan_port_get_rtnl(dev);
1644 
1645 	switch (event) {
1646 	case NETDEV_UP:
1647 	case NETDEV_DOWN:
1648 	case NETDEV_CHANGE:
1649 		list_for_each_entry(vlan, &port->vlans, list)
1650 			netif_stacked_transfer_operstate(vlan->lowerdev,
1651 							 vlan->dev);
1652 		break;
1653 	case NETDEV_FEAT_CHANGE:
1654 		list_for_each_entry(vlan, &port->vlans, list) {
1655 			vlan->dev->gso_max_size = dev->gso_max_size;
1656 			vlan->dev->gso_max_segs = dev->gso_max_segs;
1657 			netdev_update_features(vlan->dev);
1658 		}
1659 		break;
1660 	case NETDEV_CHANGEMTU:
1661 		list_for_each_entry(vlan, &port->vlans, list) {
1662 			if (vlan->dev->mtu <= dev->mtu)
1663 				continue;
1664 			dev_set_mtu(vlan->dev, dev->mtu);
1665 		}
1666 		break;
1667 	case NETDEV_CHANGEADDR:
1668 		if (!macvlan_passthru(port))
1669 			return NOTIFY_DONE;
1670 
1671 		vlan = list_first_entry_or_null(&port->vlans,
1672 						struct macvlan_dev,
1673 						list);
1674 
1675 		if (macvlan_sync_address(vlan->dev, dev->dev_addr))
1676 			return NOTIFY_BAD;
1677 
1678 		break;
1679 	case NETDEV_UNREGISTER:
1680 		/* twiddle thumbs on netns device moves */
1681 		if (dev->reg_state != NETREG_UNREGISTERING)
1682 			break;
1683 
1684 		list_for_each_entry_safe(vlan, next, &port->vlans, list)
1685 			vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1686 		unregister_netdevice_many(&list_kill);
1687 		break;
1688 	case NETDEV_PRE_TYPE_CHANGE:
1689 		/* Forbid underlaying device to change its type. */
1690 		return NOTIFY_BAD;
1691 
1692 	case NETDEV_NOTIFY_PEERS:
1693 	case NETDEV_BONDING_FAILOVER:
1694 	case NETDEV_RESEND_IGMP:
1695 		/* Propagate to all vlans */
1696 		list_for_each_entry(vlan, &port->vlans, list)
1697 			call_netdevice_notifiers(event, vlan->dev);
1698 	}
1699 	return NOTIFY_DONE;
1700 }
1701 
1702 static struct notifier_block macvlan_notifier_block __read_mostly = {
1703 	.notifier_call	= macvlan_device_event,
1704 };
1705 
1706 static int __init macvlan_init_module(void)
1707 {
1708 	int err;
1709 
1710 	register_netdevice_notifier(&macvlan_notifier_block);
1711 
1712 	err = macvlan_link_register(&macvlan_link_ops);
1713 	if (err < 0)
1714 		goto err1;
1715 	return 0;
1716 err1:
1717 	unregister_netdevice_notifier(&macvlan_notifier_block);
1718 	return err;
1719 }
1720 
1721 static void __exit macvlan_cleanup_module(void)
1722 {
1723 	rtnl_link_unregister(&macvlan_link_ops);
1724 	unregister_netdevice_notifier(&macvlan_notifier_block);
1725 }
1726 
1727 module_init(macvlan_init_module);
1728 module_exit(macvlan_cleanup_module);
1729 
1730 MODULE_LICENSE("GPL");
1731 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1732 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1733 MODULE_ALIAS_RTNL_LINK("macvlan");
1734