Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* Copyright (c) 2014 Mahesh Bandewar <maheshb@google.com>
3 */
4
5#include "ipvlan.h"
6
7static int ipvlan_set_port_mode(struct ipvl_port *port, u16 nval,
8 struct netlink_ext_ack *extack)
9{
10 struct ipvl_dev *ipvlan;
11 unsigned int flags;
12 int err;
13
14 ASSERT_RTNL();
15 if (port->mode != nval) {
16 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
17 flags = ipvlan->dev->flags;
18 if (nval == IPVLAN_MODE_L3 || nval == IPVLAN_MODE_L3S) {
19 err = dev_change_flags(ipvlan->dev,
20 flags | IFF_NOARP,
21 extack);
22 } else {
23 err = dev_change_flags(ipvlan->dev,
24 flags & ~IFF_NOARP,
25 extack);
26 }
27 if (unlikely(err))
28 goto fail;
29 }
30 if (nval == IPVLAN_MODE_L3S) {
31 /* New mode is L3S */
32 err = ipvlan_l3s_register(port);
33 if (err)
34 goto fail;
35 } else if (port->mode == IPVLAN_MODE_L3S) {
36 /* Old mode was L3S */
37 ipvlan_l3s_unregister(port);
38 }
39 port->mode = nval;
40 }
41 return 0;
42
43fail:
44 /* Undo the flags changes that have been done so far. */
45 list_for_each_entry_continue_reverse(ipvlan, &port->ipvlans, pnode) {
46 flags = ipvlan->dev->flags;
47 if (port->mode == IPVLAN_MODE_L3 ||
48 port->mode == IPVLAN_MODE_L3S)
49 dev_change_flags(ipvlan->dev, flags | IFF_NOARP,
50 NULL);
51 else
52 dev_change_flags(ipvlan->dev, flags & ~IFF_NOARP,
53 NULL);
54 }
55
56 return err;
57}
58
59static int ipvlan_port_create(struct net_device *dev)
60{
61 struct ipvl_port *port;
62 int err, idx;
63
64 port = kzalloc(sizeof(struct ipvl_port), GFP_KERNEL);
65 if (!port)
66 return -ENOMEM;
67
68 write_pnet(&port->pnet, dev_net(dev));
69 port->dev = dev;
70 port->mode = IPVLAN_MODE_L3;
71 INIT_LIST_HEAD(&port->ipvlans);
72 for (idx = 0; idx < IPVLAN_HASH_SIZE; idx++)
73 INIT_HLIST_HEAD(&port->hlhead[idx]);
74
75 skb_queue_head_init(&port->backlog);
76 INIT_WORK(&port->wq, ipvlan_process_multicast);
77 ida_init(&port->ida);
78 port->dev_id_start = 1;
79
80 err = netdev_rx_handler_register(dev, ipvlan_handle_frame, port);
81 if (err)
82 goto err;
83
84 return 0;
85
86err:
87 kfree(port);
88 return err;
89}
90
91static void ipvlan_port_destroy(struct net_device *dev)
92{
93 struct ipvl_port *port = ipvlan_port_get_rtnl(dev);
94 struct sk_buff *skb;
95
96 if (port->mode == IPVLAN_MODE_L3S)
97 ipvlan_l3s_unregister(port);
98 netdev_rx_handler_unregister(dev);
99 cancel_work_sync(&port->wq);
100 while ((skb = __skb_dequeue(&port->backlog)) != NULL) {
101 if (skb->dev)
102 dev_put(skb->dev);
103 kfree_skb(skb);
104 }
105 ida_destroy(&port->ida);
106 kfree(port);
107}
108
109#define IPVLAN_FEATURES \
110 (NETIF_F_SG | NETIF_F_CSUM_MASK | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
111 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_GSO_ROBUST | \
112 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
113 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
114
115#define IPVLAN_STATE_MASK \
116 ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
117
118static int ipvlan_init(struct net_device *dev)
119{
120 struct ipvl_dev *ipvlan = netdev_priv(dev);
121 struct net_device *phy_dev = ipvlan->phy_dev;
122 struct ipvl_port *port;
123 int err;
124
125 dev->state = (dev->state & ~IPVLAN_STATE_MASK) |
126 (phy_dev->state & IPVLAN_STATE_MASK);
127 dev->features = phy_dev->features & IPVLAN_FEATURES;
128 dev->features |= NETIF_F_LLTX | NETIF_F_VLAN_CHALLENGED;
129 dev->hw_enc_features |= dev->features;
130 dev->gso_max_size = phy_dev->gso_max_size;
131 dev->gso_max_segs = phy_dev->gso_max_segs;
132 dev->hard_header_len = phy_dev->hard_header_len;
133
134 ipvlan->pcpu_stats = netdev_alloc_pcpu_stats(struct ipvl_pcpu_stats);
135 if (!ipvlan->pcpu_stats)
136 return -ENOMEM;
137
138 if (!netif_is_ipvlan_port(phy_dev)) {
139 err = ipvlan_port_create(phy_dev);
140 if (err < 0) {
141 free_percpu(ipvlan->pcpu_stats);
142 return err;
143 }
144 }
145 port = ipvlan_port_get_rtnl(phy_dev);
146 port->count += 1;
147 return 0;
148}
149
150static void ipvlan_uninit(struct net_device *dev)
151{
152 struct ipvl_dev *ipvlan = netdev_priv(dev);
153 struct net_device *phy_dev = ipvlan->phy_dev;
154 struct ipvl_port *port;
155
156 free_percpu(ipvlan->pcpu_stats);
157
158 port = ipvlan_port_get_rtnl(phy_dev);
159 port->count -= 1;
160 if (!port->count)
161 ipvlan_port_destroy(port->dev);
162}
163
164static int ipvlan_open(struct net_device *dev)
165{
166 struct ipvl_dev *ipvlan = netdev_priv(dev);
167 struct net_device *phy_dev = ipvlan->phy_dev;
168 struct ipvl_addr *addr;
169
170 if (ipvlan->port->mode == IPVLAN_MODE_L3 ||
171 ipvlan->port->mode == IPVLAN_MODE_L3S)
172 dev->flags |= IFF_NOARP;
173 else
174 dev->flags &= ~IFF_NOARP;
175
176 rcu_read_lock();
177 list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
178 ipvlan_ht_addr_add(ipvlan, addr);
179 rcu_read_unlock();
180
181 return dev_uc_add(phy_dev, phy_dev->dev_addr);
182}
183
184static int ipvlan_stop(struct net_device *dev)
185{
186 struct ipvl_dev *ipvlan = netdev_priv(dev);
187 struct net_device *phy_dev = ipvlan->phy_dev;
188 struct ipvl_addr *addr;
189
190 dev_uc_unsync(phy_dev, dev);
191 dev_mc_unsync(phy_dev, dev);
192
193 dev_uc_del(phy_dev, phy_dev->dev_addr);
194
195 rcu_read_lock();
196 list_for_each_entry_rcu(addr, &ipvlan->addrs, anode)
197 ipvlan_ht_addr_del(addr);
198 rcu_read_unlock();
199
200 return 0;
201}
202
203static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
204 struct net_device *dev)
205{
206 const struct ipvl_dev *ipvlan = netdev_priv(dev);
207 int skblen = skb->len;
208 int ret;
209
210 ret = ipvlan_queue_xmit(skb, dev);
211 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
212 struct ipvl_pcpu_stats *pcptr;
213
214 pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
215
216 u64_stats_update_begin(&pcptr->syncp);
217 pcptr->tx_pkts++;
218 pcptr->tx_bytes += skblen;
219 u64_stats_update_end(&pcptr->syncp);
220 } else {
221 this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
222 }
223 return ret;
224}
225
226static netdev_features_t ipvlan_fix_features(struct net_device *dev,
227 netdev_features_t features)
228{
229 struct ipvl_dev *ipvlan = netdev_priv(dev);
230
231 return features & (ipvlan->sfeatures | ~IPVLAN_FEATURES);
232}
233
234static void ipvlan_change_rx_flags(struct net_device *dev, int change)
235{
236 struct ipvl_dev *ipvlan = netdev_priv(dev);
237 struct net_device *phy_dev = ipvlan->phy_dev;
238
239 if (change & IFF_ALLMULTI)
240 dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
241}
242
243static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
244{
245 struct ipvl_dev *ipvlan = netdev_priv(dev);
246
247 if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
248 bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
249 } else {
250 struct netdev_hw_addr *ha;
251 DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
252
253 bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
254 netdev_for_each_mc_addr(ha, dev)
255 __set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
256
257 /* Turn-on broadcast bit irrespective of address family,
258 * since broadcast is deferred to a work-queue, hence no
259 * impact on fast-path processing.
260 */
261 __set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
262
263 bitmap_copy(ipvlan->mac_filters, mc_filters,
264 IPVLAN_MAC_FILTER_SIZE);
265 }
266 dev_uc_sync(ipvlan->phy_dev, dev);
267 dev_mc_sync(ipvlan->phy_dev, dev);
268}
269
270static void ipvlan_get_stats64(struct net_device *dev,
271 struct rtnl_link_stats64 *s)
272{
273 struct ipvl_dev *ipvlan = netdev_priv(dev);
274
275 if (ipvlan->pcpu_stats) {
276 struct ipvl_pcpu_stats *pcptr;
277 u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
278 u32 rx_errs = 0, tx_drps = 0;
279 u32 strt;
280 int idx;
281
282 for_each_possible_cpu(idx) {
283 pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
284 do {
285 strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
286 rx_pkts = pcptr->rx_pkts;
287 rx_bytes = pcptr->rx_bytes;
288 rx_mcast = pcptr->rx_mcast;
289 tx_pkts = pcptr->tx_pkts;
290 tx_bytes = pcptr->tx_bytes;
291 } while (u64_stats_fetch_retry_irq(&pcptr->syncp,
292 strt));
293
294 s->rx_packets += rx_pkts;
295 s->rx_bytes += rx_bytes;
296 s->multicast += rx_mcast;
297 s->tx_packets += tx_pkts;
298 s->tx_bytes += tx_bytes;
299
300 /* u32 values are updated without syncp protection. */
301 rx_errs += pcptr->rx_errs;
302 tx_drps += pcptr->tx_drps;
303 }
304 s->rx_errors = rx_errs;
305 s->rx_dropped = rx_errs;
306 s->tx_dropped = tx_drps;
307 }
308}
309
310static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
311{
312 struct ipvl_dev *ipvlan = netdev_priv(dev);
313 struct net_device *phy_dev = ipvlan->phy_dev;
314
315 return vlan_vid_add(phy_dev, proto, vid);
316}
317
318static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
319 u16 vid)
320{
321 struct ipvl_dev *ipvlan = netdev_priv(dev);
322 struct net_device *phy_dev = ipvlan->phy_dev;
323
324 vlan_vid_del(phy_dev, proto, vid);
325 return 0;
326}
327
328static int ipvlan_get_iflink(const struct net_device *dev)
329{
330 struct ipvl_dev *ipvlan = netdev_priv(dev);
331
332 return ipvlan->phy_dev->ifindex;
333}
334
335static const struct net_device_ops ipvlan_netdev_ops = {
336 .ndo_init = ipvlan_init,
337 .ndo_uninit = ipvlan_uninit,
338 .ndo_open = ipvlan_open,
339 .ndo_stop = ipvlan_stop,
340 .ndo_start_xmit = ipvlan_start_xmit,
341 .ndo_fix_features = ipvlan_fix_features,
342 .ndo_change_rx_flags = ipvlan_change_rx_flags,
343 .ndo_set_rx_mode = ipvlan_set_multicast_mac_filter,
344 .ndo_get_stats64 = ipvlan_get_stats64,
345 .ndo_vlan_rx_add_vid = ipvlan_vlan_rx_add_vid,
346 .ndo_vlan_rx_kill_vid = ipvlan_vlan_rx_kill_vid,
347 .ndo_get_iflink = ipvlan_get_iflink,
348};
349
350static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
351 unsigned short type, const void *daddr,
352 const void *saddr, unsigned len)
353{
354 const struct ipvl_dev *ipvlan = netdev_priv(dev);
355 struct net_device *phy_dev = ipvlan->phy_dev;
356
357 /* TODO Probably use a different field than dev_addr so that the
358 * mac-address on the virtual device is portable and can be carried
359 * while the packets use the mac-addr on the physical device.
360 */
361 return dev_hard_header(skb, phy_dev, type, daddr,
362 saddr ? : phy_dev->dev_addr, len);
363}
364
365static const struct header_ops ipvlan_header_ops = {
366 .create = ipvlan_hard_header,
367 .parse = eth_header_parse,
368 .cache = eth_header_cache,
369 .cache_update = eth_header_cache_update,
370};
371
372static void ipvlan_adjust_mtu(struct ipvl_dev *ipvlan, struct net_device *dev)
373{
374 ipvlan->dev->mtu = dev->mtu;
375}
376
377static bool netif_is_ipvlan(const struct net_device *dev)
378{
379 /* both ipvlan and ipvtap devices use the same netdev_ops */
380 return dev->netdev_ops == &ipvlan_netdev_ops;
381}
382
383static int ipvlan_ethtool_get_link_ksettings(struct net_device *dev,
384 struct ethtool_link_ksettings *cmd)
385{
386 const struct ipvl_dev *ipvlan = netdev_priv(dev);
387
388 return __ethtool_get_link_ksettings(ipvlan->phy_dev, cmd);
389}
390
391static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
392 struct ethtool_drvinfo *drvinfo)
393{
394 strlcpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
395 strlcpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
396}
397
398static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
399{
400 const struct ipvl_dev *ipvlan = netdev_priv(dev);
401
402 return ipvlan->msg_enable;
403}
404
405static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
406{
407 struct ipvl_dev *ipvlan = netdev_priv(dev);
408
409 ipvlan->msg_enable = value;
410}
411
412static const struct ethtool_ops ipvlan_ethtool_ops = {
413 .get_link = ethtool_op_get_link,
414 .get_link_ksettings = ipvlan_ethtool_get_link_ksettings,
415 .get_drvinfo = ipvlan_ethtool_get_drvinfo,
416 .get_msglevel = ipvlan_ethtool_get_msglevel,
417 .set_msglevel = ipvlan_ethtool_set_msglevel,
418};
419
420static int ipvlan_nl_changelink(struct net_device *dev,
421 struct nlattr *tb[], struct nlattr *data[],
422 struct netlink_ext_ack *extack)
423{
424 struct ipvl_dev *ipvlan = netdev_priv(dev);
425 struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
426 int err = 0;
427
428 if (!data)
429 return 0;
430 if (!ns_capable(dev_net(ipvlan->phy_dev)->user_ns, CAP_NET_ADMIN))
431 return -EPERM;
432
433 if (data[IFLA_IPVLAN_MODE]) {
434 u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
435
436 err = ipvlan_set_port_mode(port, nmode, extack);
437 }
438
439 if (!err && data[IFLA_IPVLAN_FLAGS]) {
440 u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
441
442 if (flags & IPVLAN_F_PRIVATE)
443 ipvlan_mark_private(port);
444 else
445 ipvlan_clear_private(port);
446
447 if (flags & IPVLAN_F_VEPA)
448 ipvlan_mark_vepa(port);
449 else
450 ipvlan_clear_vepa(port);
451 }
452
453 return err;
454}
455
456static size_t ipvlan_nl_getsize(const struct net_device *dev)
457{
458 return (0
459 + nla_total_size(2) /* IFLA_IPVLAN_MODE */
460 + nla_total_size(2) /* IFLA_IPVLAN_FLAGS */
461 );
462}
463
464static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[],
465 struct netlink_ext_ack *extack)
466{
467 if (!data)
468 return 0;
469
470 if (data[IFLA_IPVLAN_MODE]) {
471 u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
472
473 if (mode >= IPVLAN_MODE_MAX)
474 return -EINVAL;
475 }
476 if (data[IFLA_IPVLAN_FLAGS]) {
477 u16 flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
478
479 /* Only two bits are used at this moment. */
480 if (flags & ~(IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
481 return -EINVAL;
482 /* Also both flags can't be active at the same time. */
483 if ((flags & (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA)) ==
484 (IPVLAN_F_PRIVATE | IPVLAN_F_VEPA))
485 return -EINVAL;
486 }
487
488 return 0;
489}
490
491static int ipvlan_nl_fillinfo(struct sk_buff *skb,
492 const struct net_device *dev)
493{
494 struct ipvl_dev *ipvlan = netdev_priv(dev);
495 struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
496 int ret = -EINVAL;
497
498 if (!port)
499 goto err;
500
501 ret = -EMSGSIZE;
502 if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
503 goto err;
504 if (nla_put_u16(skb, IFLA_IPVLAN_FLAGS, port->flags))
505 goto err;
506
507 return 0;
508
509err:
510 return ret;
511}
512
513int ipvlan_link_new(struct net *src_net, struct net_device *dev,
514 struct nlattr *tb[], struct nlattr *data[],
515 struct netlink_ext_ack *extack)
516{
517 struct ipvl_dev *ipvlan = netdev_priv(dev);
518 struct ipvl_port *port;
519 struct net_device *phy_dev;
520 int err;
521 u16 mode = IPVLAN_MODE_L3;
522
523 if (!tb[IFLA_LINK])
524 return -EINVAL;
525
526 phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
527 if (!phy_dev)
528 return -ENODEV;
529
530 if (netif_is_ipvlan(phy_dev)) {
531 struct ipvl_dev *tmp = netdev_priv(phy_dev);
532
533 phy_dev = tmp->phy_dev;
534 if (!ns_capable(dev_net(phy_dev)->user_ns, CAP_NET_ADMIN))
535 return -EPERM;
536 } else if (!netif_is_ipvlan_port(phy_dev)) {
537 /* Exit early if the underlying link is invalid or busy */
538 if (phy_dev->type != ARPHRD_ETHER ||
539 phy_dev->flags & IFF_LOOPBACK) {
540 netdev_err(phy_dev,
541 "Master is either lo or non-ether device\n");
542 return -EINVAL;
543 }
544
545 if (netdev_is_rx_handler_busy(phy_dev)) {
546 netdev_err(phy_dev, "Device is already in use.\n");
547 return -EBUSY;
548 }
549 }
550
551 ipvlan->phy_dev = phy_dev;
552 ipvlan->dev = dev;
553 ipvlan->sfeatures = IPVLAN_FEATURES;
554 if (!tb[IFLA_MTU])
555 ipvlan_adjust_mtu(ipvlan, phy_dev);
556 INIT_LIST_HEAD(&ipvlan->addrs);
557 spin_lock_init(&ipvlan->addrs_lock);
558
559 /* TODO Probably put random address here to be presented to the
560 * world but keep using the physical-dev address for the outgoing
561 * packets.
562 */
563 memcpy(dev->dev_addr, phy_dev->dev_addr, ETH_ALEN);
564
565 dev->priv_flags |= IFF_NO_RX_HANDLER;
566
567 err = register_netdevice(dev);
568 if (err < 0)
569 return err;
570
571 /* ipvlan_init() would have created the port, if required */
572 port = ipvlan_port_get_rtnl(phy_dev);
573 ipvlan->port = port;
574
575 /* If the port-id base is at the MAX value, then wrap it around and
576 * begin from 0x1 again. This may be due to a busy system where lots
577 * of slaves are getting created and deleted.
578 */
579 if (port->dev_id_start == 0xFFFE)
580 port->dev_id_start = 0x1;
581
582 /* Since L2 address is shared among all IPvlan slaves including
583 * master, use unique 16 bit dev-ids to diffentiate among them.
584 * Assign IDs between 0x1 and 0xFFFE (used by the master) to each
585 * slave link [see addrconf_ifid_eui48()].
586 */
587 err = ida_simple_get(&port->ida, port->dev_id_start, 0xFFFE,
588 GFP_KERNEL);
589 if (err < 0)
590 err = ida_simple_get(&port->ida, 0x1, port->dev_id_start,
591 GFP_KERNEL);
592 if (err < 0)
593 goto unregister_netdev;
594 dev->dev_id = err;
595
596 /* Increment id-base to the next slot for the future assignment */
597 port->dev_id_start = err + 1;
598
599 err = netdev_upper_dev_link(phy_dev, dev, extack);
600 if (err)
601 goto remove_ida;
602
603 /* Flags are per port and latest update overrides. User has
604 * to be consistent in setting it just like the mode attribute.
605 */
606 if (data && data[IFLA_IPVLAN_FLAGS])
607 port->flags = nla_get_u16(data[IFLA_IPVLAN_FLAGS]);
608
609 if (data && data[IFLA_IPVLAN_MODE])
610 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
611
612 err = ipvlan_set_port_mode(port, mode, extack);
613 if (err)
614 goto unlink_netdev;
615
616 list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
617 netif_stacked_transfer_operstate(phy_dev, dev);
618 return 0;
619
620unlink_netdev:
621 netdev_upper_dev_unlink(phy_dev, dev);
622remove_ida:
623 ida_simple_remove(&port->ida, dev->dev_id);
624unregister_netdev:
625 unregister_netdevice(dev);
626 return err;
627}
628EXPORT_SYMBOL_GPL(ipvlan_link_new);
629
630void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
631{
632 struct ipvl_dev *ipvlan = netdev_priv(dev);
633 struct ipvl_addr *addr, *next;
634
635 spin_lock_bh(&ipvlan->addrs_lock);
636 list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
637 ipvlan_ht_addr_del(addr);
638 list_del_rcu(&addr->anode);
639 kfree_rcu(addr, rcu);
640 }
641 spin_unlock_bh(&ipvlan->addrs_lock);
642
643 ida_simple_remove(&ipvlan->port->ida, dev->dev_id);
644 list_del_rcu(&ipvlan->pnode);
645 unregister_netdevice_queue(dev, head);
646 netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
647}
648EXPORT_SYMBOL_GPL(ipvlan_link_delete);
649
650void ipvlan_link_setup(struct net_device *dev)
651{
652 ether_setup(dev);
653
654 dev->max_mtu = ETH_MAX_MTU;
655 dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
656 dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
657 dev->netdev_ops = &ipvlan_netdev_ops;
658 dev->needs_free_netdev = true;
659 dev->header_ops = &ipvlan_header_ops;
660 dev->ethtool_ops = &ipvlan_ethtool_ops;
661}
662EXPORT_SYMBOL_GPL(ipvlan_link_setup);
663
664static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
665{
666 [IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
667 [IFLA_IPVLAN_FLAGS] = { .type = NLA_U16 },
668};
669
670static struct rtnl_link_ops ipvlan_link_ops = {
671 .kind = "ipvlan",
672 .priv_size = sizeof(struct ipvl_dev),
673
674 .setup = ipvlan_link_setup,
675 .newlink = ipvlan_link_new,
676 .dellink = ipvlan_link_delete,
677};
678
679int ipvlan_link_register(struct rtnl_link_ops *ops)
680{
681 ops->get_size = ipvlan_nl_getsize;
682 ops->policy = ipvlan_nl_policy;
683 ops->validate = ipvlan_nl_validate;
684 ops->fill_info = ipvlan_nl_fillinfo;
685 ops->changelink = ipvlan_nl_changelink;
686 ops->maxtype = IFLA_IPVLAN_MAX;
687 return rtnl_link_register(ops);
688}
689EXPORT_SYMBOL_GPL(ipvlan_link_register);
690
691static int ipvlan_device_event(struct notifier_block *unused,
692 unsigned long event, void *ptr)
693{
694 struct netlink_ext_ack *extack = netdev_notifier_info_to_extack(ptr);
695 struct netdev_notifier_pre_changeaddr_info *prechaddr_info;
696 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
697 struct ipvl_dev *ipvlan, *next;
698 struct ipvl_port *port;
699 LIST_HEAD(lst_kill);
700 int err;
701
702 if (!netif_is_ipvlan_port(dev))
703 return NOTIFY_DONE;
704
705 port = ipvlan_port_get_rtnl(dev);
706
707 switch (event) {
708 case NETDEV_CHANGE:
709 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
710 netif_stacked_transfer_operstate(ipvlan->phy_dev,
711 ipvlan->dev);
712 break;
713
714 case NETDEV_REGISTER: {
715 struct net *oldnet, *newnet = dev_net(dev);
716
717 oldnet = read_pnet(&port->pnet);
718 if (net_eq(newnet, oldnet))
719 break;
720
721 write_pnet(&port->pnet, newnet);
722
723 ipvlan_migrate_l3s_hook(oldnet, newnet);
724 break;
725 }
726 case NETDEV_UNREGISTER:
727 if (dev->reg_state != NETREG_UNREGISTERING)
728 break;
729
730 list_for_each_entry_safe(ipvlan, next, &port->ipvlans, pnode)
731 ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
732 &lst_kill);
733 unregister_netdevice_many(&lst_kill);
734 break;
735
736 case NETDEV_FEAT_CHANGE:
737 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
738 ipvlan->dev->features = dev->features & IPVLAN_FEATURES;
739 ipvlan->dev->gso_max_size = dev->gso_max_size;
740 ipvlan->dev->gso_max_segs = dev->gso_max_segs;
741 netdev_features_change(ipvlan->dev);
742 }
743 break;
744
745 case NETDEV_CHANGEMTU:
746 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
747 ipvlan_adjust_mtu(ipvlan, dev);
748 break;
749
750 case NETDEV_PRE_CHANGEADDR:
751 prechaddr_info = ptr;
752 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
753 err = dev_pre_changeaddr_notify(ipvlan->dev,
754 prechaddr_info->dev_addr,
755 extack);
756 if (err)
757 return notifier_from_errno(err);
758 }
759 break;
760
761 case NETDEV_CHANGEADDR:
762 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
763 ether_addr_copy(ipvlan->dev->dev_addr, dev->dev_addr);
764 call_netdevice_notifiers(NETDEV_CHANGEADDR, ipvlan->dev);
765 }
766 break;
767
768 case NETDEV_PRE_TYPE_CHANGE:
769 /* Forbid underlying device to change its type. */
770 return NOTIFY_BAD;
771 }
772 return NOTIFY_DONE;
773}
774
775/* the caller must held the addrs lock */
776static int ipvlan_add_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
777{
778 struct ipvl_addr *addr;
779
780 addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
781 if (!addr)
782 return -ENOMEM;
783
784 addr->master = ipvlan;
785 if (!is_v6) {
786 memcpy(&addr->ip4addr, iaddr, sizeof(struct in_addr));
787 addr->atype = IPVL_IPV4;
788#if IS_ENABLED(CONFIG_IPV6)
789 } else {
790 memcpy(&addr->ip6addr, iaddr, sizeof(struct in6_addr));
791 addr->atype = IPVL_IPV6;
792#endif
793 }
794
795 list_add_tail_rcu(&addr->anode, &ipvlan->addrs);
796
797 /* If the interface is not up, the address will be added to the hash
798 * list by ipvlan_open.
799 */
800 if (netif_running(ipvlan->dev))
801 ipvlan_ht_addr_add(ipvlan, addr);
802
803 return 0;
804}
805
806static void ipvlan_del_addr(struct ipvl_dev *ipvlan, void *iaddr, bool is_v6)
807{
808 struct ipvl_addr *addr;
809
810 spin_lock_bh(&ipvlan->addrs_lock);
811 addr = ipvlan_find_addr(ipvlan, iaddr, is_v6);
812 if (!addr) {
813 spin_unlock_bh(&ipvlan->addrs_lock);
814 return;
815 }
816
817 ipvlan_ht_addr_del(addr);
818 list_del_rcu(&addr->anode);
819 spin_unlock_bh(&ipvlan->addrs_lock);
820 kfree_rcu(addr, rcu);
821}
822
823static bool ipvlan_is_valid_dev(const struct net_device *dev)
824{
825 struct ipvl_dev *ipvlan = netdev_priv(dev);
826
827 if (!netif_is_ipvlan(dev))
828 return false;
829
830 if (!ipvlan || !ipvlan->port)
831 return false;
832
833 return true;
834}
835
836#if IS_ENABLED(CONFIG_IPV6)
837static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
838{
839 int ret = -EINVAL;
840
841 spin_lock_bh(&ipvlan->addrs_lock);
842 if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true))
843 netif_err(ipvlan, ifup, ipvlan->dev,
844 "Failed to add IPv6=%pI6c addr for %s intf\n",
845 ip6_addr, ipvlan->dev->name);
846 else
847 ret = ipvlan_add_addr(ipvlan, ip6_addr, true);
848 spin_unlock_bh(&ipvlan->addrs_lock);
849 return ret;
850}
851
852static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
853{
854 return ipvlan_del_addr(ipvlan, ip6_addr, true);
855}
856
857static int ipvlan_addr6_event(struct notifier_block *unused,
858 unsigned long event, void *ptr)
859{
860 struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
861 struct net_device *dev = (struct net_device *)if6->idev->dev;
862 struct ipvl_dev *ipvlan = netdev_priv(dev);
863
864 if (!ipvlan_is_valid_dev(dev))
865 return NOTIFY_DONE;
866
867 switch (event) {
868 case NETDEV_UP:
869 if (ipvlan_add_addr6(ipvlan, &if6->addr))
870 return NOTIFY_BAD;
871 break;
872
873 case NETDEV_DOWN:
874 ipvlan_del_addr6(ipvlan, &if6->addr);
875 break;
876 }
877
878 return NOTIFY_OK;
879}
880
881static int ipvlan_addr6_validator_event(struct notifier_block *unused,
882 unsigned long event, void *ptr)
883{
884 struct in6_validator_info *i6vi = (struct in6_validator_info *)ptr;
885 struct net_device *dev = (struct net_device *)i6vi->i6vi_dev->dev;
886 struct ipvl_dev *ipvlan = netdev_priv(dev);
887
888 if (!ipvlan_is_valid_dev(dev))
889 return NOTIFY_DONE;
890
891 switch (event) {
892 case NETDEV_UP:
893 if (ipvlan_addr_busy(ipvlan->port, &i6vi->i6vi_addr, true)) {
894 NL_SET_ERR_MSG(i6vi->extack,
895 "Address already assigned to an ipvlan device");
896 return notifier_from_errno(-EADDRINUSE);
897 }
898 break;
899 }
900
901 return NOTIFY_OK;
902}
903#endif
904
905static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
906{
907 int ret = -EINVAL;
908
909 spin_lock_bh(&ipvlan->addrs_lock);
910 if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false))
911 netif_err(ipvlan, ifup, ipvlan->dev,
912 "Failed to add IPv4=%pI4 on %s intf.\n",
913 ip4_addr, ipvlan->dev->name);
914 else
915 ret = ipvlan_add_addr(ipvlan, ip4_addr, false);
916 spin_unlock_bh(&ipvlan->addrs_lock);
917 return ret;
918}
919
920static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
921{
922 return ipvlan_del_addr(ipvlan, ip4_addr, false);
923}
924
925static int ipvlan_addr4_event(struct notifier_block *unused,
926 unsigned long event, void *ptr)
927{
928 struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
929 struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
930 struct ipvl_dev *ipvlan = netdev_priv(dev);
931 struct in_addr ip4_addr;
932
933 if (!ipvlan_is_valid_dev(dev))
934 return NOTIFY_DONE;
935
936 switch (event) {
937 case NETDEV_UP:
938 ip4_addr.s_addr = if4->ifa_address;
939 if (ipvlan_add_addr4(ipvlan, &ip4_addr))
940 return NOTIFY_BAD;
941 break;
942
943 case NETDEV_DOWN:
944 ip4_addr.s_addr = if4->ifa_address;
945 ipvlan_del_addr4(ipvlan, &ip4_addr);
946 break;
947 }
948
949 return NOTIFY_OK;
950}
951
952static int ipvlan_addr4_validator_event(struct notifier_block *unused,
953 unsigned long event, void *ptr)
954{
955 struct in_validator_info *ivi = (struct in_validator_info *)ptr;
956 struct net_device *dev = (struct net_device *)ivi->ivi_dev->dev;
957 struct ipvl_dev *ipvlan = netdev_priv(dev);
958
959 if (!ipvlan_is_valid_dev(dev))
960 return NOTIFY_DONE;
961
962 switch (event) {
963 case NETDEV_UP:
964 if (ipvlan_addr_busy(ipvlan->port, &ivi->ivi_addr, false)) {
965 NL_SET_ERR_MSG(ivi->extack,
966 "Address already assigned to an ipvlan device");
967 return notifier_from_errno(-EADDRINUSE);
968 }
969 break;
970 }
971
972 return NOTIFY_OK;
973}
974
975static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
976 .notifier_call = ipvlan_addr4_event,
977};
978
979static struct notifier_block ipvlan_addr4_vtor_notifier_block __read_mostly = {
980 .notifier_call = ipvlan_addr4_validator_event,
981};
982
983static struct notifier_block ipvlan_notifier_block __read_mostly = {
984 .notifier_call = ipvlan_device_event,
985};
986
987#if IS_ENABLED(CONFIG_IPV6)
988static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
989 .notifier_call = ipvlan_addr6_event,
990};
991
992static struct notifier_block ipvlan_addr6_vtor_notifier_block __read_mostly = {
993 .notifier_call = ipvlan_addr6_validator_event,
994};
995#endif
996
997static int __init ipvlan_init_module(void)
998{
999 int err;
1000
1001 ipvlan_init_secret();
1002 register_netdevice_notifier(&ipvlan_notifier_block);
1003#if IS_ENABLED(CONFIG_IPV6)
1004 register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1005 register_inet6addr_validator_notifier(
1006 &ipvlan_addr6_vtor_notifier_block);
1007#endif
1008 register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1009 register_inetaddr_validator_notifier(&ipvlan_addr4_vtor_notifier_block);
1010
1011 err = ipvlan_l3s_init();
1012 if (err < 0)
1013 goto error;
1014
1015 err = ipvlan_link_register(&ipvlan_link_ops);
1016 if (err < 0) {
1017 ipvlan_l3s_cleanup();
1018 goto error;
1019 }
1020
1021 return 0;
1022error:
1023 unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1024 unregister_inetaddr_validator_notifier(
1025 &ipvlan_addr4_vtor_notifier_block);
1026#if IS_ENABLED(CONFIG_IPV6)
1027 unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1028 unregister_inet6addr_validator_notifier(
1029 &ipvlan_addr6_vtor_notifier_block);
1030#endif
1031 unregister_netdevice_notifier(&ipvlan_notifier_block);
1032 return err;
1033}
1034
1035static void __exit ipvlan_cleanup_module(void)
1036{
1037 rtnl_link_unregister(&ipvlan_link_ops);
1038 ipvlan_l3s_cleanup();
1039 unregister_netdevice_notifier(&ipvlan_notifier_block);
1040 unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
1041 unregister_inetaddr_validator_notifier(
1042 &ipvlan_addr4_vtor_notifier_block);
1043#if IS_ENABLED(CONFIG_IPV6)
1044 unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
1045 unregister_inet6addr_validator_notifier(
1046 &ipvlan_addr6_vtor_notifier_block);
1047#endif
1048}
1049
1050module_init(ipvlan_init_module);
1051module_exit(ipvlan_cleanup_module);
1052
1053MODULE_LICENSE("GPL");
1054MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
1055MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
1056MODULE_ALIAS_RTNL_LINK("ipvlan");
1/* Copyright (c) 2014 Mahesh Bandewar <maheshb@google.com>
2 *
3 * This program is free software; you can redistribute it and/or
4 * modify it under the terms of the GNU General Public License as
5 * published by the Free Software Foundation; either version 2 of
6 * the License, or (at your option) any later version.
7 *
8 */
9
10#include "ipvlan.h"
11
12static void ipvlan_adjust_mtu(struct ipvl_dev *ipvlan, struct net_device *dev)
13{
14 ipvlan->dev->mtu = dev->mtu - ipvlan->mtu_adj;
15}
16
17static void ipvlan_set_port_mode(struct ipvl_port *port, u16 nval)
18{
19 struct ipvl_dev *ipvlan;
20
21 if (port->mode != nval) {
22 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
23 if (nval == IPVLAN_MODE_L3)
24 ipvlan->dev->flags |= IFF_NOARP;
25 else
26 ipvlan->dev->flags &= ~IFF_NOARP;
27 }
28 port->mode = nval;
29 }
30}
31
32static int ipvlan_port_create(struct net_device *dev)
33{
34 struct ipvl_port *port;
35 int err, idx;
36
37 if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK) {
38 netdev_err(dev, "Master is either lo or non-ether device\n");
39 return -EINVAL;
40 }
41
42 if (netif_is_macvlan_port(dev)) {
43 netdev_err(dev, "Master is a macvlan port.\n");
44 return -EBUSY;
45 }
46
47 port = kzalloc(sizeof(struct ipvl_port), GFP_KERNEL);
48 if (!port)
49 return -ENOMEM;
50
51 port->dev = dev;
52 port->mode = IPVLAN_MODE_L3;
53 INIT_LIST_HEAD(&port->ipvlans);
54 for (idx = 0; idx < IPVLAN_HASH_SIZE; idx++)
55 INIT_HLIST_HEAD(&port->hlhead[idx]);
56
57 skb_queue_head_init(&port->backlog);
58 INIT_WORK(&port->wq, ipvlan_process_multicast);
59
60 err = netdev_rx_handler_register(dev, ipvlan_handle_frame, port);
61 if (err)
62 goto err;
63
64 dev->priv_flags |= IFF_IPVLAN_MASTER;
65 return 0;
66
67err:
68 kfree_rcu(port, rcu);
69 return err;
70}
71
72static void ipvlan_port_destroy(struct net_device *dev)
73{
74 struct ipvl_port *port = ipvlan_port_get_rtnl(dev);
75
76 dev->priv_flags &= ~IFF_IPVLAN_MASTER;
77 netdev_rx_handler_unregister(dev);
78 cancel_work_sync(&port->wq);
79 __skb_queue_purge(&port->backlog);
80 kfree_rcu(port, rcu);
81}
82
83/* ipvlan network devices have devices nesting below it and are a special
84 * "super class" of normal network devices; split their locks off into a
85 * separate class since they always nest.
86 */
87static struct lock_class_key ipvlan_netdev_xmit_lock_key;
88static struct lock_class_key ipvlan_netdev_addr_lock_key;
89
90#define IPVLAN_FEATURES \
91 (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
92 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \
93 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
94 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
95
96#define IPVLAN_STATE_MASK \
97 ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
98
99static void ipvlan_set_lockdep_class_one(struct net_device *dev,
100 struct netdev_queue *txq,
101 void *_unused)
102{
103 lockdep_set_class(&txq->_xmit_lock, &ipvlan_netdev_xmit_lock_key);
104}
105
106static void ipvlan_set_lockdep_class(struct net_device *dev)
107{
108 lockdep_set_class(&dev->addr_list_lock, &ipvlan_netdev_addr_lock_key);
109 netdev_for_each_tx_queue(dev, ipvlan_set_lockdep_class_one, NULL);
110}
111
112static int ipvlan_init(struct net_device *dev)
113{
114 struct ipvl_dev *ipvlan = netdev_priv(dev);
115 const struct net_device *phy_dev = ipvlan->phy_dev;
116
117 dev->state = (dev->state & ~IPVLAN_STATE_MASK) |
118 (phy_dev->state & IPVLAN_STATE_MASK);
119 dev->features = phy_dev->features & IPVLAN_FEATURES;
120 dev->features |= NETIF_F_LLTX;
121 dev->gso_max_size = phy_dev->gso_max_size;
122 dev->gso_max_segs = phy_dev->gso_max_segs;
123 dev->hard_header_len = phy_dev->hard_header_len;
124
125 ipvlan_set_lockdep_class(dev);
126
127 ipvlan->pcpu_stats = alloc_percpu(struct ipvl_pcpu_stats);
128 if (!ipvlan->pcpu_stats)
129 return -ENOMEM;
130
131 return 0;
132}
133
134static void ipvlan_uninit(struct net_device *dev)
135{
136 struct ipvl_dev *ipvlan = netdev_priv(dev);
137 struct ipvl_port *port = ipvlan->port;
138
139 free_percpu(ipvlan->pcpu_stats);
140
141 port->count -= 1;
142 if (!port->count)
143 ipvlan_port_destroy(port->dev);
144}
145
146static int ipvlan_open(struct net_device *dev)
147{
148 struct ipvl_dev *ipvlan = netdev_priv(dev);
149 struct net_device *phy_dev = ipvlan->phy_dev;
150 struct ipvl_addr *addr;
151
152 if (ipvlan->port->mode == IPVLAN_MODE_L3)
153 dev->flags |= IFF_NOARP;
154 else
155 dev->flags &= ~IFF_NOARP;
156
157 list_for_each_entry(addr, &ipvlan->addrs, anode)
158 ipvlan_ht_addr_add(ipvlan, addr);
159
160 return dev_uc_add(phy_dev, phy_dev->dev_addr);
161}
162
163static int ipvlan_stop(struct net_device *dev)
164{
165 struct ipvl_dev *ipvlan = netdev_priv(dev);
166 struct net_device *phy_dev = ipvlan->phy_dev;
167 struct ipvl_addr *addr;
168
169 dev_uc_unsync(phy_dev, dev);
170 dev_mc_unsync(phy_dev, dev);
171
172 dev_uc_del(phy_dev, phy_dev->dev_addr);
173
174 list_for_each_entry(addr, &ipvlan->addrs, anode)
175 ipvlan_ht_addr_del(addr);
176
177 return 0;
178}
179
180static netdev_tx_t ipvlan_start_xmit(struct sk_buff *skb,
181 struct net_device *dev)
182{
183 const struct ipvl_dev *ipvlan = netdev_priv(dev);
184 int skblen = skb->len;
185 int ret;
186
187 ret = ipvlan_queue_xmit(skb, dev);
188 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
189 struct ipvl_pcpu_stats *pcptr;
190
191 pcptr = this_cpu_ptr(ipvlan->pcpu_stats);
192
193 u64_stats_update_begin(&pcptr->syncp);
194 pcptr->tx_pkts++;
195 pcptr->tx_bytes += skblen;
196 u64_stats_update_end(&pcptr->syncp);
197 } else {
198 this_cpu_inc(ipvlan->pcpu_stats->tx_drps);
199 }
200 return ret;
201}
202
203static netdev_features_t ipvlan_fix_features(struct net_device *dev,
204 netdev_features_t features)
205{
206 struct ipvl_dev *ipvlan = netdev_priv(dev);
207
208 return features & (ipvlan->sfeatures | ~IPVLAN_FEATURES);
209}
210
211static void ipvlan_change_rx_flags(struct net_device *dev, int change)
212{
213 struct ipvl_dev *ipvlan = netdev_priv(dev);
214 struct net_device *phy_dev = ipvlan->phy_dev;
215
216 if (change & IFF_ALLMULTI)
217 dev_set_allmulti(phy_dev, dev->flags & IFF_ALLMULTI? 1 : -1);
218}
219
220static void ipvlan_set_multicast_mac_filter(struct net_device *dev)
221{
222 struct ipvl_dev *ipvlan = netdev_priv(dev);
223
224 if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
225 bitmap_fill(ipvlan->mac_filters, IPVLAN_MAC_FILTER_SIZE);
226 } else {
227 struct netdev_hw_addr *ha;
228 DECLARE_BITMAP(mc_filters, IPVLAN_MAC_FILTER_SIZE);
229
230 bitmap_zero(mc_filters, IPVLAN_MAC_FILTER_SIZE);
231 netdev_for_each_mc_addr(ha, dev)
232 __set_bit(ipvlan_mac_hash(ha->addr), mc_filters);
233
234 /* Turn-on broadcast bit irrespective of address family,
235 * since broadcast is deferred to a work-queue, hence no
236 * impact on fast-path processing.
237 */
238 __set_bit(ipvlan_mac_hash(dev->broadcast), mc_filters);
239
240 bitmap_copy(ipvlan->mac_filters, mc_filters,
241 IPVLAN_MAC_FILTER_SIZE);
242 }
243 dev_uc_sync(ipvlan->phy_dev, dev);
244 dev_mc_sync(ipvlan->phy_dev, dev);
245}
246
247static struct rtnl_link_stats64 *ipvlan_get_stats64(struct net_device *dev,
248 struct rtnl_link_stats64 *s)
249{
250 struct ipvl_dev *ipvlan = netdev_priv(dev);
251
252 if (ipvlan->pcpu_stats) {
253 struct ipvl_pcpu_stats *pcptr;
254 u64 rx_pkts, rx_bytes, rx_mcast, tx_pkts, tx_bytes;
255 u32 rx_errs = 0, tx_drps = 0;
256 u32 strt;
257 int idx;
258
259 for_each_possible_cpu(idx) {
260 pcptr = per_cpu_ptr(ipvlan->pcpu_stats, idx);
261 do {
262 strt= u64_stats_fetch_begin_irq(&pcptr->syncp);
263 rx_pkts = pcptr->rx_pkts;
264 rx_bytes = pcptr->rx_bytes;
265 rx_mcast = pcptr->rx_mcast;
266 tx_pkts = pcptr->tx_pkts;
267 tx_bytes = pcptr->tx_bytes;
268 } while (u64_stats_fetch_retry_irq(&pcptr->syncp,
269 strt));
270
271 s->rx_packets += rx_pkts;
272 s->rx_bytes += rx_bytes;
273 s->multicast += rx_mcast;
274 s->tx_packets += tx_pkts;
275 s->tx_bytes += tx_bytes;
276
277 /* u32 values are updated without syncp protection. */
278 rx_errs += pcptr->rx_errs;
279 tx_drps += pcptr->tx_drps;
280 }
281 s->rx_errors = rx_errs;
282 s->rx_dropped = rx_errs;
283 s->tx_dropped = tx_drps;
284 }
285 return s;
286}
287
288static int ipvlan_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
289{
290 struct ipvl_dev *ipvlan = netdev_priv(dev);
291 struct net_device *phy_dev = ipvlan->phy_dev;
292
293 return vlan_vid_add(phy_dev, proto, vid);
294}
295
296static int ipvlan_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
297 u16 vid)
298{
299 struct ipvl_dev *ipvlan = netdev_priv(dev);
300 struct net_device *phy_dev = ipvlan->phy_dev;
301
302 vlan_vid_del(phy_dev, proto, vid);
303 return 0;
304}
305
306static int ipvlan_get_iflink(const struct net_device *dev)
307{
308 struct ipvl_dev *ipvlan = netdev_priv(dev);
309
310 return ipvlan->phy_dev->ifindex;
311}
312
313static const struct net_device_ops ipvlan_netdev_ops = {
314 .ndo_init = ipvlan_init,
315 .ndo_uninit = ipvlan_uninit,
316 .ndo_open = ipvlan_open,
317 .ndo_stop = ipvlan_stop,
318 .ndo_start_xmit = ipvlan_start_xmit,
319 .ndo_fix_features = ipvlan_fix_features,
320 .ndo_change_rx_flags = ipvlan_change_rx_flags,
321 .ndo_set_rx_mode = ipvlan_set_multicast_mac_filter,
322 .ndo_get_stats64 = ipvlan_get_stats64,
323 .ndo_vlan_rx_add_vid = ipvlan_vlan_rx_add_vid,
324 .ndo_vlan_rx_kill_vid = ipvlan_vlan_rx_kill_vid,
325 .ndo_get_iflink = ipvlan_get_iflink,
326};
327
328static int ipvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
329 unsigned short type, const void *daddr,
330 const void *saddr, unsigned len)
331{
332 const struct ipvl_dev *ipvlan = netdev_priv(dev);
333 struct net_device *phy_dev = ipvlan->phy_dev;
334
335 /* TODO Probably use a different field than dev_addr so that the
336 * mac-address on the virtual device is portable and can be carried
337 * while the packets use the mac-addr on the physical device.
338 */
339 return dev_hard_header(skb, phy_dev, type, daddr,
340 saddr ? : dev->dev_addr, len);
341}
342
343static const struct header_ops ipvlan_header_ops = {
344 .create = ipvlan_hard_header,
345 .parse = eth_header_parse,
346 .cache = eth_header_cache,
347 .cache_update = eth_header_cache_update,
348};
349
350static int ipvlan_ethtool_get_link_ksettings(struct net_device *dev,
351 struct ethtool_link_ksettings *cmd)
352{
353 const struct ipvl_dev *ipvlan = netdev_priv(dev);
354
355 return __ethtool_get_link_ksettings(ipvlan->phy_dev, cmd);
356}
357
358static void ipvlan_ethtool_get_drvinfo(struct net_device *dev,
359 struct ethtool_drvinfo *drvinfo)
360{
361 strlcpy(drvinfo->driver, IPVLAN_DRV, sizeof(drvinfo->driver));
362 strlcpy(drvinfo->version, IPV_DRV_VER, sizeof(drvinfo->version));
363}
364
365static u32 ipvlan_ethtool_get_msglevel(struct net_device *dev)
366{
367 const struct ipvl_dev *ipvlan = netdev_priv(dev);
368
369 return ipvlan->msg_enable;
370}
371
372static void ipvlan_ethtool_set_msglevel(struct net_device *dev, u32 value)
373{
374 struct ipvl_dev *ipvlan = netdev_priv(dev);
375
376 ipvlan->msg_enable = value;
377}
378
379static const struct ethtool_ops ipvlan_ethtool_ops = {
380 .get_link = ethtool_op_get_link,
381 .get_link_ksettings = ipvlan_ethtool_get_link_ksettings,
382 .get_drvinfo = ipvlan_ethtool_get_drvinfo,
383 .get_msglevel = ipvlan_ethtool_get_msglevel,
384 .set_msglevel = ipvlan_ethtool_set_msglevel,
385};
386
387static int ipvlan_nl_changelink(struct net_device *dev,
388 struct nlattr *tb[], struct nlattr *data[])
389{
390 struct ipvl_dev *ipvlan = netdev_priv(dev);
391 struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
392
393 if (data && data[IFLA_IPVLAN_MODE]) {
394 u16 nmode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
395
396 ipvlan_set_port_mode(port, nmode);
397 }
398 return 0;
399}
400
401static size_t ipvlan_nl_getsize(const struct net_device *dev)
402{
403 return (0
404 + nla_total_size(2) /* IFLA_IPVLAN_MODE */
405 );
406}
407
408static int ipvlan_nl_validate(struct nlattr *tb[], struct nlattr *data[])
409{
410 if (data && data[IFLA_IPVLAN_MODE]) {
411 u16 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
412
413 if (mode < IPVLAN_MODE_L2 || mode >= IPVLAN_MODE_MAX)
414 return -EINVAL;
415 }
416 return 0;
417}
418
419static int ipvlan_nl_fillinfo(struct sk_buff *skb,
420 const struct net_device *dev)
421{
422 struct ipvl_dev *ipvlan = netdev_priv(dev);
423 struct ipvl_port *port = ipvlan_port_get_rtnl(ipvlan->phy_dev);
424 int ret = -EINVAL;
425
426 if (!port)
427 goto err;
428
429 ret = -EMSGSIZE;
430 if (nla_put_u16(skb, IFLA_IPVLAN_MODE, port->mode))
431 goto err;
432
433 return 0;
434
435err:
436 return ret;
437}
438
439static int ipvlan_link_new(struct net *src_net, struct net_device *dev,
440 struct nlattr *tb[], struct nlattr *data[])
441{
442 struct ipvl_dev *ipvlan = netdev_priv(dev);
443 struct ipvl_port *port;
444 struct net_device *phy_dev;
445 int err;
446 u16 mode = IPVLAN_MODE_L3;
447
448 if (!tb[IFLA_LINK])
449 return -EINVAL;
450
451 phy_dev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
452 if (!phy_dev)
453 return -ENODEV;
454
455 if (netif_is_ipvlan(phy_dev)) {
456 struct ipvl_dev *tmp = netdev_priv(phy_dev);
457
458 phy_dev = tmp->phy_dev;
459 } else if (!netif_is_ipvlan_port(phy_dev)) {
460 err = ipvlan_port_create(phy_dev);
461 if (err < 0)
462 return err;
463 }
464
465 if (data && data[IFLA_IPVLAN_MODE])
466 mode = nla_get_u16(data[IFLA_IPVLAN_MODE]);
467
468 port = ipvlan_port_get_rtnl(phy_dev);
469 ipvlan->phy_dev = phy_dev;
470 ipvlan->dev = dev;
471 ipvlan->port = port;
472 ipvlan->sfeatures = IPVLAN_FEATURES;
473 ipvlan_adjust_mtu(ipvlan, phy_dev);
474 INIT_LIST_HEAD(&ipvlan->addrs);
475
476 /* TODO Probably put random address here to be presented to the
477 * world but keep using the physical-dev address for the outgoing
478 * packets.
479 */
480 memcpy(dev->dev_addr, phy_dev->dev_addr, ETH_ALEN);
481
482 dev->priv_flags |= IFF_IPVLAN_SLAVE;
483
484 port->count += 1;
485 err = register_netdevice(dev);
486 if (err < 0)
487 goto ipvlan_destroy_port;
488
489 err = netdev_upper_dev_link(phy_dev, dev);
490 if (err)
491 goto ipvlan_destroy_port;
492
493 list_add_tail_rcu(&ipvlan->pnode, &port->ipvlans);
494 ipvlan_set_port_mode(port, mode);
495
496 netif_stacked_transfer_operstate(phy_dev, dev);
497 return 0;
498
499ipvlan_destroy_port:
500 port->count -= 1;
501 if (!port->count)
502 ipvlan_port_destroy(phy_dev);
503
504 return err;
505}
506
507static void ipvlan_link_delete(struct net_device *dev, struct list_head *head)
508{
509 struct ipvl_dev *ipvlan = netdev_priv(dev);
510 struct ipvl_addr *addr, *next;
511
512 list_for_each_entry_safe(addr, next, &ipvlan->addrs, anode) {
513 ipvlan_ht_addr_del(addr);
514 list_del(&addr->anode);
515 kfree_rcu(addr, rcu);
516 }
517
518 list_del_rcu(&ipvlan->pnode);
519 unregister_netdevice_queue(dev, head);
520 netdev_upper_dev_unlink(ipvlan->phy_dev, dev);
521}
522
523static void ipvlan_link_setup(struct net_device *dev)
524{
525 ether_setup(dev);
526
527 dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
528 dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
529 dev->netdev_ops = &ipvlan_netdev_ops;
530 dev->destructor = free_netdev;
531 dev->header_ops = &ipvlan_header_ops;
532 dev->ethtool_ops = &ipvlan_ethtool_ops;
533}
534
535static const struct nla_policy ipvlan_nl_policy[IFLA_IPVLAN_MAX + 1] =
536{
537 [IFLA_IPVLAN_MODE] = { .type = NLA_U16 },
538};
539
540static struct rtnl_link_ops ipvlan_link_ops = {
541 .kind = "ipvlan",
542 .priv_size = sizeof(struct ipvl_dev),
543
544 .get_size = ipvlan_nl_getsize,
545 .policy = ipvlan_nl_policy,
546 .validate = ipvlan_nl_validate,
547 .fill_info = ipvlan_nl_fillinfo,
548 .changelink = ipvlan_nl_changelink,
549 .maxtype = IFLA_IPVLAN_MAX,
550
551 .setup = ipvlan_link_setup,
552 .newlink = ipvlan_link_new,
553 .dellink = ipvlan_link_delete,
554};
555
556static int ipvlan_link_register(struct rtnl_link_ops *ops)
557{
558 return rtnl_link_register(ops);
559}
560
561static int ipvlan_device_event(struct notifier_block *unused,
562 unsigned long event, void *ptr)
563{
564 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
565 struct ipvl_dev *ipvlan, *next;
566 struct ipvl_port *port;
567 LIST_HEAD(lst_kill);
568
569 if (!netif_is_ipvlan_port(dev))
570 return NOTIFY_DONE;
571
572 port = ipvlan_port_get_rtnl(dev);
573
574 switch (event) {
575 case NETDEV_CHANGE:
576 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
577 netif_stacked_transfer_operstate(ipvlan->phy_dev,
578 ipvlan->dev);
579 break;
580
581 case NETDEV_UNREGISTER:
582 if (dev->reg_state != NETREG_UNREGISTERING)
583 break;
584
585 list_for_each_entry_safe(ipvlan, next, &port->ipvlans,
586 pnode)
587 ipvlan->dev->rtnl_link_ops->dellink(ipvlan->dev,
588 &lst_kill);
589 unregister_netdevice_many(&lst_kill);
590 break;
591
592 case NETDEV_FEAT_CHANGE:
593 list_for_each_entry(ipvlan, &port->ipvlans, pnode) {
594 ipvlan->dev->features = dev->features & IPVLAN_FEATURES;
595 ipvlan->dev->gso_max_size = dev->gso_max_size;
596 ipvlan->dev->gso_max_segs = dev->gso_max_segs;
597 netdev_features_change(ipvlan->dev);
598 }
599 break;
600
601 case NETDEV_CHANGEMTU:
602 list_for_each_entry(ipvlan, &port->ipvlans, pnode)
603 ipvlan_adjust_mtu(ipvlan, dev);
604 break;
605
606 case NETDEV_PRE_TYPE_CHANGE:
607 /* Forbid underlying device to change its type. */
608 return NOTIFY_BAD;
609 }
610 return NOTIFY_DONE;
611}
612
613static int ipvlan_add_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
614{
615 struct ipvl_addr *addr;
616
617 if (ipvlan_addr_busy(ipvlan->port, ip6_addr, true)) {
618 netif_err(ipvlan, ifup, ipvlan->dev,
619 "Failed to add IPv6=%pI6c addr for %s intf\n",
620 ip6_addr, ipvlan->dev->name);
621 return -EINVAL;
622 }
623 addr = kzalloc(sizeof(struct ipvl_addr), GFP_ATOMIC);
624 if (!addr)
625 return -ENOMEM;
626
627 addr->master = ipvlan;
628 memcpy(&addr->ip6addr, ip6_addr, sizeof(struct in6_addr));
629 addr->atype = IPVL_IPV6;
630 list_add_tail(&addr->anode, &ipvlan->addrs);
631
632 /* If the interface is not up, the address will be added to the hash
633 * list by ipvlan_open.
634 */
635 if (netif_running(ipvlan->dev))
636 ipvlan_ht_addr_add(ipvlan, addr);
637
638 return 0;
639}
640
641static void ipvlan_del_addr6(struct ipvl_dev *ipvlan, struct in6_addr *ip6_addr)
642{
643 struct ipvl_addr *addr;
644
645 addr = ipvlan_find_addr(ipvlan, ip6_addr, true);
646 if (!addr)
647 return;
648
649 ipvlan_ht_addr_del(addr);
650 list_del(&addr->anode);
651 kfree_rcu(addr, rcu);
652
653 return;
654}
655
656static int ipvlan_addr6_event(struct notifier_block *unused,
657 unsigned long event, void *ptr)
658{
659 struct inet6_ifaddr *if6 = (struct inet6_ifaddr *)ptr;
660 struct net_device *dev = (struct net_device *)if6->idev->dev;
661 struct ipvl_dev *ipvlan = netdev_priv(dev);
662
663 /* FIXME IPv6 autoconf calls us from bh without RTNL */
664 if (in_softirq())
665 return NOTIFY_DONE;
666
667 if (!netif_is_ipvlan(dev))
668 return NOTIFY_DONE;
669
670 if (!ipvlan || !ipvlan->port)
671 return NOTIFY_DONE;
672
673 switch (event) {
674 case NETDEV_UP:
675 if (ipvlan_add_addr6(ipvlan, &if6->addr))
676 return NOTIFY_BAD;
677 break;
678
679 case NETDEV_DOWN:
680 ipvlan_del_addr6(ipvlan, &if6->addr);
681 break;
682 }
683
684 return NOTIFY_OK;
685}
686
687static int ipvlan_add_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
688{
689 struct ipvl_addr *addr;
690
691 if (ipvlan_addr_busy(ipvlan->port, ip4_addr, false)) {
692 netif_err(ipvlan, ifup, ipvlan->dev,
693 "Failed to add IPv4=%pI4 on %s intf.\n",
694 ip4_addr, ipvlan->dev->name);
695 return -EINVAL;
696 }
697 addr = kzalloc(sizeof(struct ipvl_addr), GFP_KERNEL);
698 if (!addr)
699 return -ENOMEM;
700
701 addr->master = ipvlan;
702 memcpy(&addr->ip4addr, ip4_addr, sizeof(struct in_addr));
703 addr->atype = IPVL_IPV4;
704 list_add_tail(&addr->anode, &ipvlan->addrs);
705
706 /* If the interface is not up, the address will be added to the hash
707 * list by ipvlan_open.
708 */
709 if (netif_running(ipvlan->dev))
710 ipvlan_ht_addr_add(ipvlan, addr);
711
712 return 0;
713}
714
715static void ipvlan_del_addr4(struct ipvl_dev *ipvlan, struct in_addr *ip4_addr)
716{
717 struct ipvl_addr *addr;
718
719 addr = ipvlan_find_addr(ipvlan, ip4_addr, false);
720 if (!addr)
721 return;
722
723 ipvlan_ht_addr_del(addr);
724 list_del(&addr->anode);
725 kfree_rcu(addr, rcu);
726
727 return;
728}
729
730static int ipvlan_addr4_event(struct notifier_block *unused,
731 unsigned long event, void *ptr)
732{
733 struct in_ifaddr *if4 = (struct in_ifaddr *)ptr;
734 struct net_device *dev = (struct net_device *)if4->ifa_dev->dev;
735 struct ipvl_dev *ipvlan = netdev_priv(dev);
736 struct in_addr ip4_addr;
737
738 if (!netif_is_ipvlan(dev))
739 return NOTIFY_DONE;
740
741 if (!ipvlan || !ipvlan->port)
742 return NOTIFY_DONE;
743
744 switch (event) {
745 case NETDEV_UP:
746 ip4_addr.s_addr = if4->ifa_address;
747 if (ipvlan_add_addr4(ipvlan, &ip4_addr))
748 return NOTIFY_BAD;
749 break;
750
751 case NETDEV_DOWN:
752 ip4_addr.s_addr = if4->ifa_address;
753 ipvlan_del_addr4(ipvlan, &ip4_addr);
754 break;
755 }
756
757 return NOTIFY_OK;
758}
759
760static struct notifier_block ipvlan_addr4_notifier_block __read_mostly = {
761 .notifier_call = ipvlan_addr4_event,
762};
763
764static struct notifier_block ipvlan_notifier_block __read_mostly = {
765 .notifier_call = ipvlan_device_event,
766};
767
768static struct notifier_block ipvlan_addr6_notifier_block __read_mostly = {
769 .notifier_call = ipvlan_addr6_event,
770};
771
772static int __init ipvlan_init_module(void)
773{
774 int err;
775
776 ipvlan_init_secret();
777 register_netdevice_notifier(&ipvlan_notifier_block);
778 register_inet6addr_notifier(&ipvlan_addr6_notifier_block);
779 register_inetaddr_notifier(&ipvlan_addr4_notifier_block);
780
781 err = ipvlan_link_register(&ipvlan_link_ops);
782 if (err < 0)
783 goto error;
784
785 return 0;
786error:
787 unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
788 unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
789 unregister_netdevice_notifier(&ipvlan_notifier_block);
790 return err;
791}
792
793static void __exit ipvlan_cleanup_module(void)
794{
795 rtnl_link_unregister(&ipvlan_link_ops);
796 unregister_netdevice_notifier(&ipvlan_notifier_block);
797 unregister_inetaddr_notifier(&ipvlan_addr4_notifier_block);
798 unregister_inet6addr_notifier(&ipvlan_addr6_notifier_block);
799}
800
801module_init(ipvlan_init_module);
802module_exit(ipvlan_cleanup_module);
803
804MODULE_LICENSE("GPL");
805MODULE_AUTHOR("Mahesh Bandewar <maheshb@google.com>");
806MODULE_DESCRIPTION("Driver for L3 (IPv6/IPv4) based VLANs");
807MODULE_ALIAS_RTNL_LINK("ipvlan");