xref: /openbmc/linux/drivers/net/macvlan.c (revision 588b48ca)
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_SIZE	(1 << BITS_PER_BYTE)
39 
40 struct macvlan_port {
41 	struct net_device	*dev;
42 	struct hlist_head	vlan_hash[MACVLAN_HASH_SIZE];
43 	struct list_head	vlans;
44 	struct rcu_head		rcu;
45 	struct sk_buff_head	bc_queue;
46 	struct work_struct	bc_work;
47 	bool 			passthru;
48 	int			count;
49 };
50 
51 struct macvlan_skb_cb {
52 	const struct macvlan_dev *src;
53 };
54 
55 #define MACVLAN_SKB_CB(__skb) ((struct macvlan_skb_cb *)&((__skb)->cb[0]))
56 
57 static void macvlan_port_destroy(struct net_device *dev);
58 
59 static struct macvlan_port *macvlan_port_get_rcu(const struct net_device *dev)
60 {
61 	return rcu_dereference(dev->rx_handler_data);
62 }
63 
64 static struct macvlan_port *macvlan_port_get_rtnl(const struct net_device *dev)
65 {
66 	return rtnl_dereference(dev->rx_handler_data);
67 }
68 
69 #define macvlan_port_exists(dev) (dev->priv_flags & IFF_MACVLAN_PORT)
70 
71 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
72 					       const unsigned char *addr)
73 {
74 	struct macvlan_dev *vlan;
75 
76 	hlist_for_each_entry_rcu(vlan, &port->vlan_hash[addr[5]], hlist) {
77 		if (ether_addr_equal_64bits(vlan->dev->dev_addr, addr))
78 			return vlan;
79 	}
80 	return NULL;
81 }
82 
83 static void macvlan_hash_add(struct macvlan_dev *vlan)
84 {
85 	struct macvlan_port *port = vlan->port;
86 	const unsigned char *addr = vlan->dev->dev_addr;
87 
88 	hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[addr[5]]);
89 }
90 
91 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync)
92 {
93 	hlist_del_rcu(&vlan->hlist);
94 	if (sync)
95 		synchronize_rcu();
96 }
97 
98 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
99 					const unsigned char *addr)
100 {
101 	macvlan_hash_del(vlan, true);
102 	/* Now that we are unhashed it is safe to change the device
103 	 * address without confusing packet delivery.
104 	 */
105 	memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
106 	macvlan_hash_add(vlan);
107 }
108 
109 static int macvlan_addr_busy(const struct macvlan_port *port,
110 				const unsigned char *addr)
111 {
112 	/* Test to see if the specified multicast address is
113 	 * currently in use by the underlying device or
114 	 * another macvlan.
115 	 */
116 	if (ether_addr_equal_64bits(port->dev->dev_addr, addr))
117 		return 1;
118 
119 	if (macvlan_hash_lookup(port, addr))
120 		return 1;
121 
122 	return 0;
123 }
124 
125 
126 static int macvlan_broadcast_one(struct sk_buff *skb,
127 				 const struct macvlan_dev *vlan,
128 				 const struct ethhdr *eth, bool local)
129 {
130 	struct net_device *dev = vlan->dev;
131 
132 	if (local)
133 		return __dev_forward_skb(dev, skb);
134 
135 	skb->dev = dev;
136 	if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
137 		skb->pkt_type = PACKET_BROADCAST;
138 	else
139 		skb->pkt_type = PACKET_MULTICAST;
140 
141 	return 0;
142 }
143 
144 static u32 macvlan_hash_mix(const struct macvlan_dev *vlan)
145 {
146 	return (u32)(((unsigned long)vlan) >> L1_CACHE_SHIFT);
147 }
148 
149 
150 static unsigned int mc_hash(const struct macvlan_dev *vlan,
151 			    const unsigned char *addr)
152 {
153 	u32 val = __get_unaligned_cpu32(addr + 2);
154 
155 	val ^= macvlan_hash_mix(vlan);
156 	return hash_32(val, MACVLAN_MC_FILTER_BITS);
157 }
158 
159 static void macvlan_broadcast(struct sk_buff *skb,
160 			      const struct macvlan_port *port,
161 			      struct net_device *src,
162 			      enum macvlan_mode mode)
163 {
164 	const struct ethhdr *eth = eth_hdr(skb);
165 	const struct macvlan_dev *vlan;
166 	struct sk_buff *nskb;
167 	unsigned int i;
168 	int err;
169 	unsigned int hash;
170 
171 	if (skb->protocol == htons(ETH_P_PAUSE))
172 		return;
173 
174 	for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
175 		hlist_for_each_entry_rcu(vlan, &port->vlan_hash[i], hlist) {
176 			if (vlan->dev == src || !(vlan->mode & mode))
177 				continue;
178 
179 			hash = mc_hash(vlan, eth->h_dest);
180 			if (!test_bit(hash, vlan->mc_filter))
181 				continue;
182 
183 			err = NET_RX_DROP;
184 			nskb = skb_clone(skb, GFP_ATOMIC);
185 			if (likely(nskb))
186 				err = macvlan_broadcast_one(
187 					nskb, vlan, eth,
188 					mode == MACVLAN_MODE_BRIDGE) ?:
189 				      netif_rx_ni(nskb);
190 			macvlan_count_rx(vlan, skb->len + ETH_HLEN,
191 					 err == NET_RX_SUCCESS, 1);
192 		}
193 	}
194 }
195 
196 static void macvlan_process_broadcast(struct work_struct *w)
197 {
198 	struct macvlan_port *port = container_of(w, struct macvlan_port,
199 						 bc_work);
200 	struct sk_buff *skb;
201 	struct sk_buff_head list;
202 
203 	skb_queue_head_init(&list);
204 
205 	spin_lock_bh(&port->bc_queue.lock);
206 	skb_queue_splice_tail_init(&port->bc_queue, &list);
207 	spin_unlock_bh(&port->bc_queue.lock);
208 
209 	while ((skb = __skb_dequeue(&list))) {
210 		const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
211 
212 		rcu_read_lock();
213 
214 		if (!src)
215 			/* frame comes from an external address */
216 			macvlan_broadcast(skb, port, NULL,
217 					  MACVLAN_MODE_PRIVATE |
218 					  MACVLAN_MODE_VEPA    |
219 					  MACVLAN_MODE_PASSTHRU|
220 					  MACVLAN_MODE_BRIDGE);
221 		else if (src->mode == MACVLAN_MODE_VEPA)
222 			/* flood to everyone except source */
223 			macvlan_broadcast(skb, port, src->dev,
224 					  MACVLAN_MODE_VEPA |
225 					  MACVLAN_MODE_BRIDGE);
226 		else
227 			/*
228 			 * flood only to VEPA ports, bridge ports
229 			 * already saw the frame on the way out.
230 			 */
231 			macvlan_broadcast(skb, port, src->dev,
232 					  MACVLAN_MODE_VEPA);
233 
234 		rcu_read_unlock();
235 
236 		kfree_skb(skb);
237 	}
238 }
239 
240 static void macvlan_broadcast_enqueue(struct macvlan_port *port,
241 				      struct sk_buff *skb)
242 {
243 	struct sk_buff *nskb;
244 	int err = -ENOMEM;
245 
246 	nskb = skb_clone(skb, GFP_ATOMIC);
247 	if (!nskb)
248 		goto err;
249 
250 	spin_lock(&port->bc_queue.lock);
251 	if (skb_queue_len(&port->bc_queue) < skb->dev->tx_queue_len) {
252 		__skb_queue_tail(&port->bc_queue, nskb);
253 		err = 0;
254 	}
255 	spin_unlock(&port->bc_queue.lock);
256 
257 	if (err)
258 		goto free_nskb;
259 
260 	schedule_work(&port->bc_work);
261 	return;
262 
263 free_nskb:
264 	kfree_skb(nskb);
265 err:
266 	atomic_long_inc(&skb->dev->rx_dropped);
267 }
268 
269 /* called under rcu_read_lock() from netif_receive_skb */
270 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
271 {
272 	struct macvlan_port *port;
273 	struct sk_buff *skb = *pskb;
274 	const struct ethhdr *eth = eth_hdr(skb);
275 	const struct macvlan_dev *vlan;
276 	const struct macvlan_dev *src;
277 	struct net_device *dev;
278 	unsigned int len = 0;
279 	int ret = NET_RX_DROP;
280 
281 	port = macvlan_port_get_rcu(skb->dev);
282 	if (is_multicast_ether_addr(eth->h_dest)) {
283 		skb = ip_check_defrag(skb, IP_DEFRAG_MACVLAN);
284 		if (!skb)
285 			return RX_HANDLER_CONSUMED;
286 		eth = eth_hdr(skb);
287 		src = macvlan_hash_lookup(port, eth->h_source);
288 		if (src && src->mode != MACVLAN_MODE_VEPA &&
289 		    src->mode != MACVLAN_MODE_BRIDGE) {
290 			/* forward to original port. */
291 			vlan = src;
292 			ret = macvlan_broadcast_one(skb, vlan, eth, 0) ?:
293 			      netif_rx(skb);
294 			goto out;
295 		}
296 
297 		MACVLAN_SKB_CB(skb)->src = src;
298 		macvlan_broadcast_enqueue(port, skb);
299 
300 		return RX_HANDLER_PASS;
301 	}
302 
303 	if (port->passthru)
304 		vlan = list_first_or_null_rcu(&port->vlans,
305 					      struct macvlan_dev, list);
306 	else
307 		vlan = macvlan_hash_lookup(port, eth->h_dest);
308 	if (vlan == NULL)
309 		return RX_HANDLER_PASS;
310 
311 	dev = vlan->dev;
312 	if (unlikely(!(dev->flags & IFF_UP))) {
313 		kfree_skb(skb);
314 		return RX_HANDLER_CONSUMED;
315 	}
316 	len = skb->len + ETH_HLEN;
317 	skb = skb_share_check(skb, GFP_ATOMIC);
318 	if (!skb)
319 		goto out;
320 
321 	skb->dev = dev;
322 	skb->pkt_type = PACKET_HOST;
323 
324 	ret = netif_rx(skb);
325 
326 out:
327 	macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, 0);
328 	return RX_HANDLER_CONSUMED;
329 }
330 
331 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
332 {
333 	const struct macvlan_dev *vlan = netdev_priv(dev);
334 	const struct macvlan_port *port = vlan->port;
335 	const struct macvlan_dev *dest;
336 
337 	if (vlan->mode == MACVLAN_MODE_BRIDGE) {
338 		const struct ethhdr *eth = (void *)skb->data;
339 
340 		/* send to other bridge ports directly */
341 		if (is_multicast_ether_addr(eth->h_dest)) {
342 			macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
343 			goto xmit_world;
344 		}
345 
346 		dest = macvlan_hash_lookup(port, eth->h_dest);
347 		if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
348 			/* send to lowerdev first for its network taps */
349 			dev_forward_skb(vlan->lowerdev, skb);
350 
351 			return NET_XMIT_SUCCESS;
352 		}
353 	}
354 
355 xmit_world:
356 	skb->dev = vlan->lowerdev;
357 	return dev_queue_xmit(skb);
358 }
359 
360 static inline netdev_tx_t macvlan_netpoll_send_skb(struct macvlan_dev *vlan, struct sk_buff *skb)
361 {
362 #ifdef CONFIG_NET_POLL_CONTROLLER
363 	if (vlan->netpoll)
364 		netpoll_send_skb(vlan->netpoll, skb);
365 #else
366 	BUG();
367 #endif
368 	return NETDEV_TX_OK;
369 }
370 
371 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
372 				      struct net_device *dev)
373 {
374 	unsigned int len = skb->len;
375 	int ret;
376 	struct macvlan_dev *vlan = netdev_priv(dev);
377 
378 	if (unlikely(netpoll_tx_running(dev)))
379 		return macvlan_netpoll_send_skb(vlan, skb);
380 
381 	if (vlan->fwd_priv) {
382 		skb->dev = vlan->lowerdev;
383 		ret = dev_queue_xmit_accel(skb, vlan->fwd_priv);
384 	} else {
385 		ret = macvlan_queue_xmit(skb, dev);
386 	}
387 
388 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
389 		struct vlan_pcpu_stats *pcpu_stats;
390 
391 		pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
392 		u64_stats_update_begin(&pcpu_stats->syncp);
393 		pcpu_stats->tx_packets++;
394 		pcpu_stats->tx_bytes += len;
395 		u64_stats_update_end(&pcpu_stats->syncp);
396 	} else {
397 		this_cpu_inc(vlan->pcpu_stats->tx_dropped);
398 	}
399 	return ret;
400 }
401 
402 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
403 			       unsigned short type, const void *daddr,
404 			       const void *saddr, unsigned len)
405 {
406 	const struct macvlan_dev *vlan = netdev_priv(dev);
407 	struct net_device *lowerdev = vlan->lowerdev;
408 
409 	return dev_hard_header(skb, lowerdev, type, daddr,
410 			       saddr ? : dev->dev_addr, len);
411 }
412 
413 static const struct header_ops macvlan_hard_header_ops = {
414 	.create  	= macvlan_hard_header,
415 	.rebuild	= eth_rebuild_header,
416 	.parse		= eth_header_parse,
417 	.cache		= eth_header_cache,
418 	.cache_update	= eth_header_cache_update,
419 };
420 
421 static struct rtnl_link_ops macvlan_link_ops;
422 
423 static int macvlan_open(struct net_device *dev)
424 {
425 	struct macvlan_dev *vlan = netdev_priv(dev);
426 	struct net_device *lowerdev = vlan->lowerdev;
427 	int err;
428 
429 	if (vlan->port->passthru) {
430 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
431 			err = dev_set_promiscuity(lowerdev, 1);
432 			if (err < 0)
433 				goto out;
434 		}
435 		goto hash_add;
436 	}
437 
438 	if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD &&
439 	    dev->rtnl_link_ops == &macvlan_link_ops) {
440 		vlan->fwd_priv =
441 		      lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
442 
443 		/* If we get a NULL pointer back, or if we get an error
444 		 * then we should just fall through to the non accelerated path
445 		 */
446 		if (IS_ERR_OR_NULL(vlan->fwd_priv)) {
447 			vlan->fwd_priv = NULL;
448 		} else
449 			return 0;
450 	}
451 
452 	err = -EBUSY;
453 	if (macvlan_addr_busy(vlan->port, dev->dev_addr))
454 		goto out;
455 
456 	err = dev_uc_add(lowerdev, dev->dev_addr);
457 	if (err < 0)
458 		goto out;
459 	if (dev->flags & IFF_ALLMULTI) {
460 		err = dev_set_allmulti(lowerdev, 1);
461 		if (err < 0)
462 			goto del_unicast;
463 	}
464 
465 hash_add:
466 	macvlan_hash_add(vlan);
467 	return 0;
468 
469 del_unicast:
470 	dev_uc_del(lowerdev, dev->dev_addr);
471 out:
472 	if (vlan->fwd_priv) {
473 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
474 							   vlan->fwd_priv);
475 		vlan->fwd_priv = NULL;
476 	}
477 	return err;
478 }
479 
480 static int macvlan_stop(struct net_device *dev)
481 {
482 	struct macvlan_dev *vlan = netdev_priv(dev);
483 	struct net_device *lowerdev = vlan->lowerdev;
484 
485 	if (vlan->fwd_priv) {
486 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
487 							   vlan->fwd_priv);
488 		vlan->fwd_priv = NULL;
489 		return 0;
490 	}
491 
492 	dev_uc_unsync(lowerdev, dev);
493 	dev_mc_unsync(lowerdev, dev);
494 
495 	if (vlan->port->passthru) {
496 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
497 			dev_set_promiscuity(lowerdev, -1);
498 		goto hash_del;
499 	}
500 
501 	if (dev->flags & IFF_ALLMULTI)
502 		dev_set_allmulti(lowerdev, -1);
503 
504 	dev_uc_del(lowerdev, dev->dev_addr);
505 
506 hash_del:
507 	macvlan_hash_del(vlan, !dev->dismantle);
508 	return 0;
509 }
510 
511 static int macvlan_sync_address(struct net_device *dev, unsigned char *addr)
512 {
513 	struct macvlan_dev *vlan = netdev_priv(dev);
514 	struct net_device *lowerdev = vlan->lowerdev;
515 	int err;
516 
517 	if (!(dev->flags & IFF_UP)) {
518 		/* Just copy in the new address */
519 		ether_addr_copy(dev->dev_addr, addr);
520 	} else {
521 		/* Rehash and update the device filters */
522 		if (macvlan_addr_busy(vlan->port, addr))
523 			return -EBUSY;
524 
525 		if (!vlan->port->passthru) {
526 			err = dev_uc_add(lowerdev, addr);
527 			if (err)
528 				return err;
529 
530 			dev_uc_del(lowerdev, dev->dev_addr);
531 		}
532 
533 		macvlan_hash_change_addr(vlan, addr);
534 	}
535 	return 0;
536 }
537 
538 static int macvlan_set_mac_address(struct net_device *dev, void *p)
539 {
540 	struct macvlan_dev *vlan = netdev_priv(dev);
541 	struct sockaddr *addr = p;
542 
543 	if (!is_valid_ether_addr(addr->sa_data))
544 		return -EADDRNOTAVAIL;
545 
546 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
547 		dev_set_mac_address(vlan->lowerdev, addr);
548 		return 0;
549 	}
550 
551 	return macvlan_sync_address(dev, addr->sa_data);
552 }
553 
554 static void macvlan_change_rx_flags(struct net_device *dev, int change)
555 {
556 	struct macvlan_dev *vlan = netdev_priv(dev);
557 	struct net_device *lowerdev = vlan->lowerdev;
558 
559 	if (dev->flags & IFF_UP) {
560 		if (change & IFF_ALLMULTI)
561 			dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
562 	}
563 }
564 
565 static void macvlan_set_mac_lists(struct net_device *dev)
566 {
567 	struct macvlan_dev *vlan = netdev_priv(dev);
568 
569 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
570 		bitmap_fill(vlan->mc_filter, MACVLAN_MC_FILTER_SZ);
571 	} else {
572 		struct netdev_hw_addr *ha;
573 		DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
574 
575 		bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
576 		netdev_for_each_mc_addr(ha, dev) {
577 			__set_bit(mc_hash(vlan, ha->addr), filter);
578 		}
579 
580 		__set_bit(mc_hash(vlan, dev->broadcast), filter);
581 
582 		bitmap_copy(vlan->mc_filter, filter, MACVLAN_MC_FILTER_SZ);
583 	}
584 	dev_uc_sync(vlan->lowerdev, dev);
585 	dev_mc_sync(vlan->lowerdev, dev);
586 }
587 
588 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
589 {
590 	struct macvlan_dev *vlan = netdev_priv(dev);
591 
592 	if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
593 		return -EINVAL;
594 	dev->mtu = new_mtu;
595 	return 0;
596 }
597 
598 /*
599  * macvlan network devices have devices nesting below it and are a special
600  * "super class" of normal network devices; split their locks off into a
601  * separate class since they always nest.
602  */
603 static struct lock_class_key macvlan_netdev_xmit_lock_key;
604 static struct lock_class_key macvlan_netdev_addr_lock_key;
605 
606 #define ALWAYS_ON_FEATURES \
607 	(NETIF_F_SG | NETIF_F_GEN_CSUM | NETIF_F_GSO_SOFTWARE | NETIF_F_LLTX)
608 
609 #define MACVLAN_FEATURES \
610 	(NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
611 	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
612 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
613 	 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
614 
615 #define MACVLAN_STATE_MASK \
616 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
617 
618 static int macvlan_get_nest_level(struct net_device *dev)
619 {
620 	return ((struct macvlan_dev *)netdev_priv(dev))->nest_level;
621 }
622 
623 static void macvlan_set_lockdep_class_one(struct net_device *dev,
624 					  struct netdev_queue *txq,
625 					  void *_unused)
626 {
627 	lockdep_set_class(&txq->_xmit_lock,
628 			  &macvlan_netdev_xmit_lock_key);
629 }
630 
631 static void macvlan_set_lockdep_class(struct net_device *dev)
632 {
633 	lockdep_set_class_and_subclass(&dev->addr_list_lock,
634 				       &macvlan_netdev_addr_lock_key,
635 				       macvlan_get_nest_level(dev));
636 	netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
637 }
638 
639 static int macvlan_init(struct net_device *dev)
640 {
641 	struct macvlan_dev *vlan = netdev_priv(dev);
642 	const struct net_device *lowerdev = vlan->lowerdev;
643 
644 	dev->state		= (dev->state & ~MACVLAN_STATE_MASK) |
645 				  (lowerdev->state & MACVLAN_STATE_MASK);
646 	dev->features 		= lowerdev->features & MACVLAN_FEATURES;
647 	dev->features		|= ALWAYS_ON_FEATURES;
648 	dev->vlan_features	= lowerdev->vlan_features & MACVLAN_FEATURES;
649 	dev->gso_max_size	= lowerdev->gso_max_size;
650 	dev->iflink		= lowerdev->ifindex;
651 	dev->hard_header_len	= lowerdev->hard_header_len;
652 
653 	macvlan_set_lockdep_class(dev);
654 
655 	vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
656 	if (!vlan->pcpu_stats)
657 		return -ENOMEM;
658 
659 	return 0;
660 }
661 
662 static void macvlan_uninit(struct net_device *dev)
663 {
664 	struct macvlan_dev *vlan = netdev_priv(dev);
665 	struct macvlan_port *port = vlan->port;
666 
667 	free_percpu(vlan->pcpu_stats);
668 
669 	port->count -= 1;
670 	if (!port->count)
671 		macvlan_port_destroy(port->dev);
672 }
673 
674 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
675 							 struct rtnl_link_stats64 *stats)
676 {
677 	struct macvlan_dev *vlan = netdev_priv(dev);
678 
679 	if (vlan->pcpu_stats) {
680 		struct vlan_pcpu_stats *p;
681 		u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
682 		u32 rx_errors = 0, tx_dropped = 0;
683 		unsigned int start;
684 		int i;
685 
686 		for_each_possible_cpu(i) {
687 			p = per_cpu_ptr(vlan->pcpu_stats, i);
688 			do {
689 				start = u64_stats_fetch_begin_irq(&p->syncp);
690 				rx_packets	= p->rx_packets;
691 				rx_bytes	= p->rx_bytes;
692 				rx_multicast	= p->rx_multicast;
693 				tx_packets	= p->tx_packets;
694 				tx_bytes	= p->tx_bytes;
695 			} while (u64_stats_fetch_retry_irq(&p->syncp, start));
696 
697 			stats->rx_packets	+= rx_packets;
698 			stats->rx_bytes		+= rx_bytes;
699 			stats->multicast	+= rx_multicast;
700 			stats->tx_packets	+= tx_packets;
701 			stats->tx_bytes		+= tx_bytes;
702 			/* rx_errors & tx_dropped are u32, updated
703 			 * without syncp protection.
704 			 */
705 			rx_errors	+= p->rx_errors;
706 			tx_dropped	+= p->tx_dropped;
707 		}
708 		stats->rx_errors	= rx_errors;
709 		stats->rx_dropped	= rx_errors;
710 		stats->tx_dropped	= tx_dropped;
711 	}
712 	return stats;
713 }
714 
715 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
716 				   __be16 proto, u16 vid)
717 {
718 	struct macvlan_dev *vlan = netdev_priv(dev);
719 	struct net_device *lowerdev = vlan->lowerdev;
720 
721 	return vlan_vid_add(lowerdev, proto, vid);
722 }
723 
724 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
725 				    __be16 proto, u16 vid)
726 {
727 	struct macvlan_dev *vlan = netdev_priv(dev);
728 	struct net_device *lowerdev = vlan->lowerdev;
729 
730 	vlan_vid_del(lowerdev, proto, vid);
731 	return 0;
732 }
733 
734 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
735 			   struct net_device *dev,
736 			   const unsigned char *addr,
737 			   u16 flags)
738 {
739 	struct macvlan_dev *vlan = netdev_priv(dev);
740 	int err = -EINVAL;
741 
742 	if (!vlan->port->passthru)
743 		return -EOPNOTSUPP;
744 
745 	if (flags & NLM_F_REPLACE)
746 		return -EOPNOTSUPP;
747 
748 	if (is_unicast_ether_addr(addr))
749 		err = dev_uc_add_excl(dev, addr);
750 	else if (is_multicast_ether_addr(addr))
751 		err = dev_mc_add_excl(dev, addr);
752 
753 	return err;
754 }
755 
756 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
757 			   struct net_device *dev,
758 			   const unsigned char *addr)
759 {
760 	struct macvlan_dev *vlan = netdev_priv(dev);
761 	int err = -EINVAL;
762 
763 	if (!vlan->port->passthru)
764 		return -EOPNOTSUPP;
765 
766 	if (is_unicast_ether_addr(addr))
767 		err = dev_uc_del(dev, addr);
768 	else if (is_multicast_ether_addr(addr))
769 		err = dev_mc_del(dev, addr);
770 
771 	return err;
772 }
773 
774 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
775 					struct ethtool_drvinfo *drvinfo)
776 {
777 	strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
778 	strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
779 }
780 
781 static int macvlan_ethtool_get_settings(struct net_device *dev,
782 					struct ethtool_cmd *cmd)
783 {
784 	const struct macvlan_dev *vlan = netdev_priv(dev);
785 
786 	return __ethtool_get_settings(vlan->lowerdev, cmd);
787 }
788 
789 static netdev_features_t macvlan_fix_features(struct net_device *dev,
790 					      netdev_features_t features)
791 {
792 	struct macvlan_dev *vlan = netdev_priv(dev);
793 	netdev_features_t mask;
794 
795 	features |= NETIF_F_ALL_FOR_ALL;
796 	features &= (vlan->set_features | ~MACVLAN_FEATURES);
797 	mask = features;
798 
799 	features = netdev_increment_features(vlan->lowerdev->features,
800 					     features,
801 					     mask);
802 	features |= ALWAYS_ON_FEATURES;
803 
804 	return features;
805 }
806 
807 #ifdef CONFIG_NET_POLL_CONTROLLER
808 static void macvlan_dev_poll_controller(struct net_device *dev)
809 {
810 	return;
811 }
812 
813 static int macvlan_dev_netpoll_setup(struct net_device *dev, struct netpoll_info *npinfo)
814 {
815 	struct macvlan_dev *vlan = netdev_priv(dev);
816 	struct net_device *real_dev = vlan->lowerdev;
817 	struct netpoll *netpoll;
818 	int err = 0;
819 
820 	netpoll = kzalloc(sizeof(*netpoll), GFP_KERNEL);
821 	err = -ENOMEM;
822 	if (!netpoll)
823 		goto out;
824 
825 	err = __netpoll_setup(netpoll, real_dev);
826 	if (err) {
827 		kfree(netpoll);
828 		goto out;
829 	}
830 
831 	vlan->netpoll = netpoll;
832 
833 out:
834 	return err;
835 }
836 
837 static void macvlan_dev_netpoll_cleanup(struct net_device *dev)
838 {
839 	struct macvlan_dev *vlan = netdev_priv(dev);
840 	struct netpoll *netpoll = vlan->netpoll;
841 
842 	if (!netpoll)
843 		return;
844 
845 	vlan->netpoll = NULL;
846 
847 	__netpoll_free_async(netpoll);
848 }
849 #endif	/* CONFIG_NET_POLL_CONTROLLER */
850 
851 static const struct ethtool_ops macvlan_ethtool_ops = {
852 	.get_link		= ethtool_op_get_link,
853 	.get_settings		= macvlan_ethtool_get_settings,
854 	.get_drvinfo		= macvlan_ethtool_get_drvinfo,
855 };
856 
857 static const struct net_device_ops macvlan_netdev_ops = {
858 	.ndo_init		= macvlan_init,
859 	.ndo_uninit		= macvlan_uninit,
860 	.ndo_open		= macvlan_open,
861 	.ndo_stop		= macvlan_stop,
862 	.ndo_start_xmit		= macvlan_start_xmit,
863 	.ndo_change_mtu		= macvlan_change_mtu,
864 	.ndo_fix_features	= macvlan_fix_features,
865 	.ndo_change_rx_flags	= macvlan_change_rx_flags,
866 	.ndo_set_mac_address	= macvlan_set_mac_address,
867 	.ndo_set_rx_mode	= macvlan_set_mac_lists,
868 	.ndo_get_stats64	= macvlan_dev_get_stats64,
869 	.ndo_validate_addr	= eth_validate_addr,
870 	.ndo_vlan_rx_add_vid	= macvlan_vlan_rx_add_vid,
871 	.ndo_vlan_rx_kill_vid	= macvlan_vlan_rx_kill_vid,
872 	.ndo_fdb_add		= macvlan_fdb_add,
873 	.ndo_fdb_del		= macvlan_fdb_del,
874 	.ndo_fdb_dump		= ndo_dflt_fdb_dump,
875 	.ndo_get_lock_subclass  = macvlan_get_nest_level,
876 #ifdef CONFIG_NET_POLL_CONTROLLER
877 	.ndo_poll_controller	= macvlan_dev_poll_controller,
878 	.ndo_netpoll_setup	= macvlan_dev_netpoll_setup,
879 	.ndo_netpoll_cleanup	= macvlan_dev_netpoll_cleanup,
880 #endif
881 };
882 
883 void macvlan_common_setup(struct net_device *dev)
884 {
885 	ether_setup(dev);
886 
887 	dev->priv_flags	       &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
888 	dev->priv_flags	       |= IFF_UNICAST_FLT;
889 	dev->netdev_ops		= &macvlan_netdev_ops;
890 	dev->destructor		= free_netdev;
891 	dev->header_ops		= &macvlan_hard_header_ops;
892 	dev->ethtool_ops	= &macvlan_ethtool_ops;
893 }
894 EXPORT_SYMBOL_GPL(macvlan_common_setup);
895 
896 static void macvlan_setup(struct net_device *dev)
897 {
898 	macvlan_common_setup(dev);
899 	dev->tx_queue_len	= 0;
900 }
901 
902 static int macvlan_port_create(struct net_device *dev)
903 {
904 	struct macvlan_port *port;
905 	unsigned int i;
906 	int err;
907 
908 	if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
909 		return -EINVAL;
910 
911 	port = kzalloc(sizeof(*port), GFP_KERNEL);
912 	if (port == NULL)
913 		return -ENOMEM;
914 
915 	port->passthru = false;
916 	port->dev = dev;
917 	INIT_LIST_HEAD(&port->vlans);
918 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
919 		INIT_HLIST_HEAD(&port->vlan_hash[i]);
920 
921 	skb_queue_head_init(&port->bc_queue);
922 	INIT_WORK(&port->bc_work, macvlan_process_broadcast);
923 
924 	err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
925 	if (err)
926 		kfree(port);
927 	else
928 		dev->priv_flags |= IFF_MACVLAN_PORT;
929 	return err;
930 }
931 
932 static void macvlan_port_destroy(struct net_device *dev)
933 {
934 	struct macvlan_port *port = macvlan_port_get_rtnl(dev);
935 
936 	cancel_work_sync(&port->bc_work);
937 	dev->priv_flags &= ~IFF_MACVLAN_PORT;
938 	netdev_rx_handler_unregister(dev);
939 	kfree_rcu(port, rcu);
940 }
941 
942 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
943 {
944 	if (tb[IFLA_ADDRESS]) {
945 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
946 			return -EINVAL;
947 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
948 			return -EADDRNOTAVAIL;
949 	}
950 
951 	if (data && data[IFLA_MACVLAN_FLAGS] &&
952 	    nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
953 		return -EINVAL;
954 
955 	if (data && data[IFLA_MACVLAN_MODE]) {
956 		switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
957 		case MACVLAN_MODE_PRIVATE:
958 		case MACVLAN_MODE_VEPA:
959 		case MACVLAN_MODE_BRIDGE:
960 		case MACVLAN_MODE_PASSTHRU:
961 			break;
962 		default:
963 			return -EINVAL;
964 		}
965 	}
966 	return 0;
967 }
968 
969 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
970 			   struct nlattr *tb[], struct nlattr *data[])
971 {
972 	struct macvlan_dev *vlan = netdev_priv(dev);
973 	struct macvlan_port *port;
974 	struct net_device *lowerdev;
975 	int err;
976 
977 	if (!tb[IFLA_LINK])
978 		return -EINVAL;
979 
980 	lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
981 	if (lowerdev == NULL)
982 		return -ENODEV;
983 
984 	/* When creating macvlans or macvtaps on top of other macvlans - use
985 	 * the real device as the lowerdev.
986 	 */
987 	if (netif_is_macvlan(lowerdev))
988 		lowerdev = macvlan_dev_real_dev(lowerdev);
989 
990 	if (!tb[IFLA_MTU])
991 		dev->mtu = lowerdev->mtu;
992 	else if (dev->mtu > lowerdev->mtu)
993 		return -EINVAL;
994 
995 	if (!tb[IFLA_ADDRESS])
996 		eth_hw_addr_random(dev);
997 
998 	if (!macvlan_port_exists(lowerdev)) {
999 		err = macvlan_port_create(lowerdev);
1000 		if (err < 0)
1001 			return err;
1002 	}
1003 	port = macvlan_port_get_rtnl(lowerdev);
1004 
1005 	/* Only 1 macvlan device can be created in passthru mode */
1006 	if (port->passthru)
1007 		return -EINVAL;
1008 
1009 	vlan->lowerdev = lowerdev;
1010 	vlan->dev      = dev;
1011 	vlan->port     = port;
1012 	vlan->set_features = MACVLAN_FEATURES;
1013 	vlan->nest_level = dev_get_nest_level(lowerdev, netif_is_macvlan) + 1;
1014 
1015 	vlan->mode     = MACVLAN_MODE_VEPA;
1016 	if (data && data[IFLA_MACVLAN_MODE])
1017 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1018 
1019 	if (data && data[IFLA_MACVLAN_FLAGS])
1020 		vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1021 
1022 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
1023 		if (port->count)
1024 			return -EINVAL;
1025 		port->passthru = true;
1026 		eth_hw_addr_inherit(dev, lowerdev);
1027 	}
1028 
1029 	port->count += 1;
1030 	err = register_netdevice(dev);
1031 	if (err < 0)
1032 		goto destroy_port;
1033 
1034 	dev->priv_flags |= IFF_MACVLAN;
1035 	err = netdev_upper_dev_link(lowerdev, dev);
1036 	if (err)
1037 		goto unregister_netdev;
1038 
1039 	list_add_tail_rcu(&vlan->list, &port->vlans);
1040 	netif_stacked_transfer_operstate(lowerdev, dev);
1041 
1042 	return 0;
1043 
1044 unregister_netdev:
1045 	unregister_netdevice(dev);
1046 destroy_port:
1047 	port->count -= 1;
1048 	if (!port->count)
1049 		macvlan_port_destroy(lowerdev);
1050 
1051 	return err;
1052 }
1053 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
1054 
1055 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
1056 			   struct nlattr *tb[], struct nlattr *data[])
1057 {
1058 	return macvlan_common_newlink(src_net, dev, tb, data);
1059 }
1060 
1061 void macvlan_dellink(struct net_device *dev, struct list_head *head)
1062 {
1063 	struct macvlan_dev *vlan = netdev_priv(dev);
1064 
1065 	list_del_rcu(&vlan->list);
1066 	unregister_netdevice_queue(dev, head);
1067 	netdev_upper_dev_unlink(vlan->lowerdev, dev);
1068 }
1069 EXPORT_SYMBOL_GPL(macvlan_dellink);
1070 
1071 static int macvlan_changelink(struct net_device *dev,
1072 		struct nlattr *tb[], struct nlattr *data[])
1073 {
1074 	struct macvlan_dev *vlan = netdev_priv(dev);
1075 	enum macvlan_mode mode;
1076 	bool set_mode = false;
1077 
1078 	/* Validate mode, but don't set yet: setting flags may fail. */
1079 	if (data && data[IFLA_MACVLAN_MODE]) {
1080 		set_mode = true;
1081 		mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1082 		/* Passthrough mode can't be set or cleared dynamically */
1083 		if ((mode == MACVLAN_MODE_PASSTHRU) !=
1084 		    (vlan->mode == MACVLAN_MODE_PASSTHRU))
1085 			return -EINVAL;
1086 	}
1087 
1088 	if (data && data[IFLA_MACVLAN_FLAGS]) {
1089 		__u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1090 		bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
1091 		if (vlan->port->passthru && promisc) {
1092 			int err;
1093 
1094 			if (flags & MACVLAN_FLAG_NOPROMISC)
1095 				err = dev_set_promiscuity(vlan->lowerdev, -1);
1096 			else
1097 				err = dev_set_promiscuity(vlan->lowerdev, 1);
1098 			if (err < 0)
1099 				return err;
1100 		}
1101 		vlan->flags = flags;
1102 	}
1103 	if (set_mode)
1104 		vlan->mode = mode;
1105 	return 0;
1106 }
1107 
1108 static size_t macvlan_get_size(const struct net_device *dev)
1109 {
1110 	return (0
1111 		+ nla_total_size(4) /* IFLA_MACVLAN_MODE */
1112 		+ nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
1113 		);
1114 }
1115 
1116 static int macvlan_fill_info(struct sk_buff *skb,
1117 				const struct net_device *dev)
1118 {
1119 	struct macvlan_dev *vlan = netdev_priv(dev);
1120 
1121 	if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
1122 		goto nla_put_failure;
1123 	if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
1124 		goto nla_put_failure;
1125 	return 0;
1126 
1127 nla_put_failure:
1128 	return -EMSGSIZE;
1129 }
1130 
1131 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
1132 	[IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
1133 	[IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
1134 };
1135 
1136 int macvlan_link_register(struct rtnl_link_ops *ops)
1137 {
1138 	/* common fields */
1139 	ops->priv_size		= sizeof(struct macvlan_dev);
1140 	ops->validate		= macvlan_validate;
1141 	ops->maxtype		= IFLA_MACVLAN_MAX;
1142 	ops->policy		= macvlan_policy;
1143 	ops->changelink		= macvlan_changelink;
1144 	ops->get_size		= macvlan_get_size;
1145 	ops->fill_info		= macvlan_fill_info;
1146 
1147 	return rtnl_link_register(ops);
1148 };
1149 EXPORT_SYMBOL_GPL(macvlan_link_register);
1150 
1151 static struct rtnl_link_ops macvlan_link_ops = {
1152 	.kind		= "macvlan",
1153 	.setup		= macvlan_setup,
1154 	.newlink	= macvlan_newlink,
1155 	.dellink	= macvlan_dellink,
1156 };
1157 
1158 static int macvlan_device_event(struct notifier_block *unused,
1159 				unsigned long event, void *ptr)
1160 {
1161 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1162 	struct macvlan_dev *vlan, *next;
1163 	struct macvlan_port *port;
1164 	LIST_HEAD(list_kill);
1165 
1166 	if (!macvlan_port_exists(dev))
1167 		return NOTIFY_DONE;
1168 
1169 	port = macvlan_port_get_rtnl(dev);
1170 
1171 	switch (event) {
1172 	case NETDEV_CHANGE:
1173 		list_for_each_entry(vlan, &port->vlans, list)
1174 			netif_stacked_transfer_operstate(vlan->lowerdev,
1175 							 vlan->dev);
1176 		break;
1177 	case NETDEV_FEAT_CHANGE:
1178 		list_for_each_entry(vlan, &port->vlans, list) {
1179 			vlan->dev->gso_max_size = dev->gso_max_size;
1180 			netdev_update_features(vlan->dev);
1181 		}
1182 		break;
1183 	case NETDEV_CHANGEMTU:
1184 		list_for_each_entry(vlan, &port->vlans, list) {
1185 			if (vlan->dev->mtu <= dev->mtu)
1186 				continue;
1187 			dev_set_mtu(vlan->dev, dev->mtu);
1188 		}
1189 		break;
1190 	case NETDEV_CHANGEADDR:
1191 		if (!port->passthru)
1192 			return NOTIFY_DONE;
1193 
1194 		vlan = list_first_entry_or_null(&port->vlans,
1195 						struct macvlan_dev,
1196 						list);
1197 
1198 		if (macvlan_sync_address(vlan->dev, dev->dev_addr))
1199 			return NOTIFY_BAD;
1200 
1201 		break;
1202 	case NETDEV_UNREGISTER:
1203 		/* twiddle thumbs on netns device moves */
1204 		if (dev->reg_state != NETREG_UNREGISTERING)
1205 			break;
1206 
1207 		list_for_each_entry_safe(vlan, next, &port->vlans, list)
1208 			vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1209 		unregister_netdevice_many(&list_kill);
1210 		break;
1211 	case NETDEV_PRE_TYPE_CHANGE:
1212 		/* Forbid underlaying device to change its type. */
1213 		return NOTIFY_BAD;
1214 
1215 	case NETDEV_NOTIFY_PEERS:
1216 	case NETDEV_BONDING_FAILOVER:
1217 	case NETDEV_RESEND_IGMP:
1218 		/* Propagate to all vlans */
1219 		list_for_each_entry(vlan, &port->vlans, list)
1220 			call_netdevice_notifiers(event, vlan->dev);
1221 	}
1222 	return NOTIFY_DONE;
1223 }
1224 
1225 static struct notifier_block macvlan_notifier_block __read_mostly = {
1226 	.notifier_call	= macvlan_device_event,
1227 };
1228 
1229 static int __init macvlan_init_module(void)
1230 {
1231 	int err;
1232 
1233 	register_netdevice_notifier(&macvlan_notifier_block);
1234 
1235 	err = macvlan_link_register(&macvlan_link_ops);
1236 	if (err < 0)
1237 		goto err1;
1238 	return 0;
1239 err1:
1240 	unregister_netdevice_notifier(&macvlan_notifier_block);
1241 	return err;
1242 }
1243 
1244 static void __exit macvlan_cleanup_module(void)
1245 {
1246 	rtnl_link_unregister(&macvlan_link_ops);
1247 	unregister_netdevice_notifier(&macvlan_notifier_block);
1248 }
1249 
1250 module_init(macvlan_init_module);
1251 module_exit(macvlan_cleanup_module);
1252 
1253 MODULE_LICENSE("GPL");
1254 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1255 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1256 MODULE_ALIAS_RTNL_LINK("macvlan");
1257