Commit | Line | Data |
---|---|---|
b863ceb7 PM |
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/list.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 | ||
33 | #define MACVLAN_HASH_SIZE (1 << BITS_PER_BYTE) | |
34 | ||
35 | struct macvlan_port { | |
36 | struct net_device *dev; | |
37 | struct hlist_head vlan_hash[MACVLAN_HASH_SIZE]; | |
38 | struct list_head vlans; | |
39 | }; | |
40 | ||
41 | struct macvlan_dev { | |
42 | struct net_device *dev; | |
43 | struct list_head list; | |
44 | struct hlist_node hlist; | |
45 | struct macvlan_port *port; | |
46 | struct net_device *lowerdev; | |
47 | }; | |
48 | ||
49 | ||
50 | static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port, | |
51 | const unsigned char *addr) | |
52 | { | |
53 | struct macvlan_dev *vlan; | |
54 | struct hlist_node *n; | |
55 | ||
56 | hlist_for_each_entry_rcu(vlan, n, &port->vlan_hash[addr[5]], hlist) { | |
57 | if (!compare_ether_addr(vlan->dev->dev_addr, addr)) | |
58 | return vlan; | |
59 | } | |
60 | return NULL; | |
61 | } | |
62 | ||
63 | static void macvlan_broadcast(struct sk_buff *skb, | |
64 | const struct macvlan_port *port) | |
65 | { | |
66 | const struct ethhdr *eth = eth_hdr(skb); | |
67 | const struct macvlan_dev *vlan; | |
68 | struct hlist_node *n; | |
69 | struct net_device *dev; | |
70 | struct sk_buff *nskb; | |
71 | unsigned int i; | |
72 | ||
73 | for (i = 0; i < MACVLAN_HASH_SIZE; i++) { | |
74 | hlist_for_each_entry_rcu(vlan, n, &port->vlan_hash[i], hlist) { | |
75 | dev = vlan->dev; | |
76 | if (unlikely(!(dev->flags & IFF_UP))) | |
77 | continue; | |
78 | ||
79 | nskb = skb_clone(skb, GFP_ATOMIC); | |
80 | if (nskb == NULL) { | |
81 | dev->stats.rx_errors++; | |
82 | dev->stats.rx_dropped++; | |
83 | continue; | |
84 | } | |
85 | ||
86 | dev->stats.rx_bytes += skb->len + ETH_HLEN; | |
87 | dev->stats.rx_packets++; | |
88 | dev->stats.multicast++; | |
89 | dev->last_rx = jiffies; | |
90 | ||
91 | nskb->dev = dev; | |
92 | if (!compare_ether_addr(eth->h_dest, dev->broadcast)) | |
93 | nskb->pkt_type = PACKET_BROADCAST; | |
94 | else | |
95 | nskb->pkt_type = PACKET_MULTICAST; | |
96 | ||
97 | netif_rx(nskb); | |
98 | } | |
99 | } | |
100 | } | |
101 | ||
102 | /* called under rcu_read_lock() from netif_receive_skb */ | |
103 | static struct sk_buff *macvlan_handle_frame(struct sk_buff *skb) | |
104 | { | |
105 | const struct ethhdr *eth = eth_hdr(skb); | |
106 | const struct macvlan_port *port; | |
107 | const struct macvlan_dev *vlan; | |
108 | struct net_device *dev; | |
109 | ||
110 | port = rcu_dereference(skb->dev->macvlan_port); | |
111 | if (port == NULL) | |
112 | return skb; | |
113 | ||
114 | if (is_multicast_ether_addr(eth->h_dest)) { | |
115 | macvlan_broadcast(skb, port); | |
116 | return skb; | |
117 | } | |
118 | ||
119 | vlan = macvlan_hash_lookup(port, eth->h_dest); | |
120 | if (vlan == NULL) | |
121 | return skb; | |
122 | ||
123 | dev = vlan->dev; | |
124 | if (unlikely(!(dev->flags & IFF_UP))) { | |
125 | kfree_skb(skb); | |
126 | return NULL; | |
127 | } | |
128 | ||
129 | skb = skb_share_check(skb, GFP_ATOMIC); | |
130 | if (skb == NULL) { | |
131 | dev->stats.rx_errors++; | |
132 | dev->stats.rx_dropped++; | |
133 | return NULL; | |
134 | } | |
135 | ||
136 | dev->stats.rx_bytes += skb->len + ETH_HLEN; | |
137 | dev->stats.rx_packets++; | |
138 | dev->last_rx = jiffies; | |
139 | ||
140 | skb->dev = dev; | |
141 | skb->pkt_type = PACKET_HOST; | |
142 | ||
143 | netif_rx(skb); | |
144 | return NULL; | |
145 | } | |
146 | ||
147 | static int macvlan_hard_start_xmit(struct sk_buff *skb, struct net_device *dev) | |
148 | { | |
149 | const struct macvlan_dev *vlan = netdev_priv(dev); | |
150 | unsigned int len = skb->len; | |
151 | int ret; | |
152 | ||
153 | skb->dev = vlan->lowerdev; | |
154 | ret = dev_queue_xmit(skb); | |
155 | ||
156 | if (likely(ret == NET_XMIT_SUCCESS)) { | |
157 | dev->stats.tx_packets++; | |
158 | dev->stats.tx_bytes += len; | |
159 | } else { | |
160 | dev->stats.tx_errors++; | |
161 | dev->stats.tx_aborted_errors++; | |
162 | } | |
163 | return NETDEV_TX_OK; | |
164 | } | |
165 | ||
166 | static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev, | |
3b04ddde SH |
167 | unsigned short type, const void *daddr, |
168 | const void *saddr, unsigned len) | |
b863ceb7 PM |
169 | { |
170 | const struct macvlan_dev *vlan = netdev_priv(dev); | |
171 | struct net_device *lowerdev = vlan->lowerdev; | |
172 | ||
0c4e8581 SH |
173 | return dev_hard_header(skb, lowerdev, type, daddr, |
174 | saddr ? : dev->dev_addr, len); | |
b863ceb7 PM |
175 | } |
176 | ||
3b04ddde SH |
177 | static const struct header_ops macvlan_hard_header_ops = { |
178 | .create = macvlan_hard_header, | |
179 | .rebuild = eth_rebuild_header, | |
180 | .parse = eth_header_parse, | |
3b04ddde SH |
181 | .cache = eth_header_cache, |
182 | .cache_update = eth_header_cache_update, | |
183 | }; | |
184 | ||
b863ceb7 PM |
185 | static int macvlan_open(struct net_device *dev) |
186 | { | |
187 | struct macvlan_dev *vlan = netdev_priv(dev); | |
188 | struct macvlan_port *port = vlan->port; | |
189 | struct net_device *lowerdev = vlan->lowerdev; | |
190 | int err; | |
191 | ||
192 | err = dev_unicast_add(lowerdev, dev->dev_addr, ETH_ALEN); | |
193 | if (err < 0) | |
194 | return err; | |
195 | if (dev->flags & IFF_ALLMULTI) | |
196 | dev_set_allmulti(lowerdev, 1); | |
197 | ||
198 | hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[dev->dev_addr[5]]); | |
199 | return 0; | |
200 | } | |
201 | ||
202 | static int macvlan_stop(struct net_device *dev) | |
203 | { | |
204 | struct macvlan_dev *vlan = netdev_priv(dev); | |
205 | struct net_device *lowerdev = vlan->lowerdev; | |
206 | ||
207 | dev_mc_unsync(lowerdev, dev); | |
208 | if (dev->flags & IFF_ALLMULTI) | |
209 | dev_set_allmulti(lowerdev, -1); | |
210 | ||
211 | dev_unicast_delete(lowerdev, dev->dev_addr, ETH_ALEN); | |
212 | ||
213 | hlist_del_rcu(&vlan->hlist); | |
214 | synchronize_rcu(); | |
215 | return 0; | |
216 | } | |
217 | ||
218 | static void macvlan_change_rx_flags(struct net_device *dev, int change) | |
219 | { | |
220 | struct macvlan_dev *vlan = netdev_priv(dev); | |
221 | struct net_device *lowerdev = vlan->lowerdev; | |
222 | ||
223 | if (change & IFF_ALLMULTI) | |
224 | dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1); | |
225 | } | |
226 | ||
227 | static void macvlan_set_multicast_list(struct net_device *dev) | |
228 | { | |
229 | struct macvlan_dev *vlan = netdev_priv(dev); | |
230 | ||
231 | dev_mc_sync(vlan->lowerdev, dev); | |
232 | } | |
233 | ||
234 | static int macvlan_change_mtu(struct net_device *dev, int new_mtu) | |
235 | { | |
236 | struct macvlan_dev *vlan = netdev_priv(dev); | |
237 | ||
238 | if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu) | |
239 | return -EINVAL; | |
240 | dev->mtu = new_mtu; | |
241 | return 0; | |
242 | } | |
243 | ||
244 | /* | |
245 | * macvlan network devices have devices nesting below it and are a special | |
246 | * "super class" of normal network devices; split their locks off into a | |
247 | * separate class since they always nest. | |
248 | */ | |
249 | static struct lock_class_key macvlan_netdev_xmit_lock_key; | |
250 | ||
251 | #define MACVLAN_FEATURES \ | |
252 | (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \ | |
253 | NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \ | |
254 | NETIF_F_TSO_ECN | NETIF_F_TSO6) | |
255 | ||
256 | #define MACVLAN_STATE_MASK \ | |
257 | ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT)) | |
258 | ||
259 | static int macvlan_init(struct net_device *dev) | |
260 | { | |
261 | struct macvlan_dev *vlan = netdev_priv(dev); | |
262 | const struct net_device *lowerdev = vlan->lowerdev; | |
263 | ||
264 | dev->state = (dev->state & ~MACVLAN_STATE_MASK) | | |
265 | (lowerdev->state & MACVLAN_STATE_MASK); | |
266 | dev->features = lowerdev->features & MACVLAN_FEATURES; | |
267 | dev->iflink = lowerdev->ifindex; | |
268 | ||
269 | lockdep_set_class(&dev->_xmit_lock, &macvlan_netdev_xmit_lock_key); | |
270 | return 0; | |
271 | } | |
272 | ||
273 | static void macvlan_ethtool_get_drvinfo(struct net_device *dev, | |
274 | struct ethtool_drvinfo *drvinfo) | |
275 | { | |
276 | snprintf(drvinfo->driver, 32, "macvlan"); | |
277 | snprintf(drvinfo->version, 32, "0.1"); | |
278 | } | |
279 | ||
280 | static u32 macvlan_ethtool_get_rx_csum(struct net_device *dev) | |
281 | { | |
282 | const struct macvlan_dev *vlan = netdev_priv(dev); | |
283 | struct net_device *lowerdev = vlan->lowerdev; | |
284 | ||
285 | if (lowerdev->ethtool_ops->get_rx_csum == NULL) | |
286 | return 0; | |
287 | return lowerdev->ethtool_ops->get_rx_csum(lowerdev); | |
288 | } | |
289 | ||
290 | static const struct ethtool_ops macvlan_ethtool_ops = { | |
291 | .get_link = ethtool_op_get_link, | |
292 | .get_rx_csum = macvlan_ethtool_get_rx_csum, | |
b863ceb7 PM |
293 | .get_drvinfo = macvlan_ethtool_get_drvinfo, |
294 | }; | |
295 | ||
296 | static void macvlan_setup(struct net_device *dev) | |
297 | { | |
298 | ether_setup(dev); | |
299 | ||
300 | dev->init = macvlan_init; | |
301 | dev->open = macvlan_open; | |
302 | dev->stop = macvlan_stop; | |
303 | dev->change_mtu = macvlan_change_mtu; | |
304 | dev->change_rx_flags = macvlan_change_rx_flags; | |
305 | dev->set_multicast_list = macvlan_set_multicast_list; | |
b863ceb7 PM |
306 | dev->hard_start_xmit = macvlan_hard_start_xmit; |
307 | dev->destructor = free_netdev; | |
3b04ddde | 308 | dev->header_ops = &macvlan_hard_header_ops, |
b863ceb7 PM |
309 | dev->ethtool_ops = &macvlan_ethtool_ops; |
310 | dev->tx_queue_len = 0; | |
311 | } | |
312 | ||
313 | static int macvlan_port_create(struct net_device *dev) | |
314 | { | |
315 | struct macvlan_port *port; | |
316 | unsigned int i; | |
317 | ||
318 | if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK) | |
319 | return -EINVAL; | |
320 | ||
321 | port = kzalloc(sizeof(*port), GFP_KERNEL); | |
322 | if (port == NULL) | |
323 | return -ENOMEM; | |
324 | ||
325 | port->dev = dev; | |
326 | INIT_LIST_HEAD(&port->vlans); | |
327 | for (i = 0; i < MACVLAN_HASH_SIZE; i++) | |
328 | INIT_HLIST_HEAD(&port->vlan_hash[i]); | |
329 | rcu_assign_pointer(dev->macvlan_port, port); | |
330 | return 0; | |
331 | } | |
332 | ||
333 | static void macvlan_port_destroy(struct net_device *dev) | |
334 | { | |
335 | struct macvlan_port *port = dev->macvlan_port; | |
336 | ||
337 | rcu_assign_pointer(dev->macvlan_port, NULL); | |
338 | synchronize_rcu(); | |
339 | kfree(port); | |
340 | } | |
341 | ||
342 | static void macvlan_transfer_operstate(struct net_device *dev) | |
343 | { | |
344 | struct macvlan_dev *vlan = netdev_priv(dev); | |
345 | const struct net_device *lowerdev = vlan->lowerdev; | |
346 | ||
347 | if (lowerdev->operstate == IF_OPER_DORMANT) | |
348 | netif_dormant_on(dev); | |
349 | else | |
350 | netif_dormant_off(dev); | |
351 | ||
352 | if (netif_carrier_ok(lowerdev)) { | |
353 | if (!netif_carrier_ok(dev)) | |
354 | netif_carrier_on(dev); | |
355 | } else { | |
356 | if (netif_carrier_ok(lowerdev)) | |
357 | netif_carrier_off(dev); | |
358 | } | |
359 | } | |
360 | ||
361 | static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[]) | |
362 | { | |
363 | if (tb[IFLA_ADDRESS]) { | |
364 | if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) | |
365 | return -EINVAL; | |
366 | if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) | |
367 | return -EADDRNOTAVAIL; | |
368 | } | |
369 | return 0; | |
370 | } | |
371 | ||
372 | static int macvlan_newlink(struct net_device *dev, | |
373 | struct nlattr *tb[], struct nlattr *data[]) | |
374 | { | |
375 | struct macvlan_dev *vlan = netdev_priv(dev); | |
376 | struct macvlan_port *port; | |
377 | struct net_device *lowerdev; | |
378 | int err; | |
379 | ||
380 | if (!tb[IFLA_LINK]) | |
381 | return -EINVAL; | |
382 | ||
881d966b | 383 | lowerdev = __dev_get_by_index(dev->nd_net, nla_get_u32(tb[IFLA_LINK])); |
b863ceb7 PM |
384 | if (lowerdev == NULL) |
385 | return -ENODEV; | |
386 | ||
a6ca5f1d PM |
387 | /* Don't allow macvlans on top of other macvlans - its not really |
388 | * wrong, but lockdep can't handle it and its not useful for anything | |
389 | * you couldn't do directly on top of the real device. | |
390 | */ | |
391 | if (lowerdev->rtnl_link_ops == dev->rtnl_link_ops) | |
392 | return -ENODEV; | |
393 | ||
b863ceb7 PM |
394 | if (!tb[IFLA_MTU]) |
395 | dev->mtu = lowerdev->mtu; | |
396 | else if (dev->mtu > lowerdev->mtu) | |
397 | return -EINVAL; | |
398 | ||
399 | if (!tb[IFLA_ADDRESS]) | |
400 | random_ether_addr(dev->dev_addr); | |
401 | ||
402 | if (lowerdev->macvlan_port == NULL) { | |
403 | err = macvlan_port_create(lowerdev); | |
404 | if (err < 0) | |
405 | return err; | |
406 | } | |
407 | port = lowerdev->macvlan_port; | |
408 | ||
409 | vlan->lowerdev = lowerdev; | |
410 | vlan->dev = dev; | |
411 | vlan->port = port; | |
412 | ||
413 | err = register_netdevice(dev); | |
414 | if (err < 0) | |
415 | return err; | |
416 | ||
417 | list_add_tail(&vlan->list, &port->vlans); | |
418 | macvlan_transfer_operstate(dev); | |
419 | return 0; | |
420 | } | |
421 | ||
422 | static void macvlan_dellink(struct net_device *dev) | |
423 | { | |
424 | struct macvlan_dev *vlan = netdev_priv(dev); | |
425 | struct macvlan_port *port = vlan->port; | |
426 | ||
427 | list_del(&vlan->list); | |
428 | unregister_netdevice(dev); | |
429 | ||
430 | if (list_empty(&port->vlans)) | |
431 | macvlan_port_destroy(dev); | |
432 | } | |
433 | ||
434 | static struct rtnl_link_ops macvlan_link_ops __read_mostly = { | |
435 | .kind = "macvlan", | |
436 | .priv_size = sizeof(struct macvlan_dev), | |
437 | .setup = macvlan_setup, | |
438 | .validate = macvlan_validate, | |
439 | .newlink = macvlan_newlink, | |
440 | .dellink = macvlan_dellink, | |
441 | }; | |
442 | ||
443 | static int macvlan_device_event(struct notifier_block *unused, | |
444 | unsigned long event, void *ptr) | |
445 | { | |
446 | struct net_device *dev = ptr; | |
447 | struct macvlan_dev *vlan, *next; | |
448 | struct macvlan_port *port; | |
449 | ||
450 | port = dev->macvlan_port; | |
451 | if (port == NULL) | |
452 | return NOTIFY_DONE; | |
453 | ||
454 | switch (event) { | |
455 | case NETDEV_CHANGE: | |
456 | list_for_each_entry(vlan, &port->vlans, list) | |
457 | macvlan_transfer_operstate(vlan->dev); | |
458 | break; | |
459 | case NETDEV_FEAT_CHANGE: | |
460 | list_for_each_entry(vlan, &port->vlans, list) { | |
461 | vlan->dev->features = dev->features & MACVLAN_FEATURES; | |
462 | netdev_features_change(vlan->dev); | |
463 | } | |
464 | break; | |
465 | case NETDEV_UNREGISTER: | |
466 | list_for_each_entry_safe(vlan, next, &port->vlans, list) | |
467 | macvlan_dellink(vlan->dev); | |
468 | break; | |
469 | } | |
470 | return NOTIFY_DONE; | |
471 | } | |
472 | ||
473 | static struct notifier_block macvlan_notifier_block __read_mostly = { | |
474 | .notifier_call = macvlan_device_event, | |
475 | }; | |
476 | ||
477 | static int __init macvlan_init_module(void) | |
478 | { | |
479 | int err; | |
480 | ||
481 | register_netdevice_notifier(&macvlan_notifier_block); | |
482 | macvlan_handle_frame_hook = macvlan_handle_frame; | |
483 | ||
484 | err = rtnl_link_register(&macvlan_link_ops); | |
485 | if (err < 0) | |
486 | goto err1; | |
487 | return 0; | |
488 | err1: | |
489 | macvlan_handle_frame_hook = macvlan_handle_frame; | |
490 | unregister_netdevice_notifier(&macvlan_notifier_block); | |
491 | return err; | |
492 | } | |
493 | ||
494 | static void __exit macvlan_cleanup_module(void) | |
495 | { | |
496 | rtnl_link_unregister(&macvlan_link_ops); | |
497 | macvlan_handle_frame_hook = NULL; | |
498 | unregister_netdevice_notifier(&macvlan_notifier_block); | |
499 | } | |
500 | ||
501 | module_init(macvlan_init_module); | |
502 | module_exit(macvlan_cleanup_module); | |
503 | ||
504 | MODULE_LICENSE("GPL"); | |
505 | MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>"); | |
506 | MODULE_DESCRIPTION("Driver for MAC address based VLANs"); | |
507 | MODULE_ALIAS_RTNL_LINK("macvlan"); |