xref: /openbmc/linux/drivers/net/macvlan.c (revision 7b73a9c8)
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/rtnetlink.h>
32 #include <net/xfrm.h>
33 #include <linux/netpoll.h>
34 #include <linux/phy.h>
35 
36 #define MACVLAN_HASH_BITS	8
37 #define MACVLAN_HASH_SIZE	(1<<MACVLAN_HASH_BITS)
38 #define MACVLAN_BC_QUEUE_LEN	1000
39 
40 #define MACVLAN_F_PASSTHRU	1
41 #define MACVLAN_F_ADDRCHANGE	2
42 
43 struct macvlan_port {
44 	struct net_device	*dev;
45 	struct hlist_head	vlan_hash[MACVLAN_HASH_SIZE];
46 	struct list_head	vlans;
47 	struct sk_buff_head	bc_queue;
48 	struct work_struct	bc_work;
49 	u32			flags;
50 	int			count;
51 	struct hlist_head	vlan_source_hash[MACVLAN_HASH_SIZE];
52 	DECLARE_BITMAP(mc_filter, MACVLAN_MC_FILTER_SZ);
53 	unsigned char           perm_addr[ETH_ALEN];
54 };
55 
56 struct macvlan_source_entry {
57 	struct hlist_node	hlist;
58 	struct macvlan_dev	*vlan;
59 	unsigned char		addr[6+2] __aligned(sizeof(u16));
60 	struct rcu_head		rcu;
61 };
62 
63 struct macvlan_skb_cb {
64 	const struct macvlan_dev *src;
65 };
66 
67 #define MACVLAN_SKB_CB(__skb) ((struct macvlan_skb_cb *)&((__skb)->cb[0]))
68 
69 static void macvlan_port_destroy(struct net_device *dev);
70 
71 static inline bool macvlan_passthru(const struct macvlan_port *port)
72 {
73 	return port->flags & MACVLAN_F_PASSTHRU;
74 }
75 
76 static inline void macvlan_set_passthru(struct macvlan_port *port)
77 {
78 	port->flags |= MACVLAN_F_PASSTHRU;
79 }
80 
81 static inline bool macvlan_addr_change(const struct macvlan_port *port)
82 {
83 	return port->flags & MACVLAN_F_ADDRCHANGE;
84 }
85 
86 static inline void macvlan_set_addr_change(struct macvlan_port *port)
87 {
88 	port->flags |= MACVLAN_F_ADDRCHANGE;
89 }
90 
91 static inline void macvlan_clear_addr_change(struct macvlan_port *port)
92 {
93 	port->flags &= ~MACVLAN_F_ADDRCHANGE;
94 }
95 
96 /* Hash Ethernet address */
97 static u32 macvlan_eth_hash(const unsigned char *addr)
98 {
99 	u64 value = get_unaligned((u64 *)addr);
100 
101 	/* only want 6 bytes */
102 #ifdef __BIG_ENDIAN
103 	value >>= 16;
104 #else
105 	value <<= 16;
106 #endif
107 	return hash_64(value, MACVLAN_HASH_BITS);
108 }
109 
110 static struct macvlan_port *macvlan_port_get_rcu(const struct net_device *dev)
111 {
112 	return rcu_dereference(dev->rx_handler_data);
113 }
114 
115 static struct macvlan_port *macvlan_port_get_rtnl(const struct net_device *dev)
116 {
117 	return rtnl_dereference(dev->rx_handler_data);
118 }
119 
120 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
121 					       const unsigned char *addr)
122 {
123 	struct macvlan_dev *vlan;
124 	u32 idx = macvlan_eth_hash(addr);
125 
126 	hlist_for_each_entry_rcu(vlan, &port->vlan_hash[idx], hlist) {
127 		if (ether_addr_equal_64bits(vlan->dev->dev_addr, addr))
128 			return vlan;
129 	}
130 	return NULL;
131 }
132 
133 static struct macvlan_source_entry *macvlan_hash_lookup_source(
134 	const struct macvlan_dev *vlan,
135 	const unsigned char *addr)
136 {
137 	struct macvlan_source_entry *entry;
138 	u32 idx = macvlan_eth_hash(addr);
139 	struct hlist_head *h = &vlan->port->vlan_source_hash[idx];
140 
141 	hlist_for_each_entry_rcu(entry, h, hlist) {
142 		if (ether_addr_equal_64bits(entry->addr, addr) &&
143 		    entry->vlan == vlan)
144 			return entry;
145 	}
146 	return NULL;
147 }
148 
149 static int macvlan_hash_add_source(struct macvlan_dev *vlan,
150 				   const unsigned char *addr)
151 {
152 	struct macvlan_port *port = vlan->port;
153 	struct macvlan_source_entry *entry;
154 	struct hlist_head *h;
155 
156 	entry = macvlan_hash_lookup_source(vlan, addr);
157 	if (entry)
158 		return 0;
159 
160 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
161 	if (!entry)
162 		return -ENOMEM;
163 
164 	ether_addr_copy(entry->addr, addr);
165 	entry->vlan = vlan;
166 	h = &port->vlan_source_hash[macvlan_eth_hash(addr)];
167 	hlist_add_head_rcu(&entry->hlist, h);
168 	vlan->macaddr_count++;
169 
170 	return 0;
171 }
172 
173 static void macvlan_hash_add(struct macvlan_dev *vlan)
174 {
175 	struct macvlan_port *port = vlan->port;
176 	const unsigned char *addr = vlan->dev->dev_addr;
177 	u32 idx = macvlan_eth_hash(addr);
178 
179 	hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[idx]);
180 }
181 
182 static void macvlan_hash_del_source(struct macvlan_source_entry *entry)
183 {
184 	hlist_del_rcu(&entry->hlist);
185 	kfree_rcu(entry, rcu);
186 }
187 
188 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync)
189 {
190 	hlist_del_rcu(&vlan->hlist);
191 	if (sync)
192 		synchronize_rcu();
193 }
194 
195 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
196 					const unsigned char *addr)
197 {
198 	macvlan_hash_del(vlan, true);
199 	/* Now that we are unhashed it is safe to change the device
200 	 * address without confusing packet delivery.
201 	 */
202 	memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
203 	macvlan_hash_add(vlan);
204 }
205 
206 static bool macvlan_addr_busy(const struct macvlan_port *port,
207 			      const unsigned char *addr)
208 {
209 	/* Test to see if the specified address is
210 	 * currently in use by the underlying device or
211 	 * another macvlan.
212 	 */
213 	if (!macvlan_passthru(port) && !macvlan_addr_change(port) &&
214 	    ether_addr_equal_64bits(port->dev->dev_addr, addr))
215 		return true;
216 
217 	if (macvlan_hash_lookup(port, addr))
218 		return true;
219 
220 	return false;
221 }
222 
223 
224 static int macvlan_broadcast_one(struct sk_buff *skb,
225 				 const struct macvlan_dev *vlan,
226 				 const struct ethhdr *eth, bool local)
227 {
228 	struct net_device *dev = vlan->dev;
229 
230 	if (local)
231 		return __dev_forward_skb(dev, skb);
232 
233 	skb->dev = dev;
234 	if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
235 		skb->pkt_type = PACKET_BROADCAST;
236 	else
237 		skb->pkt_type = PACKET_MULTICAST;
238 
239 	return 0;
240 }
241 
242 static u32 macvlan_hash_mix(const struct macvlan_dev *vlan)
243 {
244 	return (u32)(((unsigned long)vlan) >> L1_CACHE_SHIFT);
245 }
246 
247 
248 static unsigned int mc_hash(const struct macvlan_dev *vlan,
249 			    const unsigned char *addr)
250 {
251 	u32 val = __get_unaligned_cpu32(addr + 2);
252 
253 	val ^= macvlan_hash_mix(vlan);
254 	return hash_32(val, MACVLAN_MC_FILTER_BITS);
255 }
256 
257 static void macvlan_broadcast(struct sk_buff *skb,
258 			      const struct macvlan_port *port,
259 			      struct net_device *src,
260 			      enum macvlan_mode mode)
261 {
262 	const struct ethhdr *eth = skb_eth_hdr(skb);
263 	const struct macvlan_dev *vlan;
264 	struct sk_buff *nskb;
265 	unsigned int i;
266 	int err;
267 	unsigned int hash;
268 
269 	if (skb->protocol == htons(ETH_P_PAUSE))
270 		return;
271 
272 	for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
273 		hlist_for_each_entry_rcu(vlan, &port->vlan_hash[i], hlist) {
274 			if (vlan->dev == src || !(vlan->mode & mode))
275 				continue;
276 
277 			hash = mc_hash(vlan, eth->h_dest);
278 			if (!test_bit(hash, vlan->mc_filter))
279 				continue;
280 
281 			err = NET_RX_DROP;
282 			nskb = skb_clone(skb, GFP_ATOMIC);
283 			if (likely(nskb))
284 				err = macvlan_broadcast_one(
285 					nskb, vlan, eth,
286 					mode == MACVLAN_MODE_BRIDGE) ?:
287 				      netif_rx_ni(nskb);
288 			macvlan_count_rx(vlan, skb->len + ETH_HLEN,
289 					 err == NET_RX_SUCCESS, true);
290 		}
291 	}
292 }
293 
294 static void macvlan_process_broadcast(struct work_struct *w)
295 {
296 	struct macvlan_port *port = container_of(w, struct macvlan_port,
297 						 bc_work);
298 	struct sk_buff *skb;
299 	struct sk_buff_head list;
300 
301 	__skb_queue_head_init(&list);
302 
303 	spin_lock_bh(&port->bc_queue.lock);
304 	skb_queue_splice_tail_init(&port->bc_queue, &list);
305 	spin_unlock_bh(&port->bc_queue.lock);
306 
307 	while ((skb = __skb_dequeue(&list))) {
308 		const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
309 
310 		rcu_read_lock();
311 
312 		if (!src)
313 			/* frame comes from an external address */
314 			macvlan_broadcast(skb, port, NULL,
315 					  MACVLAN_MODE_PRIVATE |
316 					  MACVLAN_MODE_VEPA    |
317 					  MACVLAN_MODE_PASSTHRU|
318 					  MACVLAN_MODE_BRIDGE);
319 		else if (src->mode == MACVLAN_MODE_VEPA)
320 			/* flood to everyone except source */
321 			macvlan_broadcast(skb, port, src->dev,
322 					  MACVLAN_MODE_VEPA |
323 					  MACVLAN_MODE_BRIDGE);
324 		else
325 			/*
326 			 * flood only to VEPA ports, bridge ports
327 			 * already saw the frame on the way out.
328 			 */
329 			macvlan_broadcast(skb, port, src->dev,
330 					  MACVLAN_MODE_VEPA);
331 
332 		rcu_read_unlock();
333 
334 		if (src)
335 			dev_put(src->dev);
336 		consume_skb(skb);
337 	}
338 }
339 
340 static void macvlan_broadcast_enqueue(struct macvlan_port *port,
341 				      const struct macvlan_dev *src,
342 				      struct sk_buff *skb)
343 {
344 	struct sk_buff *nskb;
345 	int err = -ENOMEM;
346 
347 	nskb = skb_clone(skb, GFP_ATOMIC);
348 	if (!nskb)
349 		goto err;
350 
351 	MACVLAN_SKB_CB(nskb)->src = src;
352 
353 	spin_lock(&port->bc_queue.lock);
354 	if (skb_queue_len(&port->bc_queue) < MACVLAN_BC_QUEUE_LEN) {
355 		if (src)
356 			dev_hold(src->dev);
357 		__skb_queue_tail(&port->bc_queue, nskb);
358 		err = 0;
359 	}
360 	spin_unlock(&port->bc_queue.lock);
361 
362 	schedule_work(&port->bc_work);
363 
364 	if (err)
365 		goto free_nskb;
366 
367 	return;
368 
369 free_nskb:
370 	kfree_skb(nskb);
371 err:
372 	atomic_long_inc(&skb->dev->rx_dropped);
373 }
374 
375 static void macvlan_flush_sources(struct macvlan_port *port,
376 				  struct macvlan_dev *vlan)
377 {
378 	int i;
379 
380 	for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
381 		struct hlist_node *h, *n;
382 
383 		hlist_for_each_safe(h, n, &port->vlan_source_hash[i]) {
384 			struct macvlan_source_entry *entry;
385 
386 			entry = hlist_entry(h, struct macvlan_source_entry,
387 					    hlist);
388 			if (entry->vlan == vlan)
389 				macvlan_hash_del_source(entry);
390 		}
391 	}
392 	vlan->macaddr_count = 0;
393 }
394 
395 static void macvlan_forward_source_one(struct sk_buff *skb,
396 				       struct macvlan_dev *vlan)
397 {
398 	struct sk_buff *nskb;
399 	struct net_device *dev;
400 	int len;
401 	int ret;
402 
403 	dev = vlan->dev;
404 	if (unlikely(!(dev->flags & IFF_UP)))
405 		return;
406 
407 	nskb = skb_clone(skb, GFP_ATOMIC);
408 	if (!nskb)
409 		return;
410 
411 	len = nskb->len + ETH_HLEN;
412 	nskb->dev = dev;
413 
414 	if (ether_addr_equal_64bits(eth_hdr(skb)->h_dest, dev->dev_addr))
415 		nskb->pkt_type = PACKET_HOST;
416 
417 	ret = netif_rx(nskb);
418 	macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
419 }
420 
421 static void macvlan_forward_source(struct sk_buff *skb,
422 				   struct macvlan_port *port,
423 				   const unsigned char *addr)
424 {
425 	struct macvlan_source_entry *entry;
426 	u32 idx = macvlan_eth_hash(addr);
427 	struct hlist_head *h = &port->vlan_source_hash[idx];
428 
429 	hlist_for_each_entry_rcu(entry, h, hlist) {
430 		if (ether_addr_equal_64bits(entry->addr, addr))
431 			macvlan_forward_source_one(skb, entry->vlan);
432 	}
433 }
434 
435 /* called under rcu_read_lock() from netif_receive_skb */
436 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
437 {
438 	struct macvlan_port *port;
439 	struct sk_buff *skb = *pskb;
440 	const struct ethhdr *eth = eth_hdr(skb);
441 	const struct macvlan_dev *vlan;
442 	const struct macvlan_dev *src;
443 	struct net_device *dev;
444 	unsigned int len = 0;
445 	int ret;
446 	rx_handler_result_t handle_res;
447 
448 	port = macvlan_port_get_rcu(skb->dev);
449 	if (is_multicast_ether_addr(eth->h_dest)) {
450 		unsigned int hash;
451 
452 		skb = ip_check_defrag(dev_net(skb->dev), skb, IP_DEFRAG_MACVLAN);
453 		if (!skb)
454 			return RX_HANDLER_CONSUMED;
455 		*pskb = skb;
456 		eth = eth_hdr(skb);
457 		macvlan_forward_source(skb, port, eth->h_source);
458 		src = macvlan_hash_lookup(port, eth->h_source);
459 		if (src && src->mode != MACVLAN_MODE_VEPA &&
460 		    src->mode != MACVLAN_MODE_BRIDGE) {
461 			/* forward to original port. */
462 			vlan = src;
463 			ret = macvlan_broadcast_one(skb, vlan, eth, 0) ?:
464 			      netif_rx(skb);
465 			handle_res = RX_HANDLER_CONSUMED;
466 			goto out;
467 		}
468 
469 		hash = mc_hash(NULL, eth->h_dest);
470 		if (test_bit(hash, port->mc_filter))
471 			macvlan_broadcast_enqueue(port, src, skb);
472 
473 		return RX_HANDLER_PASS;
474 	}
475 
476 	macvlan_forward_source(skb, port, eth->h_source);
477 	if (macvlan_passthru(port))
478 		vlan = list_first_or_null_rcu(&port->vlans,
479 					      struct macvlan_dev, list);
480 	else
481 		vlan = macvlan_hash_lookup(port, eth->h_dest);
482 	if (!vlan || vlan->mode == MACVLAN_MODE_SOURCE)
483 		return RX_HANDLER_PASS;
484 
485 	dev = vlan->dev;
486 	if (unlikely(!(dev->flags & IFF_UP))) {
487 		kfree_skb(skb);
488 		return RX_HANDLER_CONSUMED;
489 	}
490 	len = skb->len + ETH_HLEN;
491 	skb = skb_share_check(skb, GFP_ATOMIC);
492 	if (!skb) {
493 		ret = NET_RX_DROP;
494 		handle_res = RX_HANDLER_CONSUMED;
495 		goto out;
496 	}
497 
498 	*pskb = skb;
499 	skb->dev = dev;
500 	skb->pkt_type = PACKET_HOST;
501 
502 	ret = NET_RX_SUCCESS;
503 	handle_res = RX_HANDLER_ANOTHER;
504 out:
505 	macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
506 	return handle_res;
507 }
508 
509 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
510 {
511 	const struct macvlan_dev *vlan = netdev_priv(dev);
512 	const struct macvlan_port *port = vlan->port;
513 	const struct macvlan_dev *dest;
514 
515 	if (vlan->mode == MACVLAN_MODE_BRIDGE) {
516 		const struct ethhdr *eth = (void *)skb->data;
517 
518 		/* send to other bridge ports directly */
519 		if (is_multicast_ether_addr(eth->h_dest)) {
520 			macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
521 			goto xmit_world;
522 		}
523 
524 		dest = macvlan_hash_lookup(port, eth->h_dest);
525 		if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
526 			/* send to lowerdev first for its network taps */
527 			dev_forward_skb(vlan->lowerdev, skb);
528 
529 			return NET_XMIT_SUCCESS;
530 		}
531 	}
532 xmit_world:
533 	skb->dev = vlan->lowerdev;
534 	return dev_queue_xmit_accel(skb,
535 				    netdev_get_sb_channel(dev) ? dev : NULL);
536 }
537 
538 static inline netdev_tx_t macvlan_netpoll_send_skb(struct macvlan_dev *vlan, struct sk_buff *skb)
539 {
540 #ifdef CONFIG_NET_POLL_CONTROLLER
541 	if (vlan->netpoll)
542 		netpoll_send_skb(vlan->netpoll, skb);
543 #else
544 	BUG();
545 #endif
546 	return NETDEV_TX_OK;
547 }
548 
549 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
550 				      struct net_device *dev)
551 {
552 	struct macvlan_dev *vlan = netdev_priv(dev);
553 	unsigned int len = skb->len;
554 	int ret;
555 
556 	if (unlikely(netpoll_tx_running(dev)))
557 		return macvlan_netpoll_send_skb(vlan, skb);
558 
559 	ret = macvlan_queue_xmit(skb, dev);
560 
561 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
562 		struct vlan_pcpu_stats *pcpu_stats;
563 
564 		pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
565 		u64_stats_update_begin(&pcpu_stats->syncp);
566 		pcpu_stats->tx_packets++;
567 		pcpu_stats->tx_bytes += len;
568 		u64_stats_update_end(&pcpu_stats->syncp);
569 	} else {
570 		this_cpu_inc(vlan->pcpu_stats->tx_dropped);
571 	}
572 	return ret;
573 }
574 
575 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
576 			       unsigned short type, const void *daddr,
577 			       const void *saddr, unsigned len)
578 {
579 	const struct macvlan_dev *vlan = netdev_priv(dev);
580 	struct net_device *lowerdev = vlan->lowerdev;
581 
582 	return dev_hard_header(skb, lowerdev, type, daddr,
583 			       saddr ? : dev->dev_addr, len);
584 }
585 
586 static const struct header_ops macvlan_hard_header_ops = {
587 	.create  	= macvlan_hard_header,
588 	.parse		= eth_header_parse,
589 	.cache		= eth_header_cache,
590 	.cache_update	= eth_header_cache_update,
591 };
592 
593 static int macvlan_open(struct net_device *dev)
594 {
595 	struct macvlan_dev *vlan = netdev_priv(dev);
596 	struct net_device *lowerdev = vlan->lowerdev;
597 	int err;
598 
599 	if (macvlan_passthru(vlan->port)) {
600 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
601 			err = dev_set_promiscuity(lowerdev, 1);
602 			if (err < 0)
603 				goto out;
604 		}
605 		goto hash_add;
606 	}
607 
608 	err = -EADDRINUSE;
609 	if (macvlan_addr_busy(vlan->port, dev->dev_addr))
610 		goto out;
611 
612 	/* Attempt to populate accel_priv which is used to offload the L2
613 	 * forwarding requests for unicast packets.
614 	 */
615 	if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD)
616 		vlan->accel_priv =
617 		      lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
618 
619 	/* If earlier attempt to offload failed, or accel_priv is not
620 	 * populated we must add the unicast address to the lower device.
621 	 */
622 	if (IS_ERR_OR_NULL(vlan->accel_priv)) {
623 		vlan->accel_priv = NULL;
624 		err = dev_uc_add(lowerdev, dev->dev_addr);
625 		if (err < 0)
626 			goto out;
627 	}
628 
629 	if (dev->flags & IFF_ALLMULTI) {
630 		err = dev_set_allmulti(lowerdev, 1);
631 		if (err < 0)
632 			goto del_unicast;
633 	}
634 
635 	if (dev->flags & IFF_PROMISC) {
636 		err = dev_set_promiscuity(lowerdev, 1);
637 		if (err < 0)
638 			goto clear_multi;
639 	}
640 
641 hash_add:
642 	macvlan_hash_add(vlan);
643 	return 0;
644 
645 clear_multi:
646 	if (dev->flags & IFF_ALLMULTI)
647 		dev_set_allmulti(lowerdev, -1);
648 del_unicast:
649 	if (vlan->accel_priv) {
650 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
651 							   vlan->accel_priv);
652 		vlan->accel_priv = NULL;
653 	} else {
654 		dev_uc_del(lowerdev, dev->dev_addr);
655 	}
656 out:
657 	return err;
658 }
659 
660 static int macvlan_stop(struct net_device *dev)
661 {
662 	struct macvlan_dev *vlan = netdev_priv(dev);
663 	struct net_device *lowerdev = vlan->lowerdev;
664 
665 	if (vlan->accel_priv) {
666 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
667 							   vlan->accel_priv);
668 		vlan->accel_priv = NULL;
669 	}
670 
671 	dev_uc_unsync(lowerdev, dev);
672 	dev_mc_unsync(lowerdev, dev);
673 
674 	if (macvlan_passthru(vlan->port)) {
675 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
676 			dev_set_promiscuity(lowerdev, -1);
677 		goto hash_del;
678 	}
679 
680 	if (dev->flags & IFF_ALLMULTI)
681 		dev_set_allmulti(lowerdev, -1);
682 
683 	if (dev->flags & IFF_PROMISC)
684 		dev_set_promiscuity(lowerdev, -1);
685 
686 	dev_uc_del(lowerdev, dev->dev_addr);
687 
688 hash_del:
689 	macvlan_hash_del(vlan, !dev->dismantle);
690 	return 0;
691 }
692 
693 static int macvlan_sync_address(struct net_device *dev, unsigned char *addr)
694 {
695 	struct macvlan_dev *vlan = netdev_priv(dev);
696 	struct net_device *lowerdev = vlan->lowerdev;
697 	struct macvlan_port *port = vlan->port;
698 	int err;
699 
700 	if (!(dev->flags & IFF_UP)) {
701 		/* Just copy in the new address */
702 		ether_addr_copy(dev->dev_addr, addr);
703 	} else {
704 		/* Rehash and update the device filters */
705 		if (macvlan_addr_busy(vlan->port, addr))
706 			return -EADDRINUSE;
707 
708 		if (!macvlan_passthru(port)) {
709 			err = dev_uc_add(lowerdev, addr);
710 			if (err)
711 				return err;
712 
713 			dev_uc_del(lowerdev, dev->dev_addr);
714 		}
715 
716 		macvlan_hash_change_addr(vlan, addr);
717 	}
718 	if (macvlan_passthru(port) && !macvlan_addr_change(port)) {
719 		/* Since addr_change isn't set, we are here due to lower
720 		 * device change.  Save the lower-dev address so we can
721 		 * restore it later.
722 		 */
723 		ether_addr_copy(vlan->port->perm_addr,
724 				lowerdev->dev_addr);
725 	}
726 	macvlan_clear_addr_change(port);
727 	return 0;
728 }
729 
730 static int macvlan_set_mac_address(struct net_device *dev, void *p)
731 {
732 	struct macvlan_dev *vlan = netdev_priv(dev);
733 	struct sockaddr *addr = p;
734 
735 	if (!is_valid_ether_addr(addr->sa_data))
736 		return -EADDRNOTAVAIL;
737 
738 	/* If the addresses are the same, this is a no-op */
739 	if (ether_addr_equal(dev->dev_addr, addr->sa_data))
740 		return 0;
741 
742 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
743 		macvlan_set_addr_change(vlan->port);
744 		return dev_set_mac_address(vlan->lowerdev, addr, NULL);
745 	}
746 
747 	if (macvlan_addr_busy(vlan->port, addr->sa_data))
748 		return -EADDRINUSE;
749 
750 	return macvlan_sync_address(dev, addr->sa_data);
751 }
752 
753 static void macvlan_change_rx_flags(struct net_device *dev, int change)
754 {
755 	struct macvlan_dev *vlan = netdev_priv(dev);
756 	struct net_device *lowerdev = vlan->lowerdev;
757 
758 	if (dev->flags & IFF_UP) {
759 		if (change & IFF_ALLMULTI)
760 			dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
761 		if (change & IFF_PROMISC)
762 			dev_set_promiscuity(lowerdev,
763 					    dev->flags & IFF_PROMISC ? 1 : -1);
764 
765 	}
766 }
767 
768 static void macvlan_compute_filter(unsigned long *mc_filter,
769 				   struct net_device *dev,
770 				   struct macvlan_dev *vlan)
771 {
772 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
773 		bitmap_fill(mc_filter, MACVLAN_MC_FILTER_SZ);
774 	} else {
775 		struct netdev_hw_addr *ha;
776 		DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
777 
778 		bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
779 		netdev_for_each_mc_addr(ha, dev) {
780 			__set_bit(mc_hash(vlan, ha->addr), filter);
781 		}
782 
783 		__set_bit(mc_hash(vlan, dev->broadcast), filter);
784 
785 		bitmap_copy(mc_filter, filter, MACVLAN_MC_FILTER_SZ);
786 	}
787 }
788 
789 static void macvlan_set_mac_lists(struct net_device *dev)
790 {
791 	struct macvlan_dev *vlan = netdev_priv(dev);
792 
793 	macvlan_compute_filter(vlan->mc_filter, dev, vlan);
794 
795 	dev_uc_sync(vlan->lowerdev, dev);
796 	dev_mc_sync(vlan->lowerdev, dev);
797 
798 	/* This is slightly inaccurate as we're including the subscription
799 	 * list of vlan->lowerdev too.
800 	 *
801 	 * Bug alert: This only works if everyone has the same broadcast
802 	 * address as lowerdev.  As soon as someone changes theirs this
803 	 * will break.
804 	 *
805 	 * However, this is already broken as when you change your broadcast
806 	 * address we don't get called.
807 	 *
808 	 * The solution is to maintain a list of broadcast addresses like
809 	 * we do for uc/mc, if you care.
810 	 */
811 	macvlan_compute_filter(vlan->port->mc_filter, vlan->lowerdev, NULL);
812 }
813 
814 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
815 {
816 	struct macvlan_dev *vlan = netdev_priv(dev);
817 
818 	if (vlan->lowerdev->mtu < new_mtu)
819 		return -EINVAL;
820 	dev->mtu = new_mtu;
821 	return 0;
822 }
823 
824 static int macvlan_do_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
825 {
826 	struct net_device *real_dev = macvlan_dev_real_dev(dev);
827 	const struct net_device_ops *ops = real_dev->netdev_ops;
828 	struct ifreq ifrr;
829 	int err = -EOPNOTSUPP;
830 
831 	strscpy(ifrr.ifr_name, real_dev->name, IFNAMSIZ);
832 	ifrr.ifr_ifru = ifr->ifr_ifru;
833 
834 	switch (cmd) {
835 	case SIOCSHWTSTAMP:
836 		if (!net_eq(dev_net(dev), &init_net))
837 			break;
838 		/* fall through */
839 	case SIOCGHWTSTAMP:
840 		if (netif_device_present(real_dev) && ops->ndo_do_ioctl)
841 			err = ops->ndo_do_ioctl(real_dev, &ifrr, cmd);
842 		break;
843 	}
844 
845 	if (!err)
846 		ifr->ifr_ifru = ifrr.ifr_ifru;
847 
848 	return err;
849 }
850 
851 /*
852  * macvlan network devices have devices nesting below it and are a special
853  * "super class" of normal network devices; split their locks off into a
854  * separate class since they always nest.
855  */
856 #define ALWAYS_ON_OFFLOADS \
857 	(NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE | \
858 	 NETIF_F_GSO_ROBUST | NETIF_F_GSO_ENCAP_ALL)
859 
860 #define ALWAYS_ON_FEATURES (ALWAYS_ON_OFFLOADS | NETIF_F_LLTX)
861 
862 #define MACVLAN_FEATURES \
863 	(NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
864 	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_LRO | \
865 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
866 	 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
867 
868 #define MACVLAN_STATE_MASK \
869 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
870 
871 static int macvlan_init(struct net_device *dev)
872 {
873 	struct macvlan_dev *vlan = netdev_priv(dev);
874 	const struct net_device *lowerdev = vlan->lowerdev;
875 	struct macvlan_port *port = vlan->port;
876 
877 	dev->state		= (dev->state & ~MACVLAN_STATE_MASK) |
878 				  (lowerdev->state & MACVLAN_STATE_MASK);
879 	dev->features 		= lowerdev->features & MACVLAN_FEATURES;
880 	dev->features		|= ALWAYS_ON_FEATURES;
881 	dev->hw_features	|= NETIF_F_LRO;
882 	dev->vlan_features	= lowerdev->vlan_features & MACVLAN_FEATURES;
883 	dev->vlan_features	|= ALWAYS_ON_OFFLOADS;
884 	dev->hw_enc_features    |= dev->features;
885 	dev->gso_max_size	= lowerdev->gso_max_size;
886 	dev->gso_max_segs	= lowerdev->gso_max_segs;
887 	dev->hard_header_len	= lowerdev->hard_header_len;
888 
889 	vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
890 	if (!vlan->pcpu_stats)
891 		return -ENOMEM;
892 
893 	port->count += 1;
894 
895 	return 0;
896 }
897 
898 static void macvlan_uninit(struct net_device *dev)
899 {
900 	struct macvlan_dev *vlan = netdev_priv(dev);
901 	struct macvlan_port *port = vlan->port;
902 
903 	free_percpu(vlan->pcpu_stats);
904 
905 	macvlan_flush_sources(port, vlan);
906 	port->count -= 1;
907 	if (!port->count)
908 		macvlan_port_destroy(port->dev);
909 }
910 
911 static void macvlan_dev_get_stats64(struct net_device *dev,
912 				    struct rtnl_link_stats64 *stats)
913 {
914 	struct macvlan_dev *vlan = netdev_priv(dev);
915 
916 	if (vlan->pcpu_stats) {
917 		struct vlan_pcpu_stats *p;
918 		u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
919 		u32 rx_errors = 0, tx_dropped = 0;
920 		unsigned int start;
921 		int i;
922 
923 		for_each_possible_cpu(i) {
924 			p = per_cpu_ptr(vlan->pcpu_stats, i);
925 			do {
926 				start = u64_stats_fetch_begin_irq(&p->syncp);
927 				rx_packets	= p->rx_packets;
928 				rx_bytes	= p->rx_bytes;
929 				rx_multicast	= p->rx_multicast;
930 				tx_packets	= p->tx_packets;
931 				tx_bytes	= p->tx_bytes;
932 			} while (u64_stats_fetch_retry_irq(&p->syncp, start));
933 
934 			stats->rx_packets	+= rx_packets;
935 			stats->rx_bytes		+= rx_bytes;
936 			stats->multicast	+= rx_multicast;
937 			stats->tx_packets	+= tx_packets;
938 			stats->tx_bytes		+= tx_bytes;
939 			/* rx_errors & tx_dropped are u32, updated
940 			 * without syncp protection.
941 			 */
942 			rx_errors	+= p->rx_errors;
943 			tx_dropped	+= p->tx_dropped;
944 		}
945 		stats->rx_errors	= rx_errors;
946 		stats->rx_dropped	= rx_errors;
947 		stats->tx_dropped	= tx_dropped;
948 	}
949 }
950 
951 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
952 				   __be16 proto, u16 vid)
953 {
954 	struct macvlan_dev *vlan = netdev_priv(dev);
955 	struct net_device *lowerdev = vlan->lowerdev;
956 
957 	return vlan_vid_add(lowerdev, proto, vid);
958 }
959 
960 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
961 				    __be16 proto, u16 vid)
962 {
963 	struct macvlan_dev *vlan = netdev_priv(dev);
964 	struct net_device *lowerdev = vlan->lowerdev;
965 
966 	vlan_vid_del(lowerdev, proto, vid);
967 	return 0;
968 }
969 
970 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
971 			   struct net_device *dev,
972 			   const unsigned char *addr, u16 vid,
973 			   u16 flags,
974 			   struct netlink_ext_ack *extack)
975 {
976 	struct macvlan_dev *vlan = netdev_priv(dev);
977 	int err = -EINVAL;
978 
979 	/* Support unicast filter only on passthru devices.
980 	 * Multicast filter should be allowed on all devices.
981 	 */
982 	if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr))
983 		return -EOPNOTSUPP;
984 
985 	if (flags & NLM_F_REPLACE)
986 		return -EOPNOTSUPP;
987 
988 	if (is_unicast_ether_addr(addr))
989 		err = dev_uc_add_excl(dev, addr);
990 	else if (is_multicast_ether_addr(addr))
991 		err = dev_mc_add_excl(dev, addr);
992 
993 	return err;
994 }
995 
996 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
997 			   struct net_device *dev,
998 			   const unsigned char *addr, u16 vid)
999 {
1000 	struct macvlan_dev *vlan = netdev_priv(dev);
1001 	int err = -EINVAL;
1002 
1003 	/* Support unicast filter only on passthru devices.
1004 	 * Multicast filter should be allowed on all devices.
1005 	 */
1006 	if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr))
1007 		return -EOPNOTSUPP;
1008 
1009 	if (is_unicast_ether_addr(addr))
1010 		err = dev_uc_del(dev, addr);
1011 	else if (is_multicast_ether_addr(addr))
1012 		err = dev_mc_del(dev, addr);
1013 
1014 	return err;
1015 }
1016 
1017 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
1018 					struct ethtool_drvinfo *drvinfo)
1019 {
1020 	strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
1021 	strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
1022 }
1023 
1024 static int macvlan_ethtool_get_link_ksettings(struct net_device *dev,
1025 					      struct ethtool_link_ksettings *cmd)
1026 {
1027 	const struct macvlan_dev *vlan = netdev_priv(dev);
1028 
1029 	return __ethtool_get_link_ksettings(vlan->lowerdev, cmd);
1030 }
1031 
1032 static int macvlan_ethtool_get_ts_info(struct net_device *dev,
1033 				       struct ethtool_ts_info *info)
1034 {
1035 	struct net_device *real_dev = macvlan_dev_real_dev(dev);
1036 	const struct ethtool_ops *ops = real_dev->ethtool_ops;
1037 	struct phy_device *phydev = real_dev->phydev;
1038 
1039 	if (phydev && phydev->drv && phydev->drv->ts_info) {
1040 		 return phydev->drv->ts_info(phydev, info);
1041 	} else if (ops->get_ts_info) {
1042 		return ops->get_ts_info(real_dev, info);
1043 	} else {
1044 		info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE |
1045 			SOF_TIMESTAMPING_SOFTWARE;
1046 		info->phc_index = -1;
1047 	}
1048 
1049 	return 0;
1050 }
1051 
1052 static netdev_features_t macvlan_fix_features(struct net_device *dev,
1053 					      netdev_features_t features)
1054 {
1055 	struct macvlan_dev *vlan = netdev_priv(dev);
1056 	netdev_features_t lowerdev_features = vlan->lowerdev->features;
1057 	netdev_features_t mask;
1058 
1059 	features |= NETIF_F_ALL_FOR_ALL;
1060 	features &= (vlan->set_features | ~MACVLAN_FEATURES);
1061 	mask = features;
1062 
1063 	lowerdev_features &= (features | ~NETIF_F_LRO);
1064 	features = netdev_increment_features(lowerdev_features, features, mask);
1065 	features |= ALWAYS_ON_FEATURES;
1066 	features &= (ALWAYS_ON_FEATURES | MACVLAN_FEATURES);
1067 
1068 	return features;
1069 }
1070 
1071 #ifdef CONFIG_NET_POLL_CONTROLLER
1072 static void macvlan_dev_poll_controller(struct net_device *dev)
1073 {
1074 	return;
1075 }
1076 
1077 static int macvlan_dev_netpoll_setup(struct net_device *dev, struct netpoll_info *npinfo)
1078 {
1079 	struct macvlan_dev *vlan = netdev_priv(dev);
1080 	struct net_device *real_dev = vlan->lowerdev;
1081 	struct netpoll *netpoll;
1082 	int err = 0;
1083 
1084 	netpoll = kzalloc(sizeof(*netpoll), GFP_KERNEL);
1085 	err = -ENOMEM;
1086 	if (!netpoll)
1087 		goto out;
1088 
1089 	err = __netpoll_setup(netpoll, real_dev);
1090 	if (err) {
1091 		kfree(netpoll);
1092 		goto out;
1093 	}
1094 
1095 	vlan->netpoll = netpoll;
1096 
1097 out:
1098 	return err;
1099 }
1100 
1101 static void macvlan_dev_netpoll_cleanup(struct net_device *dev)
1102 {
1103 	struct macvlan_dev *vlan = netdev_priv(dev);
1104 	struct netpoll *netpoll = vlan->netpoll;
1105 
1106 	if (!netpoll)
1107 		return;
1108 
1109 	vlan->netpoll = NULL;
1110 
1111 	__netpoll_free(netpoll);
1112 }
1113 #endif	/* CONFIG_NET_POLL_CONTROLLER */
1114 
1115 static int macvlan_dev_get_iflink(const struct net_device *dev)
1116 {
1117 	struct macvlan_dev *vlan = netdev_priv(dev);
1118 
1119 	return vlan->lowerdev->ifindex;
1120 }
1121 
1122 static const struct ethtool_ops macvlan_ethtool_ops = {
1123 	.get_link		= ethtool_op_get_link,
1124 	.get_link_ksettings	= macvlan_ethtool_get_link_ksettings,
1125 	.get_drvinfo		= macvlan_ethtool_get_drvinfo,
1126 	.get_ts_info		= macvlan_ethtool_get_ts_info,
1127 };
1128 
1129 static const struct net_device_ops macvlan_netdev_ops = {
1130 	.ndo_init		= macvlan_init,
1131 	.ndo_uninit		= macvlan_uninit,
1132 	.ndo_open		= macvlan_open,
1133 	.ndo_stop		= macvlan_stop,
1134 	.ndo_start_xmit		= macvlan_start_xmit,
1135 	.ndo_change_mtu		= macvlan_change_mtu,
1136 	.ndo_do_ioctl		= macvlan_do_ioctl,
1137 	.ndo_fix_features	= macvlan_fix_features,
1138 	.ndo_change_rx_flags	= macvlan_change_rx_flags,
1139 	.ndo_set_mac_address	= macvlan_set_mac_address,
1140 	.ndo_set_rx_mode	= macvlan_set_mac_lists,
1141 	.ndo_get_stats64	= macvlan_dev_get_stats64,
1142 	.ndo_validate_addr	= eth_validate_addr,
1143 	.ndo_vlan_rx_add_vid	= macvlan_vlan_rx_add_vid,
1144 	.ndo_vlan_rx_kill_vid	= macvlan_vlan_rx_kill_vid,
1145 	.ndo_fdb_add		= macvlan_fdb_add,
1146 	.ndo_fdb_del		= macvlan_fdb_del,
1147 	.ndo_fdb_dump		= ndo_dflt_fdb_dump,
1148 #ifdef CONFIG_NET_POLL_CONTROLLER
1149 	.ndo_poll_controller	= macvlan_dev_poll_controller,
1150 	.ndo_netpoll_setup	= macvlan_dev_netpoll_setup,
1151 	.ndo_netpoll_cleanup	= macvlan_dev_netpoll_cleanup,
1152 #endif
1153 	.ndo_get_iflink		= macvlan_dev_get_iflink,
1154 	.ndo_features_check	= passthru_features_check,
1155 	.ndo_change_proto_down  = dev_change_proto_down_generic,
1156 };
1157 
1158 void macvlan_common_setup(struct net_device *dev)
1159 {
1160 	ether_setup(dev);
1161 
1162 	dev->min_mtu		= 0;
1163 	dev->max_mtu		= ETH_MAX_MTU;
1164 	dev->priv_flags	       &= ~IFF_TX_SKB_SHARING;
1165 	netif_keep_dst(dev);
1166 	dev->priv_flags	       |= IFF_UNICAST_FLT;
1167 	dev->netdev_ops		= &macvlan_netdev_ops;
1168 	dev->needs_free_netdev	= true;
1169 	dev->header_ops		= &macvlan_hard_header_ops;
1170 	dev->ethtool_ops	= &macvlan_ethtool_ops;
1171 }
1172 EXPORT_SYMBOL_GPL(macvlan_common_setup);
1173 
1174 static void macvlan_setup(struct net_device *dev)
1175 {
1176 	macvlan_common_setup(dev);
1177 	dev->priv_flags |= IFF_NO_QUEUE;
1178 }
1179 
1180 static int macvlan_port_create(struct net_device *dev)
1181 {
1182 	struct macvlan_port *port;
1183 	unsigned int i;
1184 	int err;
1185 
1186 	if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
1187 		return -EINVAL;
1188 
1189 	if (netdev_is_rx_handler_busy(dev))
1190 		return -EBUSY;
1191 
1192 	port = kzalloc(sizeof(*port), GFP_KERNEL);
1193 	if (port == NULL)
1194 		return -ENOMEM;
1195 
1196 	port->dev = dev;
1197 	ether_addr_copy(port->perm_addr, dev->dev_addr);
1198 	INIT_LIST_HEAD(&port->vlans);
1199 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1200 		INIT_HLIST_HEAD(&port->vlan_hash[i]);
1201 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1202 		INIT_HLIST_HEAD(&port->vlan_source_hash[i]);
1203 
1204 	skb_queue_head_init(&port->bc_queue);
1205 	INIT_WORK(&port->bc_work, macvlan_process_broadcast);
1206 
1207 	err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
1208 	if (err)
1209 		kfree(port);
1210 	else
1211 		dev->priv_flags |= IFF_MACVLAN_PORT;
1212 	return err;
1213 }
1214 
1215 static void macvlan_port_destroy(struct net_device *dev)
1216 {
1217 	struct macvlan_port *port = macvlan_port_get_rtnl(dev);
1218 	struct sk_buff *skb;
1219 
1220 	dev->priv_flags &= ~IFF_MACVLAN_PORT;
1221 	netdev_rx_handler_unregister(dev);
1222 
1223 	/* After this point, no packet can schedule bc_work anymore,
1224 	 * but we need to cancel it and purge left skbs if any.
1225 	 */
1226 	cancel_work_sync(&port->bc_work);
1227 
1228 	while ((skb = __skb_dequeue(&port->bc_queue))) {
1229 		const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
1230 
1231 		if (src)
1232 			dev_put(src->dev);
1233 
1234 		kfree_skb(skb);
1235 	}
1236 
1237 	/* If the lower device address has been changed by passthru
1238 	 * macvlan, put it back.
1239 	 */
1240 	if (macvlan_passthru(port) &&
1241 	    !ether_addr_equal(port->dev->dev_addr, port->perm_addr)) {
1242 		struct sockaddr sa;
1243 
1244 		sa.sa_family = port->dev->type;
1245 		memcpy(&sa.sa_data, port->perm_addr, port->dev->addr_len);
1246 		dev_set_mac_address(port->dev, &sa, NULL);
1247 	}
1248 
1249 	kfree(port);
1250 }
1251 
1252 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[],
1253 			    struct netlink_ext_ack *extack)
1254 {
1255 	if (tb[IFLA_ADDRESS]) {
1256 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1257 			return -EINVAL;
1258 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1259 			return -EADDRNOTAVAIL;
1260 	}
1261 
1262 	if (!data)
1263 		return 0;
1264 
1265 	if (data[IFLA_MACVLAN_FLAGS] &&
1266 	    nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
1267 		return -EINVAL;
1268 
1269 	if (data[IFLA_MACVLAN_MODE]) {
1270 		switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
1271 		case MACVLAN_MODE_PRIVATE:
1272 		case MACVLAN_MODE_VEPA:
1273 		case MACVLAN_MODE_BRIDGE:
1274 		case MACVLAN_MODE_PASSTHRU:
1275 		case MACVLAN_MODE_SOURCE:
1276 			break;
1277 		default:
1278 			return -EINVAL;
1279 		}
1280 	}
1281 
1282 	if (data[IFLA_MACVLAN_MACADDR_MODE]) {
1283 		switch (nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE])) {
1284 		case MACVLAN_MACADDR_ADD:
1285 		case MACVLAN_MACADDR_DEL:
1286 		case MACVLAN_MACADDR_FLUSH:
1287 		case MACVLAN_MACADDR_SET:
1288 			break;
1289 		default:
1290 			return -EINVAL;
1291 		}
1292 	}
1293 
1294 	if (data[IFLA_MACVLAN_MACADDR]) {
1295 		if (nla_len(data[IFLA_MACVLAN_MACADDR]) != ETH_ALEN)
1296 			return -EINVAL;
1297 
1298 		if (!is_valid_ether_addr(nla_data(data[IFLA_MACVLAN_MACADDR])))
1299 			return -EADDRNOTAVAIL;
1300 	}
1301 
1302 	if (data[IFLA_MACVLAN_MACADDR_COUNT])
1303 		return -EINVAL;
1304 
1305 	return 0;
1306 }
1307 
1308 /**
1309  * reconfigure list of remote source mac address
1310  * (only for macvlan devices in source mode)
1311  * Note regarding alignment: all netlink data is aligned to 4 Byte, which
1312  * suffices for both ether_addr_copy and ether_addr_equal_64bits usage.
1313  */
1314 static int macvlan_changelink_sources(struct macvlan_dev *vlan, u32 mode,
1315 				      struct nlattr *data[])
1316 {
1317 	char *addr = NULL;
1318 	int ret, rem, len;
1319 	struct nlattr *nla, *head;
1320 	struct macvlan_source_entry *entry;
1321 
1322 	if (data[IFLA_MACVLAN_MACADDR])
1323 		addr = nla_data(data[IFLA_MACVLAN_MACADDR]);
1324 
1325 	if (mode == MACVLAN_MACADDR_ADD) {
1326 		if (!addr)
1327 			return -EINVAL;
1328 
1329 		return macvlan_hash_add_source(vlan, addr);
1330 
1331 	} else if (mode == MACVLAN_MACADDR_DEL) {
1332 		if (!addr)
1333 			return -EINVAL;
1334 
1335 		entry = macvlan_hash_lookup_source(vlan, addr);
1336 		if (entry) {
1337 			macvlan_hash_del_source(entry);
1338 			vlan->macaddr_count--;
1339 		}
1340 	} else if (mode == MACVLAN_MACADDR_FLUSH) {
1341 		macvlan_flush_sources(vlan->port, vlan);
1342 	} else if (mode == MACVLAN_MACADDR_SET) {
1343 		macvlan_flush_sources(vlan->port, vlan);
1344 
1345 		if (addr) {
1346 			ret = macvlan_hash_add_source(vlan, addr);
1347 			if (ret)
1348 				return ret;
1349 		}
1350 
1351 		if (!data || !data[IFLA_MACVLAN_MACADDR_DATA])
1352 			return 0;
1353 
1354 		head = nla_data(data[IFLA_MACVLAN_MACADDR_DATA]);
1355 		len = nla_len(data[IFLA_MACVLAN_MACADDR_DATA]);
1356 
1357 		nla_for_each_attr(nla, head, len, rem) {
1358 			if (nla_type(nla) != IFLA_MACVLAN_MACADDR ||
1359 			    nla_len(nla) != ETH_ALEN)
1360 				continue;
1361 
1362 			addr = nla_data(nla);
1363 			ret = macvlan_hash_add_source(vlan, addr);
1364 			if (ret)
1365 				return ret;
1366 		}
1367 	} else {
1368 		return -EINVAL;
1369 	}
1370 
1371 	return 0;
1372 }
1373 
1374 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
1375 			   struct nlattr *tb[], struct nlattr *data[],
1376 			   struct netlink_ext_ack *extack)
1377 {
1378 	struct macvlan_dev *vlan = netdev_priv(dev);
1379 	struct macvlan_port *port;
1380 	struct net_device *lowerdev;
1381 	int err;
1382 	int macmode;
1383 	bool create = false;
1384 
1385 	if (!tb[IFLA_LINK])
1386 		return -EINVAL;
1387 
1388 	lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
1389 	if (lowerdev == NULL)
1390 		return -ENODEV;
1391 
1392 	/* When creating macvlans or macvtaps on top of other macvlans - use
1393 	 * the real device as the lowerdev.
1394 	 */
1395 	if (netif_is_macvlan(lowerdev))
1396 		lowerdev = macvlan_dev_real_dev(lowerdev);
1397 
1398 	if (!tb[IFLA_MTU])
1399 		dev->mtu = lowerdev->mtu;
1400 	else if (dev->mtu > lowerdev->mtu)
1401 		return -EINVAL;
1402 
1403 	/* MTU range: 68 - lowerdev->max_mtu */
1404 	dev->min_mtu = ETH_MIN_MTU;
1405 	dev->max_mtu = lowerdev->max_mtu;
1406 
1407 	if (!tb[IFLA_ADDRESS])
1408 		eth_hw_addr_random(dev);
1409 
1410 	if (!netif_is_macvlan_port(lowerdev)) {
1411 		err = macvlan_port_create(lowerdev);
1412 		if (err < 0)
1413 			return err;
1414 		create = true;
1415 	}
1416 	port = macvlan_port_get_rtnl(lowerdev);
1417 
1418 	/* Only 1 macvlan device can be created in passthru mode */
1419 	if (macvlan_passthru(port)) {
1420 		/* The macvlan port must be not created this time,
1421 		 * still goto destroy_macvlan_port for readability.
1422 		 */
1423 		err = -EINVAL;
1424 		goto destroy_macvlan_port;
1425 	}
1426 
1427 	vlan->lowerdev = lowerdev;
1428 	vlan->dev      = dev;
1429 	vlan->port     = port;
1430 	vlan->set_features = MACVLAN_FEATURES;
1431 
1432 	vlan->mode     = MACVLAN_MODE_VEPA;
1433 	if (data && data[IFLA_MACVLAN_MODE])
1434 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1435 
1436 	if (data && data[IFLA_MACVLAN_FLAGS])
1437 		vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1438 
1439 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
1440 		if (port->count) {
1441 			err = -EINVAL;
1442 			goto destroy_macvlan_port;
1443 		}
1444 		macvlan_set_passthru(port);
1445 		eth_hw_addr_inherit(dev, lowerdev);
1446 	}
1447 
1448 	if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1449 		if (vlan->mode != MACVLAN_MODE_SOURCE) {
1450 			err = -EINVAL;
1451 			goto destroy_macvlan_port;
1452 		}
1453 		macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1454 		err = macvlan_changelink_sources(vlan, macmode, data);
1455 		if (err)
1456 			goto destroy_macvlan_port;
1457 	}
1458 
1459 	err = register_netdevice(dev);
1460 	if (err < 0)
1461 		goto destroy_macvlan_port;
1462 
1463 	dev->priv_flags |= IFF_MACVLAN;
1464 	err = netdev_upper_dev_link(lowerdev, dev, extack);
1465 	if (err)
1466 		goto unregister_netdev;
1467 
1468 	list_add_tail_rcu(&vlan->list, &port->vlans);
1469 	netif_stacked_transfer_operstate(lowerdev, dev);
1470 	linkwatch_fire_event(dev);
1471 
1472 	return 0;
1473 
1474 unregister_netdev:
1475 	/* macvlan_uninit would free the macvlan port */
1476 	unregister_netdevice(dev);
1477 	return err;
1478 destroy_macvlan_port:
1479 	/* the macvlan port may be freed by macvlan_uninit when fail to register.
1480 	 * so we destroy the macvlan port only when it's valid.
1481 	 */
1482 	if (create && macvlan_port_get_rtnl(lowerdev))
1483 		macvlan_port_destroy(port->dev);
1484 	return err;
1485 }
1486 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
1487 
1488 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
1489 			   struct nlattr *tb[], struct nlattr *data[],
1490 			   struct netlink_ext_ack *extack)
1491 {
1492 	return macvlan_common_newlink(src_net, dev, tb, data, extack);
1493 }
1494 
1495 void macvlan_dellink(struct net_device *dev, struct list_head *head)
1496 {
1497 	struct macvlan_dev *vlan = netdev_priv(dev);
1498 
1499 	if (vlan->mode == MACVLAN_MODE_SOURCE)
1500 		macvlan_flush_sources(vlan->port, vlan);
1501 	list_del_rcu(&vlan->list);
1502 	unregister_netdevice_queue(dev, head);
1503 	netdev_upper_dev_unlink(vlan->lowerdev, dev);
1504 }
1505 EXPORT_SYMBOL_GPL(macvlan_dellink);
1506 
1507 static int macvlan_changelink(struct net_device *dev,
1508 			      struct nlattr *tb[], struct nlattr *data[],
1509 			      struct netlink_ext_ack *extack)
1510 {
1511 	struct macvlan_dev *vlan = netdev_priv(dev);
1512 	enum macvlan_mode mode;
1513 	bool set_mode = false;
1514 	enum macvlan_macaddr_mode macmode;
1515 	int ret;
1516 
1517 	/* Validate mode, but don't set yet: setting flags may fail. */
1518 	if (data && data[IFLA_MACVLAN_MODE]) {
1519 		set_mode = true;
1520 		mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1521 		/* Passthrough mode can't be set or cleared dynamically */
1522 		if ((mode == MACVLAN_MODE_PASSTHRU) !=
1523 		    (vlan->mode == MACVLAN_MODE_PASSTHRU))
1524 			return -EINVAL;
1525 		if (vlan->mode == MACVLAN_MODE_SOURCE &&
1526 		    vlan->mode != mode)
1527 			macvlan_flush_sources(vlan->port, vlan);
1528 	}
1529 
1530 	if (data && data[IFLA_MACVLAN_FLAGS]) {
1531 		__u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1532 		bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
1533 		if (macvlan_passthru(vlan->port) && promisc) {
1534 			int err;
1535 
1536 			if (flags & MACVLAN_FLAG_NOPROMISC)
1537 				err = dev_set_promiscuity(vlan->lowerdev, -1);
1538 			else
1539 				err = dev_set_promiscuity(vlan->lowerdev, 1);
1540 			if (err < 0)
1541 				return err;
1542 		}
1543 		vlan->flags = flags;
1544 	}
1545 	if (set_mode)
1546 		vlan->mode = mode;
1547 	if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1548 		if (vlan->mode != MACVLAN_MODE_SOURCE)
1549 			return -EINVAL;
1550 		macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1551 		ret = macvlan_changelink_sources(vlan, macmode, data);
1552 		if (ret)
1553 			return ret;
1554 	}
1555 	return 0;
1556 }
1557 
1558 static size_t macvlan_get_size_mac(const struct macvlan_dev *vlan)
1559 {
1560 	if (vlan->macaddr_count == 0)
1561 		return 0;
1562 	return nla_total_size(0) /* IFLA_MACVLAN_MACADDR_DATA */
1563 		+ vlan->macaddr_count * nla_total_size(sizeof(u8) * ETH_ALEN);
1564 }
1565 
1566 static size_t macvlan_get_size(const struct net_device *dev)
1567 {
1568 	struct macvlan_dev *vlan = netdev_priv(dev);
1569 
1570 	return (0
1571 		+ nla_total_size(4) /* IFLA_MACVLAN_MODE */
1572 		+ nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
1573 		+ nla_total_size(4) /* IFLA_MACVLAN_MACADDR_COUNT */
1574 		+ macvlan_get_size_mac(vlan) /* IFLA_MACVLAN_MACADDR */
1575 		);
1576 }
1577 
1578 static int macvlan_fill_info_macaddr(struct sk_buff *skb,
1579 				     const struct macvlan_dev *vlan,
1580 				     const int i)
1581 {
1582 	struct hlist_head *h = &vlan->port->vlan_source_hash[i];
1583 	struct macvlan_source_entry *entry;
1584 
1585 	hlist_for_each_entry_rcu(entry, h, hlist) {
1586 		if (entry->vlan != vlan)
1587 			continue;
1588 		if (nla_put(skb, IFLA_MACVLAN_MACADDR, ETH_ALEN, entry->addr))
1589 			return 1;
1590 	}
1591 	return 0;
1592 }
1593 
1594 static int macvlan_fill_info(struct sk_buff *skb,
1595 				const struct net_device *dev)
1596 {
1597 	struct macvlan_dev *vlan = netdev_priv(dev);
1598 	int i;
1599 	struct nlattr *nest;
1600 
1601 	if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
1602 		goto nla_put_failure;
1603 	if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
1604 		goto nla_put_failure;
1605 	if (nla_put_u32(skb, IFLA_MACVLAN_MACADDR_COUNT, vlan->macaddr_count))
1606 		goto nla_put_failure;
1607 	if (vlan->macaddr_count > 0) {
1608 		nest = nla_nest_start_noflag(skb, IFLA_MACVLAN_MACADDR_DATA);
1609 		if (nest == NULL)
1610 			goto nla_put_failure;
1611 
1612 		for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
1613 			if (macvlan_fill_info_macaddr(skb, vlan, i))
1614 				goto nla_put_failure;
1615 		}
1616 		nla_nest_end(skb, nest);
1617 	}
1618 	return 0;
1619 
1620 nla_put_failure:
1621 	return -EMSGSIZE;
1622 }
1623 
1624 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
1625 	[IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
1626 	[IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
1627 	[IFLA_MACVLAN_MACADDR_MODE] = { .type = NLA_U32 },
1628 	[IFLA_MACVLAN_MACADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1629 	[IFLA_MACVLAN_MACADDR_DATA] = { .type = NLA_NESTED },
1630 	[IFLA_MACVLAN_MACADDR_COUNT] = { .type = NLA_U32 },
1631 };
1632 
1633 int macvlan_link_register(struct rtnl_link_ops *ops)
1634 {
1635 	/* common fields */
1636 	ops->validate		= macvlan_validate;
1637 	ops->maxtype		= IFLA_MACVLAN_MAX;
1638 	ops->policy		= macvlan_policy;
1639 	ops->changelink		= macvlan_changelink;
1640 	ops->get_size		= macvlan_get_size;
1641 	ops->fill_info		= macvlan_fill_info;
1642 
1643 	return rtnl_link_register(ops);
1644 };
1645 EXPORT_SYMBOL_GPL(macvlan_link_register);
1646 
1647 static struct net *macvlan_get_link_net(const struct net_device *dev)
1648 {
1649 	return dev_net(macvlan_dev_real_dev(dev));
1650 }
1651 
1652 static struct rtnl_link_ops macvlan_link_ops = {
1653 	.kind		= "macvlan",
1654 	.setup		= macvlan_setup,
1655 	.newlink	= macvlan_newlink,
1656 	.dellink	= macvlan_dellink,
1657 	.get_link_net	= macvlan_get_link_net,
1658 	.priv_size      = sizeof(struct macvlan_dev),
1659 };
1660 
1661 static int macvlan_device_event(struct notifier_block *unused,
1662 				unsigned long event, void *ptr)
1663 {
1664 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1665 	struct macvlan_dev *vlan, *next;
1666 	struct macvlan_port *port;
1667 	LIST_HEAD(list_kill);
1668 
1669 	if (!netif_is_macvlan_port(dev))
1670 		return NOTIFY_DONE;
1671 
1672 	port = macvlan_port_get_rtnl(dev);
1673 
1674 	switch (event) {
1675 	case NETDEV_UP:
1676 	case NETDEV_DOWN:
1677 	case NETDEV_CHANGE:
1678 		list_for_each_entry(vlan, &port->vlans, list)
1679 			netif_stacked_transfer_operstate(vlan->lowerdev,
1680 							 vlan->dev);
1681 		break;
1682 	case NETDEV_FEAT_CHANGE:
1683 		list_for_each_entry(vlan, &port->vlans, list) {
1684 			vlan->dev->gso_max_size = dev->gso_max_size;
1685 			vlan->dev->gso_max_segs = dev->gso_max_segs;
1686 			netdev_update_features(vlan->dev);
1687 		}
1688 		break;
1689 	case NETDEV_CHANGEMTU:
1690 		list_for_each_entry(vlan, &port->vlans, list) {
1691 			if (vlan->dev->mtu <= dev->mtu)
1692 				continue;
1693 			dev_set_mtu(vlan->dev, dev->mtu);
1694 		}
1695 		break;
1696 	case NETDEV_CHANGEADDR:
1697 		if (!macvlan_passthru(port))
1698 			return NOTIFY_DONE;
1699 
1700 		vlan = list_first_entry_or_null(&port->vlans,
1701 						struct macvlan_dev,
1702 						list);
1703 
1704 		if (macvlan_sync_address(vlan->dev, dev->dev_addr))
1705 			return NOTIFY_BAD;
1706 
1707 		break;
1708 	case NETDEV_UNREGISTER:
1709 		/* twiddle thumbs on netns device moves */
1710 		if (dev->reg_state != NETREG_UNREGISTERING)
1711 			break;
1712 
1713 		list_for_each_entry_safe(vlan, next, &port->vlans, list)
1714 			vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1715 		unregister_netdevice_many(&list_kill);
1716 		break;
1717 	case NETDEV_PRE_TYPE_CHANGE:
1718 		/* Forbid underlaying device to change its type. */
1719 		return NOTIFY_BAD;
1720 
1721 	case NETDEV_NOTIFY_PEERS:
1722 	case NETDEV_BONDING_FAILOVER:
1723 	case NETDEV_RESEND_IGMP:
1724 		/* Propagate to all vlans */
1725 		list_for_each_entry(vlan, &port->vlans, list)
1726 			call_netdevice_notifiers(event, vlan->dev);
1727 	}
1728 	return NOTIFY_DONE;
1729 }
1730 
1731 static struct notifier_block macvlan_notifier_block __read_mostly = {
1732 	.notifier_call	= macvlan_device_event,
1733 };
1734 
1735 static int __init macvlan_init_module(void)
1736 {
1737 	int err;
1738 
1739 	register_netdevice_notifier(&macvlan_notifier_block);
1740 
1741 	err = macvlan_link_register(&macvlan_link_ops);
1742 	if (err < 0)
1743 		goto err1;
1744 	return 0;
1745 err1:
1746 	unregister_netdevice_notifier(&macvlan_notifier_block);
1747 	return err;
1748 }
1749 
1750 static void __exit macvlan_cleanup_module(void)
1751 {
1752 	rtnl_link_unregister(&macvlan_link_ops);
1753 	unregister_netdevice_notifier(&macvlan_notifier_block);
1754 }
1755 
1756 module_init(macvlan_init_module);
1757 module_exit(macvlan_cleanup_module);
1758 
1759 MODULE_LICENSE("GPL");
1760 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1761 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1762 MODULE_ALIAS_RTNL_LINK("macvlan");
1763