]> bbs.cooldavid.org Git - net-next-2.6.git/blob - drivers/net/macvlan.c
macvlan: Introduce 'passthru' mode to takeover the underlying device
[net-next-2.6.git] / drivers / net / macvlan.c
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_link.h>
30 #include <linux/if_macvlan.h>
31 #include <net/rtnetlink.h>
32 #include <net/xfrm.h>
33
34 #define MACVLAN_HASH_SIZE       (1 << BITS_PER_BYTE)
35
36 struct macvlan_port {
37         struct net_device       *dev;
38         struct hlist_head       vlan_hash[MACVLAN_HASH_SIZE];
39         struct list_head        vlans;
40         struct rcu_head         rcu;
41         bool                    passthru;
42 };
43
44 #define macvlan_port_get_rcu(dev) \
45         ((struct macvlan_port *) rcu_dereference(dev->rx_handler_data))
46 #define macvlan_port_get(dev) ((struct macvlan_port *) dev->rx_handler_data)
47 #define macvlan_port_exists(dev) (dev->priv_flags & IFF_MACVLAN_PORT)
48
49 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
50                                                const unsigned char *addr)
51 {
52         struct macvlan_dev *vlan;
53         struct hlist_node *n;
54
55         hlist_for_each_entry_rcu(vlan, n, &port->vlan_hash[addr[5]], hlist) {
56                 if (!compare_ether_addr_64bits(vlan->dev->dev_addr, addr))
57                         return vlan;
58         }
59         return NULL;
60 }
61
62 static void macvlan_hash_add(struct macvlan_dev *vlan)
63 {
64         struct macvlan_port *port = vlan->port;
65         const unsigned char *addr = vlan->dev->dev_addr;
66
67         hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[addr[5]]);
68 }
69
70 static void macvlan_hash_del(struct macvlan_dev *vlan)
71 {
72         hlist_del_rcu(&vlan->hlist);
73         synchronize_rcu();
74 }
75
76 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
77                                         const unsigned char *addr)
78 {
79         macvlan_hash_del(vlan);
80         /* Now that we are unhashed it is safe to change the device
81          * address without confusing packet delivery.
82          */
83         memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
84         macvlan_hash_add(vlan);
85 }
86
87 static int macvlan_addr_busy(const struct macvlan_port *port,
88                                 const unsigned char *addr)
89 {
90         /* Test to see if the specified multicast address is
91          * currently in use by the underlying device or
92          * another macvlan.
93          */
94         if (!compare_ether_addr_64bits(port->dev->dev_addr, addr))
95                 return 1;
96
97         if (macvlan_hash_lookup(port, addr))
98                 return 1;
99
100         return 0;
101 }
102
103
104 static int macvlan_broadcast_one(struct sk_buff *skb,
105                                  const struct macvlan_dev *vlan,
106                                  const struct ethhdr *eth, bool local)
107 {
108         struct net_device *dev = vlan->dev;
109         if (!skb)
110                 return NET_RX_DROP;
111
112         if (local)
113                 return vlan->forward(dev, skb);
114
115         skb->dev = dev;
116         if (!compare_ether_addr_64bits(eth->h_dest,
117                                        dev->broadcast))
118                 skb->pkt_type = PACKET_BROADCAST;
119         else
120                 skb->pkt_type = PACKET_MULTICAST;
121
122         return vlan->receive(skb);
123 }
124
125 static void macvlan_broadcast(struct sk_buff *skb,
126                               const struct macvlan_port *port,
127                               struct net_device *src,
128                               enum macvlan_mode mode)
129 {
130         const struct ethhdr *eth = eth_hdr(skb);
131         const struct macvlan_dev *vlan;
132         struct hlist_node *n;
133         struct sk_buff *nskb;
134         unsigned int i;
135         int err;
136
137         if (skb->protocol == htons(ETH_P_PAUSE))
138                 return;
139
140         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
141                 hlist_for_each_entry_rcu(vlan, n, &port->vlan_hash[i], hlist) {
142                         if (vlan->dev == src || !(vlan->mode & mode))
143                                 continue;
144
145                         nskb = skb_clone(skb, GFP_ATOMIC);
146                         err = macvlan_broadcast_one(nskb, vlan, eth,
147                                          mode == MACVLAN_MODE_BRIDGE);
148                         macvlan_count_rx(vlan, skb->len + ETH_HLEN,
149                                          err == NET_RX_SUCCESS, 1);
150                 }
151         }
152 }
153
154 /* called under rcu_read_lock() from netif_receive_skb */
155 static struct sk_buff *macvlan_handle_frame(struct sk_buff *skb)
156 {
157         struct macvlan_port *port;
158         const struct ethhdr *eth = eth_hdr(skb);
159         const struct macvlan_dev *vlan;
160         const struct macvlan_dev *src;
161         struct net_device *dev;
162         unsigned int len = 0;
163         int ret = NET_RX_DROP;
164
165         port = macvlan_port_get_rcu(skb->dev);
166         if (is_multicast_ether_addr(eth->h_dest)) {
167                 src = macvlan_hash_lookup(port, eth->h_source);
168                 if (!src)
169                         /* frame comes from an external address */
170                         macvlan_broadcast(skb, port, NULL,
171                                           MACVLAN_MODE_PRIVATE |
172                                           MACVLAN_MODE_VEPA    |
173                                           MACVLAN_MODE_PASSTHRU|
174                                           MACVLAN_MODE_BRIDGE);
175                 else if (src->mode == MACVLAN_MODE_VEPA)
176                         /* flood to everyone except source */
177                         macvlan_broadcast(skb, port, src->dev,
178                                           MACVLAN_MODE_VEPA |
179                                           MACVLAN_MODE_BRIDGE);
180                 else if (src->mode == MACVLAN_MODE_BRIDGE)
181                         /*
182                          * flood only to VEPA ports, bridge ports
183                          * already saw the frame on the way out.
184                          */
185                         macvlan_broadcast(skb, port, src->dev,
186                                           MACVLAN_MODE_VEPA);
187                 return skb;
188         }
189
190         if (port->passthru)
191                 vlan = list_first_entry(&port->vlans, struct macvlan_dev, list);
192         else
193                 vlan = macvlan_hash_lookup(port, eth->h_dest);
194         if (vlan == NULL)
195                 return skb;
196
197         dev = vlan->dev;
198         if (unlikely(!(dev->flags & IFF_UP))) {
199                 kfree_skb(skb);
200                 return NULL;
201         }
202         len = skb->len + ETH_HLEN;
203         skb = skb_share_check(skb, GFP_ATOMIC);
204         if (!skb)
205                 goto out;
206
207         skb->dev = dev;
208         skb->pkt_type = PACKET_HOST;
209
210         ret = vlan->receive(skb);
211
212 out:
213         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, 0);
214         return NULL;
215 }
216
217 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
218 {
219         const struct macvlan_dev *vlan = netdev_priv(dev);
220         const struct macvlan_port *port = vlan->port;
221         const struct macvlan_dev *dest;
222
223         if (vlan->mode == MACVLAN_MODE_BRIDGE) {
224                 const struct ethhdr *eth = (void *)skb->data;
225
226                 /* send to other bridge ports directly */
227                 if (is_multicast_ether_addr(eth->h_dest)) {
228                         macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
229                         goto xmit_world;
230                 }
231
232                 dest = macvlan_hash_lookup(port, eth->h_dest);
233                 if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
234                         unsigned int length = skb->len + ETH_HLEN;
235                         int ret = dest->forward(dest->dev, skb);
236                         macvlan_count_rx(dest, length,
237                                          ret == NET_RX_SUCCESS, 0);
238
239                         return NET_XMIT_SUCCESS;
240                 }
241         }
242
243 xmit_world:
244         skb_set_dev(skb, vlan->lowerdev);
245         return dev_queue_xmit(skb);
246 }
247
248 netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
249                                struct net_device *dev)
250 {
251         unsigned int len = skb->len;
252         int ret;
253         const struct macvlan_dev *vlan = netdev_priv(dev);
254
255         ret = macvlan_queue_xmit(skb, dev);
256         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
257                 struct macvlan_pcpu_stats *pcpu_stats;
258
259                 pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
260                 u64_stats_update_begin(&pcpu_stats->syncp);
261                 pcpu_stats->tx_packets++;
262                 pcpu_stats->tx_bytes += len;
263                 u64_stats_update_end(&pcpu_stats->syncp);
264         } else {
265                 this_cpu_inc(vlan->pcpu_stats->tx_dropped);
266         }
267         return ret;
268 }
269 EXPORT_SYMBOL_GPL(macvlan_start_xmit);
270
271 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
272                                unsigned short type, const void *daddr,
273                                const void *saddr, unsigned len)
274 {
275         const struct macvlan_dev *vlan = netdev_priv(dev);
276         struct net_device *lowerdev = vlan->lowerdev;
277
278         return dev_hard_header(skb, lowerdev, type, daddr,
279                                saddr ? : dev->dev_addr, len);
280 }
281
282 static const struct header_ops macvlan_hard_header_ops = {
283         .create         = macvlan_hard_header,
284         .rebuild        = eth_rebuild_header,
285         .parse          = eth_header_parse,
286         .cache          = eth_header_cache,
287         .cache_update   = eth_header_cache_update,
288 };
289
290 static int macvlan_open(struct net_device *dev)
291 {
292         struct macvlan_dev *vlan = netdev_priv(dev);
293         struct net_device *lowerdev = vlan->lowerdev;
294         int err;
295
296         if (vlan->port->passthru) {
297                 dev_set_promiscuity(lowerdev, 1);
298                 goto hash_add;
299         }
300
301         err = -EBUSY;
302         if (macvlan_addr_busy(vlan->port, dev->dev_addr))
303                 goto out;
304
305         err = dev_uc_add(lowerdev, dev->dev_addr);
306         if (err < 0)
307                 goto out;
308         if (dev->flags & IFF_ALLMULTI) {
309                 err = dev_set_allmulti(lowerdev, 1);
310                 if (err < 0)
311                         goto del_unicast;
312         }
313
314 hash_add:
315         macvlan_hash_add(vlan);
316         return 0;
317
318 del_unicast:
319         dev_uc_del(lowerdev, dev->dev_addr);
320 out:
321         return err;
322 }
323
324 static int macvlan_stop(struct net_device *dev)
325 {
326         struct macvlan_dev *vlan = netdev_priv(dev);
327         struct net_device *lowerdev = vlan->lowerdev;
328
329         if (vlan->port->passthru) {
330                 dev_set_promiscuity(lowerdev, -1);
331                 goto hash_del;
332         }
333
334         dev_mc_unsync(lowerdev, dev);
335         if (dev->flags & IFF_ALLMULTI)
336                 dev_set_allmulti(lowerdev, -1);
337
338         dev_uc_del(lowerdev, dev->dev_addr);
339
340 hash_del:
341         macvlan_hash_del(vlan);
342         return 0;
343 }
344
345 static int macvlan_set_mac_address(struct net_device *dev, void *p)
346 {
347         struct macvlan_dev *vlan = netdev_priv(dev);
348         struct net_device *lowerdev = vlan->lowerdev;
349         struct sockaddr *addr = p;
350         int err;
351
352         if (!is_valid_ether_addr(addr->sa_data))
353                 return -EADDRNOTAVAIL;
354
355         if (!(dev->flags & IFF_UP)) {
356                 /* Just copy in the new address */
357                 memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
358         } else {
359                 /* Rehash and update the device filters */
360                 if (macvlan_addr_busy(vlan->port, addr->sa_data))
361                         return -EBUSY;
362
363                 err = dev_uc_add(lowerdev, addr->sa_data);
364                 if (err)
365                         return err;
366
367                 dev_uc_del(lowerdev, dev->dev_addr);
368
369                 macvlan_hash_change_addr(vlan, addr->sa_data);
370         }
371         return 0;
372 }
373
374 static void macvlan_change_rx_flags(struct net_device *dev, int change)
375 {
376         struct macvlan_dev *vlan = netdev_priv(dev);
377         struct net_device *lowerdev = vlan->lowerdev;
378
379         if (change & IFF_ALLMULTI)
380                 dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
381 }
382
383 static void macvlan_set_multicast_list(struct net_device *dev)
384 {
385         struct macvlan_dev *vlan = netdev_priv(dev);
386
387         dev_mc_sync(vlan->lowerdev, dev);
388 }
389
390 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
391 {
392         struct macvlan_dev *vlan = netdev_priv(dev);
393
394         if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
395                 return -EINVAL;
396         dev->mtu = new_mtu;
397         return 0;
398 }
399
400 /*
401  * macvlan network devices have devices nesting below it and are a special
402  * "super class" of normal network devices; split their locks off into a
403  * separate class since they always nest.
404  */
405 static struct lock_class_key macvlan_netdev_xmit_lock_key;
406 static struct lock_class_key macvlan_netdev_addr_lock_key;
407
408 #define MACVLAN_FEATURES \
409         (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
410          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
411          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO)
412
413 #define MACVLAN_STATE_MASK \
414         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
415
416 static void macvlan_set_lockdep_class_one(struct net_device *dev,
417                                           struct netdev_queue *txq,
418                                           void *_unused)
419 {
420         lockdep_set_class(&txq->_xmit_lock,
421                           &macvlan_netdev_xmit_lock_key);
422 }
423
424 static void macvlan_set_lockdep_class(struct net_device *dev)
425 {
426         lockdep_set_class(&dev->addr_list_lock,
427                           &macvlan_netdev_addr_lock_key);
428         netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
429 }
430
431 static int macvlan_init(struct net_device *dev)
432 {
433         struct macvlan_dev *vlan = netdev_priv(dev);
434         const struct net_device *lowerdev = vlan->lowerdev;
435
436         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
437                                   (lowerdev->state & MACVLAN_STATE_MASK);
438         dev->features           = lowerdev->features & MACVLAN_FEATURES;
439         dev->features           |= NETIF_F_LLTX;
440         dev->gso_max_size       = lowerdev->gso_max_size;
441         dev->iflink             = lowerdev->ifindex;
442         dev->hard_header_len    = lowerdev->hard_header_len;
443
444         macvlan_set_lockdep_class(dev);
445
446         vlan->pcpu_stats = alloc_percpu(struct macvlan_pcpu_stats);
447         if (!vlan->pcpu_stats)
448                 return -ENOMEM;
449
450         return 0;
451 }
452
453 static void macvlan_uninit(struct net_device *dev)
454 {
455         struct macvlan_dev *vlan = netdev_priv(dev);
456
457         free_percpu(vlan->pcpu_stats);
458 }
459
460 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
461                                                          struct rtnl_link_stats64 *stats)
462 {
463         struct macvlan_dev *vlan = netdev_priv(dev);
464
465         if (vlan->pcpu_stats) {
466                 struct macvlan_pcpu_stats *p;
467                 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
468                 u32 rx_errors = 0, tx_dropped = 0;
469                 unsigned int start;
470                 int i;
471
472                 for_each_possible_cpu(i) {
473                         p = per_cpu_ptr(vlan->pcpu_stats, i);
474                         do {
475                                 start = u64_stats_fetch_begin_bh(&p->syncp);
476                                 rx_packets      = p->rx_packets;
477                                 rx_bytes        = p->rx_bytes;
478                                 rx_multicast    = p->rx_multicast;
479                                 tx_packets      = p->tx_packets;
480                                 tx_bytes        = p->tx_bytes;
481                         } while (u64_stats_fetch_retry_bh(&p->syncp, start));
482
483                         stats->rx_packets       += rx_packets;
484                         stats->rx_bytes         += rx_bytes;
485                         stats->multicast        += rx_multicast;
486                         stats->tx_packets       += tx_packets;
487                         stats->tx_bytes         += tx_bytes;
488                         /* rx_errors & tx_dropped are u32, updated
489                          * without syncp protection.
490                          */
491                         rx_errors       += p->rx_errors;
492                         tx_dropped      += p->tx_dropped;
493                 }
494                 stats->rx_errors        = rx_errors;
495                 stats->rx_dropped       = rx_errors;
496                 stats->tx_dropped       = tx_dropped;
497         }
498         return stats;
499 }
500
501 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
502                                         struct ethtool_drvinfo *drvinfo)
503 {
504         snprintf(drvinfo->driver, 32, "macvlan");
505         snprintf(drvinfo->version, 32, "0.1");
506 }
507
508 static u32 macvlan_ethtool_get_rx_csum(struct net_device *dev)
509 {
510         const struct macvlan_dev *vlan = netdev_priv(dev);
511         return dev_ethtool_get_rx_csum(vlan->lowerdev);
512 }
513
514 static int macvlan_ethtool_get_settings(struct net_device *dev,
515                                         struct ethtool_cmd *cmd)
516 {
517         const struct macvlan_dev *vlan = netdev_priv(dev);
518         return dev_ethtool_get_settings(vlan->lowerdev, cmd);
519 }
520
521 static u32 macvlan_ethtool_get_flags(struct net_device *dev)
522 {
523         const struct macvlan_dev *vlan = netdev_priv(dev);
524         return dev_ethtool_get_flags(vlan->lowerdev);
525 }
526
527 static const struct ethtool_ops macvlan_ethtool_ops = {
528         .get_link               = ethtool_op_get_link,
529         .get_settings           = macvlan_ethtool_get_settings,
530         .get_rx_csum            = macvlan_ethtool_get_rx_csum,
531         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
532         .get_flags              = macvlan_ethtool_get_flags,
533 };
534
535 static const struct net_device_ops macvlan_netdev_ops = {
536         .ndo_init               = macvlan_init,
537         .ndo_uninit             = macvlan_uninit,
538         .ndo_open               = macvlan_open,
539         .ndo_stop               = macvlan_stop,
540         .ndo_start_xmit         = macvlan_start_xmit,
541         .ndo_change_mtu         = macvlan_change_mtu,
542         .ndo_change_rx_flags    = macvlan_change_rx_flags,
543         .ndo_set_mac_address    = macvlan_set_mac_address,
544         .ndo_set_multicast_list = macvlan_set_multicast_list,
545         .ndo_get_stats64        = macvlan_dev_get_stats64,
546         .ndo_validate_addr      = eth_validate_addr,
547 };
548
549 void macvlan_common_setup(struct net_device *dev)
550 {
551         ether_setup(dev);
552
553         dev->priv_flags        &= ~IFF_XMIT_DST_RELEASE;
554         dev->netdev_ops         = &macvlan_netdev_ops;
555         dev->destructor         = free_netdev;
556         dev->header_ops         = &macvlan_hard_header_ops,
557         dev->ethtool_ops        = &macvlan_ethtool_ops;
558 }
559 EXPORT_SYMBOL_GPL(macvlan_common_setup);
560
561 static void macvlan_setup(struct net_device *dev)
562 {
563         macvlan_common_setup(dev);
564         dev->tx_queue_len       = 0;
565 }
566
567 static int macvlan_port_create(struct net_device *dev)
568 {
569         struct macvlan_port *port;
570         unsigned int i;
571         int err;
572
573         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
574                 return -EINVAL;
575
576         port = kzalloc(sizeof(*port), GFP_KERNEL);
577         if (port == NULL)
578                 return -ENOMEM;
579
580         port->passthru = false;
581         port->dev = dev;
582         INIT_LIST_HEAD(&port->vlans);
583         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
584                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
585
586         err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
587         if (err)
588                 kfree(port);
589
590         dev->priv_flags |= IFF_MACVLAN_PORT;
591         return err;
592 }
593
594 static void macvlan_port_rcu_free(struct rcu_head *head)
595 {
596         struct macvlan_port *port;
597
598         port = container_of(head, struct macvlan_port, rcu);
599         kfree(port);
600 }
601
602 static void macvlan_port_destroy(struct net_device *dev)
603 {
604         struct macvlan_port *port = macvlan_port_get(dev);
605
606         dev->priv_flags &= ~IFF_MACVLAN_PORT;
607         netdev_rx_handler_unregister(dev);
608         call_rcu(&port->rcu, macvlan_port_rcu_free);
609 }
610
611 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
612 {
613         if (tb[IFLA_ADDRESS]) {
614                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
615                         return -EINVAL;
616                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
617                         return -EADDRNOTAVAIL;
618         }
619
620         if (data && data[IFLA_MACVLAN_MODE]) {
621                 switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
622                 case MACVLAN_MODE_PRIVATE:
623                 case MACVLAN_MODE_VEPA:
624                 case MACVLAN_MODE_BRIDGE:
625                 case MACVLAN_MODE_PASSTHRU:
626                         break;
627                 default:
628                         return -EINVAL;
629                 }
630         }
631         return 0;
632 }
633
634 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
635                            struct nlattr *tb[], struct nlattr *data[],
636                            int (*receive)(struct sk_buff *skb),
637                            int (*forward)(struct net_device *dev,
638                                           struct sk_buff *skb))
639 {
640         struct macvlan_dev *vlan = netdev_priv(dev);
641         struct macvlan_port *port;
642         struct net_device *lowerdev;
643         int err;
644
645         if (!tb[IFLA_LINK])
646                 return -EINVAL;
647
648         lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
649         if (lowerdev == NULL)
650                 return -ENODEV;
651
652         /* When creating macvlans on top of other macvlans - use
653          * the real device as the lowerdev.
654          */
655         if (lowerdev->rtnl_link_ops == dev->rtnl_link_ops) {
656                 struct macvlan_dev *lowervlan = netdev_priv(lowerdev);
657                 lowerdev = lowervlan->lowerdev;
658         }
659
660         if (!tb[IFLA_MTU])
661                 dev->mtu = lowerdev->mtu;
662         else if (dev->mtu > lowerdev->mtu)
663                 return -EINVAL;
664
665         if (!tb[IFLA_ADDRESS])
666                 random_ether_addr(dev->dev_addr);
667
668         if (!macvlan_port_exists(lowerdev)) {
669                 err = macvlan_port_create(lowerdev);
670                 if (err < 0)
671                         return err;
672         }
673         port = macvlan_port_get(lowerdev);
674
675         /* Only 1 macvlan device can be created in passthru mode */
676         if (port->passthru)
677                 return -EINVAL;
678
679         vlan->lowerdev = lowerdev;
680         vlan->dev      = dev;
681         vlan->port     = port;
682         vlan->receive  = receive;
683         vlan->forward  = forward;
684
685         vlan->mode     = MACVLAN_MODE_VEPA;
686         if (data && data[IFLA_MACVLAN_MODE])
687                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
688
689         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
690                 if (!list_empty(&port->vlans))
691                         return -EINVAL;
692                 port->passthru = true;
693                 memcpy(dev->dev_addr, lowerdev->dev_addr, ETH_ALEN);
694         }
695
696         err = register_netdevice(dev);
697         if (err < 0)
698                 goto destroy_port;
699
700         list_add_tail(&vlan->list, &port->vlans);
701         netif_stacked_transfer_operstate(lowerdev, dev);
702
703         return 0;
704
705 destroy_port:
706         if (list_empty(&port->vlans))
707                 macvlan_port_destroy(lowerdev);
708
709         return err;
710 }
711 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
712
713 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
714                            struct nlattr *tb[], struct nlattr *data[])
715 {
716         return macvlan_common_newlink(src_net, dev, tb, data,
717                                       netif_rx,
718                                       dev_forward_skb);
719 }
720
721 void macvlan_dellink(struct net_device *dev, struct list_head *head)
722 {
723         struct macvlan_dev *vlan = netdev_priv(dev);
724         struct macvlan_port *port = vlan->port;
725
726         list_del(&vlan->list);
727         unregister_netdevice_queue(dev, head);
728
729         if (list_empty(&port->vlans))
730                 macvlan_port_destroy(port->dev);
731 }
732 EXPORT_SYMBOL_GPL(macvlan_dellink);
733
734 static int macvlan_changelink(struct net_device *dev,
735                 struct nlattr *tb[], struct nlattr *data[])
736 {
737         struct macvlan_dev *vlan = netdev_priv(dev);
738         if (data && data[IFLA_MACVLAN_MODE])
739                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
740         return 0;
741 }
742
743 static size_t macvlan_get_size(const struct net_device *dev)
744 {
745         return nla_total_size(4);
746 }
747
748 static int macvlan_fill_info(struct sk_buff *skb,
749                                 const struct net_device *dev)
750 {
751         struct macvlan_dev *vlan = netdev_priv(dev);
752
753         NLA_PUT_U32(skb, IFLA_MACVLAN_MODE, vlan->mode);
754         return 0;
755
756 nla_put_failure:
757         return -EMSGSIZE;
758 }
759
760 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
761         [IFLA_MACVLAN_MODE] = { .type = NLA_U32 },
762 };
763
764 int macvlan_link_register(struct rtnl_link_ops *ops)
765 {
766         /* common fields */
767         ops->priv_size          = sizeof(struct macvlan_dev);
768         ops->validate           = macvlan_validate;
769         ops->maxtype            = IFLA_MACVLAN_MAX;
770         ops->policy             = macvlan_policy;
771         ops->changelink         = macvlan_changelink;
772         ops->get_size           = macvlan_get_size;
773         ops->fill_info          = macvlan_fill_info;
774
775         return rtnl_link_register(ops);
776 };
777 EXPORT_SYMBOL_GPL(macvlan_link_register);
778
779 static struct rtnl_link_ops macvlan_link_ops = {
780         .kind           = "macvlan",
781         .setup          = macvlan_setup,
782         .newlink        = macvlan_newlink,
783         .dellink        = macvlan_dellink,
784 };
785
786 static int macvlan_device_event(struct notifier_block *unused,
787                                 unsigned long event, void *ptr)
788 {
789         struct net_device *dev = ptr;
790         struct macvlan_dev *vlan, *next;
791         struct macvlan_port *port;
792
793         if (!macvlan_port_exists(dev))
794                 return NOTIFY_DONE;
795
796         port = macvlan_port_get(dev);
797
798         switch (event) {
799         case NETDEV_CHANGE:
800                 list_for_each_entry(vlan, &port->vlans, list)
801                         netif_stacked_transfer_operstate(vlan->lowerdev,
802                                                          vlan->dev);
803                 break;
804         case NETDEV_FEAT_CHANGE:
805                 list_for_each_entry(vlan, &port->vlans, list) {
806                         vlan->dev->features = dev->features & MACVLAN_FEATURES;
807                         vlan->dev->gso_max_size = dev->gso_max_size;
808                         netdev_features_change(vlan->dev);
809                 }
810                 break;
811         case NETDEV_UNREGISTER:
812                 /* twiddle thumbs on netns device moves */
813                 if (dev->reg_state != NETREG_UNREGISTERING)
814                         break;
815
816                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
817                         vlan->dev->rtnl_link_ops->dellink(vlan->dev, NULL);
818                 break;
819         case NETDEV_PRE_TYPE_CHANGE:
820                 /* Forbid underlaying device to change its type. */
821                 return NOTIFY_BAD;
822         }
823         return NOTIFY_DONE;
824 }
825
826 static struct notifier_block macvlan_notifier_block __read_mostly = {
827         .notifier_call  = macvlan_device_event,
828 };
829
830 static int __init macvlan_init_module(void)
831 {
832         int err;
833
834         register_netdevice_notifier(&macvlan_notifier_block);
835
836         err = macvlan_link_register(&macvlan_link_ops);
837         if (err < 0)
838                 goto err1;
839         return 0;
840 err1:
841         unregister_netdevice_notifier(&macvlan_notifier_block);
842         return err;
843 }
844
845 static void __exit macvlan_cleanup_module(void)
846 {
847         rtnl_link_unregister(&macvlan_link_ops);
848         unregister_netdevice_notifier(&macvlan_notifier_block);
849 }
850
851 module_init(macvlan_init_module);
852 module_exit(macvlan_cleanup_module);
853
854 MODULE_LICENSE("GPL");
855 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
856 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
857 MODULE_ALIAS_RTNL_LINK("macvlan");