xref: /linux/net/switchdev/switchdev.c (revision 2a10154abcb75ad0d7b6bfea6210ac743ec60897)
1 /*
2  * net/switchdev/switchdev.c - Switch device API
3  * Copyright (c) 2014 Jiri Pirko <jiri@resnulli.us>
4  * Copyright (c) 2014-2015 Scott Feldman <sfeldma@gmail.com>
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License as published by
8  * the Free Software Foundation; either version 2 of the License, or
9  * (at your option) any later version.
10  */
11 
12 #include <linux/kernel.h>
13 #include <linux/types.h>
14 #include <linux/init.h>
15 #include <linux/mutex.h>
16 #include <linux/notifier.h>
17 #include <linux/netdevice.h>
18 #include <linux/if_bridge.h>
19 #include <net/ip_fib.h>
20 #include <net/switchdev.h>
21 
22 /**
23  *	switchdev_port_attr_get - Get port attribute
24  *
25  *	@dev: port device
26  *	@attr: attribute to get
27  */
28 int switchdev_port_attr_get(struct net_device *dev, struct switchdev_attr *attr)
29 {
30 	const struct switchdev_ops *ops = dev->switchdev_ops;
31 	struct net_device *lower_dev;
32 	struct list_head *iter;
33 	struct switchdev_attr first = {
34 		.id = SWITCHDEV_ATTR_UNDEFINED
35 	};
36 	int err = -EOPNOTSUPP;
37 
38 	if (ops && ops->switchdev_port_attr_get)
39 		return ops->switchdev_port_attr_get(dev, attr);
40 
41 	if (attr->flags & SWITCHDEV_F_NO_RECURSE)
42 		return err;
43 
44 	/* Switch device port(s) may be stacked under
45 	 * bond/team/vlan dev, so recurse down to get attr on
46 	 * each port.  Return -ENODATA if attr values don't
47 	 * compare across ports.
48 	 */
49 
50 	netdev_for_each_lower_dev(dev, lower_dev, iter) {
51 		err = switchdev_port_attr_get(lower_dev, attr);
52 		if (err)
53 			break;
54 		if (first.id == SWITCHDEV_ATTR_UNDEFINED)
55 			first = *attr;
56 		else if (memcmp(&first, attr, sizeof(*attr)))
57 			return -ENODATA;
58 	}
59 
60 	return err;
61 }
62 EXPORT_SYMBOL_GPL(switchdev_port_attr_get);
63 
64 static int __switchdev_port_attr_set(struct net_device *dev,
65 				     struct switchdev_attr *attr)
66 {
67 	const struct switchdev_ops *ops = dev->switchdev_ops;
68 	struct net_device *lower_dev;
69 	struct list_head *iter;
70 	int err = -EOPNOTSUPP;
71 
72 	if (ops && ops->switchdev_port_attr_set)
73 		return ops->switchdev_port_attr_set(dev, attr);
74 
75 	if (attr->flags & SWITCHDEV_F_NO_RECURSE)
76 		return err;
77 
78 	/* Switch device port(s) may be stacked under
79 	 * bond/team/vlan dev, so recurse down to set attr on
80 	 * each port.
81 	 */
82 
83 	netdev_for_each_lower_dev(dev, lower_dev, iter) {
84 		err = __switchdev_port_attr_set(lower_dev, attr);
85 		if (err)
86 			break;
87 	}
88 
89 	return err;
90 }
91 
92 struct switchdev_attr_set_work {
93 	struct work_struct work;
94 	struct net_device *dev;
95 	struct switchdev_attr attr;
96 };
97 
98 static void switchdev_port_attr_set_work(struct work_struct *work)
99 {
100 	struct switchdev_attr_set_work *asw =
101 		container_of(work, struct switchdev_attr_set_work, work);
102 	int err;
103 
104 	rtnl_lock();
105 	err = switchdev_port_attr_set(asw->dev, &asw->attr);
106 	BUG_ON(err);
107 	rtnl_unlock();
108 
109 	dev_put(asw->dev);
110 	kfree(work);
111 }
112 
113 static int switchdev_port_attr_set_defer(struct net_device *dev,
114 					 struct switchdev_attr *attr)
115 {
116 	struct switchdev_attr_set_work *asw;
117 
118 	asw = kmalloc(sizeof(*asw), GFP_ATOMIC);
119 	if (!asw)
120 		return -ENOMEM;
121 
122 	INIT_WORK(&asw->work, switchdev_port_attr_set_work);
123 
124 	dev_hold(dev);
125 	asw->dev = dev;
126 	memcpy(&asw->attr, attr, sizeof(asw->attr));
127 
128 	schedule_work(&asw->work);
129 
130 	return 0;
131 }
132 
133 /**
134  *	switchdev_port_attr_set - Set port attribute
135  *
136  *	@dev: port device
137  *	@attr: attribute to set
138  *
139  *	Use a 2-phase prepare-commit transaction model to ensure
140  *	system is not left in a partially updated state due to
141  *	failure from driver/device.
142  */
143 int switchdev_port_attr_set(struct net_device *dev, struct switchdev_attr *attr)
144 {
145 	int err;
146 
147 	if (!rtnl_is_locked()) {
148 		/* Running prepare-commit transaction across stacked
149 		 * devices requires nothing moves, so if rtnl_lock is
150 		 * not held, schedule a worker thread to hold rtnl_lock
151 		 * while setting attr.
152 		 */
153 
154 		return switchdev_port_attr_set_defer(dev, attr);
155 	}
156 
157 	/* Phase I: prepare for attr set. Driver/device should fail
158 	 * here if there are going to be issues in the commit phase,
159 	 * such as lack of resources or support.  The driver/device
160 	 * should reserve resources needed for the commit phase here,
161 	 * but should not commit the attr.
162 	 */
163 
164 	attr->trans = SWITCHDEV_TRANS_PREPARE;
165 	err = __switchdev_port_attr_set(dev, attr);
166 	if (err) {
167 		/* Prepare phase failed: abort the transaction.  Any
168 		 * resources reserved in the prepare phase are
169 		 * released.
170 		 */
171 
172 		attr->trans = SWITCHDEV_TRANS_ABORT;
173 		__switchdev_port_attr_set(dev, attr);
174 
175 		return err;
176 	}
177 
178 	/* Phase II: commit attr set.  This cannot fail as a fault
179 	 * of driver/device.  If it does, it's a bug in the driver/device
180 	 * because the driver said everythings was OK in phase I.
181 	 */
182 
183 	attr->trans = SWITCHDEV_TRANS_COMMIT;
184 	err = __switchdev_port_attr_set(dev, attr);
185 	BUG_ON(err);
186 
187 	return err;
188 }
189 EXPORT_SYMBOL_GPL(switchdev_port_attr_set);
190 
191 static int __switchdev_port_obj_add(struct net_device *dev,
192 				    struct switchdev_obj *obj)
193 {
194 	const struct switchdev_ops *ops = dev->switchdev_ops;
195 	struct net_device *lower_dev;
196 	struct list_head *iter;
197 	int err = -EOPNOTSUPP;
198 
199 	if (ops && ops->switchdev_port_obj_add)
200 		return ops->switchdev_port_obj_add(dev, obj);
201 
202 	/* Switch device port(s) may be stacked under
203 	 * bond/team/vlan dev, so recurse down to add object on
204 	 * each port.
205 	 */
206 
207 	netdev_for_each_lower_dev(dev, lower_dev, iter) {
208 		err = __switchdev_port_obj_add(lower_dev, obj);
209 		if (err)
210 			break;
211 	}
212 
213 	return err;
214 }
215 
216 /**
217  *	switchdev_port_obj_add - Add port object
218  *
219  *	@dev: port device
220  *	@obj: object to add
221  *
222  *	Use a 2-phase prepare-commit transaction model to ensure
223  *	system is not left in a partially updated state due to
224  *	failure from driver/device.
225  *
226  *	rtnl_lock must be held.
227  */
228 int switchdev_port_obj_add(struct net_device *dev, struct switchdev_obj *obj)
229 {
230 	int err;
231 
232 	ASSERT_RTNL();
233 
234 	/* Phase I: prepare for obj add. Driver/device should fail
235 	 * here if there are going to be issues in the commit phase,
236 	 * such as lack of resources or support.  The driver/device
237 	 * should reserve resources needed for the commit phase here,
238 	 * but should not commit the obj.
239 	 */
240 
241 	obj->trans = SWITCHDEV_TRANS_PREPARE;
242 	err = __switchdev_port_obj_add(dev, obj);
243 	if (err) {
244 		/* Prepare phase failed: abort the transaction.  Any
245 		 * resources reserved in the prepare phase are
246 		 * released.
247 		 */
248 
249 		obj->trans = SWITCHDEV_TRANS_ABORT;
250 		__switchdev_port_obj_add(dev, obj);
251 
252 		return err;
253 	}
254 
255 	/* Phase II: commit obj add.  This cannot fail as a fault
256 	 * of driver/device.  If it does, it's a bug in the driver/device
257 	 * because the driver said everythings was OK in phase I.
258 	 */
259 
260 	obj->trans = SWITCHDEV_TRANS_COMMIT;
261 	err = __switchdev_port_obj_add(dev, obj);
262 	WARN(err, "%s: Commit of object (id=%d) failed.\n", dev->name, obj->id);
263 
264 	return err;
265 }
266 EXPORT_SYMBOL_GPL(switchdev_port_obj_add);
267 
268 /**
269  *	switchdev_port_obj_del - Delete port object
270  *
271  *	@dev: port device
272  *	@obj: object to delete
273  */
274 int switchdev_port_obj_del(struct net_device *dev, struct switchdev_obj *obj)
275 {
276 	const struct switchdev_ops *ops = dev->switchdev_ops;
277 	struct net_device *lower_dev;
278 	struct list_head *iter;
279 	int err = -EOPNOTSUPP;
280 
281 	if (ops && ops->switchdev_port_obj_del)
282 		return ops->switchdev_port_obj_del(dev, obj);
283 
284 	/* Switch device port(s) may be stacked under
285 	 * bond/team/vlan dev, so recurse down to delete object on
286 	 * each port.
287 	 */
288 
289 	netdev_for_each_lower_dev(dev, lower_dev, iter) {
290 		err = switchdev_port_obj_del(lower_dev, obj);
291 		if (err)
292 			break;
293 	}
294 
295 	return err;
296 }
297 EXPORT_SYMBOL_GPL(switchdev_port_obj_del);
298 
299 /**
300  *	switchdev_port_obj_dump - Dump port objects
301  *
302  *	@dev: port device
303  *	@obj: object to dump
304  */
305 int switchdev_port_obj_dump(struct net_device *dev, struct switchdev_obj *obj)
306 {
307 	const struct switchdev_ops *ops = dev->switchdev_ops;
308 	struct net_device *lower_dev;
309 	struct list_head *iter;
310 	int err = -EOPNOTSUPP;
311 
312 	if (ops && ops->switchdev_port_obj_dump)
313 		return ops->switchdev_port_obj_dump(dev, obj);
314 
315 	/* Switch device port(s) may be stacked under
316 	 * bond/team/vlan dev, so recurse down to dump objects on
317 	 * first port at bottom of stack.
318 	 */
319 
320 	netdev_for_each_lower_dev(dev, lower_dev, iter) {
321 		err = switchdev_port_obj_dump(lower_dev, obj);
322 		break;
323 	}
324 
325 	return err;
326 }
327 EXPORT_SYMBOL_GPL(switchdev_port_obj_dump);
328 
329 static DEFINE_MUTEX(switchdev_mutex);
330 static RAW_NOTIFIER_HEAD(switchdev_notif_chain);
331 
332 /**
333  *	register_switchdev_notifier - Register notifier
334  *	@nb: notifier_block
335  *
336  *	Register switch device notifier. This should be used by code
337  *	which needs to monitor events happening in particular device.
338  *	Return values are same as for atomic_notifier_chain_register().
339  */
340 int register_switchdev_notifier(struct notifier_block *nb)
341 {
342 	int err;
343 
344 	mutex_lock(&switchdev_mutex);
345 	err = raw_notifier_chain_register(&switchdev_notif_chain, nb);
346 	mutex_unlock(&switchdev_mutex);
347 	return err;
348 }
349 EXPORT_SYMBOL_GPL(register_switchdev_notifier);
350 
351 /**
352  *	unregister_switchdev_notifier - Unregister notifier
353  *	@nb: notifier_block
354  *
355  *	Unregister switch device notifier.
356  *	Return values are same as for atomic_notifier_chain_unregister().
357  */
358 int unregister_switchdev_notifier(struct notifier_block *nb)
359 {
360 	int err;
361 
362 	mutex_lock(&switchdev_mutex);
363 	err = raw_notifier_chain_unregister(&switchdev_notif_chain, nb);
364 	mutex_unlock(&switchdev_mutex);
365 	return err;
366 }
367 EXPORT_SYMBOL_GPL(unregister_switchdev_notifier);
368 
369 /**
370  *	call_switchdev_notifiers - Call notifiers
371  *	@val: value passed unmodified to notifier function
372  *	@dev: port device
373  *	@info: notifier information data
374  *
375  *	Call all network notifier blocks. This should be called by driver
376  *	when it needs to propagate hardware event.
377  *	Return values are same as for atomic_notifier_call_chain().
378  */
379 int call_switchdev_notifiers(unsigned long val, struct net_device *dev,
380 			     struct switchdev_notifier_info *info)
381 {
382 	int err;
383 
384 	info->dev = dev;
385 	mutex_lock(&switchdev_mutex);
386 	err = raw_notifier_call_chain(&switchdev_notif_chain, val, info);
387 	mutex_unlock(&switchdev_mutex);
388 	return err;
389 }
390 EXPORT_SYMBOL_GPL(call_switchdev_notifiers);
391 
392 /**
393  *	switchdev_port_bridge_getlink - Get bridge port attributes
394  *
395  *	@dev: port device
396  *
397  *	Called for SELF on rtnl_bridge_getlink to get bridge port
398  *	attributes.
399  */
400 int switchdev_port_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
401 				  struct net_device *dev, u32 filter_mask,
402 				  int nlflags)
403 {
404 	struct switchdev_attr attr = {
405 		.id = SWITCHDEV_ATTR_PORT_BRIDGE_FLAGS,
406 	};
407 	u16 mode = BRIDGE_MODE_UNDEF;
408 	u32 mask = BR_LEARNING | BR_LEARNING_SYNC;
409 	int err;
410 
411 	err = switchdev_port_attr_get(dev, &attr);
412 	if (err)
413 		return err;
414 
415 	return ndo_dflt_bridge_getlink(skb, pid, seq, dev, mode,
416 				       attr.u.brport_flags, mask, nlflags);
417 }
418 EXPORT_SYMBOL_GPL(switchdev_port_bridge_getlink);
419 
420 static int switchdev_port_br_setflag(struct net_device *dev,
421 				     struct nlattr *nlattr,
422 				     unsigned long brport_flag)
423 {
424 	struct switchdev_attr attr = {
425 		.id = SWITCHDEV_ATTR_PORT_BRIDGE_FLAGS,
426 	};
427 	u8 flag = nla_get_u8(nlattr);
428 	int err;
429 
430 	err = switchdev_port_attr_get(dev, &attr);
431 	if (err)
432 		return err;
433 
434 	if (flag)
435 		attr.u.brport_flags |= brport_flag;
436 	else
437 		attr.u.brport_flags &= ~brport_flag;
438 
439 	return switchdev_port_attr_set(dev, &attr);
440 }
441 
442 static const struct nla_policy
443 switchdev_port_bridge_policy[IFLA_BRPORT_MAX + 1] = {
444 	[IFLA_BRPORT_STATE]		= { .type = NLA_U8 },
445 	[IFLA_BRPORT_COST]		= { .type = NLA_U32 },
446 	[IFLA_BRPORT_PRIORITY]		= { .type = NLA_U16 },
447 	[IFLA_BRPORT_MODE]		= { .type = NLA_U8 },
448 	[IFLA_BRPORT_GUARD]		= { .type = NLA_U8 },
449 	[IFLA_BRPORT_PROTECT]		= { .type = NLA_U8 },
450 	[IFLA_BRPORT_FAST_LEAVE]	= { .type = NLA_U8 },
451 	[IFLA_BRPORT_LEARNING]		= { .type = NLA_U8 },
452 	[IFLA_BRPORT_LEARNING_SYNC]	= { .type = NLA_U8 },
453 	[IFLA_BRPORT_UNICAST_FLOOD]	= { .type = NLA_U8 },
454 };
455 
456 static int switchdev_port_br_setlink_protinfo(struct net_device *dev,
457 					      struct nlattr *protinfo)
458 {
459 	struct nlattr *attr;
460 	int rem;
461 	int err;
462 
463 	err = nla_validate_nested(protinfo, IFLA_BRPORT_MAX,
464 				  switchdev_port_bridge_policy);
465 	if (err)
466 		return err;
467 
468 	nla_for_each_nested(attr, protinfo, rem) {
469 		switch (nla_type(attr)) {
470 		case IFLA_BRPORT_LEARNING:
471 			err = switchdev_port_br_setflag(dev, attr,
472 							BR_LEARNING);
473 			break;
474 		case IFLA_BRPORT_LEARNING_SYNC:
475 			err = switchdev_port_br_setflag(dev, attr,
476 							BR_LEARNING_SYNC);
477 			break;
478 		default:
479 			err = -EOPNOTSUPP;
480 			break;
481 		}
482 		if (err)
483 			return err;
484 	}
485 
486 	return 0;
487 }
488 
489 static int switchdev_port_br_afspec(struct net_device *dev,
490 				    struct nlattr *afspec,
491 				    int (*f)(struct net_device *dev,
492 					     struct switchdev_obj *obj))
493 {
494 	struct nlattr *attr;
495 	struct bridge_vlan_info *vinfo;
496 	struct switchdev_obj obj = {
497 		.id = SWITCHDEV_OBJ_PORT_VLAN,
498 	};
499 	struct switchdev_obj_vlan *vlan = &obj.u.vlan;
500 	int rem;
501 	int err;
502 
503 	nla_for_each_nested(attr, afspec, rem) {
504 		if (nla_type(attr) != IFLA_BRIDGE_VLAN_INFO)
505 			continue;
506 		if (nla_len(attr) != sizeof(struct bridge_vlan_info))
507 			return -EINVAL;
508 		vinfo = nla_data(attr);
509 		vlan->flags = vinfo->flags;
510 		if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_BEGIN) {
511 			if (vlan->vid_start)
512 				return -EINVAL;
513 			vlan->vid_start = vinfo->vid;
514 		} else if (vinfo->flags & BRIDGE_VLAN_INFO_RANGE_END) {
515 			if (!vlan->vid_start)
516 				return -EINVAL;
517 			vlan->vid_end = vinfo->vid;
518 			if (vlan->vid_end <= vlan->vid_start)
519 				return -EINVAL;
520 			err = f(dev, &obj);
521 			if (err)
522 				return err;
523 			memset(vlan, 0, sizeof(*vlan));
524 		} else {
525 			if (vlan->vid_start)
526 				return -EINVAL;
527 			vlan->vid_start = vinfo->vid;
528 			vlan->vid_end = vinfo->vid;
529 			err = f(dev, &obj);
530 			if (err)
531 				return err;
532 			memset(vlan, 0, sizeof(*vlan));
533 		}
534 	}
535 
536 	return 0;
537 }
538 
539 /**
540  *	switchdev_port_bridge_setlink - Set bridge port attributes
541  *
542  *	@dev: port device
543  *	@nlh: netlink header
544  *	@flags: netlink flags
545  *
546  *	Called for SELF on rtnl_bridge_setlink to set bridge port
547  *	attributes.
548  */
549 int switchdev_port_bridge_setlink(struct net_device *dev,
550 				  struct nlmsghdr *nlh, u16 flags)
551 {
552 	struct nlattr *protinfo;
553 	struct nlattr *afspec;
554 	int err = 0;
555 
556 	protinfo = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
557 				   IFLA_PROTINFO);
558 	if (protinfo) {
559 		err = switchdev_port_br_setlink_protinfo(dev, protinfo);
560 		if (err)
561 			return err;
562 	}
563 
564 	afspec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
565 				 IFLA_AF_SPEC);
566 	if (afspec)
567 		err = switchdev_port_br_afspec(dev, afspec,
568 					       switchdev_port_obj_add);
569 
570 	return err;
571 }
572 EXPORT_SYMBOL_GPL(switchdev_port_bridge_setlink);
573 
574 /**
575  *	switchdev_port_bridge_dellink - Set bridge port attributes
576  *
577  *	@dev: port device
578  *	@nlh: netlink header
579  *	@flags: netlink flags
580  *
581  *	Called for SELF on rtnl_bridge_dellink to set bridge port
582  *	attributes.
583  */
584 int switchdev_port_bridge_dellink(struct net_device *dev,
585 				  struct nlmsghdr *nlh, u16 flags)
586 {
587 	struct nlattr *afspec;
588 
589 	afspec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg),
590 				 IFLA_AF_SPEC);
591 	if (afspec)
592 		return switchdev_port_br_afspec(dev, afspec,
593 						switchdev_port_obj_del);
594 
595 	return 0;
596 }
597 EXPORT_SYMBOL_GPL(switchdev_port_bridge_dellink);
598 
599 /**
600  *	switchdev_port_fdb_add - Add FDB (MAC/VLAN) entry to port
601  *
602  *	@ndmsg: netlink hdr
603  *	@nlattr: netlink attributes
604  *	@dev: port device
605  *	@addr: MAC address to add
606  *	@vid: VLAN to add
607  *
608  *	Add FDB entry to switch device.
609  */
610 int switchdev_port_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
611 			   struct net_device *dev, const unsigned char *addr,
612 			   u16 vid, u16 nlm_flags)
613 {
614 	struct switchdev_obj obj = {
615 		.id = SWITCHDEV_OBJ_PORT_FDB,
616 		.u.fdb = {
617 			.addr = addr,
618 			.vid = vid,
619 		},
620 	};
621 
622 	return switchdev_port_obj_add(dev, &obj);
623 }
624 EXPORT_SYMBOL_GPL(switchdev_port_fdb_add);
625 
626 /**
627  *	switchdev_port_fdb_del - Delete FDB (MAC/VLAN) entry from port
628  *
629  *	@ndmsg: netlink hdr
630  *	@nlattr: netlink attributes
631  *	@dev: port device
632  *	@addr: MAC address to delete
633  *	@vid: VLAN to delete
634  *
635  *	Delete FDB entry from switch device.
636  */
637 int switchdev_port_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
638 			   struct net_device *dev, const unsigned char *addr,
639 			   u16 vid)
640 {
641 	struct switchdev_obj obj = {
642 		.id = SWITCHDEV_OBJ_PORT_FDB,
643 		.u.fdb = {
644 			.addr = addr,
645 			.vid = vid,
646 		},
647 	};
648 
649 	return switchdev_port_obj_del(dev, &obj);
650 }
651 EXPORT_SYMBOL_GPL(switchdev_port_fdb_del);
652 
653 struct switchdev_fdb_dump {
654 	struct switchdev_obj obj;
655 	struct sk_buff *skb;
656 	struct netlink_callback *cb;
657 	struct net_device *filter_dev;
658 	int idx;
659 };
660 
661 static int switchdev_port_fdb_dump_cb(struct net_device *dev,
662 				      struct switchdev_obj *obj)
663 {
664 	struct switchdev_fdb_dump *dump =
665 		container_of(obj, struct switchdev_fdb_dump, obj);
666 	u32 portid = NETLINK_CB(dump->cb->skb).portid;
667 	u32 seq = dump->cb->nlh->nlmsg_seq;
668 	struct nlmsghdr *nlh;
669 	struct ndmsg *ndm;
670 	struct net_device *master = netdev_master_upper_dev_get(dev);
671 
672 	if (dump->idx < dump->cb->args[0])
673 		goto skip;
674 
675 	if (master && dump->filter_dev != master)
676 		goto skip;
677 
678 	nlh = nlmsg_put(dump->skb, portid, seq, RTM_NEWNEIGH,
679 			sizeof(*ndm), NLM_F_MULTI);
680 	if (!nlh)
681 		return -EMSGSIZE;
682 
683 	ndm = nlmsg_data(nlh);
684 	ndm->ndm_family  = AF_BRIDGE;
685 	ndm->ndm_pad1    = 0;
686 	ndm->ndm_pad2    = 0;
687 	ndm->ndm_flags   = NTF_SELF;
688 	ndm->ndm_type    = 0;
689 	ndm->ndm_ifindex = dev->ifindex;
690 	ndm->ndm_state   = NUD_REACHABLE;
691 
692 	if (nla_put(dump->skb, NDA_LLADDR, ETH_ALEN, obj->u.fdb.addr))
693 		goto nla_put_failure;
694 
695 	if (obj->u.fdb.vid && nla_put_u16(dump->skb, NDA_VLAN, obj->u.fdb.vid))
696 		goto nla_put_failure;
697 
698 	nlmsg_end(dump->skb, nlh);
699 
700 skip:
701 	dump->idx++;
702 	return 0;
703 
704 nla_put_failure:
705 	nlmsg_cancel(dump->skb, nlh);
706 	return -EMSGSIZE;
707 }
708 
709 /**
710  *	switchdev_port_fdb_dump - Dump port FDB (MAC/VLAN) entries
711  *
712  *	@skb: netlink skb
713  *	@cb: netlink callback
714  *	@dev: port device
715  *	@filter_dev: filter device
716  *	@idx:
717  *
718  *	Delete FDB entry from switch device.
719  */
720 int switchdev_port_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
721 			    struct net_device *dev,
722 			    struct net_device *filter_dev, int idx)
723 {
724 	struct switchdev_fdb_dump dump = {
725 		.obj = {
726 			.id = SWITCHDEV_OBJ_PORT_FDB,
727 			.cb = switchdev_port_fdb_dump_cb,
728 		},
729 		.skb = skb,
730 		.cb = cb,
731 		.filter_dev = filter_dev,
732 		.idx = idx,
733 	};
734 	int err;
735 
736 	err = switchdev_port_obj_dump(dev, &dump.obj);
737 	if (err)
738 		return err;
739 
740 	return dump.idx;
741 }
742 EXPORT_SYMBOL_GPL(switchdev_port_fdb_dump);
743 
744 static struct net_device *switchdev_get_lowest_dev(struct net_device *dev)
745 {
746 	const struct switchdev_ops *ops = dev->switchdev_ops;
747 	struct net_device *lower_dev;
748 	struct net_device *port_dev;
749 	struct list_head *iter;
750 
751 	/* Recusively search down until we find a sw port dev.
752 	 * (A sw port dev supports switchdev_port_attr_get).
753 	 */
754 
755 	if (ops && ops->switchdev_port_attr_get)
756 		return dev;
757 
758 	netdev_for_each_lower_dev(dev, lower_dev, iter) {
759 		port_dev = switchdev_get_lowest_dev(lower_dev);
760 		if (port_dev)
761 			return port_dev;
762 	}
763 
764 	return NULL;
765 }
766 
767 static struct net_device *switchdev_get_dev_by_nhs(struct fib_info *fi)
768 {
769 	struct switchdev_attr attr = {
770 		.id = SWITCHDEV_ATTR_PORT_PARENT_ID,
771 	};
772 	struct switchdev_attr prev_attr;
773 	struct net_device *dev = NULL;
774 	int nhsel;
775 
776 	/* For this route, all nexthop devs must be on the same switch. */
777 
778 	for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
779 		const struct fib_nh *nh = &fi->fib_nh[nhsel];
780 
781 		if (!nh->nh_dev)
782 			return NULL;
783 
784 		dev = switchdev_get_lowest_dev(nh->nh_dev);
785 		if (!dev)
786 			return NULL;
787 
788 		if (switchdev_port_attr_get(dev, &attr))
789 			return NULL;
790 
791 		if (nhsel > 0) {
792 			if (prev_attr.u.ppid.id_len != attr.u.ppid.id_len)
793 				return NULL;
794 			if (memcmp(prev_attr.u.ppid.id, attr.u.ppid.id,
795 				   attr.u.ppid.id_len))
796 				return NULL;
797 		}
798 
799 		prev_attr = attr;
800 	}
801 
802 	return dev;
803 }
804 
805 /**
806  *	switchdev_fib_ipv4_add - Add IPv4 route entry to switch
807  *
808  *	@dst: route's IPv4 destination address
809  *	@dst_len: destination address length (prefix length)
810  *	@fi: route FIB info structure
811  *	@tos: route TOS
812  *	@type: route type
813  *	@nlflags: netlink flags passed in (NLM_F_*)
814  *	@tb_id: route table ID
815  *
816  *	Add IPv4 route entry to switch device.
817  */
818 int switchdev_fib_ipv4_add(u32 dst, int dst_len, struct fib_info *fi,
819 			   u8 tos, u8 type, u32 nlflags, u32 tb_id)
820 {
821 	struct switchdev_obj fib_obj = {
822 		.id = SWITCHDEV_OBJ_IPV4_FIB,
823 		.u.ipv4_fib = {
824 			.dst = dst,
825 			.dst_len = dst_len,
826 			.fi = fi,
827 			.tos = tos,
828 			.type = type,
829 			.nlflags = nlflags,
830 			.tb_id = tb_id,
831 		},
832 	};
833 	struct net_device *dev;
834 	int err = 0;
835 
836 	/* Don't offload route if using custom ip rules or if
837 	 * IPv4 FIB offloading has been disabled completely.
838 	 */
839 
840 #ifdef CONFIG_IP_MULTIPLE_TABLES
841 	if (fi->fib_net->ipv4.fib_has_custom_rules)
842 		return 0;
843 #endif
844 
845 	if (fi->fib_net->ipv4.fib_offload_disabled)
846 		return 0;
847 
848 	dev = switchdev_get_dev_by_nhs(fi);
849 	if (!dev)
850 		return 0;
851 
852 	err = switchdev_port_obj_add(dev, &fib_obj);
853 	if (!err)
854 		fi->fib_flags |= RTNH_F_OFFLOAD;
855 
856 	return err;
857 }
858 EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_add);
859 
860 /**
861  *	switchdev_fib_ipv4_del - Delete IPv4 route entry from switch
862  *
863  *	@dst: route's IPv4 destination address
864  *	@dst_len: destination address length (prefix length)
865  *	@fi: route FIB info structure
866  *	@tos: route TOS
867  *	@type: route type
868  *	@tb_id: route table ID
869  *
870  *	Delete IPv4 route entry from switch device.
871  */
872 int switchdev_fib_ipv4_del(u32 dst, int dst_len, struct fib_info *fi,
873 			   u8 tos, u8 type, u32 tb_id)
874 {
875 	struct switchdev_obj fib_obj = {
876 		.id = SWITCHDEV_OBJ_IPV4_FIB,
877 		.u.ipv4_fib = {
878 			.dst = dst,
879 			.dst_len = dst_len,
880 			.fi = fi,
881 			.tos = tos,
882 			.type = type,
883 			.nlflags = 0,
884 			.tb_id = tb_id,
885 		},
886 	};
887 	struct net_device *dev;
888 	int err = 0;
889 
890 	if (!(fi->fib_flags & RTNH_F_OFFLOAD))
891 		return 0;
892 
893 	dev = switchdev_get_dev_by_nhs(fi);
894 	if (!dev)
895 		return 0;
896 
897 	err = switchdev_port_obj_del(dev, &fib_obj);
898 	if (!err)
899 		fi->fib_flags &= ~RTNH_F_OFFLOAD;
900 
901 	return err;
902 }
903 EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_del);
904 
905 /**
906  *	switchdev_fib_ipv4_abort - Abort an IPv4 FIB operation
907  *
908  *	@fi: route FIB info structure
909  */
910 void switchdev_fib_ipv4_abort(struct fib_info *fi)
911 {
912 	/* There was a problem installing this route to the offload
913 	 * device.  For now, until we come up with more refined
914 	 * policy handling, abruptly end IPv4 fib offloading for
915 	 * for entire net by flushing offload device(s) of all
916 	 * IPv4 routes, and mark IPv4 fib offloading broken from
917 	 * this point forward.
918 	 */
919 
920 	fib_flush_external(fi->fib_net);
921 	fi->fib_net->ipv4.fib_offload_disabled = true;
922 }
923 EXPORT_SYMBOL_GPL(switchdev_fib_ipv4_abort);
924