]> Pileus Git - ~andy/linux/blob - drivers/net/macvlan.c
Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/linville/wirel...
[~andy/linux] / 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_vlan.h>
30 #include <linux/if_link.h>
31 #include <linux/if_macvlan.h>
32 #include <linux/hash.h>
33 #include <net/rtnetlink.h>
34 #include <net/xfrm.h>
35
36 #define MACVLAN_HASH_SIZE       (1 << BITS_PER_BYTE)
37
38 struct macvlan_port {
39         struct net_device       *dev;
40         struct hlist_head       vlan_hash[MACVLAN_HASH_SIZE];
41         struct list_head        vlans;
42         struct rcu_head         rcu;
43         bool                    passthru;
44         int                     count;
45 };
46
47 static void macvlan_port_destroy(struct net_device *dev);
48
49 static struct macvlan_port *macvlan_port_get_rcu(const struct net_device *dev)
50 {
51         return rcu_dereference(dev->rx_handler_data);
52 }
53
54 static struct macvlan_port *macvlan_port_get_rtnl(const struct net_device *dev)
55 {
56         return rtnl_dereference(dev->rx_handler_data);
57 }
58
59 #define macvlan_port_exists(dev) (dev->priv_flags & IFF_MACVLAN_PORT)
60
61 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
62                                                const unsigned char *addr)
63 {
64         struct macvlan_dev *vlan;
65
66         hlist_for_each_entry_rcu(vlan, &port->vlan_hash[addr[5]], hlist) {
67                 if (ether_addr_equal_64bits(vlan->dev->dev_addr, addr))
68                         return vlan;
69         }
70         return NULL;
71 }
72
73 static void macvlan_hash_add(struct macvlan_dev *vlan)
74 {
75         struct macvlan_port *port = vlan->port;
76         const unsigned char *addr = vlan->dev->dev_addr;
77
78         hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[addr[5]]);
79 }
80
81 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync)
82 {
83         hlist_del_rcu(&vlan->hlist);
84         if (sync)
85                 synchronize_rcu();
86 }
87
88 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
89                                         const unsigned char *addr)
90 {
91         macvlan_hash_del(vlan, true);
92         /* Now that we are unhashed it is safe to change the device
93          * address without confusing packet delivery.
94          */
95         memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
96         macvlan_hash_add(vlan);
97 }
98
99 static int macvlan_addr_busy(const struct macvlan_port *port,
100                                 const unsigned char *addr)
101 {
102         /* Test to see if the specified multicast address is
103          * currently in use by the underlying device or
104          * another macvlan.
105          */
106         if (ether_addr_equal_64bits(port->dev->dev_addr, addr))
107                 return 1;
108
109         if (macvlan_hash_lookup(port, addr))
110                 return 1;
111
112         return 0;
113 }
114
115
116 static int macvlan_broadcast_one(struct sk_buff *skb,
117                                  const struct macvlan_dev *vlan,
118                                  const struct ethhdr *eth, bool local)
119 {
120         struct net_device *dev = vlan->dev;
121
122         if (local)
123                 return dev_forward_skb(dev, skb);
124
125         skb->dev = dev;
126         if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
127                 skb->pkt_type = PACKET_BROADCAST;
128         else
129                 skb->pkt_type = PACKET_MULTICAST;
130
131         return netif_rx(skb);
132 }
133
134 static u32 macvlan_hash_mix(const struct macvlan_dev *vlan)
135 {
136         return (u32)(((unsigned long)vlan) >> L1_CACHE_SHIFT);
137 }
138
139
140 static unsigned int mc_hash(const struct macvlan_dev *vlan,
141                             const unsigned char *addr)
142 {
143         u32 val = __get_unaligned_cpu32(addr + 2);
144
145         val ^= macvlan_hash_mix(vlan);
146         return hash_32(val, MACVLAN_MC_FILTER_BITS);
147 }
148
149 static void macvlan_broadcast(struct sk_buff *skb,
150                               const struct macvlan_port *port,
151                               struct net_device *src,
152                               enum macvlan_mode mode)
153 {
154         const struct ethhdr *eth = eth_hdr(skb);
155         const struct macvlan_dev *vlan;
156         struct sk_buff *nskb;
157         unsigned int i;
158         int err;
159         unsigned int hash;
160
161         if (skb->protocol == htons(ETH_P_PAUSE))
162                 return;
163
164         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
165                 hlist_for_each_entry_rcu(vlan, &port->vlan_hash[i], hlist) {
166                         if (vlan->dev == src || !(vlan->mode & mode))
167                                 continue;
168
169                         hash = mc_hash(vlan, eth->h_dest);
170                         if (!test_bit(hash, vlan->mc_filter))
171                                 continue;
172
173                         err = NET_RX_DROP;
174                         nskb = skb_clone(skb, GFP_ATOMIC);
175                         if (likely(nskb))
176                                 err = macvlan_broadcast_one(
177                                         nskb, vlan, eth,
178                                         mode == MACVLAN_MODE_BRIDGE);
179                         macvlan_count_rx(vlan, skb->len + ETH_HLEN,
180                                          err == NET_RX_SUCCESS, 1);
181                 }
182         }
183 }
184
185 /* called under rcu_read_lock() from netif_receive_skb */
186 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
187 {
188         struct macvlan_port *port;
189         struct sk_buff *skb = *pskb;
190         const struct ethhdr *eth = eth_hdr(skb);
191         const struct macvlan_dev *vlan;
192         const struct macvlan_dev *src;
193         struct net_device *dev;
194         unsigned int len = 0;
195         int ret = NET_RX_DROP;
196
197         port = macvlan_port_get_rcu(skb->dev);
198         if (is_multicast_ether_addr(eth->h_dest)) {
199                 skb = ip_check_defrag(skb, IP_DEFRAG_MACVLAN);
200                 if (!skb)
201                         return RX_HANDLER_CONSUMED;
202                 eth = eth_hdr(skb);
203                 src = macvlan_hash_lookup(port, eth->h_source);
204                 if (!src)
205                         /* frame comes from an external address */
206                         macvlan_broadcast(skb, port, NULL,
207                                           MACVLAN_MODE_PRIVATE |
208                                           MACVLAN_MODE_VEPA    |
209                                           MACVLAN_MODE_PASSTHRU|
210                                           MACVLAN_MODE_BRIDGE);
211                 else if (src->mode == MACVLAN_MODE_VEPA)
212                         /* flood to everyone except source */
213                         macvlan_broadcast(skb, port, src->dev,
214                                           MACVLAN_MODE_VEPA |
215                                           MACVLAN_MODE_BRIDGE);
216                 else if (src->mode == MACVLAN_MODE_BRIDGE)
217                         /*
218                          * flood only to VEPA ports, bridge ports
219                          * already saw the frame on the way out.
220                          */
221                         macvlan_broadcast(skb, port, src->dev,
222                                           MACVLAN_MODE_VEPA);
223                 else {
224                         /* forward to original port. */
225                         vlan = src;
226                         ret = macvlan_broadcast_one(skb, vlan, eth, 0);
227                         goto out;
228                 }
229
230                 return RX_HANDLER_PASS;
231         }
232
233         if (port->passthru)
234                 vlan = list_first_or_null_rcu(&port->vlans,
235                                               struct macvlan_dev, list);
236         else
237                 vlan = macvlan_hash_lookup(port, eth->h_dest);
238         if (vlan == NULL)
239                 return RX_HANDLER_PASS;
240
241         dev = vlan->dev;
242         if (unlikely(!(dev->flags & IFF_UP))) {
243                 kfree_skb(skb);
244                 return RX_HANDLER_CONSUMED;
245         }
246         len = skb->len + ETH_HLEN;
247         skb = skb_share_check(skb, GFP_ATOMIC);
248         if (!skb)
249                 goto out;
250
251         skb->dev = dev;
252         skb->pkt_type = PACKET_HOST;
253
254         ret = netif_rx(skb);
255
256 out:
257         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, 0);
258         return RX_HANDLER_CONSUMED;
259 }
260
261 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
262 {
263         const struct macvlan_dev *vlan = netdev_priv(dev);
264         const struct macvlan_port *port = vlan->port;
265         const struct macvlan_dev *dest;
266         __u8 ip_summed = skb->ip_summed;
267
268         if (vlan->mode == MACVLAN_MODE_BRIDGE) {
269                 const struct ethhdr *eth = (void *)skb->data;
270                 skb->ip_summed = CHECKSUM_UNNECESSARY;
271
272                 /* send to other bridge ports directly */
273                 if (is_multicast_ether_addr(eth->h_dest)) {
274                         macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
275                         goto xmit_world;
276                 }
277
278                 dest = macvlan_hash_lookup(port, eth->h_dest);
279                 if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
280                         /* send to lowerdev first for its network taps */
281                         dev_forward_skb(vlan->lowerdev, skb);
282
283                         return NET_XMIT_SUCCESS;
284                 }
285         }
286
287 xmit_world:
288         skb->ip_summed = ip_summed;
289         skb->dev = vlan->lowerdev;
290         return dev_queue_xmit(skb);
291 }
292
293 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
294                                       struct net_device *dev)
295 {
296         unsigned int len = skb->len;
297         int ret;
298         const struct macvlan_dev *vlan = netdev_priv(dev);
299
300         if (vlan->fwd_priv) {
301                 skb->dev = vlan->lowerdev;
302                 ret = dev_hard_start_xmit(skb, skb->dev, NULL, vlan->fwd_priv);
303         } else {
304                 ret = macvlan_queue_xmit(skb, dev);
305         }
306
307         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
308                 struct macvlan_pcpu_stats *pcpu_stats;
309
310                 pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
311                 u64_stats_update_begin(&pcpu_stats->syncp);
312                 pcpu_stats->tx_packets++;
313                 pcpu_stats->tx_bytes += len;
314                 u64_stats_update_end(&pcpu_stats->syncp);
315         } else {
316                 this_cpu_inc(vlan->pcpu_stats->tx_dropped);
317         }
318         return ret;
319 }
320
321 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
322                                unsigned short type, const void *daddr,
323                                const void *saddr, unsigned len)
324 {
325         const struct macvlan_dev *vlan = netdev_priv(dev);
326         struct net_device *lowerdev = vlan->lowerdev;
327
328         return dev_hard_header(skb, lowerdev, type, daddr,
329                                saddr ? : dev->dev_addr, len);
330 }
331
332 static const struct header_ops macvlan_hard_header_ops = {
333         .create         = macvlan_hard_header,
334         .rebuild        = eth_rebuild_header,
335         .parse          = eth_header_parse,
336         .cache          = eth_header_cache,
337         .cache_update   = eth_header_cache_update,
338 };
339
340 static int macvlan_open(struct net_device *dev)
341 {
342         struct macvlan_dev *vlan = netdev_priv(dev);
343         struct net_device *lowerdev = vlan->lowerdev;
344         int err;
345
346         if (vlan->port->passthru) {
347                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
348                         err = dev_set_promiscuity(lowerdev, 1);
349                         if (err < 0)
350                                 goto out;
351                 }
352                 goto hash_add;
353         }
354
355         if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD) {
356                 vlan->fwd_priv =
357                       lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
358
359                 /* If we get a NULL pointer back, or if we get an error
360                  * then we should just fall through to the non accelerated path
361                  */
362                 if (IS_ERR_OR_NULL(vlan->fwd_priv)) {
363                         vlan->fwd_priv = NULL;
364                 } else {
365                         dev->features &= ~NETIF_F_LLTX;
366                         return 0;
367                 }
368         }
369
370         err = -EBUSY;
371         if (macvlan_addr_busy(vlan->port, dev->dev_addr))
372                 goto out;
373
374         err = dev_uc_add(lowerdev, dev->dev_addr);
375         if (err < 0)
376                 goto out;
377         if (dev->flags & IFF_ALLMULTI) {
378                 err = dev_set_allmulti(lowerdev, 1);
379                 if (err < 0)
380                         goto del_unicast;
381         }
382
383 hash_add:
384         macvlan_hash_add(vlan);
385         return 0;
386
387 del_unicast:
388         dev_uc_del(lowerdev, dev->dev_addr);
389 out:
390         if (vlan->fwd_priv) {
391                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
392                                                            vlan->fwd_priv);
393                 vlan->fwd_priv = NULL;
394         }
395         return err;
396 }
397
398 static int macvlan_stop(struct net_device *dev)
399 {
400         struct macvlan_dev *vlan = netdev_priv(dev);
401         struct net_device *lowerdev = vlan->lowerdev;
402
403         if (vlan->fwd_priv) {
404                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
405                                                            vlan->fwd_priv);
406                 vlan->fwd_priv = NULL;
407                 return 0;
408         }
409
410         dev_uc_unsync(lowerdev, dev);
411         dev_mc_unsync(lowerdev, dev);
412
413         if (vlan->port->passthru) {
414                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
415                         dev_set_promiscuity(lowerdev, -1);
416                 goto hash_del;
417         }
418
419         if (dev->flags & IFF_ALLMULTI)
420                 dev_set_allmulti(lowerdev, -1);
421
422         dev_uc_del(lowerdev, dev->dev_addr);
423
424 hash_del:
425         macvlan_hash_del(vlan, !dev->dismantle);
426         return 0;
427 }
428
429 static int macvlan_set_mac_address(struct net_device *dev, void *p)
430 {
431         struct macvlan_dev *vlan = netdev_priv(dev);
432         struct net_device *lowerdev = vlan->lowerdev;
433         struct sockaddr *addr = p;
434         int err;
435
436         if (!is_valid_ether_addr(addr->sa_data))
437                 return -EADDRNOTAVAIL;
438
439         if (!(dev->flags & IFF_UP)) {
440                 /* Just copy in the new address */
441                 memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
442         } else {
443                 /* Rehash and update the device filters */
444                 if (macvlan_addr_busy(vlan->port, addr->sa_data))
445                         return -EBUSY;
446
447                 err = dev_uc_add(lowerdev, addr->sa_data);
448                 if (err)
449                         return err;
450
451                 dev_uc_del(lowerdev, dev->dev_addr);
452
453                 macvlan_hash_change_addr(vlan, addr->sa_data);
454         }
455         return 0;
456 }
457
458 static void macvlan_change_rx_flags(struct net_device *dev, int change)
459 {
460         struct macvlan_dev *vlan = netdev_priv(dev);
461         struct net_device *lowerdev = vlan->lowerdev;
462
463         if (change & IFF_ALLMULTI)
464                 dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
465 }
466
467 static void macvlan_set_mac_lists(struct net_device *dev)
468 {
469         struct macvlan_dev *vlan = netdev_priv(dev);
470
471         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
472                 bitmap_fill(vlan->mc_filter, MACVLAN_MC_FILTER_SZ);
473         } else {
474                 struct netdev_hw_addr *ha;
475                 DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
476
477                 bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
478                 netdev_for_each_mc_addr(ha, dev) {
479                         __set_bit(mc_hash(vlan, ha->addr), filter);
480                 }
481
482                 __set_bit(mc_hash(vlan, dev->broadcast), filter);
483
484                 bitmap_copy(vlan->mc_filter, filter, MACVLAN_MC_FILTER_SZ);
485         }
486         dev_uc_sync(vlan->lowerdev, dev);
487         dev_mc_sync(vlan->lowerdev, dev);
488 }
489
490 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
491 {
492         struct macvlan_dev *vlan = netdev_priv(dev);
493
494         if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
495                 return -EINVAL;
496         dev->mtu = new_mtu;
497         return 0;
498 }
499
500 /*
501  * macvlan network devices have devices nesting below it and are a special
502  * "super class" of normal network devices; split their locks off into a
503  * separate class since they always nest.
504  */
505 static struct lock_class_key macvlan_netdev_xmit_lock_key;
506 static struct lock_class_key macvlan_netdev_addr_lock_key;
507
508 #define MACVLAN_FEATURES \
509         (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
510          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
511          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
512          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
513
514 #define MACVLAN_STATE_MASK \
515         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
516
517 static void macvlan_set_lockdep_class_one(struct net_device *dev,
518                                           struct netdev_queue *txq,
519                                           void *_unused)
520 {
521         lockdep_set_class(&txq->_xmit_lock,
522                           &macvlan_netdev_xmit_lock_key);
523 }
524
525 static void macvlan_set_lockdep_class(struct net_device *dev)
526 {
527         lockdep_set_class(&dev->addr_list_lock,
528                           &macvlan_netdev_addr_lock_key);
529         netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
530 }
531
532 static int macvlan_init(struct net_device *dev)
533 {
534         struct macvlan_dev *vlan = netdev_priv(dev);
535         const struct net_device *lowerdev = vlan->lowerdev;
536         int i;
537
538         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
539                                   (lowerdev->state & MACVLAN_STATE_MASK);
540         dev->features           = lowerdev->features & MACVLAN_FEATURES;
541         dev->features           |= NETIF_F_LLTX;
542         dev->gso_max_size       = lowerdev->gso_max_size;
543         dev->iflink             = lowerdev->ifindex;
544         dev->hard_header_len    = lowerdev->hard_header_len;
545
546         macvlan_set_lockdep_class(dev);
547
548         vlan->pcpu_stats = alloc_percpu(struct macvlan_pcpu_stats);
549         if (!vlan->pcpu_stats)
550                 return -ENOMEM;
551
552         for_each_possible_cpu(i) {
553                 struct macvlan_pcpu_stats *mvlstats;
554                 mvlstats = per_cpu_ptr(vlan->pcpu_stats, i);
555                 u64_stats_init(&mvlstats->syncp);
556         }
557
558         return 0;
559 }
560
561 static void macvlan_uninit(struct net_device *dev)
562 {
563         struct macvlan_dev *vlan = netdev_priv(dev);
564         struct macvlan_port *port = vlan->port;
565
566         free_percpu(vlan->pcpu_stats);
567
568         port->count -= 1;
569         if (!port->count)
570                 macvlan_port_destroy(port->dev);
571 }
572
573 static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev,
574                                                          struct rtnl_link_stats64 *stats)
575 {
576         struct macvlan_dev *vlan = netdev_priv(dev);
577
578         if (vlan->pcpu_stats) {
579                 struct macvlan_pcpu_stats *p;
580                 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
581                 u32 rx_errors = 0, tx_dropped = 0;
582                 unsigned int start;
583                 int i;
584
585                 for_each_possible_cpu(i) {
586                         p = per_cpu_ptr(vlan->pcpu_stats, i);
587                         do {
588                                 start = u64_stats_fetch_begin_bh(&p->syncp);
589                                 rx_packets      = p->rx_packets;
590                                 rx_bytes        = p->rx_bytes;
591                                 rx_multicast    = p->rx_multicast;
592                                 tx_packets      = p->tx_packets;
593                                 tx_bytes        = p->tx_bytes;
594                         } while (u64_stats_fetch_retry_bh(&p->syncp, start));
595
596                         stats->rx_packets       += rx_packets;
597                         stats->rx_bytes         += rx_bytes;
598                         stats->multicast        += rx_multicast;
599                         stats->tx_packets       += tx_packets;
600                         stats->tx_bytes         += tx_bytes;
601                         /* rx_errors & tx_dropped are u32, updated
602                          * without syncp protection.
603                          */
604                         rx_errors       += p->rx_errors;
605                         tx_dropped      += p->tx_dropped;
606                 }
607                 stats->rx_errors        = rx_errors;
608                 stats->rx_dropped       = rx_errors;
609                 stats->tx_dropped       = tx_dropped;
610         }
611         return stats;
612 }
613
614 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
615                                    __be16 proto, u16 vid)
616 {
617         struct macvlan_dev *vlan = netdev_priv(dev);
618         struct net_device *lowerdev = vlan->lowerdev;
619
620         return vlan_vid_add(lowerdev, proto, vid);
621 }
622
623 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
624                                     __be16 proto, u16 vid)
625 {
626         struct macvlan_dev *vlan = netdev_priv(dev);
627         struct net_device *lowerdev = vlan->lowerdev;
628
629         vlan_vid_del(lowerdev, proto, vid);
630         return 0;
631 }
632
633 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
634                            struct net_device *dev,
635                            const unsigned char *addr,
636                            u16 flags)
637 {
638         struct macvlan_dev *vlan = netdev_priv(dev);
639         int err = -EINVAL;
640
641         if (!vlan->port->passthru)
642                 return -EOPNOTSUPP;
643
644         if (flags & NLM_F_REPLACE)
645                 return -EOPNOTSUPP;
646
647         if (is_unicast_ether_addr(addr))
648                 err = dev_uc_add_excl(dev, addr);
649         else if (is_multicast_ether_addr(addr))
650                 err = dev_mc_add_excl(dev, addr);
651
652         return err;
653 }
654
655 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
656                            struct net_device *dev,
657                            const unsigned char *addr)
658 {
659         struct macvlan_dev *vlan = netdev_priv(dev);
660         int err = -EINVAL;
661
662         if (!vlan->port->passthru)
663                 return -EOPNOTSUPP;
664
665         if (is_unicast_ether_addr(addr))
666                 err = dev_uc_del(dev, addr);
667         else if (is_multicast_ether_addr(addr))
668                 err = dev_mc_del(dev, addr);
669
670         return err;
671 }
672
673 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
674                                         struct ethtool_drvinfo *drvinfo)
675 {
676         strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
677         strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
678 }
679
680 static int macvlan_ethtool_get_settings(struct net_device *dev,
681                                         struct ethtool_cmd *cmd)
682 {
683         const struct macvlan_dev *vlan = netdev_priv(dev);
684
685         return __ethtool_get_settings(vlan->lowerdev, cmd);
686 }
687
688 static netdev_features_t macvlan_fix_features(struct net_device *dev,
689                                               netdev_features_t features)
690 {
691         struct macvlan_dev *vlan = netdev_priv(dev);
692
693         return features & (vlan->set_features | ~MACVLAN_FEATURES);
694 }
695
696 static const struct ethtool_ops macvlan_ethtool_ops = {
697         .get_link               = ethtool_op_get_link,
698         .get_settings           = macvlan_ethtool_get_settings,
699         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
700 };
701
702 static const struct net_device_ops macvlan_netdev_ops = {
703         .ndo_init               = macvlan_init,
704         .ndo_uninit             = macvlan_uninit,
705         .ndo_open               = macvlan_open,
706         .ndo_stop               = macvlan_stop,
707         .ndo_start_xmit         = macvlan_start_xmit,
708         .ndo_change_mtu         = macvlan_change_mtu,
709         .ndo_fix_features       = macvlan_fix_features,
710         .ndo_change_rx_flags    = macvlan_change_rx_flags,
711         .ndo_set_mac_address    = macvlan_set_mac_address,
712         .ndo_set_rx_mode        = macvlan_set_mac_lists,
713         .ndo_get_stats64        = macvlan_dev_get_stats64,
714         .ndo_validate_addr      = eth_validate_addr,
715         .ndo_vlan_rx_add_vid    = macvlan_vlan_rx_add_vid,
716         .ndo_vlan_rx_kill_vid   = macvlan_vlan_rx_kill_vid,
717         .ndo_fdb_add            = macvlan_fdb_add,
718         .ndo_fdb_del            = macvlan_fdb_del,
719         .ndo_fdb_dump           = ndo_dflt_fdb_dump,
720 };
721
722 void macvlan_common_setup(struct net_device *dev)
723 {
724         ether_setup(dev);
725
726         dev->priv_flags        &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
727         dev->priv_flags        |= IFF_UNICAST_FLT;
728         dev->netdev_ops         = &macvlan_netdev_ops;
729         dev->destructor         = free_netdev;
730         dev->header_ops         = &macvlan_hard_header_ops;
731         dev->ethtool_ops        = &macvlan_ethtool_ops;
732 }
733 EXPORT_SYMBOL_GPL(macvlan_common_setup);
734
735 static void macvlan_setup(struct net_device *dev)
736 {
737         macvlan_common_setup(dev);
738         dev->tx_queue_len       = 0;
739 }
740
741 static int macvlan_port_create(struct net_device *dev)
742 {
743         struct macvlan_port *port;
744         unsigned int i;
745         int err;
746
747         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
748                 return -EINVAL;
749
750         port = kzalloc(sizeof(*port), GFP_KERNEL);
751         if (port == NULL)
752                 return -ENOMEM;
753
754         port->passthru = false;
755         port->dev = dev;
756         INIT_LIST_HEAD(&port->vlans);
757         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
758                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
759
760         err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
761         if (err)
762                 kfree(port);
763         else
764                 dev->priv_flags |= IFF_MACVLAN_PORT;
765         return err;
766 }
767
768 static void macvlan_port_destroy(struct net_device *dev)
769 {
770         struct macvlan_port *port = macvlan_port_get_rtnl(dev);
771
772         dev->priv_flags &= ~IFF_MACVLAN_PORT;
773         netdev_rx_handler_unregister(dev);
774         kfree_rcu(port, rcu);
775 }
776
777 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
778 {
779         if (tb[IFLA_ADDRESS]) {
780                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
781                         return -EINVAL;
782                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
783                         return -EADDRNOTAVAIL;
784         }
785
786         if (data && data[IFLA_MACVLAN_FLAGS] &&
787             nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
788                 return -EINVAL;
789
790         if (data && data[IFLA_MACVLAN_MODE]) {
791                 switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
792                 case MACVLAN_MODE_PRIVATE:
793                 case MACVLAN_MODE_VEPA:
794                 case MACVLAN_MODE_BRIDGE:
795                 case MACVLAN_MODE_PASSTHRU:
796                         break;
797                 default:
798                         return -EINVAL;
799                 }
800         }
801         return 0;
802 }
803
804 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
805                            struct nlattr *tb[], struct nlattr *data[])
806 {
807         struct macvlan_dev *vlan = netdev_priv(dev);
808         struct macvlan_port *port;
809         struct net_device *lowerdev;
810         int err;
811
812         if (!tb[IFLA_LINK])
813                 return -EINVAL;
814
815         lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
816         if (lowerdev == NULL)
817                 return -ENODEV;
818
819         /* When creating macvlans or macvtaps on top of other macvlans - use
820          * the real device as the lowerdev.
821          */
822         if (netif_is_macvlan(lowerdev))
823                 lowerdev = macvlan_dev_real_dev(lowerdev);
824
825         if (!tb[IFLA_MTU])
826                 dev->mtu = lowerdev->mtu;
827         else if (dev->mtu > lowerdev->mtu)
828                 return -EINVAL;
829
830         if (!tb[IFLA_ADDRESS])
831                 eth_hw_addr_random(dev);
832
833         if (!macvlan_port_exists(lowerdev)) {
834                 err = macvlan_port_create(lowerdev);
835                 if (err < 0)
836                         return err;
837         }
838         port = macvlan_port_get_rtnl(lowerdev);
839
840         /* Only 1 macvlan device can be created in passthru mode */
841         if (port->passthru)
842                 return -EINVAL;
843
844         vlan->lowerdev = lowerdev;
845         vlan->dev      = dev;
846         vlan->port     = port;
847         vlan->set_features = MACVLAN_FEATURES;
848
849         vlan->mode     = MACVLAN_MODE_VEPA;
850         if (data && data[IFLA_MACVLAN_MODE])
851                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
852
853         if (data && data[IFLA_MACVLAN_FLAGS])
854                 vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
855
856         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
857                 if (port->count)
858                         return -EINVAL;
859                 port->passthru = true;
860                 eth_hw_addr_inherit(dev, lowerdev);
861         }
862
863         port->count += 1;
864         err = register_netdevice(dev);
865         if (err < 0)
866                 goto destroy_port;
867
868         dev->priv_flags |= IFF_MACVLAN;
869         err = netdev_upper_dev_link(lowerdev, dev);
870         if (err)
871                 goto destroy_port;
872
873
874         list_add_tail_rcu(&vlan->list, &port->vlans);
875         netif_stacked_transfer_operstate(lowerdev, dev);
876
877         return 0;
878
879 destroy_port:
880         port->count -= 1;
881         if (!port->count)
882                 macvlan_port_destroy(lowerdev);
883
884         return err;
885 }
886 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
887
888 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
889                            struct nlattr *tb[], struct nlattr *data[])
890 {
891         return macvlan_common_newlink(src_net, dev, tb, data);
892 }
893
894 void macvlan_dellink(struct net_device *dev, struct list_head *head)
895 {
896         struct macvlan_dev *vlan = netdev_priv(dev);
897
898         list_del_rcu(&vlan->list);
899         unregister_netdevice_queue(dev, head);
900         netdev_upper_dev_unlink(vlan->lowerdev, dev);
901 }
902 EXPORT_SYMBOL_GPL(macvlan_dellink);
903
904 static int macvlan_changelink(struct net_device *dev,
905                 struct nlattr *tb[], struct nlattr *data[])
906 {
907         struct macvlan_dev *vlan = netdev_priv(dev);
908         enum macvlan_mode mode;
909         bool set_mode = false;
910
911         /* Validate mode, but don't set yet: setting flags may fail. */
912         if (data && data[IFLA_MACVLAN_MODE]) {
913                 set_mode = true;
914                 mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
915                 /* Passthrough mode can't be set or cleared dynamically */
916                 if ((mode == MACVLAN_MODE_PASSTHRU) !=
917                     (vlan->mode == MACVLAN_MODE_PASSTHRU))
918                         return -EINVAL;
919         }
920
921         if (data && data[IFLA_MACVLAN_FLAGS]) {
922                 __u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
923                 bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
924                 if (vlan->port->passthru && promisc) {
925                         int err;
926
927                         if (flags & MACVLAN_FLAG_NOPROMISC)
928                                 err = dev_set_promiscuity(vlan->lowerdev, -1);
929                         else
930                                 err = dev_set_promiscuity(vlan->lowerdev, 1);
931                         if (err < 0)
932                                 return err;
933                 }
934                 vlan->flags = flags;
935         }
936         if (set_mode)
937                 vlan->mode = mode;
938         return 0;
939 }
940
941 static size_t macvlan_get_size(const struct net_device *dev)
942 {
943         return (0
944                 + nla_total_size(4) /* IFLA_MACVLAN_MODE */
945                 + nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
946                 );
947 }
948
949 static int macvlan_fill_info(struct sk_buff *skb,
950                                 const struct net_device *dev)
951 {
952         struct macvlan_dev *vlan = netdev_priv(dev);
953
954         if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
955                 goto nla_put_failure;
956         if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
957                 goto nla_put_failure;
958         return 0;
959
960 nla_put_failure:
961         return -EMSGSIZE;
962 }
963
964 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
965         [IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
966         [IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
967 };
968
969 int macvlan_link_register(struct rtnl_link_ops *ops)
970 {
971         /* common fields */
972         ops->priv_size          = sizeof(struct macvlan_dev);
973         ops->validate           = macvlan_validate;
974         ops->maxtype            = IFLA_MACVLAN_MAX;
975         ops->policy             = macvlan_policy;
976         ops->changelink         = macvlan_changelink;
977         ops->get_size           = macvlan_get_size;
978         ops->fill_info          = macvlan_fill_info;
979
980         return rtnl_link_register(ops);
981 };
982 EXPORT_SYMBOL_GPL(macvlan_link_register);
983
984 static struct rtnl_link_ops macvlan_link_ops = {
985         .kind           = "macvlan",
986         .setup          = macvlan_setup,
987         .newlink        = macvlan_newlink,
988         .dellink        = macvlan_dellink,
989 };
990
991 static int macvlan_device_event(struct notifier_block *unused,
992                                 unsigned long event, void *ptr)
993 {
994         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
995         struct macvlan_dev *vlan, *next;
996         struct macvlan_port *port;
997         LIST_HEAD(list_kill);
998
999         if (!macvlan_port_exists(dev))
1000                 return NOTIFY_DONE;
1001
1002         port = macvlan_port_get_rtnl(dev);
1003
1004         switch (event) {
1005         case NETDEV_CHANGE:
1006                 list_for_each_entry(vlan, &port->vlans, list)
1007                         netif_stacked_transfer_operstate(vlan->lowerdev,
1008                                                          vlan->dev);
1009                 break;
1010         case NETDEV_FEAT_CHANGE:
1011                 list_for_each_entry(vlan, &port->vlans, list) {
1012                         vlan->dev->features = dev->features & MACVLAN_FEATURES;
1013                         vlan->dev->gso_max_size = dev->gso_max_size;
1014                         netdev_features_change(vlan->dev);
1015                 }
1016                 break;
1017         case NETDEV_UNREGISTER:
1018                 /* twiddle thumbs on netns device moves */
1019                 if (dev->reg_state != NETREG_UNREGISTERING)
1020                         break;
1021
1022                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
1023                         vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1024                 unregister_netdevice_many(&list_kill);
1025                 list_del(&list_kill);
1026                 break;
1027         case NETDEV_PRE_TYPE_CHANGE:
1028                 /* Forbid underlaying device to change its type. */
1029                 return NOTIFY_BAD;
1030         }
1031         return NOTIFY_DONE;
1032 }
1033
1034 static struct notifier_block macvlan_notifier_block __read_mostly = {
1035         .notifier_call  = macvlan_device_event,
1036 };
1037
1038 static int __init macvlan_init_module(void)
1039 {
1040         int err;
1041
1042         register_netdevice_notifier(&macvlan_notifier_block);
1043
1044         err = macvlan_link_register(&macvlan_link_ops);
1045         if (err < 0)
1046                 goto err1;
1047         return 0;
1048 err1:
1049         unregister_netdevice_notifier(&macvlan_notifier_block);
1050         return err;
1051 }
1052
1053 static void __exit macvlan_cleanup_module(void)
1054 {
1055         rtnl_link_unregister(&macvlan_link_ops);
1056         unregister_netdevice_notifier(&macvlan_notifier_block);
1057 }
1058
1059 module_init(macvlan_init_module);
1060 module_exit(macvlan_cleanup_module);
1061
1062 MODULE_LICENSE("GPL");
1063 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1064 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1065 MODULE_ALIAS_RTNL_LINK("macvlan");