xref: /openbmc/linux/drivers/net/macvlan.c (revision e2c75e76)
1 /*
2  * Copyright (c) 2007 Patrick McHardy <kaber@trash.net>
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public License as
6  * published by the Free Software Foundation; either version 2 of
7  * the License, or (at your option) any later version.
8  *
9  * The code this is based on carried the following copyright notice:
10  * ---
11  * (C) Copyright 2001-2006
12  * Alex Zeffertt, Cambridge Broadband Ltd, ajz@cambridgebroadband.com
13  * Re-worked by Ben Greear <greearb@candelatech.com>
14  * ---
15  */
16 #include <linux/kernel.h>
17 #include <linux/types.h>
18 #include <linux/module.h>
19 #include <linux/init.h>
20 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/string.h>
23 #include <linux/rculist.h>
24 #include <linux/notifier.h>
25 #include <linux/netdevice.h>
26 #include <linux/etherdevice.h>
27 #include <linux/ethtool.h>
28 #include <linux/if_arp.h>
29 #include <linux/if_vlan.h>
30 #include <linux/if_link.h>
31 #include <linux/if_macvlan.h>
32 #include <linux/hash.h>
33 #include <linux/workqueue.h>
34 #include <net/rtnetlink.h>
35 #include <net/xfrm.h>
36 #include <linux/netpoll.h>
37 
38 #define MACVLAN_HASH_BITS	8
39 #define MACVLAN_HASH_SIZE	(1<<MACVLAN_HASH_BITS)
40 #define MACVLAN_BC_QUEUE_LEN	1000
41 
42 #define MACVLAN_F_PASSTHRU	1
43 #define MACVLAN_F_ADDRCHANGE	2
44 
45 struct macvlan_port {
46 	struct net_device	*dev;
47 	struct hlist_head	vlan_hash[MACVLAN_HASH_SIZE];
48 	struct list_head	vlans;
49 	struct sk_buff_head	bc_queue;
50 	struct work_struct	bc_work;
51 	u32			flags;
52 	int			count;
53 	struct hlist_head	vlan_source_hash[MACVLAN_HASH_SIZE];
54 	DECLARE_BITMAP(mc_filter, MACVLAN_MC_FILTER_SZ);
55 	unsigned char           perm_addr[ETH_ALEN];
56 };
57 
58 struct macvlan_source_entry {
59 	struct hlist_node	hlist;
60 	struct macvlan_dev	*vlan;
61 	unsigned char		addr[6+2] __aligned(sizeof(u16));
62 	struct rcu_head		rcu;
63 };
64 
65 struct macvlan_skb_cb {
66 	const struct macvlan_dev *src;
67 };
68 
69 #define MACVLAN_SKB_CB(__skb) ((struct macvlan_skb_cb *)&((__skb)->cb[0]))
70 
71 static void macvlan_port_destroy(struct net_device *dev);
72 
73 static inline bool macvlan_passthru(const struct macvlan_port *port)
74 {
75 	return port->flags & MACVLAN_F_PASSTHRU;
76 }
77 
78 static inline void macvlan_set_passthru(struct macvlan_port *port)
79 {
80 	port->flags |= MACVLAN_F_PASSTHRU;
81 }
82 
83 static inline bool macvlan_addr_change(const struct macvlan_port *port)
84 {
85 	return port->flags & MACVLAN_F_ADDRCHANGE;
86 }
87 
88 static inline void macvlan_set_addr_change(struct macvlan_port *port)
89 {
90 	port->flags |= MACVLAN_F_ADDRCHANGE;
91 }
92 
93 static inline void macvlan_clear_addr_change(struct macvlan_port *port)
94 {
95 	port->flags &= ~MACVLAN_F_ADDRCHANGE;
96 }
97 
98 /* Hash Ethernet address */
99 static u32 macvlan_eth_hash(const unsigned char *addr)
100 {
101 	u64 value = get_unaligned((u64 *)addr);
102 
103 	/* only want 6 bytes */
104 #ifdef __BIG_ENDIAN
105 	value >>= 16;
106 #else
107 	value <<= 16;
108 #endif
109 	return hash_64(value, MACVLAN_HASH_BITS);
110 }
111 
112 static struct macvlan_port *macvlan_port_get_rcu(const struct net_device *dev)
113 {
114 	return rcu_dereference(dev->rx_handler_data);
115 }
116 
117 static struct macvlan_port *macvlan_port_get_rtnl(const struct net_device *dev)
118 {
119 	return rtnl_dereference(dev->rx_handler_data);
120 }
121 
122 #define macvlan_port_exists(dev) (dev->priv_flags & IFF_MACVLAN_PORT)
123 
124 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
125 					       const unsigned char *addr)
126 {
127 	struct macvlan_dev *vlan;
128 	u32 idx = macvlan_eth_hash(addr);
129 
130 	hlist_for_each_entry_rcu(vlan, &port->vlan_hash[idx], hlist) {
131 		if (ether_addr_equal_64bits(vlan->dev->dev_addr, addr))
132 			return vlan;
133 	}
134 	return NULL;
135 }
136 
137 static struct macvlan_source_entry *macvlan_hash_lookup_source(
138 	const struct macvlan_dev *vlan,
139 	const unsigned char *addr)
140 {
141 	struct macvlan_source_entry *entry;
142 	u32 idx = macvlan_eth_hash(addr);
143 	struct hlist_head *h = &vlan->port->vlan_source_hash[idx];
144 
145 	hlist_for_each_entry_rcu(entry, h, hlist) {
146 		if (ether_addr_equal_64bits(entry->addr, addr) &&
147 		    entry->vlan == vlan)
148 			return entry;
149 	}
150 	return NULL;
151 }
152 
153 static int macvlan_hash_add_source(struct macvlan_dev *vlan,
154 				   const unsigned char *addr)
155 {
156 	struct macvlan_port *port = vlan->port;
157 	struct macvlan_source_entry *entry;
158 	struct hlist_head *h;
159 
160 	entry = macvlan_hash_lookup_source(vlan, addr);
161 	if (entry)
162 		return 0;
163 
164 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
165 	if (!entry)
166 		return -ENOMEM;
167 
168 	ether_addr_copy(entry->addr, addr);
169 	entry->vlan = vlan;
170 	h = &port->vlan_source_hash[macvlan_eth_hash(addr)];
171 	hlist_add_head_rcu(&entry->hlist, h);
172 	vlan->macaddr_count++;
173 
174 	return 0;
175 }
176 
177 static void macvlan_hash_add(struct macvlan_dev *vlan)
178 {
179 	struct macvlan_port *port = vlan->port;
180 	const unsigned char *addr = vlan->dev->dev_addr;
181 	u32 idx = macvlan_eth_hash(addr);
182 
183 	hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[idx]);
184 }
185 
186 static void macvlan_hash_del_source(struct macvlan_source_entry *entry)
187 {
188 	hlist_del_rcu(&entry->hlist);
189 	kfree_rcu(entry, rcu);
190 }
191 
192 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync)
193 {
194 	hlist_del_rcu(&vlan->hlist);
195 	if (sync)
196 		synchronize_rcu();
197 }
198 
199 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
200 					const unsigned char *addr)
201 {
202 	macvlan_hash_del(vlan, true);
203 	/* Now that we are unhashed it is safe to change the device
204 	 * address without confusing packet delivery.
205 	 */
206 	memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
207 	macvlan_hash_add(vlan);
208 }
209 
210 static bool macvlan_addr_busy(const struct macvlan_port *port,
211 			      const unsigned char *addr)
212 {
213 	/* Test to see if the specified address is
214 	 * currently in use by the underlying device or
215 	 * another macvlan.
216 	 */
217 	if (!macvlan_passthru(port) && !macvlan_addr_change(port) &&
218 	    ether_addr_equal_64bits(port->dev->dev_addr, addr))
219 		return true;
220 
221 	if (macvlan_hash_lookup(port, addr))
222 		return true;
223 
224 	return false;
225 }
226 
227 
228 static int macvlan_broadcast_one(struct sk_buff *skb,
229 				 const struct macvlan_dev *vlan,
230 				 const struct ethhdr *eth, bool local)
231 {
232 	struct net_device *dev = vlan->dev;
233 
234 	if (local)
235 		return __dev_forward_skb(dev, skb);
236 
237 	skb->dev = dev;
238 	if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
239 		skb->pkt_type = PACKET_BROADCAST;
240 	else
241 		skb->pkt_type = PACKET_MULTICAST;
242 
243 	return 0;
244 }
245 
246 static u32 macvlan_hash_mix(const struct macvlan_dev *vlan)
247 {
248 	return (u32)(((unsigned long)vlan) >> L1_CACHE_SHIFT);
249 }
250 
251 
252 static unsigned int mc_hash(const struct macvlan_dev *vlan,
253 			    const unsigned char *addr)
254 {
255 	u32 val = __get_unaligned_cpu32(addr + 2);
256 
257 	val ^= macvlan_hash_mix(vlan);
258 	return hash_32(val, MACVLAN_MC_FILTER_BITS);
259 }
260 
261 static void macvlan_broadcast(struct sk_buff *skb,
262 			      const struct macvlan_port *port,
263 			      struct net_device *src,
264 			      enum macvlan_mode mode)
265 {
266 	const struct ethhdr *eth = eth_hdr(skb);
267 	const struct macvlan_dev *vlan;
268 	struct sk_buff *nskb;
269 	unsigned int i;
270 	int err;
271 	unsigned int hash;
272 
273 	if (skb->protocol == htons(ETH_P_PAUSE))
274 		return;
275 
276 	for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
277 		hlist_for_each_entry_rcu(vlan, &port->vlan_hash[i], hlist) {
278 			if (vlan->dev == src || !(vlan->mode & mode))
279 				continue;
280 
281 			hash = mc_hash(vlan, eth->h_dest);
282 			if (!test_bit(hash, vlan->mc_filter))
283 				continue;
284 
285 			err = NET_RX_DROP;
286 			nskb = skb_clone(skb, GFP_ATOMIC);
287 			if (likely(nskb))
288 				err = macvlan_broadcast_one(
289 					nskb, vlan, eth,
290 					mode == MACVLAN_MODE_BRIDGE) ?:
291 				      netif_rx_ni(nskb);
292 			macvlan_count_rx(vlan, skb->len + ETH_HLEN,
293 					 err == NET_RX_SUCCESS, true);
294 		}
295 	}
296 }
297 
298 static void macvlan_process_broadcast(struct work_struct *w)
299 {
300 	struct macvlan_port *port = container_of(w, struct macvlan_port,
301 						 bc_work);
302 	struct sk_buff *skb;
303 	struct sk_buff_head list;
304 
305 	__skb_queue_head_init(&list);
306 
307 	spin_lock_bh(&port->bc_queue.lock);
308 	skb_queue_splice_tail_init(&port->bc_queue, &list);
309 	spin_unlock_bh(&port->bc_queue.lock);
310 
311 	while ((skb = __skb_dequeue(&list))) {
312 		const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
313 
314 		rcu_read_lock();
315 
316 		if (!src)
317 			/* frame comes from an external address */
318 			macvlan_broadcast(skb, port, NULL,
319 					  MACVLAN_MODE_PRIVATE |
320 					  MACVLAN_MODE_VEPA    |
321 					  MACVLAN_MODE_PASSTHRU|
322 					  MACVLAN_MODE_BRIDGE);
323 		else if (src->mode == MACVLAN_MODE_VEPA)
324 			/* flood to everyone except source */
325 			macvlan_broadcast(skb, port, src->dev,
326 					  MACVLAN_MODE_VEPA |
327 					  MACVLAN_MODE_BRIDGE);
328 		else
329 			/*
330 			 * flood only to VEPA ports, bridge ports
331 			 * already saw the frame on the way out.
332 			 */
333 			macvlan_broadcast(skb, port, src->dev,
334 					  MACVLAN_MODE_VEPA);
335 
336 		rcu_read_unlock();
337 
338 		if (src)
339 			dev_put(src->dev);
340 		kfree_skb(skb);
341 	}
342 }
343 
344 static void macvlan_broadcast_enqueue(struct macvlan_port *port,
345 				      const struct macvlan_dev *src,
346 				      struct sk_buff *skb)
347 {
348 	struct sk_buff *nskb;
349 	int err = -ENOMEM;
350 
351 	nskb = skb_clone(skb, GFP_ATOMIC);
352 	if (!nskb)
353 		goto err;
354 
355 	MACVLAN_SKB_CB(nskb)->src = src;
356 
357 	spin_lock(&port->bc_queue.lock);
358 	if (skb_queue_len(&port->bc_queue) < MACVLAN_BC_QUEUE_LEN) {
359 		if (src)
360 			dev_hold(src->dev);
361 		__skb_queue_tail(&port->bc_queue, nskb);
362 		err = 0;
363 	}
364 	spin_unlock(&port->bc_queue.lock);
365 
366 	if (err)
367 		goto free_nskb;
368 
369 	schedule_work(&port->bc_work);
370 	return;
371 
372 free_nskb:
373 	kfree_skb(nskb);
374 err:
375 	atomic_long_inc(&skb->dev->rx_dropped);
376 }
377 
378 static void macvlan_flush_sources(struct macvlan_port *port,
379 				  struct macvlan_dev *vlan)
380 {
381 	int i;
382 
383 	for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
384 		struct hlist_node *h, *n;
385 
386 		hlist_for_each_safe(h, n, &port->vlan_source_hash[i]) {
387 			struct macvlan_source_entry *entry;
388 
389 			entry = hlist_entry(h, struct macvlan_source_entry,
390 					    hlist);
391 			if (entry->vlan == vlan)
392 				macvlan_hash_del_source(entry);
393 		}
394 	}
395 	vlan->macaddr_count = 0;
396 }
397 
398 static void macvlan_forward_source_one(struct sk_buff *skb,
399 				       struct macvlan_dev *vlan)
400 {
401 	struct sk_buff *nskb;
402 	struct net_device *dev;
403 	int len;
404 	int ret;
405 
406 	dev = vlan->dev;
407 	if (unlikely(!(dev->flags & IFF_UP)))
408 		return;
409 
410 	nskb = skb_clone(skb, GFP_ATOMIC);
411 	if (!nskb)
412 		return;
413 
414 	len = nskb->len + ETH_HLEN;
415 	nskb->dev = dev;
416 
417 	if (ether_addr_equal_64bits(eth_hdr(skb)->h_dest, dev->dev_addr))
418 		nskb->pkt_type = PACKET_HOST;
419 
420 	ret = netif_rx(nskb);
421 	macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
422 }
423 
424 static void macvlan_forward_source(struct sk_buff *skb,
425 				   struct macvlan_port *port,
426 				   const unsigned char *addr)
427 {
428 	struct macvlan_source_entry *entry;
429 	u32 idx = macvlan_eth_hash(addr);
430 	struct hlist_head *h = &port->vlan_source_hash[idx];
431 
432 	hlist_for_each_entry_rcu(entry, h, hlist) {
433 		if (ether_addr_equal_64bits(entry->addr, addr))
434 			macvlan_forward_source_one(skb, entry->vlan);
435 	}
436 }
437 
438 /* called under rcu_read_lock() from netif_receive_skb */
439 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
440 {
441 	struct macvlan_port *port;
442 	struct sk_buff *skb = *pskb;
443 	const struct ethhdr *eth = eth_hdr(skb);
444 	const struct macvlan_dev *vlan;
445 	const struct macvlan_dev *src;
446 	struct net_device *dev;
447 	unsigned int len = 0;
448 	int ret;
449 	rx_handler_result_t handle_res;
450 
451 	port = macvlan_port_get_rcu(skb->dev);
452 	if (is_multicast_ether_addr(eth->h_dest)) {
453 		unsigned int hash;
454 
455 		skb = ip_check_defrag(dev_net(skb->dev), skb, IP_DEFRAG_MACVLAN);
456 		if (!skb)
457 			return RX_HANDLER_CONSUMED;
458 		*pskb = skb;
459 		eth = eth_hdr(skb);
460 		macvlan_forward_source(skb, port, eth->h_source);
461 		src = macvlan_hash_lookup(port, eth->h_source);
462 		if (src && src->mode != MACVLAN_MODE_VEPA &&
463 		    src->mode != MACVLAN_MODE_BRIDGE) {
464 			/* forward to original port. */
465 			vlan = src;
466 			ret = macvlan_broadcast_one(skb, vlan, eth, 0) ?:
467 			      netif_rx(skb);
468 			handle_res = RX_HANDLER_CONSUMED;
469 			goto out;
470 		}
471 
472 		hash = mc_hash(NULL, eth->h_dest);
473 		if (test_bit(hash, port->mc_filter))
474 			macvlan_broadcast_enqueue(port, src, skb);
475 
476 		return RX_HANDLER_PASS;
477 	}
478 
479 	macvlan_forward_source(skb, port, eth->h_source);
480 	if (macvlan_passthru(port))
481 		vlan = list_first_or_null_rcu(&port->vlans,
482 					      struct macvlan_dev, list);
483 	else
484 		vlan = macvlan_hash_lookup(port, eth->h_dest);
485 	if (!vlan || vlan->mode == MACVLAN_MODE_SOURCE)
486 		return RX_HANDLER_PASS;
487 
488 	dev = vlan->dev;
489 	if (unlikely(!(dev->flags & IFF_UP))) {
490 		kfree_skb(skb);
491 		return RX_HANDLER_CONSUMED;
492 	}
493 	len = skb->len + ETH_HLEN;
494 	skb = skb_share_check(skb, GFP_ATOMIC);
495 	if (!skb) {
496 		ret = NET_RX_DROP;
497 		handle_res = RX_HANDLER_CONSUMED;
498 		goto out;
499 	}
500 
501 	*pskb = skb;
502 	skb->dev = dev;
503 	skb->pkt_type = PACKET_HOST;
504 
505 	ret = NET_RX_SUCCESS;
506 	handle_res = RX_HANDLER_ANOTHER;
507 out:
508 	macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
509 	return handle_res;
510 }
511 
512 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
513 {
514 	const struct macvlan_dev *vlan = netdev_priv(dev);
515 	const struct macvlan_port *port = vlan->port;
516 	const struct macvlan_dev *dest;
517 
518 	if (vlan->mode == MACVLAN_MODE_BRIDGE) {
519 		const struct ethhdr *eth = (void *)skb->data;
520 
521 		/* send to other bridge ports directly */
522 		if (is_multicast_ether_addr(eth->h_dest)) {
523 			macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
524 			goto xmit_world;
525 		}
526 
527 		dest = macvlan_hash_lookup(port, eth->h_dest);
528 		if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
529 			/* send to lowerdev first for its network taps */
530 			dev_forward_skb(vlan->lowerdev, skb);
531 
532 			return NET_XMIT_SUCCESS;
533 		}
534 	}
535 
536 xmit_world:
537 	skb->dev = vlan->lowerdev;
538 	return dev_queue_xmit(skb);
539 }
540 
541 static inline netdev_tx_t macvlan_netpoll_send_skb(struct macvlan_dev *vlan, struct sk_buff *skb)
542 {
543 #ifdef CONFIG_NET_POLL_CONTROLLER
544 	if (vlan->netpoll)
545 		netpoll_send_skb(vlan->netpoll, skb);
546 #else
547 	BUG();
548 #endif
549 	return NETDEV_TX_OK;
550 }
551 
552 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
553 				      struct net_device *dev)
554 {
555 	unsigned int len = skb->len;
556 	int ret;
557 	struct macvlan_dev *vlan = netdev_priv(dev);
558 
559 	if (unlikely(netpoll_tx_running(dev)))
560 		return macvlan_netpoll_send_skb(vlan, skb);
561 
562 	if (vlan->fwd_priv) {
563 		skb->dev = vlan->lowerdev;
564 		ret = dev_queue_xmit_accel(skb, vlan->fwd_priv);
565 	} else {
566 		ret = macvlan_queue_xmit(skb, dev);
567 	}
568 
569 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
570 		struct vlan_pcpu_stats *pcpu_stats;
571 
572 		pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
573 		u64_stats_update_begin(&pcpu_stats->syncp);
574 		pcpu_stats->tx_packets++;
575 		pcpu_stats->tx_bytes += len;
576 		u64_stats_update_end(&pcpu_stats->syncp);
577 	} else {
578 		this_cpu_inc(vlan->pcpu_stats->tx_dropped);
579 	}
580 	return ret;
581 }
582 
583 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
584 			       unsigned short type, const void *daddr,
585 			       const void *saddr, unsigned len)
586 {
587 	const struct macvlan_dev *vlan = netdev_priv(dev);
588 	struct net_device *lowerdev = vlan->lowerdev;
589 
590 	return dev_hard_header(skb, lowerdev, type, daddr,
591 			       saddr ? : dev->dev_addr, len);
592 }
593 
594 static const struct header_ops macvlan_hard_header_ops = {
595 	.create  	= macvlan_hard_header,
596 	.parse		= eth_header_parse,
597 	.cache		= eth_header_cache,
598 	.cache_update	= eth_header_cache_update,
599 };
600 
601 static int macvlan_open(struct net_device *dev)
602 {
603 	struct macvlan_dev *vlan = netdev_priv(dev);
604 	struct net_device *lowerdev = vlan->lowerdev;
605 	int err;
606 
607 	if (macvlan_passthru(vlan->port)) {
608 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
609 			err = dev_set_promiscuity(lowerdev, 1);
610 			if (err < 0)
611 				goto out;
612 		}
613 		goto hash_add;
614 	}
615 
616 	if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD) {
617 		vlan->fwd_priv =
618 		      lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
619 
620 		/* If we get a NULL pointer back, or if we get an error
621 		 * then we should just fall through to the non accelerated path
622 		 */
623 		if (IS_ERR_OR_NULL(vlan->fwd_priv)) {
624 			vlan->fwd_priv = NULL;
625 		} else
626 			return 0;
627 	}
628 
629 	err = -EBUSY;
630 	if (macvlan_addr_busy(vlan->port, dev->dev_addr))
631 		goto out;
632 
633 	err = dev_uc_add(lowerdev, dev->dev_addr);
634 	if (err < 0)
635 		goto out;
636 	if (dev->flags & IFF_ALLMULTI) {
637 		err = dev_set_allmulti(lowerdev, 1);
638 		if (err < 0)
639 			goto del_unicast;
640 	}
641 
642 	if (dev->flags & IFF_PROMISC) {
643 		err = dev_set_promiscuity(lowerdev, 1);
644 		if (err < 0)
645 			goto clear_multi;
646 	}
647 
648 hash_add:
649 	macvlan_hash_add(vlan);
650 	return 0;
651 
652 clear_multi:
653 	if (dev->flags & IFF_ALLMULTI)
654 		dev_set_allmulti(lowerdev, -1);
655 del_unicast:
656 	dev_uc_del(lowerdev, dev->dev_addr);
657 out:
658 	if (vlan->fwd_priv) {
659 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
660 							   vlan->fwd_priv);
661 		vlan->fwd_priv = NULL;
662 	}
663 	return err;
664 }
665 
666 static int macvlan_stop(struct net_device *dev)
667 {
668 	struct macvlan_dev *vlan = netdev_priv(dev);
669 	struct net_device *lowerdev = vlan->lowerdev;
670 
671 	if (vlan->fwd_priv) {
672 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
673 							   vlan->fwd_priv);
674 		vlan->fwd_priv = NULL;
675 		return 0;
676 	}
677 
678 	dev_uc_unsync(lowerdev, dev);
679 	dev_mc_unsync(lowerdev, dev);
680 
681 	if (macvlan_passthru(vlan->port)) {
682 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
683 			dev_set_promiscuity(lowerdev, -1);
684 		goto hash_del;
685 	}
686 
687 	if (dev->flags & IFF_ALLMULTI)
688 		dev_set_allmulti(lowerdev, -1);
689 
690 	if (dev->flags & IFF_PROMISC)
691 		dev_set_promiscuity(lowerdev, -1);
692 
693 	dev_uc_del(lowerdev, dev->dev_addr);
694 
695 hash_del:
696 	macvlan_hash_del(vlan, !dev->dismantle);
697 	return 0;
698 }
699 
700 static int macvlan_sync_address(struct net_device *dev, unsigned char *addr)
701 {
702 	struct macvlan_dev *vlan = netdev_priv(dev);
703 	struct net_device *lowerdev = vlan->lowerdev;
704 	struct macvlan_port *port = vlan->port;
705 	int err;
706 
707 	if (!(dev->flags & IFF_UP)) {
708 		/* Just copy in the new address */
709 		ether_addr_copy(dev->dev_addr, addr);
710 	} else {
711 		/* Rehash and update the device filters */
712 		if (macvlan_addr_busy(vlan->port, addr))
713 			return -EBUSY;
714 
715 		if (!macvlan_passthru(port)) {
716 			err = dev_uc_add(lowerdev, addr);
717 			if (err)
718 				return err;
719 
720 			dev_uc_del(lowerdev, dev->dev_addr);
721 		}
722 
723 		macvlan_hash_change_addr(vlan, addr);
724 	}
725 	if (macvlan_passthru(port) && !macvlan_addr_change(port)) {
726 		/* Since addr_change isn't set, we are here due to lower
727 		 * device change.  Save the lower-dev address so we can
728 		 * restore it later.
729 		 */
730 		ether_addr_copy(vlan->port->perm_addr,
731 				lowerdev->dev_addr);
732 	}
733 	macvlan_clear_addr_change(port);
734 	return 0;
735 }
736 
737 static int macvlan_set_mac_address(struct net_device *dev, void *p)
738 {
739 	struct macvlan_dev *vlan = netdev_priv(dev);
740 	struct sockaddr *addr = p;
741 
742 	if (!is_valid_ether_addr(addr->sa_data))
743 		return -EADDRNOTAVAIL;
744 
745 	/* If the addresses are the same, this is a no-op */
746 	if (ether_addr_equal(dev->dev_addr, addr->sa_data))
747 		return 0;
748 
749 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
750 		macvlan_set_addr_change(vlan->port);
751 		return dev_set_mac_address(vlan->lowerdev, addr);
752 	}
753 
754 	return macvlan_sync_address(dev, addr->sa_data);
755 }
756 
757 static void macvlan_change_rx_flags(struct net_device *dev, int change)
758 {
759 	struct macvlan_dev *vlan = netdev_priv(dev);
760 	struct net_device *lowerdev = vlan->lowerdev;
761 
762 	if (dev->flags & IFF_UP) {
763 		if (change & IFF_ALLMULTI)
764 			dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
765 		if (change & IFF_PROMISC)
766 			dev_set_promiscuity(lowerdev,
767 					    dev->flags & IFF_PROMISC ? 1 : -1);
768 
769 	}
770 }
771 
772 static void macvlan_compute_filter(unsigned long *mc_filter,
773 				   struct net_device *dev,
774 				   struct macvlan_dev *vlan)
775 {
776 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
777 		bitmap_fill(mc_filter, MACVLAN_MC_FILTER_SZ);
778 	} else {
779 		struct netdev_hw_addr *ha;
780 		DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
781 
782 		bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
783 		netdev_for_each_mc_addr(ha, dev) {
784 			__set_bit(mc_hash(vlan, ha->addr), filter);
785 		}
786 
787 		__set_bit(mc_hash(vlan, dev->broadcast), filter);
788 
789 		bitmap_copy(mc_filter, filter, MACVLAN_MC_FILTER_SZ);
790 	}
791 }
792 
793 static void macvlan_set_mac_lists(struct net_device *dev)
794 {
795 	struct macvlan_dev *vlan = netdev_priv(dev);
796 
797 	macvlan_compute_filter(vlan->mc_filter, dev, vlan);
798 
799 	dev_uc_sync(vlan->lowerdev, dev);
800 	dev_mc_sync(vlan->lowerdev, dev);
801 
802 	/* This is slightly inaccurate as we're including the subscription
803 	 * list of vlan->lowerdev too.
804 	 *
805 	 * Bug alert: This only works if everyone has the same broadcast
806 	 * address as lowerdev.  As soon as someone changes theirs this
807 	 * will break.
808 	 *
809 	 * However, this is already broken as when you change your broadcast
810 	 * address we don't get called.
811 	 *
812 	 * The solution is to maintain a list of broadcast addresses like
813 	 * we do for uc/mc, if you care.
814 	 */
815 	macvlan_compute_filter(vlan->port->mc_filter, vlan->lowerdev, NULL);
816 }
817 
818 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
819 {
820 	struct macvlan_dev *vlan = netdev_priv(dev);
821 
822 	if (vlan->lowerdev->mtu < new_mtu)
823 		return -EINVAL;
824 	dev->mtu = new_mtu;
825 	return 0;
826 }
827 
828 /*
829  * macvlan network devices have devices nesting below it and are a special
830  * "super class" of normal network devices; split their locks off into a
831  * separate class since they always nest.
832  */
833 static struct lock_class_key macvlan_netdev_addr_lock_key;
834 
835 #define ALWAYS_ON_OFFLOADS \
836 	(NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE | \
837 	 NETIF_F_GSO_ROBUST | NETIF_F_GSO_ENCAP_ALL)
838 
839 #define ALWAYS_ON_FEATURES (ALWAYS_ON_OFFLOADS | NETIF_F_LLTX)
840 
841 #define MACVLAN_FEATURES \
842 	(NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
843 	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_LRO | \
844 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
845 	 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
846 
847 #define MACVLAN_STATE_MASK \
848 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
849 
850 static int macvlan_get_nest_level(struct net_device *dev)
851 {
852 	return ((struct macvlan_dev *)netdev_priv(dev))->nest_level;
853 }
854 
855 static void macvlan_set_lockdep_class(struct net_device *dev)
856 {
857 	netdev_lockdep_set_classes(dev);
858 	lockdep_set_class_and_subclass(&dev->addr_list_lock,
859 				       &macvlan_netdev_addr_lock_key,
860 				       macvlan_get_nest_level(dev));
861 }
862 
863 static int macvlan_init(struct net_device *dev)
864 {
865 	struct macvlan_dev *vlan = netdev_priv(dev);
866 	const struct net_device *lowerdev = vlan->lowerdev;
867 	struct macvlan_port *port = vlan->port;
868 
869 	dev->state		= (dev->state & ~MACVLAN_STATE_MASK) |
870 				  (lowerdev->state & MACVLAN_STATE_MASK);
871 	dev->features 		= lowerdev->features & MACVLAN_FEATURES;
872 	dev->features		|= ALWAYS_ON_FEATURES;
873 	dev->hw_features	|= NETIF_F_LRO;
874 	dev->vlan_features	= lowerdev->vlan_features & MACVLAN_FEATURES;
875 	dev->vlan_features	|= ALWAYS_ON_OFFLOADS;
876 	dev->hw_enc_features    |= dev->features;
877 	dev->gso_max_size	= lowerdev->gso_max_size;
878 	dev->gso_max_segs	= lowerdev->gso_max_segs;
879 	dev->hard_header_len	= lowerdev->hard_header_len;
880 
881 	macvlan_set_lockdep_class(dev);
882 
883 	vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
884 	if (!vlan->pcpu_stats)
885 		return -ENOMEM;
886 
887 	port->count += 1;
888 
889 	return 0;
890 }
891 
892 static void macvlan_uninit(struct net_device *dev)
893 {
894 	struct macvlan_dev *vlan = netdev_priv(dev);
895 	struct macvlan_port *port = vlan->port;
896 
897 	free_percpu(vlan->pcpu_stats);
898 
899 	macvlan_flush_sources(port, vlan);
900 	port->count -= 1;
901 	if (!port->count)
902 		macvlan_port_destroy(port->dev);
903 }
904 
905 static void macvlan_dev_get_stats64(struct net_device *dev,
906 				    struct rtnl_link_stats64 *stats)
907 {
908 	struct macvlan_dev *vlan = netdev_priv(dev);
909 
910 	if (vlan->pcpu_stats) {
911 		struct vlan_pcpu_stats *p;
912 		u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
913 		u32 rx_errors = 0, tx_dropped = 0;
914 		unsigned int start;
915 		int i;
916 
917 		for_each_possible_cpu(i) {
918 			p = per_cpu_ptr(vlan->pcpu_stats, i);
919 			do {
920 				start = u64_stats_fetch_begin_irq(&p->syncp);
921 				rx_packets	= p->rx_packets;
922 				rx_bytes	= p->rx_bytes;
923 				rx_multicast	= p->rx_multicast;
924 				tx_packets	= p->tx_packets;
925 				tx_bytes	= p->tx_bytes;
926 			} while (u64_stats_fetch_retry_irq(&p->syncp, start));
927 
928 			stats->rx_packets	+= rx_packets;
929 			stats->rx_bytes		+= rx_bytes;
930 			stats->multicast	+= rx_multicast;
931 			stats->tx_packets	+= tx_packets;
932 			stats->tx_bytes		+= tx_bytes;
933 			/* rx_errors & tx_dropped are u32, updated
934 			 * without syncp protection.
935 			 */
936 			rx_errors	+= p->rx_errors;
937 			tx_dropped	+= p->tx_dropped;
938 		}
939 		stats->rx_errors	= rx_errors;
940 		stats->rx_dropped	= rx_errors;
941 		stats->tx_dropped	= tx_dropped;
942 	}
943 }
944 
945 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
946 				   __be16 proto, u16 vid)
947 {
948 	struct macvlan_dev *vlan = netdev_priv(dev);
949 	struct net_device *lowerdev = vlan->lowerdev;
950 
951 	return vlan_vid_add(lowerdev, proto, vid);
952 }
953 
954 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
955 				    __be16 proto, u16 vid)
956 {
957 	struct macvlan_dev *vlan = netdev_priv(dev);
958 	struct net_device *lowerdev = vlan->lowerdev;
959 
960 	vlan_vid_del(lowerdev, proto, vid);
961 	return 0;
962 }
963 
964 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
965 			   struct net_device *dev,
966 			   const unsigned char *addr, u16 vid,
967 			   u16 flags)
968 {
969 	struct macvlan_dev *vlan = netdev_priv(dev);
970 	int err = -EINVAL;
971 
972 	/* Support unicast filter only on passthru devices.
973 	 * Multicast filter should be allowed on all devices.
974 	 */
975 	if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr))
976 		return -EOPNOTSUPP;
977 
978 	if (flags & NLM_F_REPLACE)
979 		return -EOPNOTSUPP;
980 
981 	if (is_unicast_ether_addr(addr))
982 		err = dev_uc_add_excl(dev, addr);
983 	else if (is_multicast_ether_addr(addr))
984 		err = dev_mc_add_excl(dev, addr);
985 
986 	return err;
987 }
988 
989 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
990 			   struct net_device *dev,
991 			   const unsigned char *addr, u16 vid)
992 {
993 	struct macvlan_dev *vlan = netdev_priv(dev);
994 	int err = -EINVAL;
995 
996 	/* Support unicast filter only on passthru devices.
997 	 * Multicast filter should be allowed on all devices.
998 	 */
999 	if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr))
1000 		return -EOPNOTSUPP;
1001 
1002 	if (is_unicast_ether_addr(addr))
1003 		err = dev_uc_del(dev, addr);
1004 	else if (is_multicast_ether_addr(addr))
1005 		err = dev_mc_del(dev, addr);
1006 
1007 	return err;
1008 }
1009 
1010 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
1011 					struct ethtool_drvinfo *drvinfo)
1012 {
1013 	strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
1014 	strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
1015 }
1016 
1017 static int macvlan_ethtool_get_link_ksettings(struct net_device *dev,
1018 					      struct ethtool_link_ksettings *cmd)
1019 {
1020 	const struct macvlan_dev *vlan = netdev_priv(dev);
1021 
1022 	return __ethtool_get_link_ksettings(vlan->lowerdev, cmd);
1023 }
1024 
1025 static netdev_features_t macvlan_fix_features(struct net_device *dev,
1026 					      netdev_features_t features)
1027 {
1028 	struct macvlan_dev *vlan = netdev_priv(dev);
1029 	netdev_features_t lowerdev_features = vlan->lowerdev->features;
1030 	netdev_features_t mask;
1031 
1032 	features |= NETIF_F_ALL_FOR_ALL;
1033 	features &= (vlan->set_features | ~MACVLAN_FEATURES);
1034 	mask = features;
1035 
1036 	lowerdev_features &= (features | ~NETIF_F_LRO);
1037 	features = netdev_increment_features(lowerdev_features, features, mask);
1038 	features |= ALWAYS_ON_FEATURES;
1039 	features &= ~NETIF_F_NETNS_LOCAL;
1040 
1041 	return features;
1042 }
1043 
1044 #ifdef CONFIG_NET_POLL_CONTROLLER
1045 static void macvlan_dev_poll_controller(struct net_device *dev)
1046 {
1047 	return;
1048 }
1049 
1050 static int macvlan_dev_netpoll_setup(struct net_device *dev, struct netpoll_info *npinfo)
1051 {
1052 	struct macvlan_dev *vlan = netdev_priv(dev);
1053 	struct net_device *real_dev = vlan->lowerdev;
1054 	struct netpoll *netpoll;
1055 	int err = 0;
1056 
1057 	netpoll = kzalloc(sizeof(*netpoll), GFP_KERNEL);
1058 	err = -ENOMEM;
1059 	if (!netpoll)
1060 		goto out;
1061 
1062 	err = __netpoll_setup(netpoll, real_dev);
1063 	if (err) {
1064 		kfree(netpoll);
1065 		goto out;
1066 	}
1067 
1068 	vlan->netpoll = netpoll;
1069 
1070 out:
1071 	return err;
1072 }
1073 
1074 static void macvlan_dev_netpoll_cleanup(struct net_device *dev)
1075 {
1076 	struct macvlan_dev *vlan = netdev_priv(dev);
1077 	struct netpoll *netpoll = vlan->netpoll;
1078 
1079 	if (!netpoll)
1080 		return;
1081 
1082 	vlan->netpoll = NULL;
1083 
1084 	__netpoll_free_async(netpoll);
1085 }
1086 #endif	/* CONFIG_NET_POLL_CONTROLLER */
1087 
1088 static int macvlan_dev_get_iflink(const struct net_device *dev)
1089 {
1090 	struct macvlan_dev *vlan = netdev_priv(dev);
1091 
1092 	return vlan->lowerdev->ifindex;
1093 }
1094 
1095 static const struct ethtool_ops macvlan_ethtool_ops = {
1096 	.get_link		= ethtool_op_get_link,
1097 	.get_link_ksettings	= macvlan_ethtool_get_link_ksettings,
1098 	.get_drvinfo		= macvlan_ethtool_get_drvinfo,
1099 };
1100 
1101 static const struct net_device_ops macvlan_netdev_ops = {
1102 	.ndo_init		= macvlan_init,
1103 	.ndo_uninit		= macvlan_uninit,
1104 	.ndo_open		= macvlan_open,
1105 	.ndo_stop		= macvlan_stop,
1106 	.ndo_start_xmit		= macvlan_start_xmit,
1107 	.ndo_change_mtu		= macvlan_change_mtu,
1108 	.ndo_fix_features	= macvlan_fix_features,
1109 	.ndo_change_rx_flags	= macvlan_change_rx_flags,
1110 	.ndo_set_mac_address	= macvlan_set_mac_address,
1111 	.ndo_set_rx_mode	= macvlan_set_mac_lists,
1112 	.ndo_get_stats64	= macvlan_dev_get_stats64,
1113 	.ndo_validate_addr	= eth_validate_addr,
1114 	.ndo_vlan_rx_add_vid	= macvlan_vlan_rx_add_vid,
1115 	.ndo_vlan_rx_kill_vid	= macvlan_vlan_rx_kill_vid,
1116 	.ndo_fdb_add		= macvlan_fdb_add,
1117 	.ndo_fdb_del		= macvlan_fdb_del,
1118 	.ndo_fdb_dump		= ndo_dflt_fdb_dump,
1119 	.ndo_get_lock_subclass  = macvlan_get_nest_level,
1120 #ifdef CONFIG_NET_POLL_CONTROLLER
1121 	.ndo_poll_controller	= macvlan_dev_poll_controller,
1122 	.ndo_netpoll_setup	= macvlan_dev_netpoll_setup,
1123 	.ndo_netpoll_cleanup	= macvlan_dev_netpoll_cleanup,
1124 #endif
1125 	.ndo_get_iflink		= macvlan_dev_get_iflink,
1126 	.ndo_features_check	= passthru_features_check,
1127 };
1128 
1129 void macvlan_common_setup(struct net_device *dev)
1130 {
1131 	ether_setup(dev);
1132 
1133 	dev->min_mtu		= 0;
1134 	dev->max_mtu		= ETH_MAX_MTU;
1135 	dev->priv_flags	       &= ~IFF_TX_SKB_SHARING;
1136 	netif_keep_dst(dev);
1137 	dev->priv_flags	       |= IFF_UNICAST_FLT;
1138 	dev->netdev_ops		= &macvlan_netdev_ops;
1139 	dev->needs_free_netdev	= true;
1140 	dev->header_ops		= &macvlan_hard_header_ops;
1141 	dev->ethtool_ops	= &macvlan_ethtool_ops;
1142 }
1143 EXPORT_SYMBOL_GPL(macvlan_common_setup);
1144 
1145 static void macvlan_setup(struct net_device *dev)
1146 {
1147 	macvlan_common_setup(dev);
1148 	dev->priv_flags |= IFF_NO_QUEUE;
1149 }
1150 
1151 static int macvlan_port_create(struct net_device *dev)
1152 {
1153 	struct macvlan_port *port;
1154 	unsigned int i;
1155 	int err;
1156 
1157 	if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
1158 		return -EINVAL;
1159 
1160 	if (netdev_is_rx_handler_busy(dev))
1161 		return -EBUSY;
1162 
1163 	port = kzalloc(sizeof(*port), GFP_KERNEL);
1164 	if (port == NULL)
1165 		return -ENOMEM;
1166 
1167 	port->dev = dev;
1168 	ether_addr_copy(port->perm_addr, dev->dev_addr);
1169 	INIT_LIST_HEAD(&port->vlans);
1170 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1171 		INIT_HLIST_HEAD(&port->vlan_hash[i]);
1172 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1173 		INIT_HLIST_HEAD(&port->vlan_source_hash[i]);
1174 
1175 	skb_queue_head_init(&port->bc_queue);
1176 	INIT_WORK(&port->bc_work, macvlan_process_broadcast);
1177 
1178 	err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
1179 	if (err)
1180 		kfree(port);
1181 	else
1182 		dev->priv_flags |= IFF_MACVLAN_PORT;
1183 	return err;
1184 }
1185 
1186 static void macvlan_port_destroy(struct net_device *dev)
1187 {
1188 	struct macvlan_port *port = macvlan_port_get_rtnl(dev);
1189 	struct sk_buff *skb;
1190 
1191 	dev->priv_flags &= ~IFF_MACVLAN_PORT;
1192 	netdev_rx_handler_unregister(dev);
1193 
1194 	/* After this point, no packet can schedule bc_work anymore,
1195 	 * but we need to cancel it and purge left skbs if any.
1196 	 */
1197 	cancel_work_sync(&port->bc_work);
1198 
1199 	while ((skb = __skb_dequeue(&port->bc_queue))) {
1200 		const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
1201 
1202 		if (src)
1203 			dev_put(src->dev);
1204 
1205 		kfree_skb(skb);
1206 	}
1207 
1208 	/* If the lower device address has been changed by passthru
1209 	 * macvlan, put it back.
1210 	 */
1211 	if (macvlan_passthru(port) &&
1212 	    !ether_addr_equal(port->dev->dev_addr, port->perm_addr)) {
1213 		struct sockaddr sa;
1214 
1215 		sa.sa_family = port->dev->type;
1216 		memcpy(&sa.sa_data, port->perm_addr, port->dev->addr_len);
1217 		dev_set_mac_address(port->dev, &sa);
1218 	}
1219 
1220 	kfree(port);
1221 }
1222 
1223 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[],
1224 			    struct netlink_ext_ack *extack)
1225 {
1226 	if (tb[IFLA_ADDRESS]) {
1227 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1228 			return -EINVAL;
1229 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1230 			return -EADDRNOTAVAIL;
1231 	}
1232 
1233 	if (!data)
1234 		return 0;
1235 
1236 	if (data[IFLA_MACVLAN_FLAGS] &&
1237 	    nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
1238 		return -EINVAL;
1239 
1240 	if (data[IFLA_MACVLAN_MODE]) {
1241 		switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
1242 		case MACVLAN_MODE_PRIVATE:
1243 		case MACVLAN_MODE_VEPA:
1244 		case MACVLAN_MODE_BRIDGE:
1245 		case MACVLAN_MODE_PASSTHRU:
1246 		case MACVLAN_MODE_SOURCE:
1247 			break;
1248 		default:
1249 			return -EINVAL;
1250 		}
1251 	}
1252 
1253 	if (data[IFLA_MACVLAN_MACADDR_MODE]) {
1254 		switch (nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE])) {
1255 		case MACVLAN_MACADDR_ADD:
1256 		case MACVLAN_MACADDR_DEL:
1257 		case MACVLAN_MACADDR_FLUSH:
1258 		case MACVLAN_MACADDR_SET:
1259 			break;
1260 		default:
1261 			return -EINVAL;
1262 		}
1263 	}
1264 
1265 	if (data[IFLA_MACVLAN_MACADDR]) {
1266 		if (nla_len(data[IFLA_MACVLAN_MACADDR]) != ETH_ALEN)
1267 			return -EINVAL;
1268 
1269 		if (!is_valid_ether_addr(nla_data(data[IFLA_MACVLAN_MACADDR])))
1270 			return -EADDRNOTAVAIL;
1271 	}
1272 
1273 	if (data[IFLA_MACVLAN_MACADDR_COUNT])
1274 		return -EINVAL;
1275 
1276 	return 0;
1277 }
1278 
1279 /**
1280  * reconfigure list of remote source mac address
1281  * (only for macvlan devices in source mode)
1282  * Note regarding alignment: all netlink data is aligned to 4 Byte, which
1283  * suffices for both ether_addr_copy and ether_addr_equal_64bits usage.
1284  */
1285 static int macvlan_changelink_sources(struct macvlan_dev *vlan, u32 mode,
1286 				      struct nlattr *data[])
1287 {
1288 	char *addr = NULL;
1289 	int ret, rem, len;
1290 	struct nlattr *nla, *head;
1291 	struct macvlan_source_entry *entry;
1292 
1293 	if (data[IFLA_MACVLAN_MACADDR])
1294 		addr = nla_data(data[IFLA_MACVLAN_MACADDR]);
1295 
1296 	if (mode == MACVLAN_MACADDR_ADD) {
1297 		if (!addr)
1298 			return -EINVAL;
1299 
1300 		return macvlan_hash_add_source(vlan, addr);
1301 
1302 	} else if (mode == MACVLAN_MACADDR_DEL) {
1303 		if (!addr)
1304 			return -EINVAL;
1305 
1306 		entry = macvlan_hash_lookup_source(vlan, addr);
1307 		if (entry) {
1308 			macvlan_hash_del_source(entry);
1309 			vlan->macaddr_count--;
1310 		}
1311 	} else if (mode == MACVLAN_MACADDR_FLUSH) {
1312 		macvlan_flush_sources(vlan->port, vlan);
1313 	} else if (mode == MACVLAN_MACADDR_SET) {
1314 		macvlan_flush_sources(vlan->port, vlan);
1315 
1316 		if (addr) {
1317 			ret = macvlan_hash_add_source(vlan, addr);
1318 			if (ret)
1319 				return ret;
1320 		}
1321 
1322 		if (!data || !data[IFLA_MACVLAN_MACADDR_DATA])
1323 			return 0;
1324 
1325 		head = nla_data(data[IFLA_MACVLAN_MACADDR_DATA]);
1326 		len = nla_len(data[IFLA_MACVLAN_MACADDR_DATA]);
1327 
1328 		nla_for_each_attr(nla, head, len, rem) {
1329 			if (nla_type(nla) != IFLA_MACVLAN_MACADDR ||
1330 			    nla_len(nla) != ETH_ALEN)
1331 				continue;
1332 
1333 			addr = nla_data(nla);
1334 			ret = macvlan_hash_add_source(vlan, addr);
1335 			if (ret)
1336 				return ret;
1337 		}
1338 	} else {
1339 		return -EINVAL;
1340 	}
1341 
1342 	return 0;
1343 }
1344 
1345 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
1346 			   struct nlattr *tb[], struct nlattr *data[],
1347 			   struct netlink_ext_ack *extack)
1348 {
1349 	struct macvlan_dev *vlan = netdev_priv(dev);
1350 	struct macvlan_port *port;
1351 	struct net_device *lowerdev;
1352 	int err;
1353 	int macmode;
1354 	bool create = false;
1355 
1356 	if (!tb[IFLA_LINK])
1357 		return -EINVAL;
1358 
1359 	lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
1360 	if (lowerdev == NULL)
1361 		return -ENODEV;
1362 
1363 	/* When creating macvlans or macvtaps on top of other macvlans - use
1364 	 * the real device as the lowerdev.
1365 	 */
1366 	if (netif_is_macvlan(lowerdev))
1367 		lowerdev = macvlan_dev_real_dev(lowerdev);
1368 
1369 	if (!tb[IFLA_MTU])
1370 		dev->mtu = lowerdev->mtu;
1371 	else if (dev->mtu > lowerdev->mtu)
1372 		return -EINVAL;
1373 
1374 	/* MTU range: 68 - lowerdev->max_mtu */
1375 	dev->min_mtu = ETH_MIN_MTU;
1376 	dev->max_mtu = lowerdev->max_mtu;
1377 
1378 	if (!tb[IFLA_ADDRESS])
1379 		eth_hw_addr_random(dev);
1380 
1381 	if (!macvlan_port_exists(lowerdev)) {
1382 		err = macvlan_port_create(lowerdev);
1383 		if (err < 0)
1384 			return err;
1385 		create = true;
1386 	}
1387 	port = macvlan_port_get_rtnl(lowerdev);
1388 
1389 	/* Only 1 macvlan device can be created in passthru mode */
1390 	if (macvlan_passthru(port)) {
1391 		/* The macvlan port must be not created this time,
1392 		 * still goto destroy_macvlan_port for readability.
1393 		 */
1394 		err = -EINVAL;
1395 		goto destroy_macvlan_port;
1396 	}
1397 
1398 	vlan->lowerdev = lowerdev;
1399 	vlan->dev      = dev;
1400 	vlan->port     = port;
1401 	vlan->set_features = MACVLAN_FEATURES;
1402 	vlan->nest_level = dev_get_nest_level(lowerdev) + 1;
1403 
1404 	vlan->mode     = MACVLAN_MODE_VEPA;
1405 	if (data && data[IFLA_MACVLAN_MODE])
1406 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1407 
1408 	if (data && data[IFLA_MACVLAN_FLAGS])
1409 		vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1410 
1411 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
1412 		if (port->count) {
1413 			err = -EINVAL;
1414 			goto destroy_macvlan_port;
1415 		}
1416 		macvlan_set_passthru(port);
1417 		eth_hw_addr_inherit(dev, lowerdev);
1418 	}
1419 
1420 	if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1421 		if (vlan->mode != MACVLAN_MODE_SOURCE) {
1422 			err = -EINVAL;
1423 			goto destroy_macvlan_port;
1424 		}
1425 		macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1426 		err = macvlan_changelink_sources(vlan, macmode, data);
1427 		if (err)
1428 			goto destroy_macvlan_port;
1429 	}
1430 
1431 	err = register_netdevice(dev);
1432 	if (err < 0)
1433 		goto destroy_macvlan_port;
1434 
1435 	dev->priv_flags |= IFF_MACVLAN;
1436 	err = netdev_upper_dev_link(lowerdev, dev, extack);
1437 	if (err)
1438 		goto unregister_netdev;
1439 
1440 	list_add_tail_rcu(&vlan->list, &port->vlans);
1441 	netif_stacked_transfer_operstate(lowerdev, dev);
1442 	linkwatch_fire_event(dev);
1443 
1444 	return 0;
1445 
1446 unregister_netdev:
1447 	/* macvlan_uninit would free the macvlan port */
1448 	unregister_netdevice(dev);
1449 	return err;
1450 destroy_macvlan_port:
1451 	/* the macvlan port may be freed by macvlan_uninit when fail to register.
1452 	 * so we destroy the macvlan port only when it's valid.
1453 	 */
1454 	if (create && macvlan_port_get_rtnl(lowerdev))
1455 		macvlan_port_destroy(port->dev);
1456 	return err;
1457 }
1458 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
1459 
1460 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
1461 			   struct nlattr *tb[], struct nlattr *data[],
1462 			   struct netlink_ext_ack *extack)
1463 {
1464 	return macvlan_common_newlink(src_net, dev, tb, data, extack);
1465 }
1466 
1467 void macvlan_dellink(struct net_device *dev, struct list_head *head)
1468 {
1469 	struct macvlan_dev *vlan = netdev_priv(dev);
1470 
1471 	if (vlan->mode == MACVLAN_MODE_SOURCE)
1472 		macvlan_flush_sources(vlan->port, vlan);
1473 	list_del_rcu(&vlan->list);
1474 	unregister_netdevice_queue(dev, head);
1475 	netdev_upper_dev_unlink(vlan->lowerdev, dev);
1476 }
1477 EXPORT_SYMBOL_GPL(macvlan_dellink);
1478 
1479 static int macvlan_changelink(struct net_device *dev,
1480 			      struct nlattr *tb[], struct nlattr *data[],
1481 			      struct netlink_ext_ack *extack)
1482 {
1483 	struct macvlan_dev *vlan = netdev_priv(dev);
1484 	enum macvlan_mode mode;
1485 	bool set_mode = false;
1486 	enum macvlan_macaddr_mode macmode;
1487 	int ret;
1488 
1489 	/* Validate mode, but don't set yet: setting flags may fail. */
1490 	if (data && data[IFLA_MACVLAN_MODE]) {
1491 		set_mode = true;
1492 		mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1493 		/* Passthrough mode can't be set or cleared dynamically */
1494 		if ((mode == MACVLAN_MODE_PASSTHRU) !=
1495 		    (vlan->mode == MACVLAN_MODE_PASSTHRU))
1496 			return -EINVAL;
1497 		if (vlan->mode == MACVLAN_MODE_SOURCE &&
1498 		    vlan->mode != mode)
1499 			macvlan_flush_sources(vlan->port, vlan);
1500 	}
1501 
1502 	if (data && data[IFLA_MACVLAN_FLAGS]) {
1503 		__u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1504 		bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
1505 		if (macvlan_passthru(vlan->port) && promisc) {
1506 			int err;
1507 
1508 			if (flags & MACVLAN_FLAG_NOPROMISC)
1509 				err = dev_set_promiscuity(vlan->lowerdev, -1);
1510 			else
1511 				err = dev_set_promiscuity(vlan->lowerdev, 1);
1512 			if (err < 0)
1513 				return err;
1514 		}
1515 		vlan->flags = flags;
1516 	}
1517 	if (set_mode)
1518 		vlan->mode = mode;
1519 	if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1520 		if (vlan->mode != MACVLAN_MODE_SOURCE)
1521 			return -EINVAL;
1522 		macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1523 		ret = macvlan_changelink_sources(vlan, macmode, data);
1524 		if (ret)
1525 			return ret;
1526 	}
1527 	return 0;
1528 }
1529 
1530 static size_t macvlan_get_size_mac(const struct macvlan_dev *vlan)
1531 {
1532 	if (vlan->macaddr_count == 0)
1533 		return 0;
1534 	return nla_total_size(0) /* IFLA_MACVLAN_MACADDR_DATA */
1535 		+ vlan->macaddr_count * nla_total_size(sizeof(u8) * ETH_ALEN);
1536 }
1537 
1538 static size_t macvlan_get_size(const struct net_device *dev)
1539 {
1540 	struct macvlan_dev *vlan = netdev_priv(dev);
1541 
1542 	return (0
1543 		+ nla_total_size(4) /* IFLA_MACVLAN_MODE */
1544 		+ nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
1545 		+ nla_total_size(4) /* IFLA_MACVLAN_MACADDR_COUNT */
1546 		+ macvlan_get_size_mac(vlan) /* IFLA_MACVLAN_MACADDR */
1547 		);
1548 }
1549 
1550 static int macvlan_fill_info_macaddr(struct sk_buff *skb,
1551 				     const struct macvlan_dev *vlan,
1552 				     const int i)
1553 {
1554 	struct hlist_head *h = &vlan->port->vlan_source_hash[i];
1555 	struct macvlan_source_entry *entry;
1556 
1557 	hlist_for_each_entry_rcu(entry, h, hlist) {
1558 		if (entry->vlan != vlan)
1559 			continue;
1560 		if (nla_put(skb, IFLA_MACVLAN_MACADDR, ETH_ALEN, entry->addr))
1561 			return 1;
1562 	}
1563 	return 0;
1564 }
1565 
1566 static int macvlan_fill_info(struct sk_buff *skb,
1567 				const struct net_device *dev)
1568 {
1569 	struct macvlan_dev *vlan = netdev_priv(dev);
1570 	int i;
1571 	struct nlattr *nest;
1572 
1573 	if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
1574 		goto nla_put_failure;
1575 	if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
1576 		goto nla_put_failure;
1577 	if (nla_put_u32(skb, IFLA_MACVLAN_MACADDR_COUNT, vlan->macaddr_count))
1578 		goto nla_put_failure;
1579 	if (vlan->macaddr_count > 0) {
1580 		nest = nla_nest_start(skb, IFLA_MACVLAN_MACADDR_DATA);
1581 		if (nest == NULL)
1582 			goto nla_put_failure;
1583 
1584 		for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
1585 			if (macvlan_fill_info_macaddr(skb, vlan, i))
1586 				goto nla_put_failure;
1587 		}
1588 		nla_nest_end(skb, nest);
1589 	}
1590 	return 0;
1591 
1592 nla_put_failure:
1593 	return -EMSGSIZE;
1594 }
1595 
1596 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
1597 	[IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
1598 	[IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
1599 	[IFLA_MACVLAN_MACADDR_MODE] = { .type = NLA_U32 },
1600 	[IFLA_MACVLAN_MACADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1601 	[IFLA_MACVLAN_MACADDR_DATA] = { .type = NLA_NESTED },
1602 	[IFLA_MACVLAN_MACADDR_COUNT] = { .type = NLA_U32 },
1603 };
1604 
1605 int macvlan_link_register(struct rtnl_link_ops *ops)
1606 {
1607 	/* common fields */
1608 	ops->validate		= macvlan_validate;
1609 	ops->maxtype		= IFLA_MACVLAN_MAX;
1610 	ops->policy		= macvlan_policy;
1611 	ops->changelink		= macvlan_changelink;
1612 	ops->get_size		= macvlan_get_size;
1613 	ops->fill_info		= macvlan_fill_info;
1614 
1615 	return rtnl_link_register(ops);
1616 };
1617 EXPORT_SYMBOL_GPL(macvlan_link_register);
1618 
1619 static struct net *macvlan_get_link_net(const struct net_device *dev)
1620 {
1621 	return dev_net(macvlan_dev_real_dev(dev));
1622 }
1623 
1624 static struct rtnl_link_ops macvlan_link_ops = {
1625 	.kind		= "macvlan",
1626 	.setup		= macvlan_setup,
1627 	.newlink	= macvlan_newlink,
1628 	.dellink	= macvlan_dellink,
1629 	.get_link_net	= macvlan_get_link_net,
1630 	.priv_size      = sizeof(struct macvlan_dev),
1631 };
1632 
1633 static int macvlan_device_event(struct notifier_block *unused,
1634 				unsigned long event, void *ptr)
1635 {
1636 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1637 	struct macvlan_dev *vlan, *next;
1638 	struct macvlan_port *port;
1639 	LIST_HEAD(list_kill);
1640 
1641 	if (!macvlan_port_exists(dev))
1642 		return NOTIFY_DONE;
1643 
1644 	port = macvlan_port_get_rtnl(dev);
1645 
1646 	switch (event) {
1647 	case NETDEV_UP:
1648 	case NETDEV_CHANGE:
1649 		list_for_each_entry(vlan, &port->vlans, list)
1650 			netif_stacked_transfer_operstate(vlan->lowerdev,
1651 							 vlan->dev);
1652 		break;
1653 	case NETDEV_FEAT_CHANGE:
1654 		list_for_each_entry(vlan, &port->vlans, list) {
1655 			vlan->dev->gso_max_size = dev->gso_max_size;
1656 			vlan->dev->gso_max_segs = dev->gso_max_segs;
1657 			netdev_update_features(vlan->dev);
1658 		}
1659 		break;
1660 	case NETDEV_CHANGEMTU:
1661 		list_for_each_entry(vlan, &port->vlans, list) {
1662 			if (vlan->dev->mtu <= dev->mtu)
1663 				continue;
1664 			dev_set_mtu(vlan->dev, dev->mtu);
1665 		}
1666 		break;
1667 	case NETDEV_CHANGEADDR:
1668 		if (!macvlan_passthru(port))
1669 			return NOTIFY_DONE;
1670 
1671 		vlan = list_first_entry_or_null(&port->vlans,
1672 						struct macvlan_dev,
1673 						list);
1674 
1675 		if (macvlan_sync_address(vlan->dev, dev->dev_addr))
1676 			return NOTIFY_BAD;
1677 
1678 		break;
1679 	case NETDEV_UNREGISTER:
1680 		/* twiddle thumbs on netns device moves */
1681 		if (dev->reg_state != NETREG_UNREGISTERING)
1682 			break;
1683 
1684 		list_for_each_entry_safe(vlan, next, &port->vlans, list)
1685 			vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1686 		unregister_netdevice_many(&list_kill);
1687 		break;
1688 	case NETDEV_PRE_TYPE_CHANGE:
1689 		/* Forbid underlaying device to change its type. */
1690 		return NOTIFY_BAD;
1691 
1692 	case NETDEV_NOTIFY_PEERS:
1693 	case NETDEV_BONDING_FAILOVER:
1694 	case NETDEV_RESEND_IGMP:
1695 		/* Propagate to all vlans */
1696 		list_for_each_entry(vlan, &port->vlans, list)
1697 			call_netdevice_notifiers(event, vlan->dev);
1698 	}
1699 	return NOTIFY_DONE;
1700 }
1701 
1702 static struct notifier_block macvlan_notifier_block __read_mostly = {
1703 	.notifier_call	= macvlan_device_event,
1704 };
1705 
1706 static int __init macvlan_init_module(void)
1707 {
1708 	int err;
1709 
1710 	register_netdevice_notifier(&macvlan_notifier_block);
1711 
1712 	err = macvlan_link_register(&macvlan_link_ops);
1713 	if (err < 0)
1714 		goto err1;
1715 	return 0;
1716 err1:
1717 	unregister_netdevice_notifier(&macvlan_notifier_block);
1718 	return err;
1719 }
1720 
1721 static void __exit macvlan_cleanup_module(void)
1722 {
1723 	rtnl_link_unregister(&macvlan_link_ops);
1724 	unregister_netdevice_notifier(&macvlan_notifier_block);
1725 }
1726 
1727 module_init(macvlan_init_module);
1728 module_exit(macvlan_cleanup_module);
1729 
1730 MODULE_LICENSE("GPL");
1731 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1732 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1733 MODULE_ALIAS_RTNL_LINK("macvlan");
1734