xref: /openbmc/linux/drivers/net/macvlan.c (revision 8a26af30)
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 };
49 
50 #define MACVLAN_PORT_IS_EMPTY(port)    list_empty(&port->vlans)
51 
52 struct macvlan_skb_cb {
53 	const struct macvlan_dev *src;
54 };
55 
56 #define MACVLAN_SKB_CB(__skb) ((struct macvlan_skb_cb *)&((__skb)->cb[0]))
57 
58 static void macvlan_port_destroy(struct net_device *dev);
59 
60 static struct macvlan_port *macvlan_port_get_rcu(const struct net_device *dev)
61 {
62 	return rcu_dereference(dev->rx_handler_data);
63 }
64 
65 static struct macvlan_port *macvlan_port_get_rtnl(const struct net_device *dev)
66 {
67 	return rtnl_dereference(dev->rx_handler_data);
68 }
69 
70 #define macvlan_port_exists(dev) (dev->priv_flags & IFF_MACVLAN_PORT)
71 
72 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
73 					       const unsigned char *addr)
74 {
75 	struct macvlan_dev *vlan;
76 
77 	hlist_for_each_entry_rcu(vlan, &port->vlan_hash[addr[5]], hlist) {
78 		if (ether_addr_equal_64bits(vlan->dev->dev_addr, addr))
79 			return vlan;
80 	}
81 	return NULL;
82 }
83 
84 static void macvlan_hash_add(struct macvlan_dev *vlan)
85 {
86 	struct macvlan_port *port = vlan->port;
87 	const unsigned char *addr = vlan->dev->dev_addr;
88 
89 	hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[addr[5]]);
90 }
91 
92 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync)
93 {
94 	hlist_del_rcu(&vlan->hlist);
95 	if (sync)
96 		synchronize_rcu();
97 }
98 
99 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
100 					const unsigned char *addr)
101 {
102 	macvlan_hash_del(vlan, true);
103 	/* Now that we are unhashed it is safe to change the device
104 	 * address without confusing packet delivery.
105 	 */
106 	memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
107 	macvlan_hash_add(vlan);
108 }
109 
110 static int macvlan_addr_busy(const struct macvlan_port *port,
111 				const unsigned char *addr)
112 {
113 	/* Test to see if the specified multicast address is
114 	 * currently in use by the underlying device or
115 	 * another macvlan.
116 	 */
117 	if (ether_addr_equal_64bits(port->dev->dev_addr, addr))
118 		return 1;
119 
120 	if (macvlan_hash_lookup(port, addr))
121 		return 1;
122 
123 	return 0;
124 }
125 
126 
127 static int macvlan_broadcast_one(struct sk_buff *skb,
128 				 const struct macvlan_dev *vlan,
129 				 const struct ethhdr *eth, bool local)
130 {
131 	struct net_device *dev = vlan->dev;
132 
133 	if (local)
134 		return __dev_forward_skb(dev, skb);
135 
136 	skb->dev = dev;
137 	if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
138 		skb->pkt_type = PACKET_BROADCAST;
139 	else
140 		skb->pkt_type = PACKET_MULTICAST;
141 
142 	return 0;
143 }
144 
145 static u32 macvlan_hash_mix(const struct macvlan_dev *vlan)
146 {
147 	return (u32)(((unsigned long)vlan) >> L1_CACHE_SHIFT);
148 }
149 
150 
151 static unsigned int mc_hash(const struct macvlan_dev *vlan,
152 			    const unsigned char *addr)
153 {
154 	u32 val = __get_unaligned_cpu32(addr + 2);
155 
156 	val ^= macvlan_hash_mix(vlan);
157 	return hash_32(val, MACVLAN_MC_FILTER_BITS);
158 }
159 
160 static void macvlan_broadcast(struct sk_buff *skb,
161 			      const struct macvlan_port *port,
162 			      struct net_device *src,
163 			      enum macvlan_mode mode)
164 {
165 	const struct ethhdr *eth = eth_hdr(skb);
166 	const struct macvlan_dev *vlan;
167 	struct sk_buff *nskb;
168 	unsigned int i;
169 	int err;
170 	unsigned int hash;
171 
172 	if (skb->protocol == htons(ETH_P_PAUSE))
173 		return;
174 
175 	for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
176 		hlist_for_each_entry_rcu(vlan, &port->vlan_hash[i], hlist) {
177 			if (vlan->dev == src || !(vlan->mode & mode))
178 				continue;
179 
180 			hash = mc_hash(vlan, eth->h_dest);
181 			if (!test_bit(hash, vlan->mc_filter))
182 				continue;
183 
184 			err = NET_RX_DROP;
185 			nskb = skb_clone(skb, GFP_ATOMIC);
186 			if (likely(nskb))
187 				err = macvlan_broadcast_one(
188 					nskb, vlan, eth,
189 					mode == MACVLAN_MODE_BRIDGE) ?:
190 				      netif_rx_ni(nskb);
191 			macvlan_count_rx(vlan, skb->len + ETH_HLEN,
192 					 err == NET_RX_SUCCESS, 1);
193 		}
194 	}
195 }
196 
197 static void macvlan_process_broadcast(struct work_struct *w)
198 {
199 	struct macvlan_port *port = container_of(w, struct macvlan_port,
200 						 bc_work);
201 	struct sk_buff *skb;
202 	struct sk_buff_head list;
203 
204 	skb_queue_head_init(&list);
205 
206 	spin_lock_bh(&port->bc_queue.lock);
207 	skb_queue_splice_tail_init(&port->bc_queue, &list);
208 	spin_unlock_bh(&port->bc_queue.lock);
209 
210 	while ((skb = __skb_dequeue(&list))) {
211 		const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
212 
213 		rcu_read_lock();
214 
215 		if (!src)
216 			/* frame comes from an external address */
217 			macvlan_broadcast(skb, port, NULL,
218 					  MACVLAN_MODE_PRIVATE |
219 					  MACVLAN_MODE_VEPA    |
220 					  MACVLAN_MODE_PASSTHRU|
221 					  MACVLAN_MODE_BRIDGE);
222 		else if (src->mode == MACVLAN_MODE_VEPA)
223 			/* flood to everyone except source */
224 			macvlan_broadcast(skb, port, src->dev,
225 					  MACVLAN_MODE_VEPA |
226 					  MACVLAN_MODE_BRIDGE);
227 		else
228 			/*
229 			 * flood only to VEPA ports, bridge ports
230 			 * already saw the frame on the way out.
231 			 */
232 			macvlan_broadcast(skb, port, src->dev,
233 					  MACVLAN_MODE_VEPA);
234 
235 		rcu_read_unlock();
236 
237 		kfree_skb(skb);
238 	}
239 }
240 
241 static void macvlan_broadcast_enqueue(struct macvlan_port *port,
242 				      struct sk_buff *skb)
243 {
244 	struct sk_buff *nskb;
245 	int err = -ENOMEM;
246 
247 	nskb = skb_clone(skb, GFP_ATOMIC);
248 	if (!nskb)
249 		goto err;
250 
251 	spin_lock(&port->bc_queue.lock);
252 	if (skb_queue_len(&port->bc_queue) < skb->dev->tx_queue_len) {
253 		__skb_queue_tail(&port->bc_queue, nskb);
254 		err = 0;
255 	}
256 	spin_unlock(&port->bc_queue.lock);
257 
258 	if (err)
259 		goto free_nskb;
260 
261 	schedule_work(&port->bc_work);
262 	return;
263 
264 free_nskb:
265 	kfree_skb(nskb);
266 err:
267 	atomic_long_inc(&skb->dev->rx_dropped);
268 }
269 
270 /* called under rcu_read_lock() from netif_receive_skb */
271 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
272 {
273 	struct macvlan_port *port;
274 	struct sk_buff *skb = *pskb;
275 	const struct ethhdr *eth = eth_hdr(skb);
276 	const struct macvlan_dev *vlan;
277 	const struct macvlan_dev *src;
278 	struct net_device *dev;
279 	unsigned int len = 0;
280 	int ret = NET_RX_DROP;
281 
282 	port = macvlan_port_get_rcu(skb->dev);
283 	if (is_multicast_ether_addr(eth->h_dest)) {
284 		skb = ip_check_defrag(skb, IP_DEFRAG_MACVLAN);
285 		if (!skb)
286 			return RX_HANDLER_CONSUMED;
287 		eth = eth_hdr(skb);
288 		src = macvlan_hash_lookup(port, eth->h_source);
289 		if (src && src->mode != MACVLAN_MODE_VEPA &&
290 		    src->mode != MACVLAN_MODE_BRIDGE) {
291 			/* forward to original port. */
292 			vlan = src;
293 			ret = macvlan_broadcast_one(skb, vlan, eth, 0) ?:
294 			      netif_rx(skb);
295 			goto out;
296 		}
297 
298 		MACVLAN_SKB_CB(skb)->src = src;
299 		macvlan_broadcast_enqueue(port, skb);
300 
301 		return RX_HANDLER_PASS;
302 	}
303 
304 	if (port->passthru)
305 		vlan = list_first_or_null_rcu(&port->vlans,
306 					      struct macvlan_dev, list);
307 	else
308 		vlan = macvlan_hash_lookup(port, eth->h_dest);
309 	if (vlan == NULL)
310 		return RX_HANDLER_PASS;
311 
312 	dev = vlan->dev;
313 	if (unlikely(!(dev->flags & IFF_UP))) {
314 		kfree_skb(skb);
315 		return RX_HANDLER_CONSUMED;
316 	}
317 	len = skb->len + ETH_HLEN;
318 	skb = skb_share_check(skb, GFP_ATOMIC);
319 	if (!skb)
320 		goto out;
321 
322 	skb->dev = dev;
323 	skb->pkt_type = PACKET_HOST;
324 
325 	ret = netif_rx(skb);
326 
327 out:
328 	macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, 0);
329 	return RX_HANDLER_CONSUMED;
330 }
331 
332 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
333 {
334 	const struct macvlan_dev *vlan = netdev_priv(dev);
335 	const struct macvlan_port *port = vlan->port;
336 	const struct macvlan_dev *dest;
337 
338 	if (vlan->mode == MACVLAN_MODE_BRIDGE) {
339 		const struct ethhdr *eth = (void *)skb->data;
340 
341 		/* send to other bridge ports directly */
342 		if (is_multicast_ether_addr(eth->h_dest)) {
343 			macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
344 			goto xmit_world;
345 		}
346 
347 		dest = macvlan_hash_lookup(port, eth->h_dest);
348 		if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
349 			/* send to lowerdev first for its network taps */
350 			dev_forward_skb(vlan->lowerdev, skb);
351 
352 			return NET_XMIT_SUCCESS;
353 		}
354 	}
355 
356 xmit_world:
357 	skb->dev = vlan->lowerdev;
358 	return dev_queue_xmit(skb);
359 }
360 
361 static inline netdev_tx_t macvlan_netpoll_send_skb(struct macvlan_dev *vlan, struct sk_buff *skb)
362 {
363 #ifdef CONFIG_NET_POLL_CONTROLLER
364 	if (vlan->netpoll)
365 		netpoll_send_skb(vlan->netpoll, skb);
366 #else
367 	BUG();
368 #endif
369 	return NETDEV_TX_OK;
370 }
371 
372 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
373 				      struct net_device *dev)
374 {
375 	unsigned int len = skb->len;
376 	int ret;
377 	struct macvlan_dev *vlan = netdev_priv(dev);
378 
379 	if (unlikely(netpoll_tx_running(dev)))
380 		return macvlan_netpoll_send_skb(vlan, skb);
381 
382 	if (vlan->fwd_priv) {
383 		skb->dev = vlan->lowerdev;
384 		ret = dev_queue_xmit_accel(skb, vlan->fwd_priv);
385 	} else {
386 		ret = macvlan_queue_xmit(skb, dev);
387 	}
388 
389 	if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
390 		struct vlan_pcpu_stats *pcpu_stats;
391 
392 		pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
393 		u64_stats_update_begin(&pcpu_stats->syncp);
394 		pcpu_stats->tx_packets++;
395 		pcpu_stats->tx_bytes += len;
396 		u64_stats_update_end(&pcpu_stats->syncp);
397 	} else {
398 		this_cpu_inc(vlan->pcpu_stats->tx_dropped);
399 	}
400 	return ret;
401 }
402 
403 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
404 			       unsigned short type, const void *daddr,
405 			       const void *saddr, unsigned len)
406 {
407 	const struct macvlan_dev *vlan = netdev_priv(dev);
408 	struct net_device *lowerdev = vlan->lowerdev;
409 
410 	return dev_hard_header(skb, lowerdev, type, daddr,
411 			       saddr ? : dev->dev_addr, len);
412 }
413 
414 static const struct header_ops macvlan_hard_header_ops = {
415 	.create  	= macvlan_hard_header,
416 	.rebuild	= eth_rebuild_header,
417 	.parse		= eth_header_parse,
418 	.cache		= eth_header_cache,
419 	.cache_update	= eth_header_cache_update,
420 };
421 
422 static struct rtnl_link_ops macvlan_link_ops;
423 
424 static int macvlan_open(struct net_device *dev)
425 {
426 	struct macvlan_dev *vlan = netdev_priv(dev);
427 	struct net_device *lowerdev = vlan->lowerdev;
428 	int err;
429 
430 	if (vlan->port->passthru) {
431 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
432 			err = dev_set_promiscuity(lowerdev, 1);
433 			if (err < 0)
434 				goto out;
435 		}
436 		goto hash_add;
437 	}
438 
439 	if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD &&
440 	    dev->rtnl_link_ops == &macvlan_link_ops) {
441 		vlan->fwd_priv =
442 		      lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
443 
444 		/* If we get a NULL pointer back, or if we get an error
445 		 * then we should just fall through to the non accelerated path
446 		 */
447 		if (IS_ERR_OR_NULL(vlan->fwd_priv)) {
448 			vlan->fwd_priv = NULL;
449 		} else
450 			return 0;
451 	}
452 
453 	err = -EBUSY;
454 	if (macvlan_addr_busy(vlan->port, dev->dev_addr))
455 		goto out;
456 
457 	err = dev_uc_add(lowerdev, dev->dev_addr);
458 	if (err < 0)
459 		goto out;
460 	if (dev->flags & IFF_ALLMULTI) {
461 		err = dev_set_allmulti(lowerdev, 1);
462 		if (err < 0)
463 			goto del_unicast;
464 	}
465 
466 hash_add:
467 	macvlan_hash_add(vlan);
468 	return 0;
469 
470 del_unicast:
471 	dev_uc_del(lowerdev, dev->dev_addr);
472 out:
473 	if (vlan->fwd_priv) {
474 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
475 							   vlan->fwd_priv);
476 		vlan->fwd_priv = NULL;
477 	}
478 	return err;
479 }
480 
481 static int macvlan_stop(struct net_device *dev)
482 {
483 	struct macvlan_dev *vlan = netdev_priv(dev);
484 	struct net_device *lowerdev = vlan->lowerdev;
485 
486 	if (vlan->fwd_priv) {
487 		lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
488 							   vlan->fwd_priv);
489 		vlan->fwd_priv = NULL;
490 		return 0;
491 	}
492 
493 	dev_uc_unsync(lowerdev, dev);
494 	dev_mc_unsync(lowerdev, dev);
495 
496 	if (vlan->port->passthru) {
497 		if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
498 			dev_set_promiscuity(lowerdev, -1);
499 		goto hash_del;
500 	}
501 
502 	if (dev->flags & IFF_ALLMULTI)
503 		dev_set_allmulti(lowerdev, -1);
504 
505 	dev_uc_del(lowerdev, dev->dev_addr);
506 
507 hash_del:
508 	macvlan_hash_del(vlan, !dev->dismantle);
509 	return 0;
510 }
511 
512 static int macvlan_sync_address(struct net_device *dev, unsigned char *addr)
513 {
514 	struct macvlan_dev *vlan = netdev_priv(dev);
515 	struct net_device *lowerdev = vlan->lowerdev;
516 	int err;
517 
518 	if (!(dev->flags & IFF_UP)) {
519 		/* Just copy in the new address */
520 		ether_addr_copy(dev->dev_addr, addr);
521 	} else {
522 		/* Rehash and update the device filters */
523 		if (macvlan_addr_busy(vlan->port, addr))
524 			return -EBUSY;
525 
526 		if (!vlan->port->passthru) {
527 			err = dev_uc_add(lowerdev, addr);
528 			if (err)
529 				return err;
530 
531 			dev_uc_del(lowerdev, dev->dev_addr);
532 		}
533 
534 		macvlan_hash_change_addr(vlan, addr);
535 	}
536 	return 0;
537 }
538 
539 static int macvlan_set_mac_address(struct net_device *dev, void *p)
540 {
541 	struct macvlan_dev *vlan = netdev_priv(dev);
542 	struct sockaddr *addr = p;
543 
544 	if (!is_valid_ether_addr(addr->sa_data))
545 		return -EADDRNOTAVAIL;
546 
547 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
548 		dev_set_mac_address(vlan->lowerdev, addr);
549 		return 0;
550 	}
551 
552 	return macvlan_sync_address(dev, addr->sa_data);
553 }
554 
555 static void macvlan_change_rx_flags(struct net_device *dev, int change)
556 {
557 	struct macvlan_dev *vlan = netdev_priv(dev);
558 	struct net_device *lowerdev = vlan->lowerdev;
559 
560 	if (dev->flags & IFF_UP) {
561 		if (change & IFF_ALLMULTI)
562 			dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
563 	}
564 }
565 
566 static void macvlan_set_mac_lists(struct net_device *dev)
567 {
568 	struct macvlan_dev *vlan = netdev_priv(dev);
569 
570 	if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
571 		bitmap_fill(vlan->mc_filter, MACVLAN_MC_FILTER_SZ);
572 	} else {
573 		struct netdev_hw_addr *ha;
574 		DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
575 
576 		bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
577 		netdev_for_each_mc_addr(ha, dev) {
578 			__set_bit(mc_hash(vlan, ha->addr), filter);
579 		}
580 
581 		__set_bit(mc_hash(vlan, dev->broadcast), filter);
582 
583 		bitmap_copy(vlan->mc_filter, filter, MACVLAN_MC_FILTER_SZ);
584 	}
585 	dev_uc_sync(vlan->lowerdev, dev);
586 	dev_mc_sync(vlan->lowerdev, dev);
587 }
588 
589 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
590 {
591 	struct macvlan_dev *vlan = netdev_priv(dev);
592 
593 	if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
594 		return -EINVAL;
595 	dev->mtu = new_mtu;
596 	return 0;
597 }
598 
599 /*
600  * macvlan network devices have devices nesting below it and are a special
601  * "super class" of normal network devices; split their locks off into a
602  * separate class since they always nest.
603  */
604 static struct lock_class_key macvlan_netdev_xmit_lock_key;
605 static struct lock_class_key macvlan_netdev_addr_lock_key;
606 
607 #define ALWAYS_ON_FEATURES \
608 	(NETIF_F_SG | NETIF_F_GEN_CSUM | NETIF_F_GSO_SOFTWARE | NETIF_F_LLTX)
609 
610 #define MACVLAN_FEATURES \
611 	(NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
612 	 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
613 	 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
614 	 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
615 
616 #define MACVLAN_STATE_MASK \
617 	((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
618 
619 static int macvlan_get_nest_level(struct net_device *dev)
620 {
621 	return ((struct macvlan_dev *)netdev_priv(dev))->nest_level;
622 }
623 
624 static void macvlan_set_lockdep_class_one(struct net_device *dev,
625 					  struct netdev_queue *txq,
626 					  void *_unused)
627 {
628 	lockdep_set_class(&txq->_xmit_lock,
629 			  &macvlan_netdev_xmit_lock_key);
630 }
631 
632 static void macvlan_set_lockdep_class(struct net_device *dev)
633 {
634 	lockdep_set_class_and_subclass(&dev->addr_list_lock,
635 				       &macvlan_netdev_addr_lock_key,
636 				       macvlan_get_nest_level(dev));
637 	netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
638 }
639 
640 static int macvlan_init(struct net_device *dev)
641 {
642 	struct macvlan_dev *vlan = netdev_priv(dev);
643 	const struct net_device *lowerdev = vlan->lowerdev;
644 
645 	dev->state		= (dev->state & ~MACVLAN_STATE_MASK) |
646 				  (lowerdev->state & MACVLAN_STATE_MASK);
647 	dev->features 		= lowerdev->features & MACVLAN_FEATURES;
648 	dev->features		|= ALWAYS_ON_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 	if (MACVLAN_PORT_IS_EMPTY(port))
670 		macvlan_port_destroy(port->dev);
671 }
672 
673 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
674 							 struct rtnl_link_stats64 *stats)
675 {
676 	struct macvlan_dev *vlan = netdev_priv(dev);
677 
678 	if (vlan->pcpu_stats) {
679 		struct vlan_pcpu_stats *p;
680 		u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
681 		u32 rx_errors = 0, tx_dropped = 0;
682 		unsigned int start;
683 		int i;
684 
685 		for_each_possible_cpu(i) {
686 			p = per_cpu_ptr(vlan->pcpu_stats, i);
687 			do {
688 				start = u64_stats_fetch_begin_irq(&p->syncp);
689 				rx_packets	= p->rx_packets;
690 				rx_bytes	= p->rx_bytes;
691 				rx_multicast	= p->rx_multicast;
692 				tx_packets	= p->tx_packets;
693 				tx_bytes	= p->tx_bytes;
694 			} while (u64_stats_fetch_retry_irq(&p->syncp, start));
695 
696 			stats->rx_packets	+= rx_packets;
697 			stats->rx_bytes		+= rx_bytes;
698 			stats->multicast	+= rx_multicast;
699 			stats->tx_packets	+= tx_packets;
700 			stats->tx_bytes		+= tx_bytes;
701 			/* rx_errors & tx_dropped are u32, updated
702 			 * without syncp protection.
703 			 */
704 			rx_errors	+= p->rx_errors;
705 			tx_dropped	+= p->tx_dropped;
706 		}
707 		stats->rx_errors	= rx_errors;
708 		stats->rx_dropped	= rx_errors;
709 		stats->tx_dropped	= tx_dropped;
710 	}
711 	return stats;
712 }
713 
714 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
715 				   __be16 proto, u16 vid)
716 {
717 	struct macvlan_dev *vlan = netdev_priv(dev);
718 	struct net_device *lowerdev = vlan->lowerdev;
719 
720 	return vlan_vid_add(lowerdev, proto, vid);
721 }
722 
723 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
724 				    __be16 proto, u16 vid)
725 {
726 	struct macvlan_dev *vlan = netdev_priv(dev);
727 	struct net_device *lowerdev = vlan->lowerdev;
728 
729 	vlan_vid_del(lowerdev, proto, vid);
730 	return 0;
731 }
732 
733 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
734 			   struct net_device *dev,
735 			   const unsigned char *addr,
736 			   u16 flags)
737 {
738 	struct macvlan_dev *vlan = netdev_priv(dev);
739 	int err = -EINVAL;
740 
741 	if (!vlan->port->passthru)
742 		return -EOPNOTSUPP;
743 
744 	if (flags & NLM_F_REPLACE)
745 		return -EOPNOTSUPP;
746 
747 	if (is_unicast_ether_addr(addr))
748 		err = dev_uc_add_excl(dev, addr);
749 	else if (is_multicast_ether_addr(addr))
750 		err = dev_mc_add_excl(dev, addr);
751 
752 	return err;
753 }
754 
755 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
756 			   struct net_device *dev,
757 			   const unsigned char *addr)
758 {
759 	struct macvlan_dev *vlan = netdev_priv(dev);
760 	int err = -EINVAL;
761 
762 	if (!vlan->port->passthru)
763 		return -EOPNOTSUPP;
764 
765 	if (is_unicast_ether_addr(addr))
766 		err = dev_uc_del(dev, addr);
767 	else if (is_multicast_ether_addr(addr))
768 		err = dev_mc_del(dev, addr);
769 
770 	return err;
771 }
772 
773 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
774 					struct ethtool_drvinfo *drvinfo)
775 {
776 	strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
777 	strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
778 }
779 
780 static int macvlan_ethtool_get_settings(struct net_device *dev,
781 					struct ethtool_cmd *cmd)
782 {
783 	const struct macvlan_dev *vlan = netdev_priv(dev);
784 
785 	return __ethtool_get_settings(vlan->lowerdev, cmd);
786 }
787 
788 static netdev_features_t macvlan_fix_features(struct net_device *dev,
789 					      netdev_features_t features)
790 {
791 	struct macvlan_dev *vlan = netdev_priv(dev);
792 	netdev_features_t mask;
793 
794 	features |= NETIF_F_ALL_FOR_ALL;
795 	features &= (vlan->set_features | ~MACVLAN_FEATURES);
796 	mask = features;
797 
798 	features = netdev_increment_features(vlan->lowerdev->features,
799 					     features,
800 					     mask);
801 	features |= ALWAYS_ON_FEATURES;
802 
803 	return features;
804 }
805 
806 #ifdef CONFIG_NET_POLL_CONTROLLER
807 static void macvlan_dev_poll_controller(struct net_device *dev)
808 {
809 	return;
810 }
811 
812 static int macvlan_dev_netpoll_setup(struct net_device *dev, struct netpoll_info *npinfo)
813 {
814 	struct macvlan_dev *vlan = netdev_priv(dev);
815 	struct net_device *real_dev = vlan->lowerdev;
816 	struct netpoll *netpoll;
817 	int err = 0;
818 
819 	netpoll = kzalloc(sizeof(*netpoll), GFP_KERNEL);
820 	err = -ENOMEM;
821 	if (!netpoll)
822 		goto out;
823 
824 	err = __netpoll_setup(netpoll, real_dev);
825 	if (err) {
826 		kfree(netpoll);
827 		goto out;
828 	}
829 
830 	vlan->netpoll = netpoll;
831 
832 out:
833 	return err;
834 }
835 
836 static void macvlan_dev_netpoll_cleanup(struct net_device *dev)
837 {
838 	struct macvlan_dev *vlan = netdev_priv(dev);
839 	struct netpoll *netpoll = vlan->netpoll;
840 
841 	if (!netpoll)
842 		return;
843 
844 	vlan->netpoll = NULL;
845 
846 	__netpoll_free_async(netpoll);
847 }
848 #endif	/* CONFIG_NET_POLL_CONTROLLER */
849 
850 static const struct ethtool_ops macvlan_ethtool_ops = {
851 	.get_link		= ethtool_op_get_link,
852 	.get_settings		= macvlan_ethtool_get_settings,
853 	.get_drvinfo		= macvlan_ethtool_get_drvinfo,
854 };
855 
856 static const struct net_device_ops macvlan_netdev_ops = {
857 	.ndo_init		= macvlan_init,
858 	.ndo_uninit		= macvlan_uninit,
859 	.ndo_open		= macvlan_open,
860 	.ndo_stop		= macvlan_stop,
861 	.ndo_start_xmit		= macvlan_start_xmit,
862 	.ndo_change_mtu		= macvlan_change_mtu,
863 	.ndo_fix_features	= macvlan_fix_features,
864 	.ndo_change_rx_flags	= macvlan_change_rx_flags,
865 	.ndo_set_mac_address	= macvlan_set_mac_address,
866 	.ndo_set_rx_mode	= macvlan_set_mac_lists,
867 	.ndo_get_stats64	= macvlan_dev_get_stats64,
868 	.ndo_validate_addr	= eth_validate_addr,
869 	.ndo_vlan_rx_add_vid	= macvlan_vlan_rx_add_vid,
870 	.ndo_vlan_rx_kill_vid	= macvlan_vlan_rx_kill_vid,
871 	.ndo_fdb_add		= macvlan_fdb_add,
872 	.ndo_fdb_del		= macvlan_fdb_del,
873 	.ndo_fdb_dump		= ndo_dflt_fdb_dump,
874 	.ndo_get_lock_subclass  = macvlan_get_nest_level,
875 #ifdef CONFIG_NET_POLL_CONTROLLER
876 	.ndo_poll_controller	= macvlan_dev_poll_controller,
877 	.ndo_netpoll_setup	= macvlan_dev_netpoll_setup,
878 	.ndo_netpoll_cleanup	= macvlan_dev_netpoll_cleanup,
879 #endif
880 };
881 
882 void macvlan_common_setup(struct net_device *dev)
883 {
884 	ether_setup(dev);
885 
886 	dev->priv_flags	       &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
887 	dev->priv_flags	       |= IFF_UNICAST_FLT;
888 	dev->netdev_ops		= &macvlan_netdev_ops;
889 	dev->destructor		= free_netdev;
890 	dev->header_ops		= &macvlan_hard_header_ops;
891 	dev->ethtool_ops	= &macvlan_ethtool_ops;
892 }
893 EXPORT_SYMBOL_GPL(macvlan_common_setup);
894 
895 static void macvlan_setup(struct net_device *dev)
896 {
897 	macvlan_common_setup(dev);
898 	dev->tx_queue_len	= 0;
899 }
900 
901 static int macvlan_port_create(struct net_device *dev)
902 {
903 	struct macvlan_port *port;
904 	unsigned int i;
905 	int err;
906 
907 	if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
908 		return -EINVAL;
909 
910 	port = kzalloc(sizeof(*port), GFP_KERNEL);
911 	if (port == NULL)
912 		return -ENOMEM;
913 
914 	port->passthru = false;
915 	port->dev = dev;
916 	INIT_LIST_HEAD(&port->vlans);
917 	for (i = 0; i < MACVLAN_HASH_SIZE; i++)
918 		INIT_HLIST_HEAD(&port->vlan_hash[i]);
919 
920 	skb_queue_head_init(&port->bc_queue);
921 	INIT_WORK(&port->bc_work, macvlan_process_broadcast);
922 
923 	err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
924 	if (err)
925 		kfree(port);
926 	else
927 		dev->priv_flags |= IFF_MACVLAN_PORT;
928 	return err;
929 }
930 
931 static void macvlan_port_destroy(struct net_device *dev)
932 {
933 	struct macvlan_port *port = macvlan_port_get_rtnl(dev);
934 
935 	cancel_work_sync(&port->bc_work);
936 	dev->priv_flags &= ~IFF_MACVLAN_PORT;
937 	netdev_rx_handler_unregister(dev);
938 	kfree_rcu(port, rcu);
939 }
940 
941 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
942 {
943 	if (tb[IFLA_ADDRESS]) {
944 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
945 			return -EINVAL;
946 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
947 			return -EADDRNOTAVAIL;
948 	}
949 
950 	if (data && data[IFLA_MACVLAN_FLAGS] &&
951 	    nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
952 		return -EINVAL;
953 
954 	if (data && data[IFLA_MACVLAN_MODE]) {
955 		switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
956 		case MACVLAN_MODE_PRIVATE:
957 		case MACVLAN_MODE_VEPA:
958 		case MACVLAN_MODE_BRIDGE:
959 		case MACVLAN_MODE_PASSTHRU:
960 			break;
961 		default:
962 			return -EINVAL;
963 		}
964 	}
965 	return 0;
966 }
967 
968 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
969 			   struct nlattr *tb[], struct nlattr *data[])
970 {
971 	struct macvlan_dev *vlan = netdev_priv(dev);
972 	struct macvlan_port *port;
973 	struct net_device *lowerdev;
974 	int err;
975 
976 	if (!tb[IFLA_LINK])
977 		return -EINVAL;
978 
979 	lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
980 	if (lowerdev == NULL)
981 		return -ENODEV;
982 
983 	/* When creating macvlans or macvtaps on top of other macvlans - use
984 	 * the real device as the lowerdev.
985 	 */
986 	if (netif_is_macvlan(lowerdev))
987 		lowerdev = macvlan_dev_real_dev(lowerdev);
988 
989 	if (!tb[IFLA_MTU])
990 		dev->mtu = lowerdev->mtu;
991 	else if (dev->mtu > lowerdev->mtu)
992 		return -EINVAL;
993 
994 	if (!tb[IFLA_ADDRESS])
995 		eth_hw_addr_random(dev);
996 
997 	if (!macvlan_port_exists(lowerdev)) {
998 		err = macvlan_port_create(lowerdev);
999 		if (err < 0)
1000 			return err;
1001 	}
1002 	port = macvlan_port_get_rtnl(lowerdev);
1003 
1004 	/* Only 1 macvlan device can be created in passthru mode */
1005 	if (port->passthru)
1006 		return -EINVAL;
1007 
1008 	vlan->lowerdev = lowerdev;
1009 	vlan->dev      = dev;
1010 	vlan->port     = port;
1011 	vlan->set_features = MACVLAN_FEATURES;
1012 	vlan->nest_level = dev_get_nest_level(lowerdev, netif_is_macvlan) + 1;
1013 
1014 	vlan->mode     = MACVLAN_MODE_VEPA;
1015 	if (data && data[IFLA_MACVLAN_MODE])
1016 		vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1017 
1018 	if (data && data[IFLA_MACVLAN_FLAGS])
1019 		vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1020 
1021 	if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
1022 		if (!MACVLAN_PORT_IS_EMPTY(port))
1023 			return -EINVAL;
1024 		port->passthru = true;
1025 		eth_hw_addr_inherit(dev, lowerdev);
1026 	}
1027 
1028 	err = register_netdevice(dev);
1029 	if (err < 0)
1030 		goto destroy_port;
1031 
1032 	dev->priv_flags |= IFF_MACVLAN;
1033 	err = netdev_upper_dev_link(lowerdev, dev);
1034 	if (err)
1035 		goto unregister_netdev;
1036 
1037 	list_add_tail_rcu(&vlan->list, &port->vlans);
1038 	netif_stacked_transfer_operstate(lowerdev, dev);
1039 
1040 	return 0;
1041 
1042 unregister_netdev:
1043 	unregister_netdevice(dev);
1044 destroy_port:
1045 	if (MACVLAN_PORT_IS_EMPTY(port))
1046 		macvlan_port_destroy(lowerdev);
1047 
1048 	return err;
1049 }
1050 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
1051 
1052 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
1053 			   struct nlattr *tb[], struct nlattr *data[])
1054 {
1055 	return macvlan_common_newlink(src_net, dev, tb, data);
1056 }
1057 
1058 void macvlan_dellink(struct net_device *dev, struct list_head *head)
1059 {
1060 	struct macvlan_dev *vlan = netdev_priv(dev);
1061 
1062 	list_del_rcu(&vlan->list);
1063 	unregister_netdevice_queue(dev, head);
1064 	netdev_upper_dev_unlink(vlan->lowerdev, dev);
1065 }
1066 EXPORT_SYMBOL_GPL(macvlan_dellink);
1067 
1068 static int macvlan_changelink(struct net_device *dev,
1069 		struct nlattr *tb[], struct nlattr *data[])
1070 {
1071 	struct macvlan_dev *vlan = netdev_priv(dev);
1072 	enum macvlan_mode mode;
1073 	bool set_mode = false;
1074 
1075 	/* Validate mode, but don't set yet: setting flags may fail. */
1076 	if (data && data[IFLA_MACVLAN_MODE]) {
1077 		set_mode = true;
1078 		mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1079 		/* Passthrough mode can't be set or cleared dynamically */
1080 		if ((mode == MACVLAN_MODE_PASSTHRU) !=
1081 		    (vlan->mode == MACVLAN_MODE_PASSTHRU))
1082 			return -EINVAL;
1083 	}
1084 
1085 	if (data && data[IFLA_MACVLAN_FLAGS]) {
1086 		__u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1087 		bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
1088 		if (vlan->port->passthru && promisc) {
1089 			int err;
1090 
1091 			if (flags & MACVLAN_FLAG_NOPROMISC)
1092 				err = dev_set_promiscuity(vlan->lowerdev, -1);
1093 			else
1094 				err = dev_set_promiscuity(vlan->lowerdev, 1);
1095 			if (err < 0)
1096 				return err;
1097 		}
1098 		vlan->flags = flags;
1099 	}
1100 	if (set_mode)
1101 		vlan->mode = mode;
1102 	return 0;
1103 }
1104 
1105 static size_t macvlan_get_size(const struct net_device *dev)
1106 {
1107 	return (0
1108 		+ nla_total_size(4) /* IFLA_MACVLAN_MODE */
1109 		+ nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
1110 		);
1111 }
1112 
1113 static int macvlan_fill_info(struct sk_buff *skb,
1114 				const struct net_device *dev)
1115 {
1116 	struct macvlan_dev *vlan = netdev_priv(dev);
1117 
1118 	if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
1119 		goto nla_put_failure;
1120 	if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
1121 		goto nla_put_failure;
1122 	return 0;
1123 
1124 nla_put_failure:
1125 	return -EMSGSIZE;
1126 }
1127 
1128 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
1129 	[IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
1130 	[IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
1131 };
1132 
1133 int macvlan_link_register(struct rtnl_link_ops *ops)
1134 {
1135 	/* common fields */
1136 	ops->priv_size		= sizeof(struct macvlan_dev);
1137 	ops->validate		= macvlan_validate;
1138 	ops->maxtype		= IFLA_MACVLAN_MAX;
1139 	ops->policy		= macvlan_policy;
1140 	ops->changelink		= macvlan_changelink;
1141 	ops->get_size		= macvlan_get_size;
1142 	ops->fill_info		= macvlan_fill_info;
1143 
1144 	return rtnl_link_register(ops);
1145 };
1146 EXPORT_SYMBOL_GPL(macvlan_link_register);
1147 
1148 static struct rtnl_link_ops macvlan_link_ops = {
1149 	.kind		= "macvlan",
1150 	.setup		= macvlan_setup,
1151 	.newlink	= macvlan_newlink,
1152 	.dellink	= macvlan_dellink,
1153 };
1154 
1155 static int macvlan_device_event(struct notifier_block *unused,
1156 				unsigned long event, void *ptr)
1157 {
1158 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1159 	struct macvlan_dev *vlan, *next;
1160 	struct macvlan_port *port;
1161 	LIST_HEAD(list_kill);
1162 
1163 	if (!macvlan_port_exists(dev))
1164 		return NOTIFY_DONE;
1165 
1166 	port = macvlan_port_get_rtnl(dev);
1167 
1168 	switch (event) {
1169 	case NETDEV_CHANGE:
1170 		list_for_each_entry(vlan, &port->vlans, list)
1171 			netif_stacked_transfer_operstate(vlan->lowerdev,
1172 							 vlan->dev);
1173 		break;
1174 	case NETDEV_FEAT_CHANGE:
1175 		list_for_each_entry(vlan, &port->vlans, list) {
1176 			vlan->dev->gso_max_size = dev->gso_max_size;
1177 			netdev_update_features(vlan->dev);
1178 		}
1179 		break;
1180 	case NETDEV_CHANGEMTU:
1181 		list_for_each_entry(vlan, &port->vlans, list) {
1182 			if (vlan->dev->mtu <= dev->mtu)
1183 				continue;
1184 			dev_set_mtu(vlan->dev, dev->mtu);
1185 		}
1186 		break;
1187 	case NETDEV_CHANGEADDR:
1188 		if (!port->passthru)
1189 			return NOTIFY_DONE;
1190 
1191 		vlan = list_first_entry_or_null(&port->vlans,
1192 						struct macvlan_dev,
1193 						list);
1194 
1195 		if (macvlan_sync_address(vlan->dev, dev->dev_addr))
1196 			return NOTIFY_BAD;
1197 
1198 		break;
1199 	case NETDEV_UNREGISTER:
1200 		/* twiddle thumbs on netns device moves */
1201 		if (dev->reg_state != NETREG_UNREGISTERING)
1202 			break;
1203 
1204 		list_for_each_entry_safe(vlan, next, &port->vlans, list)
1205 			vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1206 		unregister_netdevice_many(&list_kill);
1207 		break;
1208 	case NETDEV_PRE_TYPE_CHANGE:
1209 		/* Forbid underlaying device to change its type. */
1210 		return NOTIFY_BAD;
1211 
1212 	case NETDEV_NOTIFY_PEERS:
1213 	case NETDEV_BONDING_FAILOVER:
1214 	case NETDEV_RESEND_IGMP:
1215 		/* Propagate to all vlans */
1216 		list_for_each_entry(vlan, &port->vlans, list)
1217 			call_netdevice_notifiers(event, vlan->dev);
1218 	}
1219 	return NOTIFY_DONE;
1220 }
1221 
1222 static struct notifier_block macvlan_notifier_block __read_mostly = {
1223 	.notifier_call	= macvlan_device_event,
1224 };
1225 
1226 static int __init macvlan_init_module(void)
1227 {
1228 	int err;
1229 
1230 	register_netdevice_notifier(&macvlan_notifier_block);
1231 
1232 	err = macvlan_link_register(&macvlan_link_ops);
1233 	if (err < 0)
1234 		goto err1;
1235 	return 0;
1236 err1:
1237 	unregister_netdevice_notifier(&macvlan_notifier_block);
1238 	return err;
1239 }
1240 
1241 static void __exit macvlan_cleanup_module(void)
1242 {
1243 	rtnl_link_unregister(&macvlan_link_ops);
1244 	unregister_netdevice_notifier(&macvlan_notifier_block);
1245 }
1246 
1247 module_init(macvlan_init_module);
1248 module_exit(macvlan_cleanup_module);
1249 
1250 MODULE_LICENSE("GPL");
1251 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1252 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1253 MODULE_ALIAS_RTNL_LINK("macvlan");
1254