xref: /openbmc/linux/drivers/net/net_failover.c (revision 2e6ae11dd0d1c37f44cec51a58fb2092e55ed0f5)
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,
119 				     struct net_device *sb_dev,
120 				     select_queue_fallback_t fallback)
121 {
122 	struct net_failover_info *nfo_info = netdev_priv(dev);
123 	struct net_device *primary_dev;
124 	u16 txq;
125 
126 	primary_dev = rcu_dereference(nfo_info->primary_dev);
127 	if (primary_dev) {
128 		const struct net_device_ops *ops = primary_dev->netdev_ops;
129 
130 		if (ops->ndo_select_queue)
131 			txq = ops->ndo_select_queue(primary_dev, skb,
132 						    sb_dev, fallback);
133 		else
134 			txq = fallback(primary_dev, skb, NULL);
135 
136 		qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
137 
138 		return txq;
139 	}
140 
141 	txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
142 
143 	/* Save the original txq to restore before passing to the driver */
144 	qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
145 
146 	if (unlikely(txq >= dev->real_num_tx_queues)) {
147 		do {
148 			txq -= dev->real_num_tx_queues;
149 		} while (txq >= dev->real_num_tx_queues);
150 	}
151 
152 	return txq;
153 }
154 
155 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
156  * that some drivers can provide 32bit values only.
157  */
158 static void net_failover_fold_stats(struct rtnl_link_stats64 *_res,
159 				    const struct rtnl_link_stats64 *_new,
160 				    const struct rtnl_link_stats64 *_old)
161 {
162 	const u64 *new = (const u64 *)_new;
163 	const u64 *old = (const u64 *)_old;
164 	u64 *res = (u64 *)_res;
165 	int i;
166 
167 	for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
168 		u64 nv = new[i];
169 		u64 ov = old[i];
170 		s64 delta = nv - ov;
171 
172 		/* detects if this particular field is 32bit only */
173 		if (((nv | ov) >> 32) == 0)
174 			delta = (s64)(s32)((u32)nv - (u32)ov);
175 
176 		/* filter anomalies, some drivers reset their stats
177 		 * at down/up events.
178 		 */
179 		if (delta > 0)
180 			res[i] += delta;
181 	}
182 }
183 
184 static void net_failover_get_stats(struct net_device *dev,
185 				   struct rtnl_link_stats64 *stats)
186 {
187 	struct net_failover_info *nfo_info = netdev_priv(dev);
188 	const struct rtnl_link_stats64 *new;
189 	struct rtnl_link_stats64 temp;
190 	struct net_device *slave_dev;
191 
192 	spin_lock(&nfo_info->stats_lock);
193 	memcpy(stats, &nfo_info->failover_stats, sizeof(*stats));
194 
195 	rcu_read_lock();
196 
197 	slave_dev = rcu_dereference(nfo_info->primary_dev);
198 	if (slave_dev) {
199 		new = dev_get_stats(slave_dev, &temp);
200 		net_failover_fold_stats(stats, new, &nfo_info->primary_stats);
201 		memcpy(&nfo_info->primary_stats, new, sizeof(*new));
202 	}
203 
204 	slave_dev = rcu_dereference(nfo_info->standby_dev);
205 	if (slave_dev) {
206 		new = dev_get_stats(slave_dev, &temp);
207 		net_failover_fold_stats(stats, new, &nfo_info->standby_stats);
208 		memcpy(&nfo_info->standby_stats, new, sizeof(*new));
209 	}
210 
211 	rcu_read_unlock();
212 
213 	memcpy(&nfo_info->failover_stats, stats, sizeof(*stats));
214 	spin_unlock(&nfo_info->stats_lock);
215 }
216 
217 static int net_failover_change_mtu(struct net_device *dev, int new_mtu)
218 {
219 	struct net_failover_info *nfo_info = netdev_priv(dev);
220 	struct net_device *primary_dev, *standby_dev;
221 	int ret = 0;
222 
223 	primary_dev = rtnl_dereference(nfo_info->primary_dev);
224 	if (primary_dev) {
225 		ret = dev_set_mtu(primary_dev, new_mtu);
226 		if (ret)
227 			return ret;
228 	}
229 
230 	standby_dev = rtnl_dereference(nfo_info->standby_dev);
231 	if (standby_dev) {
232 		ret = dev_set_mtu(standby_dev, new_mtu);
233 		if (ret) {
234 			if (primary_dev)
235 				dev_set_mtu(primary_dev, dev->mtu);
236 			return ret;
237 		}
238 	}
239 
240 	dev->mtu = new_mtu;
241 
242 	return 0;
243 }
244 
245 static void net_failover_set_rx_mode(struct net_device *dev)
246 {
247 	struct net_failover_info *nfo_info = netdev_priv(dev);
248 	struct net_device *slave_dev;
249 
250 	rcu_read_lock();
251 
252 	slave_dev = rcu_dereference(nfo_info->primary_dev);
253 	if (slave_dev) {
254 		dev_uc_sync_multiple(slave_dev, dev);
255 		dev_mc_sync_multiple(slave_dev, dev);
256 	}
257 
258 	slave_dev = rcu_dereference(nfo_info->standby_dev);
259 	if (slave_dev) {
260 		dev_uc_sync_multiple(slave_dev, dev);
261 		dev_mc_sync_multiple(slave_dev, dev);
262 	}
263 
264 	rcu_read_unlock();
265 }
266 
267 static int net_failover_vlan_rx_add_vid(struct net_device *dev, __be16 proto,
268 					u16 vid)
269 {
270 	struct net_failover_info *nfo_info = netdev_priv(dev);
271 	struct net_device *primary_dev, *standby_dev;
272 	int ret = 0;
273 
274 	primary_dev = rcu_dereference(nfo_info->primary_dev);
275 	if (primary_dev) {
276 		ret = vlan_vid_add(primary_dev, proto, vid);
277 		if (ret)
278 			return ret;
279 	}
280 
281 	standby_dev = rcu_dereference(nfo_info->standby_dev);
282 	if (standby_dev) {
283 		ret = vlan_vid_add(standby_dev, proto, vid);
284 		if (ret)
285 			if (primary_dev)
286 				vlan_vid_del(primary_dev, proto, vid);
287 	}
288 
289 	return ret;
290 }
291 
292 static int net_failover_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,
293 					 u16 vid)
294 {
295 	struct net_failover_info *nfo_info = netdev_priv(dev);
296 	struct net_device *slave_dev;
297 
298 	slave_dev = rcu_dereference(nfo_info->primary_dev);
299 	if (slave_dev)
300 		vlan_vid_del(slave_dev, proto, vid);
301 
302 	slave_dev = rcu_dereference(nfo_info->standby_dev);
303 	if (slave_dev)
304 		vlan_vid_del(slave_dev, proto, vid);
305 
306 	return 0;
307 }
308 
309 static const struct net_device_ops failover_dev_ops = {
310 	.ndo_open		= net_failover_open,
311 	.ndo_stop		= net_failover_close,
312 	.ndo_start_xmit		= net_failover_start_xmit,
313 	.ndo_select_queue	= net_failover_select_queue,
314 	.ndo_get_stats64	= net_failover_get_stats,
315 	.ndo_change_mtu		= net_failover_change_mtu,
316 	.ndo_set_rx_mode	= net_failover_set_rx_mode,
317 	.ndo_vlan_rx_add_vid	= net_failover_vlan_rx_add_vid,
318 	.ndo_vlan_rx_kill_vid	= net_failover_vlan_rx_kill_vid,
319 	.ndo_validate_addr	= eth_validate_addr,
320 	.ndo_features_check	= passthru_features_check,
321 };
322 
323 #define FAILOVER_NAME "net_failover"
324 #define FAILOVER_VERSION "0.1"
325 
326 static void nfo_ethtool_get_drvinfo(struct net_device *dev,
327 				    struct ethtool_drvinfo *drvinfo)
328 {
329 	strlcpy(drvinfo->driver, FAILOVER_NAME, sizeof(drvinfo->driver));
330 	strlcpy(drvinfo->version, FAILOVER_VERSION, sizeof(drvinfo->version));
331 }
332 
333 static int nfo_ethtool_get_link_ksettings(struct net_device *dev,
334 					  struct ethtool_link_ksettings *cmd)
335 {
336 	struct net_failover_info *nfo_info = netdev_priv(dev);
337 	struct net_device *slave_dev;
338 
339 	slave_dev = rtnl_dereference(nfo_info->primary_dev);
340 	if (!slave_dev || !net_failover_xmit_ready(slave_dev)) {
341 		slave_dev = rtnl_dereference(nfo_info->standby_dev);
342 		if (!slave_dev || !net_failover_xmit_ready(slave_dev)) {
343 			cmd->base.duplex = DUPLEX_UNKNOWN;
344 			cmd->base.port = PORT_OTHER;
345 			cmd->base.speed = SPEED_UNKNOWN;
346 
347 			return 0;
348 		}
349 	}
350 
351 	return __ethtool_get_link_ksettings(slave_dev, cmd);
352 }
353 
354 static const struct ethtool_ops failover_ethtool_ops = {
355 	.get_drvinfo            = nfo_ethtool_get_drvinfo,
356 	.get_link               = ethtool_op_get_link,
357 	.get_link_ksettings     = nfo_ethtool_get_link_ksettings,
358 };
359 
360 /* Called when slave dev is injecting data into network stack.
361  * Change the associated network device from lower dev to failover dev.
362  * note: already called with rcu_read_lock
363  */
364 static rx_handler_result_t net_failover_handle_frame(struct sk_buff **pskb)
365 {
366 	struct sk_buff *skb = *pskb;
367 	struct net_device *dev = rcu_dereference(skb->dev->rx_handler_data);
368 	struct net_failover_info *nfo_info = netdev_priv(dev);
369 	struct net_device *primary_dev, *standby_dev;
370 
371 	primary_dev = rcu_dereference(nfo_info->primary_dev);
372 	standby_dev = rcu_dereference(nfo_info->standby_dev);
373 
374 	if (primary_dev && skb->dev == standby_dev)
375 		return RX_HANDLER_EXACT;
376 
377 	skb->dev = dev;
378 
379 	return RX_HANDLER_ANOTHER;
380 }
381 
382 static void net_failover_compute_features(struct net_device *dev)
383 {
384 	netdev_features_t vlan_features = FAILOVER_VLAN_FEATURES &
385 					  NETIF_F_ALL_FOR_ALL;
386 	netdev_features_t enc_features  = FAILOVER_ENC_FEATURES;
387 	unsigned short max_hard_header_len = ETH_HLEN;
388 	unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
389 					IFF_XMIT_DST_RELEASE_PERM;
390 	struct net_failover_info *nfo_info = netdev_priv(dev);
391 	struct net_device *primary_dev, *standby_dev;
392 
393 	primary_dev = rcu_dereference(nfo_info->primary_dev);
394 	if (primary_dev) {
395 		vlan_features =
396 			netdev_increment_features(vlan_features,
397 						  primary_dev->vlan_features,
398 						  FAILOVER_VLAN_FEATURES);
399 		enc_features =
400 			netdev_increment_features(enc_features,
401 						  primary_dev->hw_enc_features,
402 						  FAILOVER_ENC_FEATURES);
403 
404 		dst_release_flag &= primary_dev->priv_flags;
405 		if (primary_dev->hard_header_len > max_hard_header_len)
406 			max_hard_header_len = primary_dev->hard_header_len;
407 	}
408 
409 	standby_dev = rcu_dereference(nfo_info->standby_dev);
410 	if (standby_dev) {
411 		vlan_features =
412 			netdev_increment_features(vlan_features,
413 						  standby_dev->vlan_features,
414 						  FAILOVER_VLAN_FEATURES);
415 		enc_features =
416 			netdev_increment_features(enc_features,
417 						  standby_dev->hw_enc_features,
418 						  FAILOVER_ENC_FEATURES);
419 
420 		dst_release_flag &= standby_dev->priv_flags;
421 		if (standby_dev->hard_header_len > max_hard_header_len)
422 			max_hard_header_len = standby_dev->hard_header_len;
423 	}
424 
425 	dev->vlan_features = vlan_features;
426 	dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
427 	dev->hard_header_len = max_hard_header_len;
428 
429 	dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
430 	if (dst_release_flag == (IFF_XMIT_DST_RELEASE |
431 				 IFF_XMIT_DST_RELEASE_PERM))
432 		dev->priv_flags |= IFF_XMIT_DST_RELEASE;
433 
434 	netdev_change_features(dev);
435 }
436 
437 static void net_failover_lower_state_changed(struct net_device *slave_dev,
438 					     struct net_device *primary_dev,
439 					     struct net_device *standby_dev)
440 {
441 	struct netdev_lag_lower_state_info info;
442 
443 	if (netif_carrier_ok(slave_dev))
444 		info.link_up = true;
445 	else
446 		info.link_up = false;
447 
448 	if (slave_dev == primary_dev) {
449 		if (netif_running(primary_dev))
450 			info.tx_enabled = true;
451 		else
452 			info.tx_enabled = false;
453 	} else {
454 		if ((primary_dev && netif_running(primary_dev)) ||
455 		    (!netif_running(standby_dev)))
456 			info.tx_enabled = false;
457 		else
458 			info.tx_enabled = true;
459 	}
460 
461 	netdev_lower_state_changed(slave_dev, &info);
462 }
463 
464 static int net_failover_slave_pre_register(struct net_device *slave_dev,
465 					   struct net_device *failover_dev)
466 {
467 	struct net_device *standby_dev, *primary_dev;
468 	struct net_failover_info *nfo_info;
469 	bool slave_is_standby;
470 
471 	nfo_info = netdev_priv(failover_dev);
472 	standby_dev = rtnl_dereference(nfo_info->standby_dev);
473 	primary_dev = rtnl_dereference(nfo_info->primary_dev);
474 	slave_is_standby = slave_dev->dev.parent == failover_dev->dev.parent;
475 	if (slave_is_standby ? standby_dev : primary_dev) {
476 		netdev_err(failover_dev, "%s attempting to register as slave dev when %s already present\n",
477 			   slave_dev->name,
478 			   slave_is_standby ? "standby" : "primary");
479 		return -EINVAL;
480 	}
481 
482 	/* We want to allow only a direct attached VF device as a primary
483 	 * netdev. As there is no easy way to check for a VF device, restrict
484 	 * this to a pci device.
485 	 */
486 	if (!slave_is_standby && (!slave_dev->dev.parent ||
487 				  !dev_is_pci(slave_dev->dev.parent)))
488 		return -EINVAL;
489 
490 	if (failover_dev->features & NETIF_F_VLAN_CHALLENGED &&
491 	    vlan_uses_dev(failover_dev)) {
492 		netdev_err(failover_dev, "Device %s is VLAN challenged and failover device has VLAN set up\n",
493 			   failover_dev->name);
494 		return -EINVAL;
495 	}
496 
497 	return 0;
498 }
499 
500 static int net_failover_slave_register(struct net_device *slave_dev,
501 				       struct net_device *failover_dev)
502 {
503 	struct net_device *standby_dev, *primary_dev;
504 	struct net_failover_info *nfo_info;
505 	bool slave_is_standby;
506 	u32 orig_mtu;
507 	int err;
508 
509 	/* Align MTU of slave with failover dev */
510 	orig_mtu = slave_dev->mtu;
511 	err = dev_set_mtu(slave_dev, failover_dev->mtu);
512 	if (err) {
513 		netdev_err(failover_dev, "unable to change mtu of %s to %u register failed\n",
514 			   slave_dev->name, failover_dev->mtu);
515 		goto done;
516 	}
517 
518 	dev_hold(slave_dev);
519 
520 	if (netif_running(failover_dev)) {
521 		err = dev_open(slave_dev);
522 		if (err && (err != -EBUSY)) {
523 			netdev_err(failover_dev, "Opening slave %s failed err:%d\n",
524 				   slave_dev->name, err);
525 			goto err_dev_open;
526 		}
527 	}
528 
529 	netif_addr_lock_bh(failover_dev);
530 	dev_uc_sync_multiple(slave_dev, failover_dev);
531 	dev_mc_sync_multiple(slave_dev, failover_dev);
532 	netif_addr_unlock_bh(failover_dev);
533 
534 	err = vlan_vids_add_by_dev(slave_dev, failover_dev);
535 	if (err) {
536 		netdev_err(failover_dev, "Failed to add vlan ids to device %s err:%d\n",
537 			   slave_dev->name, err);
538 		goto err_vlan_add;
539 	}
540 
541 	nfo_info = netdev_priv(failover_dev);
542 	standby_dev = rtnl_dereference(nfo_info->standby_dev);
543 	primary_dev = rtnl_dereference(nfo_info->primary_dev);
544 	slave_is_standby = slave_dev->dev.parent == failover_dev->dev.parent;
545 
546 	if (slave_is_standby) {
547 		rcu_assign_pointer(nfo_info->standby_dev, slave_dev);
548 		standby_dev = slave_dev;
549 		dev_get_stats(standby_dev, &nfo_info->standby_stats);
550 	} else {
551 		rcu_assign_pointer(nfo_info->primary_dev, slave_dev);
552 		primary_dev = slave_dev;
553 		dev_get_stats(primary_dev, &nfo_info->primary_stats);
554 		failover_dev->min_mtu = slave_dev->min_mtu;
555 		failover_dev->max_mtu = slave_dev->max_mtu;
556 	}
557 
558 	net_failover_lower_state_changed(slave_dev, primary_dev, standby_dev);
559 	net_failover_compute_features(failover_dev);
560 
561 	call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
562 
563 	netdev_info(failover_dev, "failover %s slave:%s registered\n",
564 		    slave_is_standby ? "standby" : "primary", slave_dev->name);
565 
566 	return 0;
567 
568 err_vlan_add:
569 	dev_uc_unsync(slave_dev, failover_dev);
570 	dev_mc_unsync(slave_dev, failover_dev);
571 	dev_close(slave_dev);
572 err_dev_open:
573 	dev_put(slave_dev);
574 	dev_set_mtu(slave_dev, orig_mtu);
575 done:
576 	return err;
577 }
578 
579 static int net_failover_slave_pre_unregister(struct net_device *slave_dev,
580 					     struct net_device *failover_dev)
581 {
582 	struct net_device *standby_dev, *primary_dev;
583 	struct net_failover_info *nfo_info;
584 
585 	nfo_info = netdev_priv(failover_dev);
586 	primary_dev = rtnl_dereference(nfo_info->primary_dev);
587 	standby_dev = rtnl_dereference(nfo_info->standby_dev);
588 
589 	if (slave_dev != primary_dev && slave_dev != standby_dev)
590 		return -ENODEV;
591 
592 	return 0;
593 }
594 
595 static int net_failover_slave_unregister(struct net_device *slave_dev,
596 					 struct net_device *failover_dev)
597 {
598 	struct net_device *standby_dev, *primary_dev;
599 	struct net_failover_info *nfo_info;
600 	bool slave_is_standby;
601 
602 	nfo_info = netdev_priv(failover_dev);
603 	primary_dev = rtnl_dereference(nfo_info->primary_dev);
604 	standby_dev = rtnl_dereference(nfo_info->standby_dev);
605 
606 	vlan_vids_del_by_dev(slave_dev, failover_dev);
607 	dev_uc_unsync(slave_dev, failover_dev);
608 	dev_mc_unsync(slave_dev, failover_dev);
609 	dev_close(slave_dev);
610 
611 	nfo_info = netdev_priv(failover_dev);
612 	dev_get_stats(failover_dev, &nfo_info->failover_stats);
613 
614 	slave_is_standby = slave_dev->dev.parent == failover_dev->dev.parent;
615 	if (slave_is_standby) {
616 		RCU_INIT_POINTER(nfo_info->standby_dev, NULL);
617 	} else {
618 		RCU_INIT_POINTER(nfo_info->primary_dev, NULL);
619 		if (standby_dev) {
620 			failover_dev->min_mtu = standby_dev->min_mtu;
621 			failover_dev->max_mtu = standby_dev->max_mtu;
622 		}
623 	}
624 
625 	dev_put(slave_dev);
626 
627 	net_failover_compute_features(failover_dev);
628 
629 	netdev_info(failover_dev, "failover %s slave:%s unregistered\n",
630 		    slave_is_standby ? "standby" : "primary", slave_dev->name);
631 
632 	return 0;
633 }
634 
635 static int net_failover_slave_link_change(struct net_device *slave_dev,
636 					  struct net_device *failover_dev)
637 {
638 	struct net_device *primary_dev, *standby_dev;
639 	struct net_failover_info *nfo_info;
640 
641 	nfo_info = netdev_priv(failover_dev);
642 
643 	primary_dev = rtnl_dereference(nfo_info->primary_dev);
644 	standby_dev = rtnl_dereference(nfo_info->standby_dev);
645 
646 	if (slave_dev != primary_dev && slave_dev != standby_dev)
647 		return -ENODEV;
648 
649 	if ((primary_dev && net_failover_xmit_ready(primary_dev)) ||
650 	    (standby_dev && net_failover_xmit_ready(standby_dev))) {
651 		netif_carrier_on(failover_dev);
652 		netif_tx_wake_all_queues(failover_dev);
653 	} else {
654 		dev_get_stats(failover_dev, &nfo_info->failover_stats);
655 		netif_carrier_off(failover_dev);
656 		netif_tx_stop_all_queues(failover_dev);
657 	}
658 
659 	net_failover_lower_state_changed(slave_dev, primary_dev, standby_dev);
660 
661 	return 0;
662 }
663 
664 static int net_failover_slave_name_change(struct net_device *slave_dev,
665 					  struct net_device *failover_dev)
666 {
667 	struct net_device *primary_dev, *standby_dev;
668 	struct net_failover_info *nfo_info;
669 
670 	nfo_info = netdev_priv(failover_dev);
671 
672 	primary_dev = rtnl_dereference(nfo_info->primary_dev);
673 	standby_dev = rtnl_dereference(nfo_info->standby_dev);
674 
675 	if (slave_dev != primary_dev && slave_dev != standby_dev)
676 		return -ENODEV;
677 
678 	/* We need to bring up the slave after the rename by udev in case
679 	 * open failed with EBUSY when it was registered.
680 	 */
681 	dev_open(slave_dev);
682 
683 	return 0;
684 }
685 
686 static struct failover_ops net_failover_ops = {
687 	.slave_pre_register	= net_failover_slave_pre_register,
688 	.slave_register		= net_failover_slave_register,
689 	.slave_pre_unregister	= net_failover_slave_pre_unregister,
690 	.slave_unregister	= net_failover_slave_unregister,
691 	.slave_link_change	= net_failover_slave_link_change,
692 	.slave_name_change	= net_failover_slave_name_change,
693 	.slave_handle_frame	= net_failover_handle_frame,
694 };
695 
696 /**
697  * net_failover_create - Create and register a failover instance
698  *
699  * @dev: standby netdev
700  *
701  * Creates a failover netdev and registers a failover instance for a standby
702  * netdev. Used by paravirtual drivers that use 3-netdev model.
703  * The failover netdev acts as a master device and controls 2 slave devices -
704  * the original standby netdev and a VF netdev with the same MAC gets
705  * registered as primary netdev.
706  *
707  * Return: pointer to failover instance
708  */
709 struct failover *net_failover_create(struct net_device *standby_dev)
710 {
711 	struct device *dev = standby_dev->dev.parent;
712 	struct net_device *failover_dev;
713 	struct failover *failover;
714 	int err;
715 
716 	/* Alloc at least 2 queues, for now we are going with 16 assuming
717 	 * that VF devices being enslaved won't have too many queues.
718 	 */
719 	failover_dev = alloc_etherdev_mq(sizeof(struct net_failover_info), 16);
720 	if (!failover_dev) {
721 		dev_err(dev, "Unable to allocate failover_netdev!\n");
722 		return ERR_PTR(-ENOMEM);
723 	}
724 
725 	dev_net_set(failover_dev, dev_net(standby_dev));
726 	SET_NETDEV_DEV(failover_dev, dev);
727 
728 	failover_dev->netdev_ops = &failover_dev_ops;
729 	failover_dev->ethtool_ops = &failover_ethtool_ops;
730 
731 	/* Initialize the device options */
732 	failover_dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE;
733 	failover_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE |
734 				       IFF_TX_SKB_SHARING);
735 
736 	/* don't acquire failover netdev's netif_tx_lock when transmitting */
737 	failover_dev->features |= NETIF_F_LLTX;
738 
739 	/* Don't allow failover devices to change network namespaces. */
740 	failover_dev->features |= NETIF_F_NETNS_LOCAL;
741 
742 	failover_dev->hw_features = FAILOVER_VLAN_FEATURES |
743 				    NETIF_F_HW_VLAN_CTAG_TX |
744 				    NETIF_F_HW_VLAN_CTAG_RX |
745 				    NETIF_F_HW_VLAN_CTAG_FILTER;
746 
747 	failover_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
748 	failover_dev->features |= failover_dev->hw_features;
749 
750 	memcpy(failover_dev->dev_addr, standby_dev->dev_addr,
751 	       failover_dev->addr_len);
752 
753 	failover_dev->min_mtu = standby_dev->min_mtu;
754 	failover_dev->max_mtu = standby_dev->max_mtu;
755 
756 	err = register_netdev(failover_dev);
757 	if (err) {
758 		dev_err(dev, "Unable to register failover_dev!\n");
759 		goto err_register_netdev;
760 	}
761 
762 	netif_carrier_off(failover_dev);
763 
764 	failover = failover_register(failover_dev, &net_failover_ops);
765 	if (IS_ERR(failover))
766 		goto err_failover_register;
767 
768 	return failover;
769 
770 err_failover_register:
771 	unregister_netdev(failover_dev);
772 err_register_netdev:
773 	free_netdev(failover_dev);
774 
775 	return ERR_PTR(err);
776 }
777 EXPORT_SYMBOL_GPL(net_failover_create);
778 
779 /**
780  * net_failover_destroy - Destroy a failover instance
781  *
782  * @failover: pointer to failover instance
783  *
784  * Unregisters any slave netdevs associated with the failover instance by
785  * calling failover_slave_unregister().
786  * unregisters the failover instance itself and finally frees the failover
787  * netdev. Used by paravirtual drivers that use 3-netdev model.
788  *
789  */
790 void net_failover_destroy(struct failover *failover)
791 {
792 	struct net_failover_info *nfo_info;
793 	struct net_device *failover_dev;
794 	struct net_device *slave_dev;
795 
796 	if (!failover)
797 		return;
798 
799 	failover_dev = rcu_dereference(failover->failover_dev);
800 	nfo_info = netdev_priv(failover_dev);
801 
802 	netif_device_detach(failover_dev);
803 
804 	rtnl_lock();
805 
806 	slave_dev = rtnl_dereference(nfo_info->primary_dev);
807 	if (slave_dev)
808 		failover_slave_unregister(slave_dev);
809 
810 	slave_dev = rtnl_dereference(nfo_info->standby_dev);
811 	if (slave_dev)
812 		failover_slave_unregister(slave_dev);
813 
814 	failover_unregister(failover);
815 
816 	unregister_netdevice(failover_dev);
817 
818 	rtnl_unlock();
819 
820 	free_netdev(failover_dev);
821 }
822 EXPORT_SYMBOL_GPL(net_failover_destroy);
823 
824 static __init int
825 net_failover_init(void)
826 {
827 	return 0;
828 }
829 module_init(net_failover_init);
830 
831 static __exit
832 void net_failover_exit(void)
833 {
834 }
835 module_exit(net_failover_exit);
836 
837 MODULE_DESCRIPTION("Failover driver for Paravirtual drivers");
838 MODULE_LICENSE("GPL v2");
839