2 * Copyright (c) 2007 Patrick McHardy <kaber@trash.net>
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.
9 * The code this is based on carried the following copyright notice:
11 * (C) Copyright 2001-2006
12 * Alex Zeffertt, Cambridge Broadband Ltd, ajz@cambridgebroadband.com
13 * Re-worked by Ben Greear <greearb@candelatech.com>
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>
34 #define MACVLAN_HASH_SIZE (1 << BITS_PER_BYTE)
37 MACVLAN_MODE_PRIVATE = 1,
38 MACVLAN_MODE_VEPA = 2,
39 MACVLAN_MODE_BRIDGE = 4,
43 struct net_device *dev;
44 struct hlist_head vlan_hash[MACVLAN_HASH_SIZE];
45 struct list_head vlans;
49 * struct macvlan_rx_stats - MACVLAN percpu rx stats
50 * @rx_packets: number of received packets
51 * @rx_bytes: number of received bytes
52 * @multicast: number of received multicast packets
53 * @rx_errors: number of errors
55 struct macvlan_rx_stats {
56 unsigned long rx_packets;
57 unsigned long rx_bytes;
58 unsigned long multicast;
59 unsigned long rx_errors;
63 struct net_device *dev;
64 struct list_head list;
65 struct hlist_node hlist;
66 struct macvlan_port *port;
67 struct net_device *lowerdev;
68 struct macvlan_rx_stats *rx_stats;
69 enum macvlan_mode mode;
73 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
74 const unsigned char *addr)
76 struct macvlan_dev *vlan;
79 hlist_for_each_entry_rcu(vlan, n, &port->vlan_hash[addr[5]], hlist) {
80 if (!compare_ether_addr_64bits(vlan->dev->dev_addr, addr))
86 static void macvlan_hash_add(struct macvlan_dev *vlan)
88 struct macvlan_port *port = vlan->port;
89 const unsigned char *addr = vlan->dev->dev_addr;
91 hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[addr[5]]);
94 static void macvlan_hash_del(struct macvlan_dev *vlan)
96 hlist_del_rcu(&vlan->hlist);
100 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
101 const unsigned char *addr)
103 macvlan_hash_del(vlan);
104 /* Now that we are unhashed it is safe to change the device
105 * address without confusing packet delivery.
107 memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
108 macvlan_hash_add(vlan);
111 static int macvlan_addr_busy(const struct macvlan_port *port,
112 const unsigned char *addr)
114 /* Test to see if the specified multicast address is
115 * currently in use by the underlying device or
118 if (!compare_ether_addr_64bits(port->dev->dev_addr, addr))
121 if (macvlan_hash_lookup(port, addr))
127 static inline void macvlan_count_rx(const struct macvlan_dev *vlan,
128 unsigned int len, bool success,
131 struct macvlan_rx_stats *rx_stats;
133 rx_stats = per_cpu_ptr(vlan->rx_stats, smp_processor_id());
134 if (likely(success)) {
135 rx_stats->rx_packets++;;
136 rx_stats->rx_bytes += len;
138 rx_stats->multicast++;
140 rx_stats->rx_errors++;
144 static int macvlan_broadcast_one(struct sk_buff *skb, struct net_device *dev,
145 const struct ethhdr *eth, bool local)
151 return dev_forward_skb(dev, skb);
154 if (!compare_ether_addr_64bits(eth->h_dest,
156 skb->pkt_type = PACKET_BROADCAST;
158 skb->pkt_type = PACKET_MULTICAST;
160 return netif_rx(skb);
163 static void macvlan_broadcast(struct sk_buff *skb,
164 const struct macvlan_port *port,
165 struct net_device *src,
166 enum macvlan_mode mode)
168 const struct ethhdr *eth = eth_hdr(skb);
169 const struct macvlan_dev *vlan;
170 struct hlist_node *n;
171 struct sk_buff *nskb;
175 if (skb->protocol == htons(ETH_P_PAUSE))
178 for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
179 hlist_for_each_entry_rcu(vlan, n, &port->vlan_hash[i], hlist) {
180 if (vlan->dev == src || !(vlan->mode & mode))
183 nskb = skb_clone(skb, GFP_ATOMIC);
184 err = macvlan_broadcast_one(nskb, vlan->dev, eth,
185 mode == MACVLAN_MODE_BRIDGE);
186 macvlan_count_rx(vlan, skb->len + ETH_HLEN,
187 err == NET_RX_SUCCESS, 1);
192 /* called under rcu_read_lock() from netif_receive_skb */
193 static struct sk_buff *macvlan_handle_frame(struct sk_buff *skb)
195 const struct ethhdr *eth = eth_hdr(skb);
196 const struct macvlan_port *port;
197 const struct macvlan_dev *vlan;
198 const struct macvlan_dev *src;
199 struct net_device *dev;
202 port = rcu_dereference(skb->dev->macvlan_port);
206 if (is_multicast_ether_addr(eth->h_dest)) {
207 src = macvlan_hash_lookup(port, eth->h_source);
209 /* frame comes from an external address */
210 macvlan_broadcast(skb, port, NULL,
211 MACVLAN_MODE_PRIVATE |
213 MACVLAN_MODE_BRIDGE);
214 else if (src->mode == MACVLAN_MODE_VEPA)
215 /* flood to everyone except source */
216 macvlan_broadcast(skb, port, src->dev,
218 MACVLAN_MODE_BRIDGE);
219 else if (src->mode == MACVLAN_MODE_BRIDGE)
221 * flood only to VEPA ports, bridge ports
222 * already saw the frame on the way out.
224 macvlan_broadcast(skb, port, src->dev,
229 vlan = macvlan_hash_lookup(port, eth->h_dest);
234 if (unlikely(!(dev->flags & IFF_UP))) {
238 len = skb->len + ETH_HLEN;
239 skb = skb_share_check(skb, GFP_ATOMIC);
240 macvlan_count_rx(vlan, len, skb != NULL, 0);
245 skb->pkt_type = PACKET_HOST;
251 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
253 const struct macvlan_dev *vlan = netdev_priv(dev);
254 const struct macvlan_port *port = vlan->port;
255 const struct macvlan_dev *dest;
257 if (vlan->mode == MACVLAN_MODE_BRIDGE) {
258 const struct ethhdr *eth = (void *)skb->data;
260 /* send to other bridge ports directly */
261 if (is_multicast_ether_addr(eth->h_dest)) {
262 macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
266 dest = macvlan_hash_lookup(port, eth->h_dest);
267 if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
268 unsigned int length = skb->len + ETH_HLEN;
269 int ret = dev_forward_skb(dest->dev, skb);
270 macvlan_count_rx(dest, length,
271 ret == NET_RX_SUCCESS, 0);
273 return NET_XMIT_SUCCESS;
278 skb->dev = vlan->lowerdev;
279 return dev_queue_xmit(skb);
282 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
283 struct net_device *dev)
285 int i = skb_get_queue_mapping(skb);
286 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
287 unsigned int len = skb->len;
290 ret = macvlan_queue_xmit(skb, dev);
291 if (likely(ret == NET_XMIT_SUCCESS)) {
293 txq->tx_bytes += len;
300 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
301 unsigned short type, const void *daddr,
302 const void *saddr, unsigned len)
304 const struct macvlan_dev *vlan = netdev_priv(dev);
305 struct net_device *lowerdev = vlan->lowerdev;
307 return dev_hard_header(skb, lowerdev, type, daddr,
308 saddr ? : dev->dev_addr, len);
311 static const struct header_ops macvlan_hard_header_ops = {
312 .create = macvlan_hard_header,
313 .rebuild = eth_rebuild_header,
314 .parse = eth_header_parse,
315 .cache = eth_header_cache,
316 .cache_update = eth_header_cache_update,
319 static int macvlan_open(struct net_device *dev)
321 struct macvlan_dev *vlan = netdev_priv(dev);
322 struct net_device *lowerdev = vlan->lowerdev;
326 if (macvlan_addr_busy(vlan->port, dev->dev_addr))
329 err = dev_unicast_add(lowerdev, dev->dev_addr);
332 if (dev->flags & IFF_ALLMULTI) {
333 err = dev_set_allmulti(lowerdev, 1);
337 macvlan_hash_add(vlan);
341 dev_unicast_delete(lowerdev, dev->dev_addr);
346 static int macvlan_stop(struct net_device *dev)
348 struct macvlan_dev *vlan = netdev_priv(dev);
349 struct net_device *lowerdev = vlan->lowerdev;
351 dev_mc_unsync(lowerdev, dev);
352 if (dev->flags & IFF_ALLMULTI)
353 dev_set_allmulti(lowerdev, -1);
355 dev_unicast_delete(lowerdev, dev->dev_addr);
357 macvlan_hash_del(vlan);
361 static int macvlan_set_mac_address(struct net_device *dev, void *p)
363 struct macvlan_dev *vlan = netdev_priv(dev);
364 struct net_device *lowerdev = vlan->lowerdev;
365 struct sockaddr *addr = p;
368 if (!is_valid_ether_addr(addr->sa_data))
369 return -EADDRNOTAVAIL;
371 if (!(dev->flags & IFF_UP)) {
372 /* Just copy in the new address */
373 memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
375 /* Rehash and update the device filters */
376 if (macvlan_addr_busy(vlan->port, addr->sa_data))
379 err = dev_unicast_add(lowerdev, addr->sa_data);
383 dev_unicast_delete(lowerdev, dev->dev_addr);
385 macvlan_hash_change_addr(vlan, addr->sa_data);
390 static void macvlan_change_rx_flags(struct net_device *dev, int change)
392 struct macvlan_dev *vlan = netdev_priv(dev);
393 struct net_device *lowerdev = vlan->lowerdev;
395 if (change & IFF_ALLMULTI)
396 dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
399 static void macvlan_set_multicast_list(struct net_device *dev)
401 struct macvlan_dev *vlan = netdev_priv(dev);
403 dev_mc_sync(vlan->lowerdev, dev);
406 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
408 struct macvlan_dev *vlan = netdev_priv(dev);
410 if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu)
417 * macvlan network devices have devices nesting below it and are a special
418 * "super class" of normal network devices; split their locks off into a
419 * separate class since they always nest.
421 static struct lock_class_key macvlan_netdev_xmit_lock_key;
422 static struct lock_class_key macvlan_netdev_addr_lock_key;
424 #define MACVLAN_FEATURES \
425 (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
426 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
427 NETIF_F_TSO_ECN | NETIF_F_TSO6)
429 #define MACVLAN_STATE_MASK \
430 ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
432 static void macvlan_set_lockdep_class_one(struct net_device *dev,
433 struct netdev_queue *txq,
436 lockdep_set_class(&txq->_xmit_lock,
437 &macvlan_netdev_xmit_lock_key);
440 static void macvlan_set_lockdep_class(struct net_device *dev)
442 lockdep_set_class(&dev->addr_list_lock,
443 &macvlan_netdev_addr_lock_key);
444 netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL);
447 static int macvlan_init(struct net_device *dev)
449 struct macvlan_dev *vlan = netdev_priv(dev);
450 const struct net_device *lowerdev = vlan->lowerdev;
452 dev->state = (dev->state & ~MACVLAN_STATE_MASK) |
453 (lowerdev->state & MACVLAN_STATE_MASK);
454 dev->features = lowerdev->features & MACVLAN_FEATURES;
455 dev->iflink = lowerdev->ifindex;
456 dev->hard_header_len = lowerdev->hard_header_len;
458 macvlan_set_lockdep_class(dev);
460 vlan->rx_stats = alloc_percpu(struct macvlan_rx_stats);
467 static void macvlan_uninit(struct net_device *dev)
469 struct macvlan_dev *vlan = netdev_priv(dev);
471 free_percpu(vlan->rx_stats);
474 static struct net_device_stats *macvlan_dev_get_stats(struct net_device *dev)
476 struct net_device_stats *stats = &dev->stats;
477 struct macvlan_dev *vlan = netdev_priv(dev);
479 dev_txq_stats_fold(dev, stats);
481 if (vlan->rx_stats) {
482 struct macvlan_rx_stats *p, rx = {0};
485 for_each_possible_cpu(i) {
486 p = per_cpu_ptr(vlan->rx_stats, i);
487 rx.rx_packets += p->rx_packets;
488 rx.rx_bytes += p->rx_bytes;
489 rx.rx_errors += p->rx_errors;
490 rx.multicast += p->multicast;
492 stats->rx_packets = rx.rx_packets;
493 stats->rx_bytes = rx.rx_bytes;
494 stats->rx_errors = rx.rx_errors;
495 stats->rx_dropped = rx.rx_errors;
496 stats->multicast = rx.multicast;
501 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
502 struct ethtool_drvinfo *drvinfo)
504 snprintf(drvinfo->driver, 32, "macvlan");
505 snprintf(drvinfo->version, 32, "0.1");
508 static u32 macvlan_ethtool_get_rx_csum(struct net_device *dev)
510 const struct macvlan_dev *vlan = netdev_priv(dev);
511 return dev_ethtool_get_rx_csum(vlan->lowerdev);
514 static int macvlan_ethtool_get_settings(struct net_device *dev,
515 struct ethtool_cmd *cmd)
517 const struct macvlan_dev *vlan = netdev_priv(dev);
518 return dev_ethtool_get_settings(vlan->lowerdev, cmd);
521 static u32 macvlan_ethtool_get_flags(struct net_device *dev)
523 const struct macvlan_dev *vlan = netdev_priv(dev);
524 return dev_ethtool_get_flags(vlan->lowerdev);
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,
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_stats = macvlan_dev_get_stats,
546 .ndo_validate_addr = eth_validate_addr,
549 static void macvlan_setup(struct net_device *dev)
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 dev->tx_queue_len = 0;
561 static int macvlan_port_create(struct net_device *dev)
563 struct macvlan_port *port;
566 if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
569 port = kzalloc(sizeof(*port), GFP_KERNEL);
574 INIT_LIST_HEAD(&port->vlans);
575 for (i = 0; i < MACVLAN_HASH_SIZE; i++)
576 INIT_HLIST_HEAD(&port->vlan_hash[i]);
577 rcu_assign_pointer(dev->macvlan_port, port);
581 static void macvlan_port_destroy(struct net_device *dev)
583 struct macvlan_port *port = dev->macvlan_port;
585 rcu_assign_pointer(dev->macvlan_port, NULL);
590 static void macvlan_transfer_operstate(struct net_device *dev)
592 struct macvlan_dev *vlan = netdev_priv(dev);
593 const struct net_device *lowerdev = vlan->lowerdev;
595 if (lowerdev->operstate == IF_OPER_DORMANT)
596 netif_dormant_on(dev);
598 netif_dormant_off(dev);
600 if (netif_carrier_ok(lowerdev)) {
601 if (!netif_carrier_ok(dev))
602 netif_carrier_on(dev);
604 if (netif_carrier_ok(dev))
605 netif_carrier_off(dev);
609 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[])
611 if (tb[IFLA_ADDRESS]) {
612 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
614 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
615 return -EADDRNOTAVAIL;
620 static int macvlan_get_tx_queues(struct net *net,
622 unsigned int *num_tx_queues,
623 unsigned int *real_num_tx_queues)
625 struct net_device *real_dev;
630 real_dev = __dev_get_by_index(net, nla_get_u32(tb[IFLA_LINK]));
634 *num_tx_queues = real_dev->num_tx_queues;
635 *real_num_tx_queues = real_dev->real_num_tx_queues;
639 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
640 struct nlattr *tb[], struct nlattr *data[])
642 struct macvlan_dev *vlan = netdev_priv(dev);
643 struct macvlan_port *port;
644 struct net_device *lowerdev;
650 lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
651 if (lowerdev == NULL)
654 /* When creating macvlans on top of other macvlans - use
655 * the real device as the lowerdev.
657 if (lowerdev->rtnl_link_ops == dev->rtnl_link_ops) {
658 struct macvlan_dev *lowervlan = netdev_priv(lowerdev);
659 lowerdev = lowervlan->lowerdev;
663 dev->mtu = lowerdev->mtu;
664 else if (dev->mtu > lowerdev->mtu)
667 if (!tb[IFLA_ADDRESS])
668 random_ether_addr(dev->dev_addr);
670 if (lowerdev->macvlan_port == NULL) {
671 err = macvlan_port_create(lowerdev);
675 port = lowerdev->macvlan_port;
677 vlan->lowerdev = lowerdev;
681 err = register_netdevice(dev);
685 list_add_tail(&vlan->list, &port->vlans);
686 macvlan_transfer_operstate(dev);
690 static void macvlan_dellink(struct net_device *dev, struct list_head *head)
692 struct macvlan_dev *vlan = netdev_priv(dev);
693 struct macvlan_port *port = vlan->port;
695 list_del(&vlan->list);
696 unregister_netdevice_queue(dev, head);
698 if (list_empty(&port->vlans))
699 macvlan_port_destroy(port->dev);
702 static struct rtnl_link_ops macvlan_link_ops __read_mostly = {
704 .priv_size = sizeof(struct macvlan_dev),
705 .get_tx_queues = macvlan_get_tx_queues,
706 .setup = macvlan_setup,
707 .validate = macvlan_validate,
708 .newlink = macvlan_newlink,
709 .dellink = macvlan_dellink,
712 static int macvlan_device_event(struct notifier_block *unused,
713 unsigned long event, void *ptr)
715 struct net_device *dev = ptr;
716 struct macvlan_dev *vlan, *next;
717 struct macvlan_port *port;
719 port = dev->macvlan_port;
725 list_for_each_entry(vlan, &port->vlans, list)
726 macvlan_transfer_operstate(vlan->dev);
728 case NETDEV_FEAT_CHANGE:
729 list_for_each_entry(vlan, &port->vlans, list) {
730 vlan->dev->features = dev->features & MACVLAN_FEATURES;
731 netdev_features_change(vlan->dev);
734 case NETDEV_UNREGISTER:
735 list_for_each_entry_safe(vlan, next, &port->vlans, list)
736 macvlan_dellink(vlan->dev, NULL);
742 static struct notifier_block macvlan_notifier_block __read_mostly = {
743 .notifier_call = macvlan_device_event,
746 static int __init macvlan_init_module(void)
750 register_netdevice_notifier(&macvlan_notifier_block);
751 macvlan_handle_frame_hook = macvlan_handle_frame;
753 err = rtnl_link_register(&macvlan_link_ops);
758 macvlan_handle_frame_hook = NULL;
759 unregister_netdevice_notifier(&macvlan_notifier_block);
763 static void __exit macvlan_cleanup_module(void)
765 rtnl_link_unregister(&macvlan_link_ops);
766 macvlan_handle_frame_hook = NULL;
767 unregister_netdevice_notifier(&macvlan_notifier_block);
770 module_init(macvlan_init_module);
771 module_exit(macvlan_cleanup_module);
773 MODULE_LICENSE("GPL");
774 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
775 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
776 MODULE_ALIAS_RTNL_LINK("macvlan");