xref: /linux/drivers/net/macvlan.c (revision cf85f810f911234a06a4ef2439e8694b93b717fc)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (c) 2007 Patrick McHardy <kaber@trash.net>
4  *
5  * The code this is based on carried the following copyright notice:
6  * ---
7  * (C) Copyright 2001-2006
8  * Alex Zeffertt, Cambridge Broadband Ltd, ajz@cambridgebroadband.com
9  * Re-worked by Ben Greear <greearb@candelatech.com>
10  * ---
11  */
12 #include <linux/kernel.h>
13 #include <linux/types.h>
14 #include <linux/module.h>
15 #include <linux/init.h>
16 #include <linux/errno.h>
17 #include <linux/slab.h>
18 #include <linux/string.h>
19 #include <linux/rculist.h>
20 #include <linux/notifier.h>
21 #include <linux/netdevice.h>
22 #include <linux/etherdevice.h>
23 #include <linux/net_tstamp.h>
24 #include <linux/ethtool.h>
25 #include <linux/if_arp.h>
26 #include <linux/if_vlan.h>
27 #include <linux/if_link.h>
28 #include <linux/if_macvlan.h>
29 #include <linux/hash.h>
30 #include <linux/workqueue.h>
31 #include <net/netdev_lock.h>
32 #include <net/rtnetlink.h>
33 #include <net/xfrm.h>
34 #include <linux/netpoll.h>
35 #include <linux/phy.h>
36 
37 #define MACVLAN_HASH_BITS	8
38 #define MACVLAN_HASH_SIZE	(1<<MACVLAN_HASH_BITS)
39 #define MACVLAN_DEFAULT_BC_QUEUE_LEN	1000
40 
41 #define MACVLAN_F_PASSTHRU	1
42 #define MACVLAN_F_ADDRCHANGE	2
43 
44 struct macvlan_port {
45 	struct net_device	*dev;
46 	struct hlist_head	vlan_hash[MACVLAN_HASH_SIZE];
47 	struct list_head	vlans;
48 	struct sk_buff_head	bc_queue;
49 	struct work_struct	bc_work;
50 	u32			bc_queue_len_used;
51 	int			bc_cutoff;
52 	u32			flags;
53 	int			count;
54 	struct hlist_head	vlan_source_hash[MACVLAN_HASH_SIZE];
55 	DECLARE_BITMAP(bc_filter, MACVLAN_MC_FILTER_SZ);
56 	DECLARE_BITMAP(mc_filter, MACVLAN_MC_FILTER_SZ);
57 	unsigned char           perm_addr[ETH_ALEN];
58 };
59 
60 struct macvlan_source_entry {
61 	struct hlist_node	hlist;
62 	struct macvlan_dev __rcu *vlan;
63 	unsigned char		addr[6+2] __aligned(sizeof(u16));
64 	struct rcu_head		rcu;
65 };
66 
67 struct macvlan_skb_cb {
68 	const struct macvlan_dev *src;
69 };
70 
71 #define MACVLAN_SKB_CB(__skb) ((struct macvlan_skb_cb *)&((__skb)->cb[0]))
72 
73 static void macvlan_port_destroy(struct net_device *dev);
74 static void update_port_bc_queue_len(struct macvlan_port *port);
75 
76 static inline bool macvlan_passthru(const struct macvlan_port *port)
77 {
78 	return port->flags & MACVLAN_F_PASSTHRU;
79 }
80 
81 static inline void macvlan_set_passthru(struct macvlan_port *port)
82 {
83 	port->flags |= MACVLAN_F_PASSTHRU;
84 }
85 
86 static inline bool macvlan_addr_change(const struct macvlan_port *port)
87 {
88 	return port->flags & MACVLAN_F_ADDRCHANGE;
89 }
90 
91 static inline void macvlan_set_addr_change(struct macvlan_port *port)
92 {
93 	port->flags |= MACVLAN_F_ADDRCHANGE;
94 }
95 
96 static inline void macvlan_clear_addr_change(struct macvlan_port *port)
97 {
98 	port->flags &= ~MACVLAN_F_ADDRCHANGE;
99 }
100 
101 /* Hash Ethernet address */
102 static u32 macvlan_eth_hash(const unsigned char *addr)
103 {
104 	u64 value = get_unaligned((u64 *)addr);
105 
106 	/* only want 6 bytes */
107 #ifdef __BIG_ENDIAN
108 	value >>= 16;
109 #else
110 	value <<= 16;
111 #endif
112 	return hash_64(value, MACVLAN_HASH_BITS);
113 }
114 
115 static struct macvlan_port *macvlan_port_get_rcu(const struct net_device *dev)
116 {
117 	return rcu_dereference(dev->rx_handler_data);
118 }
119 
120 static struct macvlan_port *macvlan_port_get_rtnl(const struct net_device *dev)
121 {
122 	return rtnl_dereference(dev->rx_handler_data);
123 }
124 
125 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
126 					       const unsigned char *addr)
127 {
128 	struct macvlan_dev *vlan;
129 	u32 idx = macvlan_eth_hash(addr);
130 
131 	hlist_for_each_entry_rcu(vlan, &port->vlan_hash[idx], hlist,
132 				 lockdep_rtnl_is_held()) {
133 		if (ether_addr_equal_64bits(vlan->dev->dev_addr, addr))
134 			return vlan;
135 	}
136 	return NULL;
137 }
138 
139 static struct macvlan_source_entry *macvlan_hash_lookup_source(
140 	const struct macvlan_dev *vlan,
141 	const unsigned char *addr)
142 {
143 	struct macvlan_source_entry *entry;
144 	u32 idx = macvlan_eth_hash(addr);
145 	struct hlist_head *h = &vlan->port->vlan_source_hash[idx];
146 
147 	hlist_for_each_entry_rcu(entry, h, hlist, lockdep_rtnl_is_held()) {
148 		if (ether_addr_equal_64bits(entry->addr, addr) &&
149 		    rcu_access_pointer(entry->vlan) == vlan)
150 			return entry;
151 	}
152 	return NULL;
153 }
154 
155 static int macvlan_hash_add_source(struct macvlan_dev *vlan,
156 				   const unsigned char *addr)
157 {
158 	struct macvlan_port *port = vlan->port;
159 	struct macvlan_source_entry *entry;
160 	struct hlist_head *h;
161 
162 	entry = macvlan_hash_lookup_source(vlan, addr);
163 	if (entry)
164 		return 0;
165 
166 	entry = kmalloc_obj(*entry);
167 	if (!entry)
168 		return -ENOMEM;
169 
170 	ether_addr_copy(entry->addr, addr);
171 	RCU_INIT_POINTER(entry->vlan, vlan);
172 	h = &port->vlan_source_hash[macvlan_eth_hash(addr)];
173 	hlist_add_head_rcu(&entry->hlist, h);
174 	WRITE_ONCE(vlan->macaddr_count, vlan->macaddr_count + 1);
175 
176 	return 0;
177 }
178 
179 static void macvlan_hash_add(struct macvlan_dev *vlan)
180 {
181 	struct macvlan_port *port = vlan->port;
182 	const unsigned char *addr = vlan->dev->dev_addr;
183 	u32 idx = macvlan_eth_hash(addr);
184 
185 	hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[idx]);
186 }
187 
188 static void macvlan_hash_del_source(struct macvlan_source_entry *entry)
189 {
190 	RCU_INIT_POINTER(entry->vlan, NULL);
191 	hlist_del_rcu(&entry->hlist);
192 	kfree_rcu(entry, rcu);
193 }
194 
195 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync)
196 {
197 	hlist_del_rcu(&vlan->hlist);
198 	if (sync)
199 		synchronize_rcu();
200 }
201 
202 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
203 					const unsigned char *addr)
204 {
205 	macvlan_hash_del(vlan, true);
206 	/* Now that we are unhashed it is safe to change the device
207 	 * address without confusing packet delivery.
208 	 */
209 	eth_hw_addr_set(vlan->dev, addr);
210 	macvlan_hash_add(vlan);
211 }
212 
213 static bool macvlan_addr_busy(const struct macvlan_port *port,
214 			      const unsigned char *addr)
215 {
216 	/* Test to see if the specified address is
217 	 * currently in use by the underlying device or
218 	 * another macvlan.
219 	 */
220 	if (!macvlan_passthru(port) && !macvlan_addr_change(port) &&
221 	    ether_addr_equal_64bits(port->dev->dev_addr, addr))
222 		return true;
223 
224 	if (macvlan_hash_lookup(port, addr))
225 		return true;
226 
227 	return false;
228 }
229 
230 
231 static int macvlan_broadcast_one(struct sk_buff *skb,
232 				 const struct macvlan_dev *vlan,
233 				 const struct ethhdr *eth, bool local)
234 {
235 	struct net_device *dev = vlan->dev;
236 
237 	if (local)
238 		return __dev_forward_skb(dev, skb);
239 
240 	skb->dev = dev;
241 	if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
242 		skb->pkt_type = PACKET_BROADCAST;
243 	else
244 		skb->pkt_type = PACKET_MULTICAST;
245 
246 	return 0;
247 }
248 
249 static u32 macvlan_hash_mix(const struct macvlan_dev *vlan)
250 {
251 	return (u32)(((unsigned long)vlan) >> L1_CACHE_SHIFT);
252 }
253 
254 
255 static unsigned int mc_hash(const struct macvlan_dev *vlan,
256 			    const unsigned char *addr)
257 {
258 	u32 val = get_unaligned((u32 *)(addr + 2));
259 
260 	val ^= macvlan_hash_mix(vlan);
261 	return hash_32(val, MACVLAN_MC_FILTER_BITS);
262 }
263 
264 static void macvlan_broadcast(struct sk_buff *skb,
265 			      const struct macvlan_port *port,
266 			      struct net_device *src,
267 			      enum macvlan_mode mode)
268 {
269 	const struct ethhdr *eth = eth_hdr(skb);
270 	const struct macvlan_dev *vlan;
271 	struct sk_buff *nskb;
272 	unsigned int i;
273 	int err;
274 	unsigned int hash;
275 
276 	if (skb->protocol == htons(ETH_P_PAUSE))
277 		return;
278 
279 	hash_for_each_rcu(port->vlan_hash, i, vlan, hlist) {
280 		if (vlan->dev == src || !(READ_ONCE(vlan->mode) & mode))
281 			continue;
282 
283 		hash = mc_hash(vlan, eth->h_dest);
284 		if (!test_bit(hash, vlan->mc_filter))
285 			continue;
286 
287 		err = NET_RX_DROP;
288 		nskb = skb_clone(skb, GFP_ATOMIC);
289 		if (likely(nskb))
290 			err = macvlan_broadcast_one(nskb, vlan, eth,
291 					mode == MACVLAN_MODE_BRIDGE) ?:
292 			      netif_rx(nskb);
293 		macvlan_count_rx(vlan, skb->len + ETH_HLEN,
294 				 err == NET_RX_SUCCESS, true);
295 	}
296 }
297 
298 static void macvlan_multicast_rx(const struct macvlan_port *port,
299 				 const struct macvlan_dev *src,
300 				 struct sk_buff *skb)
301 {
302 	if (!src)
303 		/* frame comes from an external address */
304 		macvlan_broadcast(skb, port, NULL,
305 				  MACVLAN_MODE_PRIVATE |
306 				  MACVLAN_MODE_VEPA    |
307 				  MACVLAN_MODE_PASSTHRU|
308 				  MACVLAN_MODE_BRIDGE);
309 	else if (READ_ONCE(src->mode) == MACVLAN_MODE_VEPA)
310 		/* flood to everyone except source */
311 		macvlan_broadcast(skb, port, src->dev,
312 				  MACVLAN_MODE_VEPA |
313 				  MACVLAN_MODE_BRIDGE);
314 	else
315 		/*
316 		 * Flood to VEPA and bridge ports. We cannot distinguish
317 		 * a looped-back locally-originated multicast from one
318 		 * sent by an external source sharing the same source MAC
319 		 * (e.g., VRRP virtual MAC), so deliver to bridge ports
320 		 * as well to ensure correct reception in all cases.
321 		 */
322 		macvlan_broadcast(skb, port, NULL,
323 				  MACVLAN_MODE_VEPA |
324 				  MACVLAN_MODE_BRIDGE);
325 }
326 
327 static void macvlan_process_broadcast(struct work_struct *w)
328 {
329 	struct macvlan_port *port = container_of(w, struct macvlan_port,
330 						 bc_work);
331 	struct sk_buff *skb;
332 	struct sk_buff_head list;
333 
334 	__skb_queue_head_init(&list);
335 
336 	spin_lock_bh(&port->bc_queue.lock);
337 	skb_queue_splice_tail_init(&port->bc_queue, &list);
338 	spin_unlock_bh(&port->bc_queue.lock);
339 
340 	while ((skb = __skb_dequeue(&list))) {
341 		const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
342 
343 		rcu_read_lock();
344 		macvlan_multicast_rx(port, src, skb);
345 		rcu_read_unlock();
346 
347 		if (src)
348 			dev_put(src->dev);
349 		consume_skb(skb);
350 
351 		cond_resched();
352 	}
353 }
354 
355 static void macvlan_broadcast_enqueue(struct macvlan_port *port,
356 				      const struct macvlan_dev *src,
357 				      struct sk_buff *skb)
358 {
359 	u32 bc_queue_len_used = READ_ONCE(port->bc_queue_len_used);
360 	struct sk_buff *nskb;
361 	int err = -ENOMEM;
362 
363 	if (skb_queue_len_lockless(&port->bc_queue) >= bc_queue_len_used)
364 		goto err;
365 
366 	nskb = skb_clone(skb, GFP_ATOMIC);
367 	if (!nskb)
368 		goto err;
369 
370 	MACVLAN_SKB_CB(nskb)->src = src;
371 
372 	spin_lock(&port->bc_queue.lock);
373 	if (skb_queue_len(&port->bc_queue) < bc_queue_len_used) {
374 		if (src)
375 			dev_hold(src->dev);
376 		__skb_queue_tail(&port->bc_queue, nskb);
377 		err = 0;
378 	}
379 	spin_unlock(&port->bc_queue.lock);
380 
381 	queue_work(system_dfl_wq, &port->bc_work);
382 
383 	if (err)
384 		goto free_nskb;
385 
386 	return;
387 
388 free_nskb:
389 	kfree_skb_reason(nskb, SKB_DROP_REASON_MACVLAN_BROADCAST_BACKLOG);
390 err:
391 	dev_core_stats_rx_dropped_inc(skb->dev);
392 }
393 
394 static void macvlan_flush_sources(struct macvlan_port *port,
395 				  struct macvlan_dev *vlan)
396 {
397 	struct macvlan_source_entry *entry;
398 	struct hlist_node *next;
399 	int i;
400 
401 	hash_for_each_safe(port->vlan_source_hash, i, next, entry, hlist)
402 		if (rcu_access_pointer(entry->vlan) == vlan)
403 			macvlan_hash_del_source(entry);
404 
405 	WRITE_ONCE(vlan->macaddr_count, 0);
406 }
407 
408 static void macvlan_forward_source_one(struct sk_buff *skb,
409 				       struct macvlan_dev *vlan)
410 {
411 	struct sk_buff *nskb;
412 	struct net_device *dev;
413 	int len;
414 	int ret;
415 
416 	dev = vlan->dev;
417 	if (unlikely(!(dev->flags & IFF_UP)))
418 		return;
419 
420 	nskb = skb_clone(skb, GFP_ATOMIC);
421 	if (!nskb)
422 		return;
423 
424 	len = nskb->len + ETH_HLEN;
425 	nskb->dev = dev;
426 
427 	if (ether_addr_equal_64bits(eth_hdr(skb)->h_dest, dev->dev_addr))
428 		nskb->pkt_type = PACKET_HOST;
429 
430 	ret = __netif_rx(nskb);
431 	macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
432 }
433 
434 static bool macvlan_forward_source(struct sk_buff *skb,
435 				   struct macvlan_port *port,
436 				   const unsigned char *addr)
437 {
438 	struct macvlan_source_entry *entry;
439 	u32 idx = macvlan_eth_hash(addr);
440 	struct hlist_head *h = &port->vlan_source_hash[idx];
441 	bool consume = false;
442 
443 	hlist_for_each_entry_rcu(entry, h, hlist) {
444 		if (ether_addr_equal_64bits(entry->addr, addr)) {
445 			struct macvlan_dev *vlan = rcu_dereference(entry->vlan);
446 
447 			if (!vlan)
448 				continue;
449 
450 			if (READ_ONCE(vlan->flags) & MACVLAN_FLAG_NODST)
451 				consume = true;
452 			macvlan_forward_source_one(skb, vlan);
453 		}
454 	}
455 
456 	return consume;
457 }
458 
459 /* called under rcu_read_lock() from netif_receive_skb */
460 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
461 {
462 	struct macvlan_port *port;
463 	struct sk_buff *skb = *pskb;
464 	const struct ethhdr *eth = eth_hdr(skb);
465 	const struct macvlan_dev *vlan;
466 	const struct macvlan_dev *src;
467 	struct net_device *dev;
468 	unsigned int len = 0;
469 	int ret;
470 	rx_handler_result_t handle_res;
471 
472 	/* Packets from dev_loopback_xmit() do not have L2 header, bail out */
473 	if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
474 		return RX_HANDLER_PASS;
475 
476 	port = macvlan_port_get_rcu(skb->dev);
477 	if (is_multicast_ether_addr(eth->h_dest)) {
478 		unsigned int hash;
479 
480 		skb = ip_check_defrag(dev_net(skb->dev), skb, IP_DEFRAG_MACVLAN);
481 		if (!skb)
482 			return RX_HANDLER_CONSUMED;
483 		*pskb = skb;
484 		eth = eth_hdr(skb);
485 		if (macvlan_forward_source(skb, port, eth->h_source)) {
486 			kfree_skb(skb);
487 			return RX_HANDLER_CONSUMED;
488 		}
489 		src = macvlan_hash_lookup(port, eth->h_source);
490 		if (src) {
491 			enum macvlan_mode mode = READ_ONCE(src->mode);
492 
493 			if (mode != MACVLAN_MODE_VEPA &&
494 			    mode != MACVLAN_MODE_BRIDGE) {
495 				/* forward to original port. */
496 				vlan = src;
497 				ret = macvlan_broadcast_one(skb, vlan, eth, 0) ?:
498 				      __netif_rx(skb);
499 				handle_res = RX_HANDLER_CONSUMED;
500 				goto out;
501 			}
502 		}
503 
504 		hash = mc_hash(NULL, eth->h_dest);
505 		if (test_bit(hash, port->bc_filter))
506 			macvlan_broadcast_enqueue(port, src, skb);
507 		else if (test_bit(hash, port->mc_filter))
508 			macvlan_multicast_rx(port, src, skb);
509 
510 		return RX_HANDLER_PASS;
511 	}
512 
513 	if (macvlan_forward_source(skb, port, eth->h_source)) {
514 		kfree_skb(skb);
515 		return RX_HANDLER_CONSUMED;
516 	}
517 	if (macvlan_passthru(port))
518 		vlan = list_first_or_null_rcu(&port->vlans,
519 					      struct macvlan_dev, list);
520 	else
521 		vlan = macvlan_hash_lookup(port, eth->h_dest);
522 	if (!vlan || READ_ONCE(vlan->mode) == MACVLAN_MODE_SOURCE)
523 		return RX_HANDLER_PASS;
524 
525 	dev = vlan->dev;
526 	if (unlikely(!(dev->flags & IFF_UP))) {
527 		kfree_skb(skb);
528 		return RX_HANDLER_CONSUMED;
529 	}
530 	len = skb->len + ETH_HLEN;
531 	skb = skb_share_check(skb, GFP_ATOMIC);
532 	if (!skb) {
533 		ret = NET_RX_DROP;
534 		handle_res = RX_HANDLER_CONSUMED;
535 		goto out;
536 	}
537 
538 	*pskb = skb;
539 	skb->dev = dev;
540 	skb->pkt_type = PACKET_HOST;
541 
542 	ret = NET_RX_SUCCESS;
543 	handle_res = RX_HANDLER_ANOTHER;
544 out:
545 	macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
546 	return handle_res;
547 }
548 
549 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
550 {
551 	const struct macvlan_dev *vlan = netdev_priv(dev);
552 	const struct macvlan_port *port = vlan->port;
553 	const struct macvlan_dev *dest;
554 
555 	if (READ_ONCE(vlan->mode) == MACVLAN_MODE_BRIDGE) {
556 		const struct ethhdr *eth = skb_eth_hdr(skb);
557 
558 		/* send to other bridge ports directly */
559 		if (is_multicast_ether_addr(eth->h_dest)) {
560 			skb_reset_mac_header(skb);
561 			macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
562 			goto xmit_world;
563 		}
564 
565 		dest = macvlan_hash_lookup(port, eth->h_dest);
566 		if (dest && READ_ONCE(dest->mode) == MACVLAN_MODE_BRIDGE) {
567 			/* send to lowerdev first for its network taps */
568 			dev_forward_skb(vlan->lowerdev, skb);
569 
570 			return NET_XMIT_SUCCESS;
571 		}
572 	}
573 xmit_world:
574 	skb->dev = vlan->lowerdev;
575 	return dev_queue_xmit_accel(skb,
576 				    netdev_get_sb_channel(dev) ? dev : NULL);
577 }
578 
579 static inline netdev_tx_t macvlan_netpoll_send_skb(struct macvlan_dev *vlan, struct sk_buff *skb)
580 {
581 #ifdef CONFIG_NET_POLL_CONTROLLER
582 	return netpoll_send_skb(vlan->netpoll, skb);
583 #else
584 	BUG();
585 	return NETDEV_TX_OK;
586 #endif
587 }
588 
589 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
590 				      struct net_device *dev)
591 {
592 	struct macvlan_dev *vlan = netdev_priv(dev);
593 	unsigned int len = skb->len;
594 	int ret;
595 
596 	if (unlikely(netpoll_tx_running(dev)))
597 		return macvlan_netpoll_send_skb(vlan, skb);
598 
599 	ret = macvlan_queue_xmit(skb, dev);
600 
601 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
602 		struct vlan_pcpu_stats *pcpu_stats;
603 
604 		pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
605 		u64_stats_update_begin(&pcpu_stats->syncp);
606 		u64_stats_inc(&pcpu_stats->tx_packets);
607 		u64_stats_add(&pcpu_stats->tx_bytes, len);
608 		u64_stats_update_end(&pcpu_stats->syncp);
609 	} else {
610 		this_cpu_inc(vlan->pcpu_stats->tx_dropped);
611 	}
612 	return ret;
613 }
614 
615 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
616 			       unsigned short type, const void *daddr,
617 			       const void *saddr, unsigned len)
618 {
619 	const struct macvlan_dev *vlan = netdev_priv(dev);
620 	struct net_device *lowerdev = vlan->lowerdev;
621 
622 	return dev_hard_header(skb, lowerdev, type, daddr,
623 			       saddr ? : dev->dev_addr, len);
624 }
625 
626 static const struct header_ops macvlan_hard_header_ops = {
627 	.create  	= macvlan_hard_header,
628 	.parse		= eth_header_parse,
629 	.cache		= eth_header_cache,
630 	.cache_update	= eth_header_cache_update,
631 	.parse_protocol	= eth_header_parse_protocol,
632 };
633 
634 static int macvlan_open(struct net_device *dev)
635 {
636 	struct macvlan_dev *vlan = netdev_priv(dev);
637 	struct net_device *lowerdev = vlan->lowerdev;
638 	int err;
639 
640 	if (macvlan_passthru(vlan->port)) {
641 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
642 			err = dev_set_promiscuity(lowerdev, 1);
643 			if (err < 0)
644 				goto out;
645 		}
646 		goto hash_add;
647 	}
648 
649 	err = -EADDRINUSE;
650 	if (macvlan_addr_busy(vlan->port, dev->dev_addr))
651 		goto out;
652 
653 	/* Attempt to populate accel_priv which is used to offload the L2
654 	 * forwarding requests for unicast packets.
655 	 */
656 	if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD)
657 		vlan->accel_priv =
658 		      lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
659 
660 	/* If earlier attempt to offload failed, or accel_priv is not
661 	 * populated we must add the unicast address to the lower device.
662 	 */
663 	if (IS_ERR_OR_NULL(vlan->accel_priv)) {
664 		vlan->accel_priv = NULL;
665 		err = dev_uc_add(lowerdev, dev->dev_addr);
666 		if (err < 0)
667 			goto out;
668 	}
669 
670 	if (dev->flags & IFF_ALLMULTI) {
671 		err = dev_set_allmulti(lowerdev, 1);
672 		if (err < 0)
673 			goto del_unicast;
674 	}
675 
676 	if (dev->flags & IFF_PROMISC) {
677 		err = dev_set_promiscuity(lowerdev, 1);
678 		if (err < 0)
679 			goto clear_multi;
680 	}
681 
682 hash_add:
683 	macvlan_hash_add(vlan);
684 	return 0;
685 
686 clear_multi:
687 	if (dev->flags & IFF_ALLMULTI)
688 		dev_set_allmulti(lowerdev, -1);
689 del_unicast:
690 	if (vlan->accel_priv) {
691 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
692 							   vlan->accel_priv);
693 		vlan->accel_priv = NULL;
694 	} else {
695 		dev_uc_del(lowerdev, dev->dev_addr);
696 	}
697 out:
698 	return err;
699 }
700 
701 static int macvlan_stop(struct net_device *dev)
702 {
703 	struct macvlan_dev *vlan = netdev_priv(dev);
704 	struct net_device *lowerdev = vlan->lowerdev;
705 
706 	if (vlan->accel_priv) {
707 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
708 							   vlan->accel_priv);
709 		vlan->accel_priv = NULL;
710 	}
711 
712 	dev_uc_unsync(lowerdev, dev);
713 	dev_mc_unsync(lowerdev, dev);
714 
715 	if (macvlan_passthru(vlan->port)) {
716 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
717 			dev_set_promiscuity(lowerdev, -1);
718 		goto hash_del;
719 	}
720 
721 	if (dev->flags & IFF_ALLMULTI)
722 		dev_set_allmulti(lowerdev, -1);
723 
724 	if (dev->flags & IFF_PROMISC)
725 		dev_set_promiscuity(lowerdev, -1);
726 
727 	dev_uc_del(lowerdev, dev->dev_addr);
728 
729 hash_del:
730 	macvlan_hash_del(vlan, !dev->dismantle);
731 	return 0;
732 }
733 
734 static int macvlan_sync_address(struct net_device *dev,
735 				const unsigned char *addr)
736 {
737 	struct macvlan_dev *vlan = netdev_priv(dev);
738 	struct net_device *lowerdev = vlan->lowerdev;
739 	struct macvlan_port *port = vlan->port;
740 	int err;
741 
742 	if (!(dev->flags & IFF_UP)) {
743 		/* Just copy in the new address */
744 		eth_hw_addr_set(dev, addr);
745 	} else {
746 		/* Rehash and update the device filters */
747 		if (macvlan_addr_busy(vlan->port, addr))
748 			return -EADDRINUSE;
749 
750 		if (!macvlan_passthru(port)) {
751 			err = dev_uc_add(lowerdev, addr);
752 			if (err)
753 				return err;
754 
755 			dev_uc_del(lowerdev, dev->dev_addr);
756 		}
757 
758 		macvlan_hash_change_addr(vlan, addr);
759 	}
760 	if (macvlan_passthru(port) && !macvlan_addr_change(port)) {
761 		/* Since addr_change isn't set, we are here due to lower
762 		 * device change.  Save the lower-dev address so we can
763 		 * restore it later.
764 		 */
765 		ether_addr_copy(vlan->port->perm_addr,
766 				lowerdev->dev_addr);
767 	}
768 	macvlan_clear_addr_change(port);
769 	return 0;
770 }
771 
772 static int macvlan_set_mac_address(struct net_device *dev, void *p)
773 {
774 	struct macvlan_dev *vlan = netdev_priv(dev);
775 	struct sockaddr_storage *addr = p;
776 
777 	if (!is_valid_ether_addr(addr->__data))
778 		return -EADDRNOTAVAIL;
779 
780 	/* If the addresses are the same, this is a no-op */
781 	if (ether_addr_equal(dev->dev_addr, addr->__data))
782 		return 0;
783 
784 	if (READ_ONCE(vlan->mode) == MACVLAN_MODE_PASSTHRU) {
785 		macvlan_set_addr_change(vlan->port);
786 		return dev_set_mac_address(vlan->lowerdev, addr, NULL);
787 	}
788 
789 	if (macvlan_addr_busy(vlan->port, addr->__data))
790 		return -EADDRINUSE;
791 
792 	return macvlan_sync_address(dev, addr->__data);
793 }
794 
795 static void macvlan_change_rx_flags(struct net_device *dev, int change)
796 {
797 	struct macvlan_dev *vlan = netdev_priv(dev);
798 	struct net_device *lowerdev = vlan->lowerdev;
799 
800 	if (dev->flags & IFF_UP) {
801 		if (change & IFF_ALLMULTI)
802 			dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
803 		if (!macvlan_passthru(vlan->port) && change & IFF_PROMISC)
804 			dev_set_promiscuity(lowerdev,
805 					    dev->flags & IFF_PROMISC ? 1 : -1);
806 
807 	}
808 }
809 
810 static void macvlan_compute_filter(unsigned long *mc_filter,
811 				   struct net_device *dev,
812 				   struct macvlan_dev *vlan, int cutoff)
813 {
814 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
815 		bitmap_fill(mc_filter, MACVLAN_MC_FILTER_SZ);
816 	} else {
817 		DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
818 		struct netdev_hw_addr *ha;
819 
820 		bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
821 		netdev_for_each_mc_addr(ha, dev) {
822 			if (!vlan && ha->synced <= cutoff)
823 				continue;
824 
825 			__set_bit(mc_hash(vlan, ha->addr), filter);
826 		}
827 
828 		__set_bit(mc_hash(vlan, dev->broadcast), filter);
829 
830 		bitmap_copy(mc_filter, filter, MACVLAN_MC_FILTER_SZ);
831 	}
832 }
833 
834 static void macvlan_recompute_bc_filter(struct macvlan_dev *vlan)
835 {
836 	if (vlan->port->bc_cutoff < 0) {
837 		bitmap_zero(vlan->port->bc_filter, MACVLAN_MC_FILTER_SZ);
838 		return;
839 	}
840 
841 	macvlan_compute_filter(vlan->port->bc_filter, vlan->lowerdev, NULL,
842 			       vlan->port->bc_cutoff);
843 }
844 
845 static void macvlan_set_mac_lists(struct net_device *dev)
846 {
847 	struct macvlan_dev *vlan = netdev_priv(dev);
848 
849 	macvlan_compute_filter(vlan->mc_filter, dev, vlan, 0);
850 
851 	dev_uc_sync(vlan->lowerdev, dev);
852 	dev_mc_sync(vlan->lowerdev, dev);
853 
854 	/* This is slightly inaccurate as we're including the subscription
855 	 * list of vlan->lowerdev too.
856 	 *
857 	 * Bug alert: This only works if everyone has the same broadcast
858 	 * address as lowerdev.  As soon as someone changes theirs this
859 	 * will break.
860 	 *
861 	 * However, this is already broken as when you change your broadcast
862 	 * address we don't get called.
863 	 *
864 	 * The solution is to maintain a list of broadcast addresses like
865 	 * we do for uc/mc, if you care.
866 	 */
867 	macvlan_compute_filter(vlan->port->mc_filter, vlan->lowerdev, NULL,
868 			       0);
869 	macvlan_recompute_bc_filter(vlan);
870 }
871 
872 static void update_port_bc_cutoff(struct macvlan_dev *vlan, int cutoff)
873 {
874 	if (vlan->port->bc_cutoff == cutoff)
875 		return;
876 
877 	WRITE_ONCE(vlan->port->bc_cutoff, cutoff);
878 	macvlan_recompute_bc_filter(vlan);
879 }
880 
881 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
882 {
883 	struct macvlan_dev *vlan = netdev_priv(dev);
884 
885 	if (vlan->lowerdev->mtu < new_mtu)
886 		return -EINVAL;
887 	WRITE_ONCE(dev->mtu, new_mtu);
888 	return 0;
889 }
890 
891 static int macvlan_hwtstamp_get(struct net_device *dev,
892 				struct kernel_hwtstamp_config *cfg)
893 {
894 	struct net_device *real_dev = macvlan_dev_real_dev(dev);
895 
896 	return generic_hwtstamp_get_lower(real_dev, cfg);
897 }
898 
899 static int macvlan_hwtstamp_set(struct net_device *dev,
900 				struct kernel_hwtstamp_config *cfg,
901 				struct netlink_ext_ack *extack)
902 {
903 	struct net_device *real_dev = macvlan_dev_real_dev(dev);
904 
905 	if (!net_eq(dev_net(dev), &init_net))
906 		return -EOPNOTSUPP;
907 
908 	return generic_hwtstamp_set_lower(real_dev, cfg, extack);
909 }
910 
911 /*
912  * macvlan network devices have devices nesting below it and are a special
913  * "super class" of normal network devices; split their locks off into a
914  * separate class since they always nest.
915  */
916 static struct lock_class_key macvlan_netdev_addr_lock_key;
917 
918 #define ALWAYS_ON_OFFLOADS \
919 	(NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE | \
920 	 NETIF_F_GSO_ROBUST | NETIF_F_GSO_ENCAP_ALL)
921 
922 #define ALWAYS_ON_FEATURES ALWAYS_ON_OFFLOADS
923 
924 #define MACVLAN_FEATURES \
925 	(NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
926 	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_LRO | \
927 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
928 	 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
929 
930 #define MACVLAN_STATE_MASK \
931 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
932 
933 static void macvlan_set_lockdep_class(struct net_device *dev)
934 {
935 	netdev_lockdep_set_classes(dev);
936 	lockdep_set_class(&dev->addr_list_lock,
937 			  &macvlan_netdev_addr_lock_key);
938 }
939 
940 static int macvlan_init(struct net_device *dev)
941 {
942 	struct macvlan_dev *vlan = netdev_priv(dev);
943 	struct net_device *lowerdev = vlan->lowerdev;
944 	struct macvlan_port *port = vlan->port;
945 
946 	dev->state		= (dev->state & ~MACVLAN_STATE_MASK) |
947 				  (lowerdev->state & MACVLAN_STATE_MASK);
948 	dev->features 		= lowerdev->features & MACVLAN_FEATURES;
949 	dev->features		|= ALWAYS_ON_FEATURES;
950 	dev->hw_features	|= NETIF_F_LRO;
951 	dev->vlan_features	= lowerdev->vlan_features & MACVLAN_FEATURES;
952 	dev->vlan_features	|= ALWAYS_ON_OFFLOADS;
953 	dev->hw_enc_features    |= dev->features;
954 	dev->lltx		= true;
955 	netif_inherit_tso_max(dev, lowerdev);
956 	dev->hard_header_len	= lowerdev->hard_header_len;
957 	dev->needed_headroom	= lowerdev->needed_headroom;
958 	dev->needed_tailroom	= lowerdev->needed_tailroom;
959 	macvlan_set_lockdep_class(dev);
960 
961 	vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
962 	if (!vlan->pcpu_stats)
963 		return -ENOMEM;
964 
965 	port->count += 1;
966 
967 	/* Get macvlan's reference to lowerdev */
968 	netdev_hold(lowerdev, &vlan->dev_tracker, GFP_KERNEL);
969 
970 	return 0;
971 }
972 
973 static void macvlan_uninit(struct net_device *dev)
974 {
975 	struct macvlan_dev *vlan = netdev_priv(dev);
976 	struct macvlan_port *port = vlan->port;
977 
978 	free_percpu(vlan->pcpu_stats);
979 
980 	macvlan_flush_sources(port, vlan);
981 	port->count -= 1;
982 	if (!port->count)
983 		macvlan_port_destroy(port->dev);
984 }
985 
986 static void macvlan_dev_get_stats64(struct net_device *dev,
987 				    struct rtnl_link_stats64 *stats)
988 {
989 	struct macvlan_dev *vlan = netdev_priv(dev);
990 
991 	if (vlan->pcpu_stats) {
992 		struct vlan_pcpu_stats *p;
993 		u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
994 		u32 rx_errors = 0, tx_dropped = 0;
995 		unsigned int start;
996 		int i;
997 
998 		for_each_possible_cpu(i) {
999 			p = per_cpu_ptr(vlan->pcpu_stats, i);
1000 			do {
1001 				start = u64_stats_fetch_begin(&p->syncp);
1002 				rx_packets	= u64_stats_read(&p->rx_packets);
1003 				rx_bytes	= u64_stats_read(&p->rx_bytes);
1004 				rx_multicast	= u64_stats_read(&p->rx_multicast);
1005 				tx_packets	= u64_stats_read(&p->tx_packets);
1006 				tx_bytes	= u64_stats_read(&p->tx_bytes);
1007 			} while (u64_stats_fetch_retry(&p->syncp, start));
1008 
1009 			stats->rx_packets	+= rx_packets;
1010 			stats->rx_bytes		+= rx_bytes;
1011 			stats->multicast	+= rx_multicast;
1012 			stats->tx_packets	+= tx_packets;
1013 			stats->tx_bytes		+= tx_bytes;
1014 			/* rx_errors & tx_dropped are u32, updated
1015 			 * without syncp protection.
1016 			 */
1017 			rx_errors	+= READ_ONCE(p->rx_errors);
1018 			tx_dropped	+= READ_ONCE(p->tx_dropped);
1019 		}
1020 		stats->rx_errors	= rx_errors;
1021 		stats->rx_dropped	= rx_errors;
1022 		stats->tx_dropped	= tx_dropped;
1023 	}
1024 }
1025 
1026 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
1027 				   __be16 proto, u16 vid)
1028 {
1029 	struct macvlan_dev *vlan = netdev_priv(dev);
1030 	struct net_device *lowerdev = vlan->lowerdev;
1031 
1032 	return vlan_vid_add(lowerdev, proto, vid);
1033 }
1034 
1035 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
1036 				    __be16 proto, u16 vid)
1037 {
1038 	struct macvlan_dev *vlan = netdev_priv(dev);
1039 	struct net_device *lowerdev = vlan->lowerdev;
1040 
1041 	vlan_vid_del(lowerdev, proto, vid);
1042 	return 0;
1043 }
1044 
1045 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
1046 			   struct net_device *dev,
1047 			   const unsigned char *addr, u16 vid,
1048 			   u16 flags, bool *notified,
1049 			   struct netlink_ext_ack *extack)
1050 {
1051 	struct macvlan_dev *vlan = netdev_priv(dev);
1052 	int err = -EINVAL;
1053 
1054 	/* Support unicast filter only on passthru devices.
1055 	 * Multicast filter should be allowed on all devices.
1056 	 */
1057 	if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr))
1058 		return -EOPNOTSUPP;
1059 
1060 	if (flags & NLM_F_REPLACE)
1061 		return -EOPNOTSUPP;
1062 
1063 	if (is_unicast_ether_addr(addr))
1064 		err = dev_uc_add_excl(dev, addr);
1065 	else if (is_multicast_ether_addr(addr))
1066 		err = dev_mc_add_excl(dev, addr);
1067 
1068 	return err;
1069 }
1070 
1071 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
1072 			   struct net_device *dev,
1073 			   const unsigned char *addr, u16 vid, bool *notified,
1074 			   struct netlink_ext_ack *extack)
1075 {
1076 	struct macvlan_dev *vlan = netdev_priv(dev);
1077 	int err = -EINVAL;
1078 
1079 	/* Support unicast filter only on passthru devices.
1080 	 * Multicast filter should be allowed on all devices.
1081 	 */
1082 	if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr))
1083 		return -EOPNOTSUPP;
1084 
1085 	if (is_unicast_ether_addr(addr))
1086 		err = dev_uc_del(dev, addr);
1087 	else if (is_multicast_ether_addr(addr))
1088 		err = dev_mc_del(dev, addr);
1089 
1090 	return err;
1091 }
1092 
1093 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
1094 					struct ethtool_drvinfo *drvinfo)
1095 {
1096 	strscpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
1097 	strscpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
1098 }
1099 
1100 static int macvlan_ethtool_get_link_ksettings(struct net_device *dev,
1101 					      struct ethtool_link_ksettings *cmd)
1102 {
1103 	const struct macvlan_dev *vlan = netdev_priv(dev);
1104 
1105 	return __ethtool_get_link_ksettings(vlan->lowerdev, cmd);
1106 }
1107 
1108 static int macvlan_ethtool_get_ts_info(struct net_device *dev,
1109 				       struct kernel_ethtool_ts_info *info)
1110 {
1111 	struct net_device *real_dev = macvlan_dev_real_dev(dev);
1112 
1113 	return ethtool_get_ts_info_by_layer(real_dev, info);
1114 }
1115 
1116 static netdev_features_t macvlan_fix_features(struct net_device *dev,
1117 					      netdev_features_t features)
1118 {
1119 	struct macvlan_dev *vlan = netdev_priv(dev);
1120 	netdev_features_t lowerdev_features = vlan->lowerdev->features;
1121 	netdev_features_t mask;
1122 
1123 	features |= NETIF_F_ALL_FOR_ALL;
1124 	features &= (vlan->set_features | ~MACVLAN_FEATURES);
1125 	mask = features;
1126 
1127 	lowerdev_features &= (features | ~NETIF_F_LRO);
1128 	features = netdev_increment_features(lowerdev_features, features, mask);
1129 	features |= ALWAYS_ON_FEATURES;
1130 	features &= (ALWAYS_ON_FEATURES | MACVLAN_FEATURES);
1131 
1132 	return features;
1133 }
1134 
1135 #ifdef CONFIG_NET_POLL_CONTROLLER
1136 static void macvlan_dev_poll_controller(struct net_device *dev)
1137 {
1138 	return;
1139 }
1140 
1141 static int macvlan_dev_netpoll_setup(struct net_device *dev)
1142 {
1143 	struct macvlan_dev *vlan = netdev_priv(dev);
1144 	struct net_device *real_dev = vlan->lowerdev;
1145 	struct netpoll *netpoll;
1146 	int err;
1147 
1148 	netpoll = kzalloc_obj(*netpoll);
1149 	err = -ENOMEM;
1150 	if (!netpoll)
1151 		goto out;
1152 
1153 	err = __netpoll_setup(netpoll, real_dev);
1154 	if (err) {
1155 		kfree(netpoll);
1156 		goto out;
1157 	}
1158 
1159 	vlan->netpoll = netpoll;
1160 
1161 out:
1162 	return err;
1163 }
1164 
1165 static void macvlan_dev_netpoll_cleanup(struct net_device *dev)
1166 {
1167 	struct macvlan_dev *vlan = netdev_priv(dev);
1168 	struct netpoll *netpoll = vlan->netpoll;
1169 
1170 	if (!netpoll)
1171 		return;
1172 
1173 	vlan->netpoll = NULL;
1174 
1175 	__netpoll_free(netpoll);
1176 }
1177 #endif	/* CONFIG_NET_POLL_CONTROLLER */
1178 
1179 static int macvlan_dev_get_iflink(const struct net_device *dev)
1180 {
1181 	struct macvlan_dev *vlan = netdev_priv(dev);
1182 
1183 	return READ_ONCE(vlan->lowerdev->ifindex);
1184 }
1185 
1186 static const struct ethtool_ops macvlan_ethtool_ops = {
1187 	.get_link		= ethtool_op_get_link,
1188 	.get_link_ksettings	= macvlan_ethtool_get_link_ksettings,
1189 	.get_drvinfo		= macvlan_ethtool_get_drvinfo,
1190 	.get_ts_info		= macvlan_ethtool_get_ts_info,
1191 };
1192 
1193 static const struct net_device_ops macvlan_netdev_ops = {
1194 	.ndo_init		= macvlan_init,
1195 	.ndo_uninit		= macvlan_uninit,
1196 	.ndo_open		= macvlan_open,
1197 	.ndo_stop		= macvlan_stop,
1198 	.ndo_start_xmit		= macvlan_start_xmit,
1199 	.ndo_change_mtu		= macvlan_change_mtu,
1200 	.ndo_fix_features	= macvlan_fix_features,
1201 	.ndo_change_rx_flags	= macvlan_change_rx_flags,
1202 	.ndo_set_mac_address	= macvlan_set_mac_address,
1203 	.ndo_set_rx_mode	= macvlan_set_mac_lists,
1204 	.ndo_get_stats64	= macvlan_dev_get_stats64,
1205 	.ndo_validate_addr	= eth_validate_addr,
1206 	.ndo_vlan_rx_add_vid	= macvlan_vlan_rx_add_vid,
1207 	.ndo_vlan_rx_kill_vid	= macvlan_vlan_rx_kill_vid,
1208 	.ndo_fdb_add		= macvlan_fdb_add,
1209 	.ndo_fdb_del		= macvlan_fdb_del,
1210 	.ndo_fdb_dump		= ndo_dflt_fdb_dump,
1211 #ifdef CONFIG_NET_POLL_CONTROLLER
1212 	.ndo_poll_controller	= macvlan_dev_poll_controller,
1213 	.ndo_netpoll_setup	= macvlan_dev_netpoll_setup,
1214 	.ndo_netpoll_cleanup	= macvlan_dev_netpoll_cleanup,
1215 #endif
1216 	.ndo_get_iflink		= macvlan_dev_get_iflink,
1217 	.ndo_features_check	= passthru_features_check,
1218 	.ndo_hwtstamp_get	= macvlan_hwtstamp_get,
1219 	.ndo_hwtstamp_set	= macvlan_hwtstamp_set,
1220 };
1221 
1222 static void macvlan_dev_free(struct net_device *dev)
1223 {
1224 	struct macvlan_dev *vlan = netdev_priv(dev);
1225 
1226 	/* Get rid of the macvlan's reference to lowerdev */
1227 	netdev_put(vlan->lowerdev, &vlan->dev_tracker);
1228 }
1229 
1230 void macvlan_common_setup(struct net_device *dev)
1231 {
1232 	ether_setup(dev);
1233 
1234 	/* ether_setup() has set dev->min_mtu to ETH_MIN_MTU. */
1235 	dev->max_mtu		= ETH_MAX_MTU;
1236 	dev->priv_flags	       &= ~IFF_TX_SKB_SHARING;
1237 	netif_keep_dst(dev);
1238 	dev->priv_flags	       |= IFF_UNICAST_FLT;
1239 	dev->change_proto_down	= true;
1240 	dev->netdev_ops		= &macvlan_netdev_ops;
1241 	dev->needs_free_netdev	= true;
1242 	dev->priv_destructor	= macvlan_dev_free;
1243 	dev->header_ops		= &macvlan_hard_header_ops;
1244 	dev->ethtool_ops	= &macvlan_ethtool_ops;
1245 }
1246 EXPORT_SYMBOL_GPL(macvlan_common_setup);
1247 
1248 static void macvlan_setup(struct net_device *dev)
1249 {
1250 	macvlan_common_setup(dev);
1251 	dev->priv_flags |= IFF_NO_QUEUE;
1252 }
1253 
1254 static int macvlan_port_create(struct net_device *dev)
1255 {
1256 	struct macvlan_port *port;
1257 	unsigned int i;
1258 	int err;
1259 
1260 	if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
1261 		return -EINVAL;
1262 
1263 	if (netdev_is_rx_handler_busy(dev))
1264 		return -EBUSY;
1265 
1266 	port = kzalloc_obj(*port);
1267 	if (port == NULL)
1268 		return -ENOMEM;
1269 
1270 	port->dev = dev;
1271 	ether_addr_copy(port->perm_addr, dev->dev_addr);
1272 	INIT_LIST_HEAD(&port->vlans);
1273 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1274 		INIT_HLIST_HEAD(&port->vlan_hash[i]);
1275 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1276 		INIT_HLIST_HEAD(&port->vlan_source_hash[i]);
1277 
1278 	port->bc_queue_len_used = 0;
1279 	port->bc_cutoff = 1;
1280 	skb_queue_head_init(&port->bc_queue);
1281 	INIT_WORK(&port->bc_work, macvlan_process_broadcast);
1282 
1283 	err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
1284 	if (err)
1285 		kfree(port);
1286 	else
1287 		dev->priv_flags |= IFF_MACVLAN_PORT;
1288 	return err;
1289 }
1290 
1291 static void macvlan_port_destroy(struct net_device *dev)
1292 {
1293 	struct macvlan_port *port = macvlan_port_get_rtnl(dev);
1294 	struct sk_buff *skb;
1295 
1296 	dev->priv_flags &= ~IFF_MACVLAN_PORT;
1297 	netdev_rx_handler_unregister(dev);
1298 
1299 	/* After this point, no packet can schedule bc_work anymore,
1300 	 * but we need to cancel it and purge left skbs if any.
1301 	 */
1302 	cancel_work_sync(&port->bc_work);
1303 
1304 	while ((skb = __skb_dequeue(&port->bc_queue))) {
1305 		const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
1306 
1307 		if (src)
1308 			dev_put(src->dev);
1309 
1310 		kfree_skb(skb);
1311 	}
1312 
1313 	/* If the lower device address has been changed by passthru
1314 	 * macvlan, put it back.
1315 	 */
1316 	if (macvlan_passthru(port) &&
1317 	    !ether_addr_equal(port->dev->dev_addr, port->perm_addr)) {
1318 		struct sockaddr_storage ss;
1319 
1320 		ss.ss_family = port->dev->type;
1321 		memcpy(&ss.__data, port->perm_addr, port->dev->addr_len);
1322 		dev_set_mac_address(port->dev, &ss, NULL);
1323 	}
1324 
1325 	kfree(port);
1326 }
1327 
1328 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[],
1329 			    struct netlink_ext_ack *extack)
1330 {
1331 	struct nlattr *nla, *head;
1332 	int rem, len;
1333 
1334 	if (tb[IFLA_ADDRESS]) {
1335 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1336 			return -EINVAL;
1337 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1338 			return -EADDRNOTAVAIL;
1339 	}
1340 
1341 	if (!data)
1342 		return 0;
1343 
1344 	if (data[IFLA_MACVLAN_BC_QUEUE_LEN] &&
1345 	    nla_get_u32(data[IFLA_MACVLAN_BC_QUEUE_LEN]) >
1346 			MACVLAN_DEFAULT_BC_QUEUE_LEN &&
1347 	    !capable(CAP_NET_ADMIN)) {
1348 		NL_SET_ERR_MSG_ATTR(extack, data[IFLA_MACVLAN_BC_QUEUE_LEN],
1349 				    "bc_queue_len above the default requires CAP_NET_ADMIN in the initial user namespace");
1350 		return -EPERM;
1351 	}
1352 
1353 	if (data[IFLA_MACVLAN_FLAGS] &&
1354 	    nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~(MACVLAN_FLAG_NOPROMISC |
1355 						      MACVLAN_FLAG_NODST))
1356 		return -EINVAL;
1357 
1358 	if (data[IFLA_MACVLAN_MODE]) {
1359 		switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
1360 		case MACVLAN_MODE_PRIVATE:
1361 		case MACVLAN_MODE_VEPA:
1362 		case MACVLAN_MODE_BRIDGE:
1363 		case MACVLAN_MODE_PASSTHRU:
1364 		case MACVLAN_MODE_SOURCE:
1365 			break;
1366 		default:
1367 			return -EINVAL;
1368 		}
1369 	}
1370 
1371 	if (data[IFLA_MACVLAN_MACADDR_MODE]) {
1372 		switch (nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE])) {
1373 		case MACVLAN_MACADDR_ADD:
1374 		case MACVLAN_MACADDR_DEL:
1375 		case MACVLAN_MACADDR_FLUSH:
1376 		case MACVLAN_MACADDR_SET:
1377 			break;
1378 		default:
1379 			return -EINVAL;
1380 		}
1381 	}
1382 
1383 	if (data[IFLA_MACVLAN_MACADDR]) {
1384 		if (nla_len(data[IFLA_MACVLAN_MACADDR]) != ETH_ALEN)
1385 			return -EINVAL;
1386 
1387 		if (!is_valid_ether_addr(nla_data(data[IFLA_MACVLAN_MACADDR])))
1388 			return -EADDRNOTAVAIL;
1389 	}
1390 
1391 	if (data[IFLA_MACVLAN_MACADDR_DATA]) {
1392 		head = nla_data(data[IFLA_MACVLAN_MACADDR_DATA]);
1393 		len = nla_len(data[IFLA_MACVLAN_MACADDR_DATA]);
1394 
1395 		nla_for_each_attr(nla, head, len, rem) {
1396 			if (nla_type(nla) != IFLA_MACVLAN_MACADDR ||
1397 			    nla_len(nla) != ETH_ALEN)
1398 				return -EINVAL;
1399 
1400 			if (!is_valid_ether_addr(nla_data(nla)))
1401 				return -EADDRNOTAVAIL;
1402 		}
1403 	}
1404 
1405 	if (data[IFLA_MACVLAN_MACADDR_COUNT])
1406 		return -EINVAL;
1407 
1408 	return 0;
1409 }
1410 
1411 /*
1412  * reconfigure list of remote source mac address
1413  * (only for macvlan devices in source mode)
1414  * Note regarding alignment: all netlink data is aligned to 4 Byte, which
1415  * suffices for both ether_addr_copy and ether_addr_equal_64bits usage.
1416  */
1417 static int macvlan_changelink_sources(struct macvlan_dev *vlan, u32 mode,
1418 				      struct nlattr *data[])
1419 {
1420 	char *addr = NULL;
1421 	int ret, rem, len;
1422 	struct nlattr *nla, *head;
1423 	struct macvlan_source_entry *entry;
1424 
1425 	if (data[IFLA_MACVLAN_MACADDR])
1426 		addr = nla_data(data[IFLA_MACVLAN_MACADDR]);
1427 
1428 	if (mode == MACVLAN_MACADDR_ADD) {
1429 		if (!addr)
1430 			return -EINVAL;
1431 
1432 		return macvlan_hash_add_source(vlan, addr);
1433 
1434 	} else if (mode == MACVLAN_MACADDR_DEL) {
1435 		if (!addr)
1436 			return -EINVAL;
1437 
1438 		entry = macvlan_hash_lookup_source(vlan, addr);
1439 		if (entry) {
1440 			macvlan_hash_del_source(entry);
1441 			WRITE_ONCE(vlan->macaddr_count, vlan->macaddr_count - 1);
1442 		}
1443 	} else if (mode == MACVLAN_MACADDR_FLUSH) {
1444 		macvlan_flush_sources(vlan->port, vlan);
1445 	} else if (mode == MACVLAN_MACADDR_SET) {
1446 		macvlan_flush_sources(vlan->port, vlan);
1447 
1448 		if (addr) {
1449 			ret = macvlan_hash_add_source(vlan, addr);
1450 			if (ret)
1451 				return ret;
1452 		}
1453 
1454 		if (!data[IFLA_MACVLAN_MACADDR_DATA])
1455 			return 0;
1456 
1457 		head = nla_data(data[IFLA_MACVLAN_MACADDR_DATA]);
1458 		len = nla_len(data[IFLA_MACVLAN_MACADDR_DATA]);
1459 
1460 		nla_for_each_attr(nla, head, len, rem) {
1461 			addr = nla_data(nla);
1462 			ret = macvlan_hash_add_source(vlan, addr);
1463 			if (ret)
1464 				return ret;
1465 		}
1466 	} else {
1467 		return -EINVAL;
1468 	}
1469 
1470 	return 0;
1471 }
1472 
1473 int macvlan_common_newlink(struct net_device *dev,
1474 			   struct rtnl_newlink_params *params,
1475 			   struct netlink_ext_ack *extack)
1476 {
1477 	struct net *link_net = rtnl_newlink_link_net(params);
1478 	struct macvlan_dev *vlan = netdev_priv(dev);
1479 	struct nlattr **data = params->data;
1480 	struct nlattr **tb = params->tb;
1481 	struct net_device *lowerdev;
1482 	struct macvlan_port *port;
1483 	bool create = false;
1484 	int macmode;
1485 	int err;
1486 
1487 	if (!tb[IFLA_LINK])
1488 		return -EINVAL;
1489 
1490 	lowerdev = __dev_get_by_index(link_net, nla_get_u32(tb[IFLA_LINK]));
1491 	if (lowerdev == NULL)
1492 		return -ENODEV;
1493 
1494 	/* When creating macvlans or macvtaps on top of other macvlans - use
1495 	 * the real device as the lowerdev.
1496 	 */
1497 	if (netif_is_macvlan(lowerdev)) {
1498 		lowerdev = macvlan_dev_real_dev(lowerdev);
1499 		if (!rtnl_dev_link_net_capable(dev, dev_net(lowerdev))) {
1500 			NL_SET_ERR_MSG(extack,
1501 				       "Creating a macvlan on a lower device in another network namespace requires CAP_NET_ADMIN in that namespace");
1502 			return -EPERM;
1503 		}
1504 	}
1505 
1506 	if (!tb[IFLA_MTU])
1507 		dev->mtu = lowerdev->mtu;
1508 	else if (dev->mtu > lowerdev->mtu)
1509 		return -EINVAL;
1510 
1511 	/* MTU range: 68 - lowerdev->max_mtu */
1512 	dev->min_mtu = ETH_MIN_MTU;
1513 	dev->max_mtu = lowerdev->max_mtu;
1514 
1515 	if (!tb[IFLA_ADDRESS])
1516 		eth_hw_addr_random(dev);
1517 
1518 	if (!netif_is_macvlan_port(lowerdev)) {
1519 		err = macvlan_port_create(lowerdev);
1520 		if (err < 0)
1521 			return err;
1522 		create = true;
1523 	}
1524 	port = macvlan_port_get_rtnl(lowerdev);
1525 
1526 	/* Only 1 macvlan device can be created in passthru mode */
1527 	if (macvlan_passthru(port)) {
1528 		/* The macvlan port must be not created this time,
1529 		 * still goto destroy_macvlan_port for readability.
1530 		 */
1531 		err = -EINVAL;
1532 		goto destroy_macvlan_port;
1533 	}
1534 
1535 	vlan->lowerdev = lowerdev;
1536 	vlan->dev      = dev;
1537 	vlan->port     = port;
1538 	vlan->set_features = MACVLAN_FEATURES;
1539 
1540 	vlan->mode     = MACVLAN_MODE_VEPA;
1541 	if (data && data[IFLA_MACVLAN_MODE])
1542 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1543 
1544 	if (data && data[IFLA_MACVLAN_FLAGS])
1545 		vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1546 
1547 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
1548 		if (port->count) {
1549 			err = -EINVAL;
1550 			goto destroy_macvlan_port;
1551 		}
1552 		macvlan_set_passthru(port);
1553 		eth_hw_addr_inherit(dev, lowerdev);
1554 	}
1555 
1556 	if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1557 		if (vlan->mode != MACVLAN_MODE_SOURCE) {
1558 			err = -EINVAL;
1559 			goto destroy_macvlan_port;
1560 		}
1561 		macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1562 		err = macvlan_changelink_sources(vlan, macmode, data);
1563 		if (err)
1564 			goto destroy_macvlan_port;
1565 	}
1566 
1567 	vlan->bc_queue_len_req = MACVLAN_DEFAULT_BC_QUEUE_LEN;
1568 	if (data && data[IFLA_MACVLAN_BC_QUEUE_LEN])
1569 		vlan->bc_queue_len_req = nla_get_u32(data[IFLA_MACVLAN_BC_QUEUE_LEN]);
1570 
1571 	if (data && data[IFLA_MACVLAN_BC_CUTOFF])
1572 		update_port_bc_cutoff(
1573 			vlan, nla_get_s32(data[IFLA_MACVLAN_BC_CUTOFF]));
1574 
1575 	err = register_netdevice(dev);
1576 	if (err < 0)
1577 		goto destroy_macvlan_port;
1578 
1579 	dev->priv_flags |= IFF_MACVLAN;
1580 	err = netdev_upper_dev_link(lowerdev, dev, extack);
1581 	if (err)
1582 		goto unregister_netdev;
1583 
1584 	list_add_tail_rcu(&vlan->list, &port->vlans);
1585 	update_port_bc_queue_len(vlan->port);
1586 	netif_stacked_transfer_operstate(lowerdev, dev);
1587 	linkwatch_fire_event(dev);
1588 
1589 	return 0;
1590 
1591 unregister_netdev:
1592 	/* macvlan_uninit would free the macvlan port */
1593 	unregister_netdevice(dev);
1594 	return err;
1595 destroy_macvlan_port:
1596 	/* the macvlan port may be freed by macvlan_uninit when fail to register.
1597 	 * so we destroy the macvlan port only when it's valid.
1598 	 */
1599 	if (macvlan_port_get_rtnl(lowerdev)) {
1600 		macvlan_flush_sources(port, vlan);
1601 		if (create)
1602 			macvlan_port_destroy(port->dev);
1603 	}
1604 	/* @dev might have been made visible before an error was detected.
1605 	 * Make sure to observe an RCU grace period before our caller
1606 	 * (rtnl_newlink()) frees it.
1607 	 */
1608 	synchronize_net();
1609 	return err;
1610 }
1611 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
1612 
1613 static int macvlan_newlink(struct net_device *dev,
1614 			   struct rtnl_newlink_params *params,
1615 			   struct netlink_ext_ack *extack)
1616 {
1617 	return macvlan_common_newlink(dev, params, extack);
1618 }
1619 
1620 void macvlan_dellink(struct net_device *dev, struct list_head *head)
1621 {
1622 	struct macvlan_dev *vlan = netdev_priv(dev);
1623 
1624 	if (vlan->mode == MACVLAN_MODE_SOURCE)
1625 		macvlan_flush_sources(vlan->port, vlan);
1626 	list_del_rcu(&vlan->list);
1627 	update_port_bc_queue_len(vlan->port);
1628 	unregister_netdevice_queue(dev, head);
1629 	netdev_upper_dev_unlink(vlan->lowerdev, dev);
1630 }
1631 EXPORT_SYMBOL_GPL(macvlan_dellink);
1632 
1633 static int macvlan_changelink(struct net_device *dev,
1634 			      struct nlattr *tb[], struct nlattr *data[],
1635 			      struct netlink_ext_ack *extack)
1636 {
1637 	struct macvlan_dev *vlan = netdev_priv(dev);
1638 	enum macvlan_mode mode;
1639 	bool set_mode = false;
1640 	enum macvlan_macaddr_mode macmode;
1641 	int ret;
1642 
1643 	if (data &&
1644 	    (data[IFLA_MACVLAN_BC_QUEUE_LEN] || data[IFLA_MACVLAN_BC_CUTOFF]) &&
1645 	    !rtnl_dev_link_net_capable(dev, dev_net(vlan->lowerdev))) {
1646 		NL_SET_ERR_MSG(extack,
1647 			       "Changing shared macvlan port settings requires CAP_NET_ADMIN in the lower device network namespace");
1648 		return -EPERM;
1649 	}
1650 
1651 	/* Validate mode, but don't set yet: setting flags may fail. */
1652 	if (data && data[IFLA_MACVLAN_MODE]) {
1653 		set_mode = true;
1654 		mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1655 		/* Passthrough mode can't be set or cleared dynamically */
1656 		if ((mode == MACVLAN_MODE_PASSTHRU) !=
1657 		    (vlan->mode == MACVLAN_MODE_PASSTHRU))
1658 			return -EINVAL;
1659 		if (vlan->mode == MACVLAN_MODE_SOURCE &&
1660 		    vlan->mode != mode)
1661 			macvlan_flush_sources(vlan->port, vlan);
1662 	}
1663 
1664 	if (data && data[IFLA_MACVLAN_FLAGS]) {
1665 		__u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1666 		bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
1667 		if (macvlan_passthru(vlan->port) && promisc) {
1668 			int err;
1669 
1670 			if (flags & MACVLAN_FLAG_NOPROMISC)
1671 				err = dev_set_promiscuity(vlan->lowerdev, -1);
1672 			else
1673 				err = dev_set_promiscuity(vlan->lowerdev, 1);
1674 			if (err < 0)
1675 				return err;
1676 		}
1677 		WRITE_ONCE(vlan->flags, flags);
1678 	}
1679 
1680 	if (data && data[IFLA_MACVLAN_BC_QUEUE_LEN]) {
1681 		WRITE_ONCE(vlan->bc_queue_len_req,
1682 			   nla_get_u32(data[IFLA_MACVLAN_BC_QUEUE_LEN]));
1683 		update_port_bc_queue_len(vlan->port);
1684 	}
1685 
1686 	if (data && data[IFLA_MACVLAN_BC_CUTOFF])
1687 		update_port_bc_cutoff(
1688 			vlan, nla_get_s32(data[IFLA_MACVLAN_BC_CUTOFF]));
1689 
1690 	if (set_mode)
1691 		WRITE_ONCE(vlan->mode, mode);
1692 	if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1693 		if (vlan->mode != MACVLAN_MODE_SOURCE)
1694 			return -EINVAL;
1695 		macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1696 		ret = macvlan_changelink_sources(vlan, macmode, data);
1697 		if (ret)
1698 			return ret;
1699 	}
1700 	return 0;
1701 }
1702 
1703 static size_t macvlan_get_size_mac(const struct macvlan_dev *vlan)
1704 {
1705 	unsigned int macaddr_count = READ_ONCE(vlan->macaddr_count);
1706 
1707 	if (!macaddr_count)
1708 		return 0;
1709 	return nla_total_size(0) /* IFLA_MACVLAN_MACADDR_DATA */
1710 		+ macaddr_count * nla_total_size(sizeof(u8) * ETH_ALEN);
1711 }
1712 
1713 static size_t macvlan_get_size(const struct net_device *dev)
1714 {
1715 	struct macvlan_dev *vlan = netdev_priv(dev);
1716 
1717 	return (0
1718 		+ nla_total_size(4) /* IFLA_MACVLAN_MODE */
1719 		+ nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
1720 		+ nla_total_size(4) /* IFLA_MACVLAN_MACADDR_COUNT */
1721 		+ macvlan_get_size_mac(vlan) /* IFLA_MACVLAN_MACADDR */
1722 		+ nla_total_size(4) /* IFLA_MACVLAN_BC_QUEUE_LEN */
1723 		+ nla_total_size(4) /* IFLA_MACVLAN_BC_QUEUE_LEN_USED */
1724 		+ nla_total_size(4) /* IFLA_MACVLAN_BC_CUTOFF */
1725 		);
1726 }
1727 
1728 static int macvlan_fill_info_macaddr(struct sk_buff *skb,
1729 				     const struct macvlan_dev *vlan,
1730 				     const int i)
1731 {
1732 	struct hlist_head *h = &vlan->port->vlan_source_hash[i];
1733 	const struct macvlan_source_entry *entry;
1734 	int cnt = 0;
1735 
1736 	hlist_for_each_entry_rcu(entry, h, hlist) {
1737 		if (rcu_access_pointer(entry->vlan) != vlan)
1738 			continue;
1739 		if (nla_put(skb, IFLA_MACVLAN_MACADDR, ETH_ALEN, entry->addr))
1740 			return -EMSGSIZE;
1741 		cnt++;
1742 	}
1743 	return cnt;
1744 }
1745 
1746 static int macvlan_fill_info(struct sk_buff *skb,
1747 				const struct net_device *dev)
1748 {
1749 	const struct macvlan_dev *vlan = netdev_priv(dev);
1750 	struct macvlan_port *port = vlan->port;
1751 	unsigned int macaddr_count = 0;
1752 	struct nlattr *nest, *attr;
1753 	int bc_cutoff, cnt, i;
1754 
1755 	rcu_read_lock();
1756 	if (nla_put_u32(skb, IFLA_MACVLAN_MODE, READ_ONCE(vlan->mode)))
1757 		goto nla_put_failure;
1758 
1759 	if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, READ_ONCE(vlan->flags)))
1760 		goto nla_put_failure;
1761 
1762 	attr = nla_reserve(skb, IFLA_MACVLAN_MACADDR_COUNT, sizeof(u32));
1763 	if (!attr)
1764 		goto nla_put_failure;
1765 
1766 	if (READ_ONCE(vlan->macaddr_count) > 0) {
1767 		nest = nla_nest_start_noflag(skb, IFLA_MACVLAN_MACADDR_DATA);
1768 		if (nest == NULL)
1769 			goto nla_put_failure;
1770 
1771 		for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
1772 			cnt = macvlan_fill_info_macaddr(skb, vlan, i);
1773 			if (cnt < 0)
1774 				goto nla_put_failure;
1775 			macaddr_count += cnt;
1776 		}
1777 		if (!macaddr_count)
1778 			nla_nest_cancel(skb, nest);
1779 		else if (nla_nest_end_safe(skb, nest) < 0)
1780 			goto nla_put_failure;
1781 	}
1782 	*(u32 *)nla_data(attr) = macaddr_count;
1783 
1784 	if (nla_put_u32(skb, IFLA_MACVLAN_BC_QUEUE_LEN,
1785 			READ_ONCE(vlan->bc_queue_len_req)))
1786 		goto nla_put_failure;
1787 
1788 	if (nla_put_u32(skb, IFLA_MACVLAN_BC_QUEUE_LEN_USED,
1789 			READ_ONCE(port->bc_queue_len_used)))
1790 		goto nla_put_failure;
1791 
1792 	bc_cutoff = READ_ONCE(port->bc_cutoff);
1793 	if (bc_cutoff != 1 &&
1794 	    nla_put_s32(skb, IFLA_MACVLAN_BC_CUTOFF, bc_cutoff))
1795 		goto nla_put_failure;
1796 
1797 	rcu_read_unlock();
1798 	return 0;
1799 
1800 nla_put_failure:
1801 	rcu_read_unlock();
1802 	return -EMSGSIZE;
1803 }
1804 
1805 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
1806 	[IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
1807 	[IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
1808 	[IFLA_MACVLAN_MACADDR_MODE] = { .type = NLA_U32 },
1809 	[IFLA_MACVLAN_MACADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1810 	[IFLA_MACVLAN_MACADDR_DATA] = { .type = NLA_NESTED },
1811 	[IFLA_MACVLAN_MACADDR_COUNT] = { .type = NLA_U32 },
1812 	[IFLA_MACVLAN_BC_QUEUE_LEN] = { .type = NLA_U32 },
1813 	[IFLA_MACVLAN_BC_QUEUE_LEN_USED] = { .type = NLA_REJECT },
1814 	[IFLA_MACVLAN_BC_CUTOFF] = { .type = NLA_S32 },
1815 };
1816 
1817 int macvlan_link_register(struct rtnl_link_ops *ops)
1818 {
1819 	/* common fields */
1820 	ops->validate		= macvlan_validate;
1821 	ops->maxtype		= IFLA_MACVLAN_MAX;
1822 	ops->policy		= macvlan_policy;
1823 	ops->changelink		= macvlan_changelink;
1824 	ops->get_size		= macvlan_get_size;
1825 	ops->fill_info		= macvlan_fill_info;
1826 
1827 	return rtnl_link_register(ops);
1828 };
1829 EXPORT_SYMBOL_GPL(macvlan_link_register);
1830 
1831 static struct net *macvlan_get_link_net(const struct net_device *dev)
1832 {
1833 	return dev_net(macvlan_dev_real_dev(dev));
1834 }
1835 
1836 static struct rtnl_link_ops macvlan_link_ops = {
1837 	.kind		= "macvlan",
1838 	.setup		= macvlan_setup,
1839 	.newlink	= macvlan_newlink,
1840 	.dellink	= macvlan_dellink,
1841 	.get_link_net	= macvlan_get_link_net,
1842 	.priv_size      = sizeof(struct macvlan_dev),
1843 };
1844 
1845 static void update_port_bc_queue_len(struct macvlan_port *port)
1846 {
1847 	u32 max_bc_queue_len_req = 0;
1848 	struct macvlan_dev *vlan;
1849 
1850 	list_for_each_entry(vlan, &port->vlans, list) {
1851 		if (vlan->bc_queue_len_req > max_bc_queue_len_req)
1852 			max_bc_queue_len_req = vlan->bc_queue_len_req;
1853 	}
1854 	WRITE_ONCE(port->bc_queue_len_used, max_bc_queue_len_req);
1855 }
1856 
1857 static int macvlan_device_event(struct notifier_block *unused,
1858 				unsigned long event, void *ptr)
1859 {
1860 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1861 	struct macvlan_dev *vlan, *next;
1862 	struct macvlan_port *port;
1863 	LIST_HEAD(list_kill);
1864 
1865 	if (!netif_is_macvlan_port(dev))
1866 		return NOTIFY_DONE;
1867 
1868 	port = macvlan_port_get_rtnl(dev);
1869 
1870 	switch (event) {
1871 	case NETDEV_UP:
1872 	case NETDEV_DOWN:
1873 	case NETDEV_CHANGE:
1874 		list_for_each_entry(vlan, &port->vlans, list)
1875 			netif_stacked_transfer_operstate(vlan->lowerdev,
1876 							 vlan->dev);
1877 		break;
1878 	case NETDEV_FEAT_CHANGE:
1879 		list_for_each_entry(vlan, &port->vlans, list) {
1880 			netif_inherit_tso_max(vlan->dev, dev);
1881 			vlan->dev->needed_headroom = dev->needed_headroom;
1882 			vlan->dev->needed_tailroom = dev->needed_tailroom;
1883 			netdev_update_features(vlan->dev);
1884 		}
1885 		break;
1886 	case NETDEV_CHANGEMTU:
1887 		list_for_each_entry(vlan, &port->vlans, list) {
1888 			if (vlan->dev->mtu <= dev->mtu)
1889 				continue;
1890 			dev_set_mtu(vlan->dev, dev->mtu);
1891 		}
1892 		break;
1893 	case NETDEV_CHANGEADDR:
1894 		if (!macvlan_passthru(port))
1895 			return NOTIFY_DONE;
1896 
1897 		vlan = list_first_entry_or_null(&port->vlans,
1898 						struct macvlan_dev,
1899 						list);
1900 
1901 		if (vlan && macvlan_sync_address(vlan->dev, dev->dev_addr))
1902 			return NOTIFY_BAD;
1903 
1904 		break;
1905 	case NETDEV_UNREGISTER:
1906 		/* twiddle thumbs on netns device moves */
1907 		if (dev->reg_state != NETREG_UNREGISTERING)
1908 			break;
1909 
1910 		list_for_each_entry_safe(vlan, next, &port->vlans, list)
1911 			vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1912 		unregister_netdevice_many(&list_kill);
1913 		break;
1914 	case NETDEV_PRE_TYPE_CHANGE:
1915 		/* Forbid underlying device to change its type. */
1916 		return NOTIFY_BAD;
1917 
1918 	case NETDEV_NOTIFY_PEERS:
1919 	case NETDEV_BONDING_FAILOVER:
1920 	case NETDEV_RESEND_IGMP:
1921 		/* Propagate to all vlans */
1922 		list_for_each_entry(vlan, &port->vlans, list)
1923 			call_netdevice_notifiers(event, vlan->dev);
1924 	}
1925 	return NOTIFY_DONE;
1926 }
1927 
1928 static struct notifier_block macvlan_notifier_block __read_mostly = {
1929 	.notifier_call	= macvlan_device_event,
1930 };
1931 
1932 static int __init macvlan_init_module(void)
1933 {
1934 	int err;
1935 
1936 	err = register_netdevice_notifier(&macvlan_notifier_block);
1937 	if (err)
1938 		return err;
1939 
1940 	err = macvlan_link_register(&macvlan_link_ops);
1941 	if (err < 0)
1942 		goto err1;
1943 	return 0;
1944 err1:
1945 	unregister_netdevice_notifier(&macvlan_notifier_block);
1946 	return err;
1947 }
1948 
1949 static void __exit macvlan_cleanup_module(void)
1950 {
1951 	rtnl_link_unregister(&macvlan_link_ops);
1952 	unregister_netdevice_notifier(&macvlan_notifier_block);
1953 }
1954 
1955 module_init(macvlan_init_module);
1956 module_exit(macvlan_cleanup_module);
1957 
1958 MODULE_LICENSE("GPL");
1959 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1960 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1961 MODULE_ALIAS_RTNL_LINK("macvlan");
1962