xref: /openbmc/linux/drivers/net/net_failover.c (revision 77ab8d5d)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2018, Intel Corporation. */
3 
4 /* This provides a net_failover interface for paravirtual drivers to
5  * provide an alternate datapath by exporting APIs to create and
6  * destroy a upper 'net_failover' netdev. The upper dev manages the
7  * original paravirtual interface as a 'standby' netdev and uses the
8  * generic failover infrastructure to register and manage a direct
9  * attached VF as a 'primary' netdev. This enables live migration of
10  * a VM with direct attached VF by failing over to the paravirtual
11  * datapath when the VF is unplugged.
12  *
13  * Some of the netdev management routines are based on bond/team driver as
14  * this driver provides active-backup functionality similar to those drivers.
15  */
16 
17 #include <linux/netdevice.h>
18 #include <linux/etherdevice.h>
19 #include <linux/ethtool.h>
20 #include <linux/module.h>
21 #include <linux/slab.h>
22 #include <linux/netdevice.h>
23 #include <linux/netpoll.h>
24 #include <linux/rtnetlink.h>
25 #include <linux/if_vlan.h>
26 #include <linux/pci.h>
27 #include <net/sch_generic.h>
28 #include <uapi/linux/if_arp.h>
29 #include <net/net_failover.h>
30 
31 static bool net_failover_xmit_ready(struct net_device *dev)
32 {
33 	return netif_running(dev) && netif_carrier_ok(dev);
34 }
35 
36 static int net_failover_open(struct net_device *dev)
37 {
38 	struct net_failover_info *nfo_info = netdev_priv(dev);
39 	struct net_device *primary_dev, *standby_dev;
40 	int err;
41 
42 	primary_dev = rtnl_dereference(nfo_info->primary_dev);
43 	if (primary_dev) {
44 		err = dev_open(primary_dev);
45 		if (err)
46 			goto err_primary_open;
47 	}
48 
49 	standby_dev = rtnl_dereference(nfo_info->standby_dev);
50 	if (standby_dev) {
51 		err = dev_open(standby_dev);
52 		if (err)
53 			goto err_standby_open;
54 	}
55 
56 	if ((primary_dev && net_failover_xmit_ready(primary_dev)) ||
57 	    (standby_dev && net_failover_xmit_ready(standby_dev))) {
58 		netif_carrier_on(dev);
59 		netif_tx_wake_all_queues(dev);
60 	}
61 
62 	return 0;
63 
64 err_standby_open:
65 	dev_close(primary_dev);
66 err_primary_open:
67 	netif_tx_disable(dev);
68 	return err;
69 }
70 
71 static int net_failover_close(struct net_device *dev)
72 {
73 	struct net_failover_info *nfo_info = netdev_priv(dev);
74 	struct net_device *slave_dev;
75 
76 	netif_tx_disable(dev);
77 
78 	slave_dev = rtnl_dereference(nfo_info->primary_dev);
79 	if (slave_dev)
80 		dev_close(slave_dev);
81 
82 	slave_dev = rtnl_dereference(nfo_info->standby_dev);
83 	if (slave_dev)
84 		dev_close(slave_dev);
85 
86 	return 0;
87 }
88 
89 static netdev_tx_t net_failover_drop_xmit(struct sk_buff *skb,
90 					  struct net_device *dev)
91 {
92 	atomic_long_inc(&dev->tx_dropped);
93 	dev_kfree_skb_any(skb);
94 	return NETDEV_TX_OK;
95 }
96 
97 static netdev_tx_t net_failover_start_xmit(struct sk_buff *skb,
98 					   struct net_device *dev)
99 {
100 	struct net_failover_info *nfo_info = netdev_priv(dev);
101 	struct net_device *xmit_dev;
102 
103 	/* Try xmit via primary netdev followed by standby netdev */
104 	xmit_dev = rcu_dereference_bh(nfo_info->primary_dev);
105 	if (!xmit_dev || !net_failover_xmit_ready(xmit_dev)) {
106 		xmit_dev = rcu_dereference_bh(nfo_info->standby_dev);
107 		if (!xmit_dev || !net_failover_xmit_ready(xmit_dev))
108 			return net_failover_drop_xmit(skb, dev);
109 	}
110 
111 	skb->dev = xmit_dev;
112 	skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
113 
114 	return dev_queue_xmit(skb);
115 }
116 
117 static u16 net_failover_select_queue(struct net_device *dev,
118 				     struct sk_buff *skb, void *accel_priv,
119 				     select_queue_fallback_t fallback)
120 {
121 	struct net_failover_info *nfo_info = netdev_priv(dev);
122 	struct net_device *primary_dev;
123 	u16 txq;
124 
125 	primary_dev = rcu_dereference(nfo_info->primary_dev);
126 	if (primary_dev) {
127 		const struct net_device_ops *ops = primary_dev->netdev_ops;
128 
129 		if (ops->ndo_select_queue)
130 			txq = ops->ndo_select_queue(primary_dev, skb,
131 						    accel_priv, fallback);
132 		else
133 			txq = fallback(primary_dev, skb);
134 
135 		qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
136 
137 		return txq;
138 	}
139 
140 	txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
141 
142 	/* Save the original txq to restore before passing to the driver */
143 	qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
144 
145 	if (unlikely(txq >= dev->real_num_tx_queues)) {
146 		do {
147 			txq -= dev->real_num_tx_queues;
148 		} while (txq >= dev->real_num_tx_queues);
149 	}
150 
151 	return txq;
152 }
153 
154 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
155  * that some drivers can provide 32bit values only.
156  */
157 static void net_failover_fold_stats(struct rtnl_link_stats64 *_res,
158 				    const struct rtnl_link_stats64 *_new,
159 				    const struct rtnl_link_stats64 *_old)
160 {
161 	const u64 *new = (const u64 *)_new;
162 	const u64 *old = (const u64 *)_old;
163 	u64 *res = (u64 *)_res;
164 	int i;
165 
166 	for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
167 		u64 nv = new[i];
168 		u64 ov = old[i];
169 		s64 delta = nv - ov;
170 
171 		/* detects if this particular field is 32bit only */
172 		if (((nv | ov) >> 32) == 0)
173 			delta = (s64)(s32)((u32)nv - (u32)ov);
174 
175 		/* filter anomalies, some drivers reset their stats
176 		 * at down/up events.
177 		 */
178 		if (delta > 0)
179 			res[i] += delta;
180 	}
181 }
182 
183 static void net_failover_get_stats(struct net_device *dev,
184 				   struct rtnl_link_stats64 *stats)
185 {
186 	struct net_failover_info *nfo_info = netdev_priv(dev);
187 	const struct rtnl_link_stats64 *new;
188 	struct rtnl_link_stats64 temp;
189 	struct net_device *slave_dev;
190 
191 	spin_lock(&nfo_info->stats_lock);
192 	memcpy(stats, &nfo_info->failover_stats, sizeof(*stats));
193 
194 	rcu_read_lock();
195 
196 	slave_dev = rcu_dereference(nfo_info->primary_dev);
197 	if (slave_dev) {
198 		new = dev_get_stats(slave_dev, &temp);
199 		net_failover_fold_stats(stats, new, &nfo_info->primary_stats);
200 		memcpy(&nfo_info->primary_stats, new, sizeof(*new));
201 	}
202 
203 	slave_dev = rcu_dereference(nfo_info->standby_dev);
204 	if (slave_dev) {
205 		new = dev_get_stats(slave_dev, &temp);
206 		net_failover_fold_stats(stats, new, &nfo_info->standby_stats);
207 		memcpy(&nfo_info->standby_stats, new, sizeof(*new));
208 	}
209 
210 	rcu_read_unlock();
211 
212 	memcpy(&nfo_info->failover_stats, stats, sizeof(*stats));
213 	spin_unlock(&nfo_info->stats_lock);
214 }
215 
216 static int net_failover_change_mtu(struct net_device *dev, int new_mtu)
217 {
218 	struct net_failover_info *nfo_info = netdev_priv(dev);
219 	struct net_device *primary_dev, *standby_dev;
220 	int ret = 0;
221 
222 	primary_dev = rcu_dereference(nfo_info->primary_dev);
223 	if (primary_dev) {
224 		ret = dev_set_mtu(primary_dev, new_mtu);
225 		if (ret)
226 			return ret;
227 	}
228 
229 	standby_dev = rcu_dereference(nfo_info->standby_dev);
230 	if (standby_dev) {
231 		ret = dev_set_mtu(standby_dev, new_mtu);
232 		if (ret) {
233 			if (primary_dev)
234 				dev_set_mtu(primary_dev, dev->mtu);
235 			return ret;
236 		}
237 	}
238 
239 	dev->mtu = new_mtu;
240 
241 	return 0;
242 }
243 
244 static void net_failover_set_rx_mode(struct net_device *dev)
245 {
246 	struct net_failover_info *nfo_info = netdev_priv(dev);
247 	struct net_device *slave_dev;
248 
249 	rcu_read_lock();
250 
251 	slave_dev = rcu_dereference(nfo_info->primary_dev);
252 	if (slave_dev) {
253 		dev_uc_sync_multiple(slave_dev, dev);
254 		dev_mc_sync_multiple(slave_dev, dev);
255 	}
256 
257 	slave_dev = rcu_dereference(nfo_info->standby_dev);
258 	if (slave_dev) {
259 		dev_uc_sync_multiple(slave_dev, dev);
260 		dev_mc_sync_multiple(slave_dev, dev);
261 	}
262 
263 	rcu_read_unlock();
264 }
265 
266 static int net_failover_vlan_rx_add_vid(struct net_device *dev, __be16 proto,
267 					u16 vid)
268 {
269 	struct net_failover_info *nfo_info = netdev_priv(dev);
270 	struct net_device *primary_dev, *standby_dev;
271 	int ret = 0;
272 
273 	primary_dev = rcu_dereference(nfo_info->primary_dev);
274 	if (primary_dev) {
275 		ret = vlan_vid_add(primary_dev, proto, vid);
276 		if (ret)
277 			return ret;
278 	}
279 
280 	standby_dev = rcu_dereference(nfo_info->standby_dev);
281 	if (standby_dev) {
282 		ret = vlan_vid_add(standby_dev, proto, vid);
283 		if (ret)
284 			if (primary_dev)
285 				vlan_vid_del(primary_dev, proto, vid);
286 	}
287 
288 	return ret;
289 }
290 
291 static int net_failover_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
292 					 u16 vid)
293 {
294 	struct net_failover_info *nfo_info = netdev_priv(dev);
295 	struct net_device *slave_dev;
296 
297 	slave_dev = rcu_dereference(nfo_info->primary_dev);
298 	if (slave_dev)
299 		vlan_vid_del(slave_dev, proto, vid);
300 
301 	slave_dev = rcu_dereference(nfo_info->standby_dev);
302 	if (slave_dev)
303 		vlan_vid_del(slave_dev, proto, vid);
304 
305 	return 0;
306 }
307 
308 static const struct net_device_ops failover_dev_ops = {
309 	.ndo_open		= net_failover_open,
310 	.ndo_stop		= net_failover_close,
311 	.ndo_start_xmit		= net_failover_start_xmit,
312 	.ndo_select_queue	= net_failover_select_queue,
313 	.ndo_get_stats64	= net_failover_get_stats,
314 	.ndo_change_mtu		= net_failover_change_mtu,
315 	.ndo_set_rx_mode	= net_failover_set_rx_mode,
316 	.ndo_vlan_rx_add_vid	= net_failover_vlan_rx_add_vid,
317 	.ndo_vlan_rx_kill_vid	= net_failover_vlan_rx_kill_vid,
318 	.ndo_validate_addr	= eth_validate_addr,
319 	.ndo_features_check	= passthru_features_check,
320 };
321 
322 #define FAILOVER_NAME "net_failover"
323 #define FAILOVER_VERSION "0.1"
324 
325 static void nfo_ethtool_get_drvinfo(struct net_device *dev,
326 				    struct ethtool_drvinfo *drvinfo)
327 {
328 	strlcpy(drvinfo->driver, FAILOVER_NAME, sizeof(drvinfo->driver));
329 	strlcpy(drvinfo->version, FAILOVER_VERSION, sizeof(drvinfo->version));
330 }
331 
332 static int nfo_ethtool_get_link_ksettings(struct net_device *dev,
333 					  struct ethtool_link_ksettings *cmd)
334 {
335 	struct net_failover_info *nfo_info = netdev_priv(dev);
336 	struct net_device *slave_dev;
337 
338 	slave_dev = rtnl_dereference(nfo_info->primary_dev);
339 	if (!slave_dev || !net_failover_xmit_ready(slave_dev)) {
340 		slave_dev = rtnl_dereference(nfo_info->standby_dev);
341 		if (!slave_dev || !net_failover_xmit_ready(slave_dev)) {
342 			cmd->base.duplex = DUPLEX_UNKNOWN;
343 			cmd->base.port = PORT_OTHER;
344 			cmd->base.speed = SPEED_UNKNOWN;
345 
346 			return 0;
347 		}
348 	}
349 
350 	return __ethtool_get_link_ksettings(slave_dev, cmd);
351 }
352 
353 static const struct ethtool_ops failover_ethtool_ops = {
354 	.get_drvinfo            = nfo_ethtool_get_drvinfo,
355 	.get_link               = ethtool_op_get_link,
356 	.get_link_ksettings     = nfo_ethtool_get_link_ksettings,
357 };
358 
359 /* Called when slave dev is injecting data into network stack.
360  * Change the associated network device from lower dev to failover dev.
361  * note: already called with rcu_read_lock
362  */
363 static rx_handler_result_t net_failover_handle_frame(struct sk_buff **pskb)
364 {
365 	struct sk_buff *skb = *pskb;
366 	struct net_device *dev = rcu_dereference(skb->dev->rx_handler_data);
367 	struct net_failover_info *nfo_info = netdev_priv(dev);
368 	struct net_device *primary_dev, *standby_dev;
369 
370 	primary_dev = rcu_dereference(nfo_info->primary_dev);
371 	standby_dev = rcu_dereference(nfo_info->standby_dev);
372 
373 	if (primary_dev && skb->dev == standby_dev)
374 		return RX_HANDLER_EXACT;
375 
376 	skb->dev = dev;
377 
378 	return RX_HANDLER_ANOTHER;
379 }
380 
381 static void net_failover_compute_features(struct net_device *dev)
382 {
383 	u32 vlan_features = FAILOVER_VLAN_FEATURES & NETIF_F_ALL_FOR_ALL;
384 	netdev_features_t enc_features  = FAILOVER_ENC_FEATURES;
385 	unsigned short max_hard_header_len = ETH_HLEN;
386 	unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
387 					IFF_XMIT_DST_RELEASE_PERM;
388 	struct net_failover_info *nfo_info = netdev_priv(dev);
389 	struct net_device *primary_dev, *standby_dev;
390 
391 	primary_dev = rcu_dereference(nfo_info->primary_dev);
392 	if (primary_dev) {
393 		vlan_features =
394 			netdev_increment_features(vlan_features,
395 						  primary_dev->vlan_features,
396 						  FAILOVER_VLAN_FEATURES);
397 		enc_features =
398 			netdev_increment_features(enc_features,
399 						  primary_dev->hw_enc_features,
400 						  FAILOVER_ENC_FEATURES);
401 
402 		dst_release_flag &= primary_dev->priv_flags;
403 		if (primary_dev->hard_header_len > max_hard_header_len)
404 			max_hard_header_len = primary_dev->hard_header_len;
405 	}
406 
407 	standby_dev = rcu_dereference(nfo_info->standby_dev);
408 	if (standby_dev) {
409 		vlan_features =
410 			netdev_increment_features(vlan_features,
411 						  standby_dev->vlan_features,
412 						  FAILOVER_VLAN_FEATURES);
413 		enc_features =
414 			netdev_increment_features(enc_features,
415 						  standby_dev->hw_enc_features,
416 						  FAILOVER_ENC_FEATURES);
417 
418 		dst_release_flag &= standby_dev->priv_flags;
419 		if (standby_dev->hard_header_len > max_hard_header_len)
420 			max_hard_header_len = standby_dev->hard_header_len;
421 	}
422 
423 	dev->vlan_features = vlan_features;
424 	dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
425 	dev->hard_header_len = max_hard_header_len;
426 
427 	dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
428 	if (dst_release_flag == (IFF_XMIT_DST_RELEASE |
429 				 IFF_XMIT_DST_RELEASE_PERM))
430 		dev->priv_flags |= IFF_XMIT_DST_RELEASE;
431 
432 	netdev_change_features(dev);
433 }
434 
435 static void net_failover_lower_state_changed(struct net_device *slave_dev,
436 					     struct net_device *primary_dev,
437 					     struct net_device *standby_dev)
438 {
439 	struct netdev_lag_lower_state_info info;
440 
441 	if (netif_carrier_ok(slave_dev))
442 		info.link_up = true;
443 	else
444 		info.link_up = false;
445 
446 	if (slave_dev == primary_dev) {
447 		if (netif_running(primary_dev))
448 			info.tx_enabled = true;
449 		else
450 			info.tx_enabled = false;
451 	} else {
452 		if ((primary_dev && netif_running(primary_dev)) ||
453 		    (!netif_running(standby_dev)))
454 			info.tx_enabled = false;
455 		else
456 			info.tx_enabled = true;
457 	}
458 
459 	netdev_lower_state_changed(slave_dev, &info);
460 }
461 
462 static int net_failover_slave_pre_register(struct net_device *slave_dev,
463 					   struct net_device *failover_dev)
464 {
465 	struct net_device *standby_dev, *primary_dev;
466 	struct net_failover_info *nfo_info;
467 	bool slave_is_standby;
468 
469 	nfo_info = netdev_priv(failover_dev);
470 	standby_dev = rtnl_dereference(nfo_info->standby_dev);
471 	primary_dev = rtnl_dereference(nfo_info->primary_dev);
472 	slave_is_standby = slave_dev->dev.parent == failover_dev->dev.parent;
473 	if (slave_is_standby ? standby_dev : primary_dev) {
474 		netdev_err(failover_dev, "%s attempting to register as slave dev when %s already present\n",
475 			   slave_dev->name,
476 			   slave_is_standby ? "standby" : "primary");
477 		return -EINVAL;
478 	}
479 
480 	/* We want to allow only a direct attached VF device as a primary
481 	 * netdev. As there is no easy way to check for a VF device, restrict
482 	 * this to a pci device.
483 	 */
484 	if (!slave_is_standby && (!slave_dev->dev.parent ||
485 				  !dev_is_pci(slave_dev->dev.parent)))
486 		return -EINVAL;
487 
488 	if (failover_dev->features & NETIF_F_VLAN_CHALLENGED &&
489 	    vlan_uses_dev(failover_dev)) {
490 		netdev_err(failover_dev, "Device %s is VLAN challenged and failover device has VLAN set up\n",
491 			   failover_dev->name);
492 		return -EINVAL;
493 	}
494 
495 	return 0;
496 }
497 
498 static int net_failover_slave_register(struct net_device *slave_dev,
499 				       struct net_device *failover_dev)
500 {
501 	struct net_device *standby_dev, *primary_dev;
502 	struct net_failover_info *nfo_info;
503 	bool slave_is_standby;
504 	u32 orig_mtu;
505 	int err;
506 
507 	/* Align MTU of slave with failover dev */
508 	orig_mtu = slave_dev->mtu;
509 	err = dev_set_mtu(slave_dev, failover_dev->mtu);
510 	if (err) {
511 		netdev_err(failover_dev, "unable to change mtu of %s to %u register failed\n",
512 			   slave_dev->name, failover_dev->mtu);
513 		goto done;
514 	}
515 
516 	dev_hold(slave_dev);
517 
518 	if (netif_running(failover_dev)) {
519 		err = dev_open(slave_dev);
520 		if (err && (err != -EBUSY)) {
521 			netdev_err(failover_dev, "Opening slave %s failed err:%d\n",
522 				   slave_dev->name, err);
523 			goto err_dev_open;
524 		}
525 	}
526 
527 	netif_addr_lock_bh(failover_dev);
528 	dev_uc_sync_multiple(slave_dev, failover_dev);
529 	dev_uc_sync_multiple(slave_dev, failover_dev);
530 	netif_addr_unlock_bh(failover_dev);
531 
532 	err = vlan_vids_add_by_dev(slave_dev, failover_dev);
533 	if (err) {
534 		netdev_err(failover_dev, "Failed to add vlan ids to device %s err:%d\n",
535 			   slave_dev->name, err);
536 		goto err_vlan_add;
537 	}
538 
539 	nfo_info = netdev_priv(failover_dev);
540 	standby_dev = rtnl_dereference(nfo_info->standby_dev);
541 	primary_dev = rtnl_dereference(nfo_info->primary_dev);
542 	slave_is_standby = slave_dev->dev.parent == failover_dev->dev.parent;
543 
544 	if (slave_is_standby) {
545 		rcu_assign_pointer(nfo_info->standby_dev, slave_dev);
546 		standby_dev = slave_dev;
547 		dev_get_stats(standby_dev, &nfo_info->standby_stats);
548 	} else {
549 		rcu_assign_pointer(nfo_info->primary_dev, slave_dev);
550 		primary_dev = slave_dev;
551 		dev_get_stats(primary_dev, &nfo_info->primary_stats);
552 		failover_dev->min_mtu = slave_dev->min_mtu;
553 		failover_dev->max_mtu = slave_dev->max_mtu;
554 	}
555 
556 	net_failover_lower_state_changed(slave_dev, primary_dev, standby_dev);
557 	net_failover_compute_features(failover_dev);
558 
559 	call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
560 
561 	netdev_info(failover_dev, "failover %s slave:%s registered\n",
562 		    slave_is_standby ? "standby" : "primary", slave_dev->name);
563 
564 	return 0;
565 
566 err_vlan_add:
567 	dev_uc_unsync(slave_dev, failover_dev);
568 	dev_mc_unsync(slave_dev, failover_dev);
569 	dev_close(slave_dev);
570 err_dev_open:
571 	dev_put(slave_dev);
572 	dev_set_mtu(slave_dev, orig_mtu);
573 done:
574 	return err;
575 }
576 
577 static int net_failover_slave_pre_unregister(struct net_device *slave_dev,
578 					     struct net_device *failover_dev)
579 {
580 	struct net_device *standby_dev, *primary_dev;
581 	struct net_failover_info *nfo_info;
582 
583 	nfo_info = netdev_priv(failover_dev);
584 	primary_dev = rtnl_dereference(nfo_info->primary_dev);
585 	standby_dev = rtnl_dereference(nfo_info->standby_dev);
586 
587 	if (slave_dev != primary_dev && slave_dev != standby_dev)
588 		return -ENODEV;
589 
590 	return 0;
591 }
592 
593 static int net_failover_slave_unregister(struct net_device *slave_dev,
594 					 struct net_device *failover_dev)
595 {
596 	struct net_device *standby_dev, *primary_dev;
597 	struct net_failover_info *nfo_info;
598 	bool slave_is_standby;
599 
600 	nfo_info = netdev_priv(failover_dev);
601 	primary_dev = rtnl_dereference(nfo_info->primary_dev);
602 	standby_dev = rtnl_dereference(nfo_info->standby_dev);
603 
604 	vlan_vids_del_by_dev(slave_dev, failover_dev);
605 	dev_uc_unsync(slave_dev, failover_dev);
606 	dev_mc_unsync(slave_dev, failover_dev);
607 	dev_close(slave_dev);
608 
609 	nfo_info = netdev_priv(failover_dev);
610 	dev_get_stats(failover_dev, &nfo_info->failover_stats);
611 
612 	slave_is_standby = slave_dev->dev.parent == failover_dev->dev.parent;
613 	if (slave_is_standby) {
614 		RCU_INIT_POINTER(nfo_info->standby_dev, NULL);
615 	} else {
616 		RCU_INIT_POINTER(nfo_info->primary_dev, NULL);
617 		if (standby_dev) {
618 			failover_dev->min_mtu = standby_dev->min_mtu;
619 			failover_dev->max_mtu = standby_dev->max_mtu;
620 		}
621 	}
622 
623 	dev_put(slave_dev);
624 
625 	net_failover_compute_features(failover_dev);
626 
627 	netdev_info(failover_dev, "failover %s slave:%s unregistered\n",
628 		    slave_is_standby ? "standby" : "primary", slave_dev->name);
629 
630 	return 0;
631 }
632 
633 static int net_failover_slave_link_change(struct net_device *slave_dev,
634 					  struct net_device *failover_dev)
635 {
636 	struct net_device *primary_dev, *standby_dev;
637 	struct net_failover_info *nfo_info;
638 
639 	nfo_info = netdev_priv(failover_dev);
640 
641 	primary_dev = rtnl_dereference(nfo_info->primary_dev);
642 	standby_dev = rtnl_dereference(nfo_info->standby_dev);
643 
644 	if (slave_dev != primary_dev && slave_dev != standby_dev)
645 		return -ENODEV;
646 
647 	if ((primary_dev && net_failover_xmit_ready(primary_dev)) ||
648 	    (standby_dev && net_failover_xmit_ready(standby_dev))) {
649 		netif_carrier_on(failover_dev);
650 		netif_tx_wake_all_queues(failover_dev);
651 	} else {
652 		dev_get_stats(failover_dev, &nfo_info->failover_stats);
653 		netif_carrier_off(failover_dev);
654 		netif_tx_stop_all_queues(failover_dev);
655 	}
656 
657 	net_failover_lower_state_changed(slave_dev, primary_dev, standby_dev);
658 
659 	return 0;
660 }
661 
662 static int net_failover_slave_name_change(struct net_device *slave_dev,
663 					  struct net_device *failover_dev)
664 {
665 	struct net_device *primary_dev, *standby_dev;
666 	struct net_failover_info *nfo_info;
667 
668 	nfo_info = netdev_priv(failover_dev);
669 
670 	primary_dev = rtnl_dereference(nfo_info->primary_dev);
671 	standby_dev = rtnl_dereference(nfo_info->standby_dev);
672 
673 	if (slave_dev != primary_dev && slave_dev != standby_dev)
674 		return -ENODEV;
675 
676 	/* We need to bring up the slave after the rename by udev in case
677 	 * open failed with EBUSY when it was registered.
678 	 */
679 	dev_open(slave_dev);
680 
681 	return 0;
682 }
683 
684 static struct failover_ops net_failover_ops = {
685 	.slave_pre_register	= net_failover_slave_pre_register,
686 	.slave_register		= net_failover_slave_register,
687 	.slave_pre_unregister	= net_failover_slave_pre_unregister,
688 	.slave_unregister	= net_failover_slave_unregister,
689 	.slave_link_change	= net_failover_slave_link_change,
690 	.slave_name_change	= net_failover_slave_name_change,
691 	.slave_handle_frame	= net_failover_handle_frame,
692 };
693 
694 /**
695  * net_failover_create - Create and register a failover instance
696  *
697  * @dev: standby netdev
698  *
699  * Creates a failover netdev and registers a failover instance for a standby
700  * netdev. Used by paravirtual drivers that use 3-netdev model.
701  * The failover netdev acts as a master device and controls 2 slave devices -
702  * the original standby netdev and a VF netdev with the same MAC gets
703  * registered as primary netdev.
704  *
705  * Return: pointer to failover instance
706  */
707 struct failover *net_failover_create(struct net_device *standby_dev)
708 {
709 	struct device *dev = standby_dev->dev.parent;
710 	struct net_device *failover_dev;
711 	struct failover *failover;
712 	int err;
713 
714 	/* Alloc at least 2 queues, for now we are going with 16 assuming
715 	 * that VF devices being enslaved won't have too many queues.
716 	 */
717 	failover_dev = alloc_etherdev_mq(sizeof(struct net_failover_info), 16);
718 	if (!failover_dev) {
719 		dev_err(dev, "Unable to allocate failover_netdev!\n");
720 		return ERR_PTR(-ENOMEM);
721 	}
722 
723 	dev_net_set(failover_dev, dev_net(standby_dev));
724 	SET_NETDEV_DEV(failover_dev, dev);
725 
726 	failover_dev->netdev_ops = &failover_dev_ops;
727 	failover_dev->ethtool_ops = &failover_ethtool_ops;
728 
729 	/* Initialize the device options */
730 	failover_dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
731 	failover_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE |
732 				       IFF_TX_SKB_SHARING);
733 
734 	/* don't acquire failover netdev's netif_tx_lock when transmitting */
735 	failover_dev->features |= NETIF_F_LLTX;
736 
737 	/* Don't allow failover devices to change network namespaces. */
738 	failover_dev->features |= NETIF_F_NETNS_LOCAL;
739 
740 	failover_dev->hw_features = FAILOVER_VLAN_FEATURES |
741 				    NETIF_F_HW_VLAN_CTAG_TX |
742 				    NETIF_F_HW_VLAN_CTAG_RX |
743 				    NETIF_F_HW_VLAN_CTAG_FILTER;
744 
745 	failover_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
746 	failover_dev->features |= failover_dev->hw_features;
747 
748 	memcpy(failover_dev->dev_addr, standby_dev->dev_addr,
749 	       failover_dev->addr_len);
750 
751 	failover_dev->min_mtu = standby_dev->min_mtu;
752 	failover_dev->max_mtu = standby_dev->max_mtu;
753 
754 	err = register_netdev(failover_dev);
755 	if (err) {
756 		dev_err(dev, "Unable to register failover_dev!\n");
757 		goto err_register_netdev;
758 	}
759 
760 	netif_carrier_off(failover_dev);
761 
762 	failover = failover_register(failover_dev, &net_failover_ops);
763 	if (IS_ERR(failover))
764 		goto err_failover_register;
765 
766 	return failover;
767 
768 err_failover_register:
769 	unregister_netdev(failover_dev);
770 err_register_netdev:
771 	free_netdev(failover_dev);
772 
773 	return ERR_PTR(err);
774 }
775 EXPORT_SYMBOL_GPL(net_failover_create);
776 
777 /**
778  * net_failover_destroy - Destroy a failover instance
779  *
780  * @failover: pointer to failover instance
781  *
782  * Unregisters any slave netdevs associated with the failover instance by
783  * calling failover_slave_unregister().
784  * unregisters the failover instance itself and finally frees the failover
785  * netdev. Used by paravirtual drivers that use 3-netdev model.
786  *
787  */
788 void net_failover_destroy(struct failover *failover)
789 {
790 	struct net_failover_info *nfo_info;
791 	struct net_device *failover_dev;
792 	struct net_device *slave_dev;
793 
794 	if (!failover)
795 		return;
796 
797 	failover_dev = rcu_dereference(failover->failover_dev);
798 	nfo_info = netdev_priv(failover_dev);
799 
800 	netif_device_detach(failover_dev);
801 
802 	rtnl_lock();
803 
804 	slave_dev = rtnl_dereference(nfo_info->primary_dev);
805 	if (slave_dev)
806 		failover_slave_unregister(slave_dev);
807 
808 	slave_dev = rtnl_dereference(nfo_info->standby_dev);
809 	if (slave_dev)
810 		failover_slave_unregister(slave_dev);
811 
812 	failover_unregister(failover);
813 
814 	unregister_netdevice(failover_dev);
815 
816 	rtnl_unlock();
817 
818 	free_netdev(failover_dev);
819 }
820 EXPORT_SYMBOL_GPL(net_failover_destroy);
821 
822 static __init int
823 net_failover_init(void)
824 {
825 	return 0;
826 }
827 module_init(net_failover_init);
828 
829 static __exit
830 void net_failover_exit(void)
831 {
832 }
833 module_exit(net_failover_exit);
834 
835 MODULE_DESCRIPTION("Failover driver for Paravirtual drivers");
836 MODULE_LICENSE("GPL v2");
837