Merge branch 'core-rcu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-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         int                     count;
43 };
44
45 static void macvlan_port_destroy(struct net_device *dev);
46
47 #define macvlan_port_get_rcu(dev) \
48         ((struct macvlan_port *) rcu_dereference(dev->rx_handler_data))
49 #define macvlan_port_get(dev) ((struct macvlan_port *) dev->rx_handler_data)
50 #define macvlan_port_exists(dev) (dev->priv_flags & IFF_MACVLAN_PORT)
51
52 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
53                                                const unsigned char *addr)
54 {
55         struct macvlan_dev *vlan;
56         struct hlist_node *n;
57
58         hlist_for_each_entry_rcu(vlan, n, &port->vlan_hash[addr[5]], hlist) {
59                 if (!compare_ether_addr_64bits(vlan->dev->dev_addr, addr))
60                         return vlan;
61         }
62         return NULL;
63 }
64
65 static void macvlan_hash_add(struct macvlan_dev *vlan)
66 {
67         struct macvlan_port *port = vlan->port;
68         const unsigned char *addr = vlan->dev->dev_addr;
69
70         hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[addr[5]]);
71 }
72
73 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync)
74 {
75         hlist_del_rcu(&vlan->hlist);
76         if (sync)
77                 synchronize_rcu();
78 }
79
80 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
81                                         const unsigned char *addr)
82 {
83         macvlan_hash_del(vlan, true);
84         /* Now that we are unhashed it is safe to change the device
85          * address without confusing packet delivery.
86          */
87         memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
88         macvlan_hash_add(vlan);
89 }
90
91 static int macvlan_addr_busy(const struct macvlan_port *port,
92                                 const unsigned char *addr)
93 {
94         /* Test to see if the specified multicast address is
95          * currently in use by the underlying device or
96          * another macvlan.
97          */
98         if (!compare_ether_addr_64bits(port->dev->dev_addr, addr))
99                 return 1;
100
101         if (macvlan_hash_lookup(port, addr))
102                 return 1;
103
104         return 0;
105 }
106
107
108 static int macvlan_broadcast_one(struct sk_buff *skb,
109                                  const struct macvlan_dev *vlan,
110                                  const struct ethhdr *eth, bool local)
111 {
112         struct net_device *dev = vlan->dev;
113         if (!skb)
114                 return NET_RX_DROP;
115
116         if (local)
117                 return vlan->forward(dev, skb);
118
119         skb->dev = dev;
120         if (!compare_ether_addr_64bits(eth->h_dest,
121                                        dev->broadcast))
122                 skb->pkt_type = PACKET_BROADCAST;
123         else
124                 skb->pkt_type = PACKET_MULTICAST;
125
126         return vlan->receive(skb);
127 }
128
129 static void macvlan_broadcast(struct sk_buff *skb,
130                               const struct macvlan_port *port,
131                               struct net_device *src,
132                               enum macvlan_mode mode)
133 {
134         const struct ethhdr *eth = eth_hdr(skb);
135         const struct macvlan_dev *vlan;
136         struct hlist_node *n;
137         struct sk_buff *nskb;
138         unsigned int i;
139         int err;
140
141         if (skb->protocol == htons(ETH_P_PAUSE))
142                 return;
143
144         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
145                 hlist_for_each_entry_rcu(vlan, n, &port->vlan_hash[i], hlist) {
146                         if (vlan->dev == src || !(vlan->mode & mode))
147                                 continue;
148
149                         nskb = skb_clone(skb, GFP_ATOMIC);
150                         err = macvlan_broadcast_one(nskb, vlan, eth,
151                                          mode == MACVLAN_MODE_BRIDGE);
152                         macvlan_count_rx(vlan, skb->len + ETH_HLEN,
153                                          err == NET_RX_SUCCESS, 1);
154                 }
155         }
156 }
157
158 /* called under rcu_read_lock() from netif_receive_skb */
159 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
160 {
161         struct macvlan_port *port;
162         struct sk_buff *skb = *pskb;
163         const struct ethhdr *eth = eth_hdr(skb);
164         const struct macvlan_dev *vlan;
165         const struct macvlan_dev *src;
166         struct net_device *dev;
167         unsigned int len = 0;
168         int ret = NET_RX_DROP;
169
170         port = macvlan_port_get_rcu(skb->dev);
171         if (is_multicast_ether_addr(eth->h_dest)) {
172                 src = macvlan_hash_lookup(port, eth->h_source);
173                 if (!src)
174                         /* frame comes from an external address */
175                         macvlan_broadcast(skb, port, NULL,
176                                           MACVLAN_MODE_PRIVATE |
177                                           MACVLAN_MODE_VEPA    |
178                                           MACVLAN_MODE_PASSTHRU|
179                                           MACVLAN_MODE_BRIDGE);
180                 else if (src->mode == MACVLAN_MODE_VEPA)
181                         /* flood to everyone except source */
182                         macvlan_broadcast(skb, port, src->dev,
183                                           MACVLAN_MODE_VEPA |
184                                           MACVLAN_MODE_BRIDGE);
185                 else if (src->mode == MACVLAN_MODE_BRIDGE)
186                         /*
187                          * flood only to VEPA ports, bridge ports
188                          * already saw the frame on the way out.
189                          */
190                         macvlan_broadcast(skb, port, src->dev,
191                                           MACVLAN_MODE_VEPA);
192                 return RX_HANDLER_PASS;
193         }
194
195         if (port->passthru)
196                 vlan = list_first_entry(&port->vlans, struct macvlan_dev, list);
197         else
198                 vlan = macvlan_hash_lookup(port, eth->h_dest);
199         if (vlan == NULL)
200                 return RX_HANDLER_PASS;
201
202         dev = vlan->dev;
203         if (unlikely(!(dev->flags & IFF_UP))) {
204                 kfree_skb(skb);
205                 return RX_HANDLER_CONSUMED;
206         }
207         len = skb->len + ETH_HLEN;
208         skb = skb_share_check(skb, GFP_ATOMIC);
209         if (!skb)
210                 goto out;
211
212         skb->dev = dev;
213         skb->pkt_type = PACKET_HOST;
214
215         ret = vlan->receive(skb);
216
217 out:
218         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, 0);
219         return RX_HANDLER_CONSUMED;
220 }
221
222 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
223 {
224         const struct macvlan_dev *vlan = netdev_priv(dev);
225         const struct macvlan_port *port = vlan->port;
226         const struct macvlan_dev *dest;
227         __u8 ip_summed = skb->ip_summed;
228
229         if (vlan->mode == MACVLAN_MODE_BRIDGE) {
230                 const struct ethhdr *eth = (void *)skb->data;
231                 skb->ip_summed = CHECKSUM_UNNECESSARY;
232
233                 /* send to other bridge ports directly */
234                 if (is_multicast_ether_addr(eth->h_dest)) {
235                         macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
236                         goto xmit_world;
237                 }
238
239                 dest = macvlan_hash_lookup(port, eth->h_dest);
240                 if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
241                         /* send to lowerdev first for its network taps */
242                         vlan->forward(vlan->lowerdev, skb);
243
244                         return NET_XMIT_SUCCESS;
245                 }
246         }
247
248 xmit_world:
249         skb->ip_summed = ip_summed;
250         skb_set_dev(skb, vlan->lowerdev);
251         return dev_queue_xmit(skb);
252 }
253
254 netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
255                                struct net_device *dev)
256 {
257         unsigned int len = skb->len;
258         int ret;
259         const struct macvlan_dev *vlan = netdev_priv(dev);
260
261         ret = macvlan_queue_xmit(skb, dev);
262         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
263                 struct macvlan_pcpu_stats *pcpu_stats;
264
265                 pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
266                 u64_stats_update_begin(&pcpu_stats->syncp);
267                 pcpu_stats->tx_packets++;
268                 pcpu_stats->tx_bytes += len;
269                 u64_stats_update_end(&pcpu_stats->syncp);
270         } else {
271                 this_cpu_inc(vlan->pcpu_stats->tx_dropped);
272         }
273         return ret;
274 }
275 EXPORT_SYMBOL_GPL(macvlan_start_xmit);
276
277 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
278                                unsigned short type, const void *daddr,
279                                const void *saddr, unsigned len)
280 {
281         const struct macvlan_dev *vlan = netdev_priv(dev);
282         struct net_device *lowerdev = vlan->lowerdev;
283
284         return dev_hard_header(skb, lowerdev, type, daddr,
285                                saddr ? : dev->dev_addr, len);
286 }
287
288 static const struct header_ops macvlan_hard_header_ops = {
289         .create         = macvlan_hard_header,
290         .rebuild        = eth_rebuild_header,
291         .parse          = eth_header_parse,
292         .cache          = eth_header_cache,
293         .cache_update   = eth_header_cache_update,
294 };
295
296 static int macvlan_open(struct net_device *dev)
297 {
298         struct macvlan_dev *vlan = netdev_priv(dev);
299         struct net_device *lowerdev = vlan->lowerdev;
300         int err;
301
302         if (vlan->port->passthru) {
303                 dev_set_promiscuity(lowerdev, 1);
304                 goto hash_add;
305         }
306
307         err = -EBUSY;
308         if (macvlan_addr_busy(vlan->port, dev->dev_addr))
309                 goto out;
310
311         err = dev_uc_add(lowerdev, dev->dev_addr);
312         if (err < 0)
313                 goto out;
314         if (dev->flags & IFF_ALLMULTI) {
315                 err = dev_set_allmulti(lowerdev, 1);
316                 if (err < 0)
317                         goto del_unicast;
318         }
319
320 hash_add:
321         macvlan_hash_add(vlan);
322         return 0;
323
324 del_unicast:
325         dev_uc_del(lowerdev, dev->dev_addr);
326 out:
327         return err;
328 }
329
330 static int macvlan_stop(struct net_device *dev)
331 {
332         struct macvlan_dev *vlan = netdev_priv(dev);
333         struct net_device *lowerdev = vlan->lowerdev;
334
335         if (vlan->port->passthru) {
336                 dev_set_promiscuity(lowerdev, -1);
337                 goto hash_del;
338         }
339
340         dev_mc_unsync(lowerdev, dev);
341         if (dev->flags & IFF_ALLMULTI)
342                 dev_set_allmulti(lowerdev, -1);
343
344         dev_uc_del(lowerdev, dev->dev_addr);
345
346 hash_del:
347         macvlan_hash_del(vlan, !dev->dismantle);
348         return 0;
349 }
350
351 static int macvlan_set_mac_address(struct net_device *dev, void *p)
352 {
353         struct macvlan_dev *vlan = netdev_priv(dev);
354         struct net_device *lowerdev = vlan->lowerdev;
355         struct sockaddr *addr = p;
356         int err;
357
358         if (!is_valid_ether_addr(addr->sa_data))
359                 return -EADDRNOTAVAIL;
360
361         if (!(dev->flags & IFF_UP)) {
362                 /* Just copy in the new address */
363                 memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
364         } else {
365                 /* Rehash and update the device filters */
366                 if (macvlan_addr_busy(vlan->port, addr->sa_data))
367                         return -EBUSY;
368
369                 err = dev_uc_add(lowerdev, addr->sa_data);
370                 if (err)
371                         return err;
372
373                 dev_uc_del(lowerdev, dev->dev_addr);
374
375                 macvlan_hash_change_addr(vlan, addr->sa_data);
376         }
377         return 0;
378 }
379
380 static void macvlan_change_rx_flags(struct net_device *dev, int change)
381 {
382         struct macvlan_dev *vlan = netdev_priv(dev);
383         struct net_device *lowerdev = vlan->lowerdev;
384
385         if (change & IFF_ALLMULTI)
386                 dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
387 }
388
389 static void macvlan_set_multicast_list(struct net_device *dev)
390 {
391         struct macvlan_dev *vlan = netdev_priv(dev);
392
393         dev_mc_sync(vlan->lowerdev, dev);
394 }
395
396 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
397 {
398         struct macvlan_dev *vlan = netdev_priv(dev);
399
400         if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
401                 return -EINVAL;
402         dev->mtu = new_mtu;
403         return 0;
404 }
405
406 /*
407  * macvlan network devices have devices nesting below it and are a special
408  * "super class" of normal network devices; split their locks off into a
409  * separate class since they always nest.
410  */
411 static struct lock_class_key macvlan_netdev_xmit_lock_key;
412 static struct lock_class_key macvlan_netdev_addr_lock_key;
413
414 #define MACVLAN_FEATURES \
415         (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
416          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
417          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
418          NETIF_F_HW_VLAN_FILTER)
419
420 #define MACVLAN_STATE_MASK \
421         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
422
423 static void macvlan_set_lockdep_class_one(struct net_device *dev,
424                                           struct netdev_queue *txq,
425                                           void *_unused)
426 {
427         lockdep_set_class(&txq->_xmit_lock,
428                           &macvlan_netdev_xmit_lock_key);
429 }
430
431 static void macvlan_set_lockdep_class(struct net_device *dev)
432 {
433         lockdep_set_class(&dev->addr_list_lock,
434                           &macvlan_netdev_addr_lock_key);
435         netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
436 }
437
438 static int macvlan_init(struct net_device *dev)
439 {
440         struct macvlan_dev *vlan = netdev_priv(dev);
441         const struct net_device *lowerdev = vlan->lowerdev;
442
443         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
444                                   (lowerdev->state & MACVLAN_STATE_MASK);
445         dev->features           = lowerdev->features & MACVLAN_FEATURES;
446         dev->features           |= NETIF_F_LLTX;
447         dev->gso_max_size       = lowerdev->gso_max_size;
448         dev->iflink             = lowerdev->ifindex;
449         dev->hard_header_len    = lowerdev->hard_header_len;
450
451         macvlan_set_lockdep_class(dev);
452
453         vlan->pcpu_stats = alloc_percpu(struct macvlan_pcpu_stats);
454         if (!vlan->pcpu_stats)
455                 return -ENOMEM;
456
457         return 0;
458 }
459
460 static void macvlan_uninit(struct net_device *dev)
461 {
462         struct macvlan_dev *vlan = netdev_priv(dev);
463         struct macvlan_port *port = vlan->port;
464
465         free_percpu(vlan->pcpu_stats);
466
467         port->count -= 1;
468         if (!port->count)
469                 macvlan_port_destroy(port->dev);
470 }
471
472 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
473                                                          struct rtnl_link_stats64 *stats)
474 {
475         struct macvlan_dev *vlan = netdev_priv(dev);
476
477         if (vlan->pcpu_stats) {
478                 struct macvlan_pcpu_stats *p;
479                 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
480                 u32 rx_errors = 0, tx_dropped = 0;
481                 unsigned int start;
482                 int i;
483
484                 for_each_possible_cpu(i) {
485                         p = per_cpu_ptr(vlan->pcpu_stats, i);
486                         do {
487                                 start = u64_stats_fetch_begin_bh(&p->syncp);
488                                 rx_packets      = p->rx_packets;
489                                 rx_bytes        = p->rx_bytes;
490                                 rx_multicast    = p->rx_multicast;
491                                 tx_packets      = p->tx_packets;
492                                 tx_bytes        = p->tx_bytes;
493                         } while (u64_stats_fetch_retry_bh(&p->syncp, start));
494
495                         stats->rx_packets       += rx_packets;
496                         stats->rx_bytes         += rx_bytes;
497                         stats->multicast        += rx_multicast;
498                         stats->tx_packets       += tx_packets;
499                         stats->tx_bytes         += tx_bytes;
500                         /* rx_errors & tx_dropped are u32, updated
501                          * without syncp protection.
502                          */
503                         rx_errors       += p->rx_errors;
504                         tx_dropped      += p->tx_dropped;
505                 }
506                 stats->rx_errors        = rx_errors;
507                 stats->rx_dropped       = rx_errors;
508                 stats->tx_dropped       = tx_dropped;
509         }
510         return stats;
511 }
512
513 static void macvlan_vlan_rx_add_vid(struct net_device *dev,
514                                     unsigned short vid)
515 {
516         struct macvlan_dev *vlan = netdev_priv(dev);
517         struct net_device *lowerdev = vlan->lowerdev;
518         const struct net_device_ops *ops = lowerdev->netdev_ops;
519
520         if (ops->ndo_vlan_rx_add_vid)
521                 ops->ndo_vlan_rx_add_vid(lowerdev, vid);
522 }
523
524 static void macvlan_vlan_rx_kill_vid(struct net_device *dev,
525                                      unsigned short vid)
526 {
527         struct macvlan_dev *vlan = netdev_priv(dev);
528         struct net_device *lowerdev = vlan->lowerdev;
529         const struct net_device_ops *ops = lowerdev->netdev_ops;
530
531         if (ops->ndo_vlan_rx_kill_vid)
532                 ops->ndo_vlan_rx_kill_vid(lowerdev, vid);
533 }
534
535 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
536                                         struct ethtool_drvinfo *drvinfo)
537 {
538         snprintf(drvinfo->driver, 32, "macvlan");
539         snprintf(drvinfo->version, 32, "0.1");
540 }
541
542 static int macvlan_ethtool_get_settings(struct net_device *dev,
543                                         struct ethtool_cmd *cmd)
544 {
545         const struct macvlan_dev *vlan = netdev_priv(dev);
546         return dev_ethtool_get_settings(vlan->lowerdev, cmd);
547 }
548
549 static const struct ethtool_ops macvlan_ethtool_ops = {
550         .get_link               = ethtool_op_get_link,
551         .get_settings           = macvlan_ethtool_get_settings,
552         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
553 };
554
555 static const struct net_device_ops macvlan_netdev_ops = {
556         .ndo_init               = macvlan_init,
557         .ndo_uninit             = macvlan_uninit,
558         .ndo_open               = macvlan_open,
559         .ndo_stop               = macvlan_stop,
560         .ndo_start_xmit         = macvlan_start_xmit,
561         .ndo_change_mtu         = macvlan_change_mtu,
562         .ndo_change_rx_flags    = macvlan_change_rx_flags,
563         .ndo_set_mac_address    = macvlan_set_mac_address,
564         .ndo_set_multicast_list = macvlan_set_multicast_list,
565         .ndo_get_stats64        = macvlan_dev_get_stats64,
566         .ndo_validate_addr      = eth_validate_addr,
567         .ndo_vlan_rx_add_vid    = macvlan_vlan_rx_add_vid,
568         .ndo_vlan_rx_kill_vid   = macvlan_vlan_rx_kill_vid,
569 };
570
571 void macvlan_common_setup(struct net_device *dev)
572 {
573         ether_setup(dev);
574
575         dev->priv_flags        &= ~IFF_XMIT_DST_RELEASE;
576         dev->netdev_ops         = &macvlan_netdev_ops;
577         dev->destructor         = free_netdev;
578         dev->header_ops         = &macvlan_hard_header_ops,
579         dev->ethtool_ops        = &macvlan_ethtool_ops;
580 }
581 EXPORT_SYMBOL_GPL(macvlan_common_setup);
582
583 static void macvlan_setup(struct net_device *dev)
584 {
585         macvlan_common_setup(dev);
586         dev->tx_queue_len       = 0;
587 }
588
589 static int macvlan_port_create(struct net_device *dev)
590 {
591         struct macvlan_port *port;
592         unsigned int i;
593         int err;
594
595         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
596                 return -EINVAL;
597
598         port = kzalloc(sizeof(*port), GFP_KERNEL);
599         if (port == NULL)
600                 return -ENOMEM;
601
602         port->passthru = false;
603         port->dev = dev;
604         INIT_LIST_HEAD(&port->vlans);
605         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
606                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
607
608         err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
609         if (err)
610                 kfree(port);
611         else
612                 dev->priv_flags |= IFF_MACVLAN_PORT;
613         return err;
614 }
615
616 static void macvlan_port_destroy(struct net_device *dev)
617 {
618         struct macvlan_port *port = macvlan_port_get(dev);
619
620         dev->priv_flags &= ~IFF_MACVLAN_PORT;
621         netdev_rx_handler_unregister(dev);
622         kfree_rcu(port, rcu);
623 }
624
625 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
626 {
627         if (tb[IFLA_ADDRESS]) {
628                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
629                         return -EINVAL;
630                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
631                         return -EADDRNOTAVAIL;
632         }
633
634         if (data && data[IFLA_MACVLAN_MODE]) {
635                 switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
636                 case MACVLAN_MODE_PRIVATE:
637                 case MACVLAN_MODE_VEPA:
638                 case MACVLAN_MODE_BRIDGE:
639                 case MACVLAN_MODE_PASSTHRU:
640                         break;
641                 default:
642                         return -EINVAL;
643                 }
644         }
645         return 0;
646 }
647
648 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
649                            struct nlattr *tb[], struct nlattr *data[],
650                            int (*receive)(struct sk_buff *skb),
651                            int (*forward)(struct net_device *dev,
652                                           struct sk_buff *skb))
653 {
654         struct macvlan_dev *vlan = netdev_priv(dev);
655         struct macvlan_port *port;
656         struct net_device *lowerdev;
657         int err;
658
659         if (!tb[IFLA_LINK])
660                 return -EINVAL;
661
662         lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
663         if (lowerdev == NULL)
664                 return -ENODEV;
665
666         /* When creating macvlans on top of other macvlans - use
667          * the real device as the lowerdev.
668          */
669         if (lowerdev->rtnl_link_ops == dev->rtnl_link_ops) {
670                 struct macvlan_dev *lowervlan = netdev_priv(lowerdev);
671                 lowerdev = lowervlan->lowerdev;
672         }
673
674         if (!tb[IFLA_MTU])
675                 dev->mtu = lowerdev->mtu;
676         else if (dev->mtu > lowerdev->mtu)
677                 return -EINVAL;
678
679         if (!tb[IFLA_ADDRESS])
680                 random_ether_addr(dev->dev_addr);
681
682         if (!macvlan_port_exists(lowerdev)) {
683                 err = macvlan_port_create(lowerdev);
684                 if (err < 0)
685                         return err;
686         }
687         port = macvlan_port_get(lowerdev);
688
689         /* Only 1 macvlan device can be created in passthru mode */
690         if (port->passthru)
691                 return -EINVAL;
692
693         vlan->lowerdev = lowerdev;
694         vlan->dev      = dev;
695         vlan->port     = port;
696         vlan->receive  = receive;
697         vlan->forward  = forward;
698
699         vlan->mode     = MACVLAN_MODE_VEPA;
700         if (data && data[IFLA_MACVLAN_MODE])
701                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
702
703         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
704                 if (port->count)
705                         return -EINVAL;
706                 port->passthru = true;
707                 memcpy(dev->dev_addr, lowerdev->dev_addr, ETH_ALEN);
708         }
709
710         port->count += 1;
711         err = register_netdevice(dev);
712         if (err < 0)
713                 goto destroy_port;
714
715         list_add_tail(&vlan->list, &port->vlans);
716         netif_stacked_transfer_operstate(lowerdev, dev);
717
718         return 0;
719
720 destroy_port:
721         port->count -= 1;
722         if (!port->count)
723                 macvlan_port_destroy(lowerdev);
724
725         return err;
726 }
727 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
728
729 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
730                            struct nlattr *tb[], struct nlattr *data[])
731 {
732         return macvlan_common_newlink(src_net, dev, tb, data,
733                                       netif_rx,
734                                       dev_forward_skb);
735 }
736
737 void macvlan_dellink(struct net_device *dev, struct list_head *head)
738 {
739         struct macvlan_dev *vlan = netdev_priv(dev);
740
741         list_del(&vlan->list);
742         unregister_netdevice_queue(dev, head);
743 }
744 EXPORT_SYMBOL_GPL(macvlan_dellink);
745
746 static int macvlan_changelink(struct net_device *dev,
747                 struct nlattr *tb[], struct nlattr *data[])
748 {
749         struct macvlan_dev *vlan = netdev_priv(dev);
750         if (data && data[IFLA_MACVLAN_MODE])
751                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
752         return 0;
753 }
754
755 static size_t macvlan_get_size(const struct net_device *dev)
756 {
757         return nla_total_size(4);
758 }
759
760 static int macvlan_fill_info(struct sk_buff *skb,
761                                 const struct net_device *dev)
762 {
763         struct macvlan_dev *vlan = netdev_priv(dev);
764
765         NLA_PUT_U32(skb, IFLA_MACVLAN_MODE, vlan->mode);
766         return 0;
767
768 nla_put_failure:
769         return -EMSGSIZE;
770 }
771
772 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
773         [IFLA_MACVLAN_MODE] = { .type = NLA_U32 },
774 };
775
776 int macvlan_link_register(struct rtnl_link_ops *ops)
777 {
778         /* common fields */
779         ops->priv_size          = sizeof(struct macvlan_dev);
780         ops->validate           = macvlan_validate;
781         ops->maxtype            = IFLA_MACVLAN_MAX;
782         ops->policy             = macvlan_policy;
783         ops->changelink         = macvlan_changelink;
784         ops->get_size           = macvlan_get_size;
785         ops->fill_info          = macvlan_fill_info;
786
787         return rtnl_link_register(ops);
788 };
789 EXPORT_SYMBOL_GPL(macvlan_link_register);
790
791 static struct rtnl_link_ops macvlan_link_ops = {
792         .kind           = "macvlan",
793         .setup          = macvlan_setup,
794         .newlink        = macvlan_newlink,
795         .dellink        = macvlan_dellink,
796 };
797
798 static int macvlan_device_event(struct notifier_block *unused,
799                                 unsigned long event, void *ptr)
800 {
801         struct net_device *dev = ptr;
802         struct macvlan_dev *vlan, *next;
803         struct macvlan_port *port;
804         LIST_HEAD(list_kill);
805
806         if (!macvlan_port_exists(dev))
807                 return NOTIFY_DONE;
808
809         port = macvlan_port_get(dev);
810
811         switch (event) {
812         case NETDEV_CHANGE:
813                 list_for_each_entry(vlan, &port->vlans, list)
814                         netif_stacked_transfer_operstate(vlan->lowerdev,
815                                                          vlan->dev);
816                 break;
817         case NETDEV_FEAT_CHANGE:
818                 list_for_each_entry(vlan, &port->vlans, list) {
819                         vlan->dev->features = dev->features & MACVLAN_FEATURES;
820                         vlan->dev->gso_max_size = dev->gso_max_size;
821                         netdev_features_change(vlan->dev);
822                 }
823                 break;
824         case NETDEV_UNREGISTER:
825                 /* twiddle thumbs on netns device moves */
826                 if (dev->reg_state != NETREG_UNREGISTERING)
827                         break;
828
829                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
830                         vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
831                 unregister_netdevice_many(&list_kill);
832                 list_del(&list_kill);
833                 break;
834         case NETDEV_PRE_TYPE_CHANGE:
835                 /* Forbid underlaying device to change its type. */
836                 return NOTIFY_BAD;
837         }
838         return NOTIFY_DONE;
839 }
840
841 static struct notifier_block macvlan_notifier_block __read_mostly = {
842         .notifier_call  = macvlan_device_event,
843 };
844
845 static int __init macvlan_init_module(void)
846 {
847         int err;
848
849         register_netdevice_notifier(&macvlan_notifier_block);
850
851         err = macvlan_link_register(&macvlan_link_ops);
852         if (err < 0)
853                 goto err1;
854         return 0;
855 err1:
856         unregister_netdevice_notifier(&macvlan_notifier_block);
857         return err;
858 }
859
860 static void __exit macvlan_cleanup_module(void)
861 {
862         rtnl_link_unregister(&macvlan_link_ops);
863         unregister_netdevice_notifier(&macvlan_notifier_block);
864 }
865
866 module_init(macvlan_init_module);
867 module_exit(macvlan_cleanup_module);
868
869 MODULE_LICENSE("GPL");
870 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
871 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
872 MODULE_ALIAS_RTNL_LINK("macvlan");