[go: nahoru, domu]

1/*
2 * drivers/net/team/team.c - Network team device driver
3 * Copyright (c) 2011 Jiri Pirko <jpirko@redhat.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 */
10
11#include <linux/kernel.h>
12#include <linux/types.h>
13#include <linux/module.h>
14#include <linux/init.h>
15#include <linux/slab.h>
16#include <linux/rcupdate.h>
17#include <linux/errno.h>
18#include <linux/ctype.h>
19#include <linux/notifier.h>
20#include <linux/netdevice.h>
21#include <linux/netpoll.h>
22#include <linux/if_vlan.h>
23#include <linux/if_arp.h>
24#include <linux/socket.h>
25#include <linux/etherdevice.h>
26#include <linux/rtnetlink.h>
27#include <net/rtnetlink.h>
28#include <net/genetlink.h>
29#include <net/netlink.h>
30#include <net/sch_generic.h>
31#include <generated/utsrelease.h>
32#include <linux/if_team.h>
33
34#define DRV_NAME "team"
35
36
37/**********
38 * Helpers
39 **********/
40
41#define team_port_exists(dev) (dev->priv_flags & IFF_TEAM_PORT)
42
43static struct team_port *team_port_get_rcu(const struct net_device *dev)
44{
45	struct team_port *port = rcu_dereference(dev->rx_handler_data);
46
47	return team_port_exists(dev) ? port : NULL;
48}
49
50static struct team_port *team_port_get_rtnl(const struct net_device *dev)
51{
52	struct team_port *port = rtnl_dereference(dev->rx_handler_data);
53
54	return team_port_exists(dev) ? port : NULL;
55}
56
57/*
58 * Since the ability to change device address for open port device is tested in
59 * team_port_add, this function can be called without control of return value
60 */
61static int __set_port_dev_addr(struct net_device *port_dev,
62			       const unsigned char *dev_addr)
63{
64	struct sockaddr addr;
65
66	memcpy(addr.sa_data, dev_addr, port_dev->addr_len);
67	addr.sa_family = port_dev->type;
68	return dev_set_mac_address(port_dev, &addr);
69}
70
71static int team_port_set_orig_dev_addr(struct team_port *port)
72{
73	return __set_port_dev_addr(port->dev, port->orig.dev_addr);
74}
75
76static int team_port_set_team_dev_addr(struct team *team,
77				       struct team_port *port)
78{
79	return __set_port_dev_addr(port->dev, team->dev->dev_addr);
80}
81
82int team_modeop_port_enter(struct team *team, struct team_port *port)
83{
84	return team_port_set_team_dev_addr(team, port);
85}
86EXPORT_SYMBOL(team_modeop_port_enter);
87
88void team_modeop_port_change_dev_addr(struct team *team,
89				      struct team_port *port)
90{
91	team_port_set_team_dev_addr(team, port);
92}
93EXPORT_SYMBOL(team_modeop_port_change_dev_addr);
94
95static void team_refresh_port_linkup(struct team_port *port)
96{
97	port->linkup = port->user.linkup_enabled ? port->user.linkup :
98						   port->state.linkup;
99}
100
101
102/*******************
103 * Options handling
104 *******************/
105
106struct team_option_inst { /* One for each option instance */
107	struct list_head list;
108	struct list_head tmp_list;
109	struct team_option *option;
110	struct team_option_inst_info info;
111	bool changed;
112	bool removed;
113};
114
115static struct team_option *__team_find_option(struct team *team,
116					      const char *opt_name)
117{
118	struct team_option *option;
119
120	list_for_each_entry(option, &team->option_list, list) {
121		if (strcmp(option->name, opt_name) == 0)
122			return option;
123	}
124	return NULL;
125}
126
127static void __team_option_inst_del(struct team_option_inst *opt_inst)
128{
129	list_del(&opt_inst->list);
130	kfree(opt_inst);
131}
132
133static void __team_option_inst_del_option(struct team *team,
134					  struct team_option *option)
135{
136	struct team_option_inst *opt_inst, *tmp;
137
138	list_for_each_entry_safe(opt_inst, tmp, &team->option_inst_list, list) {
139		if (opt_inst->option == option)
140			__team_option_inst_del(opt_inst);
141	}
142}
143
144static int __team_option_inst_add(struct team *team, struct team_option *option,
145				  struct team_port *port)
146{
147	struct team_option_inst *opt_inst;
148	unsigned int array_size;
149	unsigned int i;
150	int err;
151
152	array_size = option->array_size;
153	if (!array_size)
154		array_size = 1; /* No array but still need one instance */
155
156	for (i = 0; i < array_size; i++) {
157		opt_inst = kmalloc(sizeof(*opt_inst), GFP_KERNEL);
158		if (!opt_inst)
159			return -ENOMEM;
160		opt_inst->option = option;
161		opt_inst->info.port = port;
162		opt_inst->info.array_index = i;
163		opt_inst->changed = true;
164		opt_inst->removed = false;
165		list_add_tail(&opt_inst->list, &team->option_inst_list);
166		if (option->init) {
167			err = option->init(team, &opt_inst->info);
168			if (err)
169				return err;
170		}
171
172	}
173	return 0;
174}
175
176static int __team_option_inst_add_option(struct team *team,
177					 struct team_option *option)
178{
179	struct team_port *port;
180	int err;
181
182	if (!option->per_port) {
183		err = __team_option_inst_add(team, option, NULL);
184		if (err)
185			goto inst_del_option;
186	}
187
188	list_for_each_entry(port, &team->port_list, list) {
189		err = __team_option_inst_add(team, option, port);
190		if (err)
191			goto inst_del_option;
192	}
193	return 0;
194
195inst_del_option:
196	__team_option_inst_del_option(team, option);
197	return err;
198}
199
200static void __team_option_inst_mark_removed_option(struct team *team,
201						   struct team_option *option)
202{
203	struct team_option_inst *opt_inst;
204
205	list_for_each_entry(opt_inst, &team->option_inst_list, list) {
206		if (opt_inst->option == option) {
207			opt_inst->changed = true;
208			opt_inst->removed = true;
209		}
210	}
211}
212
213static void __team_option_inst_del_port(struct team *team,
214					struct team_port *port)
215{
216	struct team_option_inst *opt_inst, *tmp;
217
218	list_for_each_entry_safe(opt_inst, tmp, &team->option_inst_list, list) {
219		if (opt_inst->option->per_port &&
220		    opt_inst->info.port == port)
221			__team_option_inst_del(opt_inst);
222	}
223}
224
225static int __team_option_inst_add_port(struct team *team,
226				       struct team_port *port)
227{
228	struct team_option *option;
229	int err;
230
231	list_for_each_entry(option, &team->option_list, list) {
232		if (!option->per_port)
233			continue;
234		err = __team_option_inst_add(team, option, port);
235		if (err)
236			goto inst_del_port;
237	}
238	return 0;
239
240inst_del_port:
241	__team_option_inst_del_port(team, port);
242	return err;
243}
244
245static void __team_option_inst_mark_removed_port(struct team *team,
246						 struct team_port *port)
247{
248	struct team_option_inst *opt_inst;
249
250	list_for_each_entry(opt_inst, &team->option_inst_list, list) {
251		if (opt_inst->info.port == port) {
252			opt_inst->changed = true;
253			opt_inst->removed = true;
254		}
255	}
256}
257
258static int __team_options_register(struct team *team,
259				   const struct team_option *option,
260				   size_t option_count)
261{
262	int i;
263	struct team_option **dst_opts;
264	int err;
265
266	dst_opts = kzalloc(sizeof(struct team_option *) * option_count,
267			   GFP_KERNEL);
268	if (!dst_opts)
269		return -ENOMEM;
270	for (i = 0; i < option_count; i++, option++) {
271		if (__team_find_option(team, option->name)) {
272			err = -EEXIST;
273			goto alloc_rollback;
274		}
275		dst_opts[i] = kmemdup(option, sizeof(*option), GFP_KERNEL);
276		if (!dst_opts[i]) {
277			err = -ENOMEM;
278			goto alloc_rollback;
279		}
280	}
281
282	for (i = 0; i < option_count; i++) {
283		err = __team_option_inst_add_option(team, dst_opts[i]);
284		if (err)
285			goto inst_rollback;
286		list_add_tail(&dst_opts[i]->list, &team->option_list);
287	}
288
289	kfree(dst_opts);
290	return 0;
291
292inst_rollback:
293	for (i--; i >= 0; i--)
294		__team_option_inst_del_option(team, dst_opts[i]);
295
296	i = option_count - 1;
297alloc_rollback:
298	for (i--; i >= 0; i--)
299		kfree(dst_opts[i]);
300
301	kfree(dst_opts);
302	return err;
303}
304
305static void __team_options_mark_removed(struct team *team,
306					const struct team_option *option,
307					size_t option_count)
308{
309	int i;
310
311	for (i = 0; i < option_count; i++, option++) {
312		struct team_option *del_opt;
313
314		del_opt = __team_find_option(team, option->name);
315		if (del_opt)
316			__team_option_inst_mark_removed_option(team, del_opt);
317	}
318}
319
320static void __team_options_unregister(struct team *team,
321				      const struct team_option *option,
322				      size_t option_count)
323{
324	int i;
325
326	for (i = 0; i < option_count; i++, option++) {
327		struct team_option *del_opt;
328
329		del_opt = __team_find_option(team, option->name);
330		if (del_opt) {
331			__team_option_inst_del_option(team, del_opt);
332			list_del(&del_opt->list);
333			kfree(del_opt);
334		}
335	}
336}
337
338static void __team_options_change_check(struct team *team);
339
340int team_options_register(struct team *team,
341			  const struct team_option *option,
342			  size_t option_count)
343{
344	int err;
345
346	err = __team_options_register(team, option, option_count);
347	if (err)
348		return err;
349	__team_options_change_check(team);
350	return 0;
351}
352EXPORT_SYMBOL(team_options_register);
353
354void team_options_unregister(struct team *team,
355			     const struct team_option *option,
356			     size_t option_count)
357{
358	__team_options_mark_removed(team, option, option_count);
359	__team_options_change_check(team);
360	__team_options_unregister(team, option, option_count);
361}
362EXPORT_SYMBOL(team_options_unregister);
363
364static int team_option_get(struct team *team,
365			   struct team_option_inst *opt_inst,
366			   struct team_gsetter_ctx *ctx)
367{
368	if (!opt_inst->option->getter)
369		return -EOPNOTSUPP;
370	return opt_inst->option->getter(team, ctx);
371}
372
373static int team_option_set(struct team *team,
374			   struct team_option_inst *opt_inst,
375			   struct team_gsetter_ctx *ctx)
376{
377	if (!opt_inst->option->setter)
378		return -EOPNOTSUPP;
379	return opt_inst->option->setter(team, ctx);
380}
381
382void team_option_inst_set_change(struct team_option_inst_info *opt_inst_info)
383{
384	struct team_option_inst *opt_inst;
385
386	opt_inst = container_of(opt_inst_info, struct team_option_inst, info);
387	opt_inst->changed = true;
388}
389EXPORT_SYMBOL(team_option_inst_set_change);
390
391void team_options_change_check(struct team *team)
392{
393	__team_options_change_check(team);
394}
395EXPORT_SYMBOL(team_options_change_check);
396
397
398/****************
399 * Mode handling
400 ****************/
401
402static LIST_HEAD(mode_list);
403static DEFINE_SPINLOCK(mode_list_lock);
404
405struct team_mode_item {
406	struct list_head list;
407	const struct team_mode *mode;
408};
409
410static struct team_mode_item *__find_mode(const char *kind)
411{
412	struct team_mode_item *mitem;
413
414	list_for_each_entry(mitem, &mode_list, list) {
415		if (strcmp(mitem->mode->kind, kind) == 0)
416			return mitem;
417	}
418	return NULL;
419}
420
421static bool is_good_mode_name(const char *name)
422{
423	while (*name != '\0') {
424		if (!isalpha(*name) && !isdigit(*name) && *name != '_')
425			return false;
426		name++;
427	}
428	return true;
429}
430
431int team_mode_register(const struct team_mode *mode)
432{
433	int err = 0;
434	struct team_mode_item *mitem;
435
436	if (!is_good_mode_name(mode->kind) ||
437	    mode->priv_size > TEAM_MODE_PRIV_SIZE)
438		return -EINVAL;
439
440	mitem = kmalloc(sizeof(*mitem), GFP_KERNEL);
441	if (!mitem)
442		return -ENOMEM;
443
444	spin_lock(&mode_list_lock);
445	if (__find_mode(mode->kind)) {
446		err = -EEXIST;
447		kfree(mitem);
448		goto unlock;
449	}
450	mitem->mode = mode;
451	list_add_tail(&mitem->list, &mode_list);
452unlock:
453	spin_unlock(&mode_list_lock);
454	return err;
455}
456EXPORT_SYMBOL(team_mode_register);
457
458void team_mode_unregister(const struct team_mode *mode)
459{
460	struct team_mode_item *mitem;
461
462	spin_lock(&mode_list_lock);
463	mitem = __find_mode(mode->kind);
464	if (mitem) {
465		list_del_init(&mitem->list);
466		kfree(mitem);
467	}
468	spin_unlock(&mode_list_lock);
469}
470EXPORT_SYMBOL(team_mode_unregister);
471
472static const struct team_mode *team_mode_get(const char *kind)
473{
474	struct team_mode_item *mitem;
475	const struct team_mode *mode = NULL;
476
477	spin_lock(&mode_list_lock);
478	mitem = __find_mode(kind);
479	if (!mitem) {
480		spin_unlock(&mode_list_lock);
481		request_module("team-mode-%s", kind);
482		spin_lock(&mode_list_lock);
483		mitem = __find_mode(kind);
484	}
485	if (mitem) {
486		mode = mitem->mode;
487		if (!try_module_get(mode->owner))
488			mode = NULL;
489	}
490
491	spin_unlock(&mode_list_lock);
492	return mode;
493}
494
495static void team_mode_put(const struct team_mode *mode)
496{
497	module_put(mode->owner);
498}
499
500static bool team_dummy_transmit(struct team *team, struct sk_buff *skb)
501{
502	dev_kfree_skb_any(skb);
503	return false;
504}
505
506static rx_handler_result_t team_dummy_receive(struct team *team,
507					      struct team_port *port,
508					      struct sk_buff *skb)
509{
510	return RX_HANDLER_ANOTHER;
511}
512
513static const struct team_mode __team_no_mode = {
514	.kind		= "*NOMODE*",
515};
516
517static bool team_is_mode_set(struct team *team)
518{
519	return team->mode != &__team_no_mode;
520}
521
522static void team_set_no_mode(struct team *team)
523{
524	team->user_carrier_enabled = false;
525	team->mode = &__team_no_mode;
526}
527
528static void team_adjust_ops(struct team *team)
529{
530	/*
531	 * To avoid checks in rx/tx skb paths, ensure here that non-null and
532	 * correct ops are always set.
533	 */
534
535	if (!team->en_port_count || !team_is_mode_set(team) ||
536	    !team->mode->ops->transmit)
537		team->ops.transmit = team_dummy_transmit;
538	else
539		team->ops.transmit = team->mode->ops->transmit;
540
541	if (!team->en_port_count || !team_is_mode_set(team) ||
542	    !team->mode->ops->receive)
543		team->ops.receive = team_dummy_receive;
544	else
545		team->ops.receive = team->mode->ops->receive;
546}
547
548/*
549 * We can benefit from the fact that it's ensured no port is present
550 * at the time of mode change. Therefore no packets are in fly so there's no
551 * need to set mode operations in any special way.
552 */
553static int __team_change_mode(struct team *team,
554			      const struct team_mode *new_mode)
555{
556	/* Check if mode was previously set and do cleanup if so */
557	if (team_is_mode_set(team)) {
558		void (*exit_op)(struct team *team) = team->ops.exit;
559
560		/* Clear ops area so no callback is called any longer */
561		memset(&team->ops, 0, sizeof(struct team_mode_ops));
562		team_adjust_ops(team);
563
564		if (exit_op)
565			exit_op(team);
566		team_mode_put(team->mode);
567		team_set_no_mode(team);
568		/* zero private data area */
569		memset(&team->mode_priv, 0,
570		       sizeof(struct team) - offsetof(struct team, mode_priv));
571	}
572
573	if (!new_mode)
574		return 0;
575
576	if (new_mode->ops->init) {
577		int err;
578
579		err = new_mode->ops->init(team);
580		if (err)
581			return err;
582	}
583
584	team->mode = new_mode;
585	memcpy(&team->ops, new_mode->ops, sizeof(struct team_mode_ops));
586	team_adjust_ops(team);
587
588	return 0;
589}
590
591static int team_change_mode(struct team *team, const char *kind)
592{
593	const struct team_mode *new_mode;
594	struct net_device *dev = team->dev;
595	int err;
596
597	if (!list_empty(&team->port_list)) {
598		netdev_err(dev, "No ports can be present during mode change\n");
599		return -EBUSY;
600	}
601
602	if (team_is_mode_set(team) && strcmp(team->mode->kind, kind) == 0) {
603		netdev_err(dev, "Unable to change to the same mode the team is in\n");
604		return -EINVAL;
605	}
606
607	new_mode = team_mode_get(kind);
608	if (!new_mode) {
609		netdev_err(dev, "Mode \"%s\" not found\n", kind);
610		return -EINVAL;
611	}
612
613	err = __team_change_mode(team, new_mode);
614	if (err) {
615		netdev_err(dev, "Failed to change to mode \"%s\"\n", kind);
616		team_mode_put(new_mode);
617		return err;
618	}
619
620	netdev_info(dev, "Mode changed to \"%s\"\n", kind);
621	return 0;
622}
623
624
625/*********************
626 * Peers notification
627 *********************/
628
629static void team_notify_peers_work(struct work_struct *work)
630{
631	struct team *team;
632
633	team = container_of(work, struct team, notify_peers.dw.work);
634
635	if (!rtnl_trylock()) {
636		schedule_delayed_work(&team->notify_peers.dw, 0);
637		return;
638	}
639	call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, team->dev);
640	rtnl_unlock();
641	if (!atomic_dec_and_test(&team->notify_peers.count_pending))
642		schedule_delayed_work(&team->notify_peers.dw,
643				      msecs_to_jiffies(team->notify_peers.interval));
644}
645
646static void team_notify_peers(struct team *team)
647{
648	if (!team->notify_peers.count || !netif_running(team->dev))
649		return;
650	atomic_add(team->notify_peers.count, &team->notify_peers.count_pending);
651	schedule_delayed_work(&team->notify_peers.dw, 0);
652}
653
654static void team_notify_peers_init(struct team *team)
655{
656	INIT_DELAYED_WORK(&team->notify_peers.dw, team_notify_peers_work);
657}
658
659static void team_notify_peers_fini(struct team *team)
660{
661	cancel_delayed_work_sync(&team->notify_peers.dw);
662}
663
664
665/*******************************
666 * Send multicast group rejoins
667 *******************************/
668
669static void team_mcast_rejoin_work(struct work_struct *work)
670{
671	struct team *team;
672
673	team = container_of(work, struct team, mcast_rejoin.dw.work);
674
675	if (!rtnl_trylock()) {
676		schedule_delayed_work(&team->mcast_rejoin.dw, 0);
677		return;
678	}
679	call_netdevice_notifiers(NETDEV_RESEND_IGMP, team->dev);
680	rtnl_unlock();
681	if (!atomic_dec_and_test(&team->mcast_rejoin.count_pending))
682		schedule_delayed_work(&team->mcast_rejoin.dw,
683				      msecs_to_jiffies(team->mcast_rejoin.interval));
684}
685
686static void team_mcast_rejoin(struct team *team)
687{
688	if (!team->mcast_rejoin.count || !netif_running(team->dev))
689		return;
690	atomic_add(team->mcast_rejoin.count, &team->mcast_rejoin.count_pending);
691	schedule_delayed_work(&team->mcast_rejoin.dw, 0);
692}
693
694static void team_mcast_rejoin_init(struct team *team)
695{
696	INIT_DELAYED_WORK(&team->mcast_rejoin.dw, team_mcast_rejoin_work);
697}
698
699static void team_mcast_rejoin_fini(struct team *team)
700{
701	cancel_delayed_work_sync(&team->mcast_rejoin.dw);
702}
703
704
705/************************
706 * Rx path frame handler
707 ************************/
708
709/* note: already called with rcu_read_lock */
710static rx_handler_result_t team_handle_frame(struct sk_buff **pskb)
711{
712	struct sk_buff *skb = *pskb;
713	struct team_port *port;
714	struct team *team;
715	rx_handler_result_t res;
716
717	skb = skb_share_check(skb, GFP_ATOMIC);
718	if (!skb)
719		return RX_HANDLER_CONSUMED;
720
721	*pskb = skb;
722
723	port = team_port_get_rcu(skb->dev);
724	team = port->team;
725	if (!team_port_enabled(port)) {
726		/* allow exact match delivery for disabled ports */
727		res = RX_HANDLER_EXACT;
728	} else {
729		res = team->ops.receive(team, port, skb);
730	}
731	if (res == RX_HANDLER_ANOTHER) {
732		struct team_pcpu_stats *pcpu_stats;
733
734		pcpu_stats = this_cpu_ptr(team->pcpu_stats);
735		u64_stats_update_begin(&pcpu_stats->syncp);
736		pcpu_stats->rx_packets++;
737		pcpu_stats->rx_bytes += skb->len;
738		if (skb->pkt_type == PACKET_MULTICAST)
739			pcpu_stats->rx_multicast++;
740		u64_stats_update_end(&pcpu_stats->syncp);
741
742		skb->dev = team->dev;
743	} else {
744		this_cpu_inc(team->pcpu_stats->rx_dropped);
745	}
746
747	return res;
748}
749
750
751/*************************************
752 * Multiqueue Tx port select override
753 *************************************/
754
755static int team_queue_override_init(struct team *team)
756{
757	struct list_head *listarr;
758	unsigned int queue_cnt = team->dev->num_tx_queues - 1;
759	unsigned int i;
760
761	if (!queue_cnt)
762		return 0;
763	listarr = kmalloc(sizeof(struct list_head) * queue_cnt, GFP_KERNEL);
764	if (!listarr)
765		return -ENOMEM;
766	team->qom_lists = listarr;
767	for (i = 0; i < queue_cnt; i++)
768		INIT_LIST_HEAD(listarr++);
769	return 0;
770}
771
772static void team_queue_override_fini(struct team *team)
773{
774	kfree(team->qom_lists);
775}
776
777static struct list_head *__team_get_qom_list(struct team *team, u16 queue_id)
778{
779	return &team->qom_lists[queue_id - 1];
780}
781
782/*
783 * note: already called with rcu_read_lock
784 */
785static bool team_queue_override_transmit(struct team *team, struct sk_buff *skb)
786{
787	struct list_head *qom_list;
788	struct team_port *port;
789
790	if (!team->queue_override_enabled || !skb->queue_mapping)
791		return false;
792	qom_list = __team_get_qom_list(team, skb->queue_mapping);
793	list_for_each_entry_rcu(port, qom_list, qom_list) {
794		if (!team_dev_queue_xmit(team, port, skb))
795			return true;
796	}
797	return false;
798}
799
800static void __team_queue_override_port_del(struct team *team,
801					   struct team_port *port)
802{
803	if (!port->queue_id)
804		return;
805	list_del_rcu(&port->qom_list);
806}
807
808static bool team_queue_override_port_has_gt_prio_than(struct team_port *port,
809						      struct team_port *cur)
810{
811	if (port->priority < cur->priority)
812		return true;
813	if (port->priority > cur->priority)
814		return false;
815	if (port->index < cur->index)
816		return true;
817	return false;
818}
819
820static void __team_queue_override_port_add(struct team *team,
821					   struct team_port *port)
822{
823	struct team_port *cur;
824	struct list_head *qom_list;
825	struct list_head *node;
826
827	if (!port->queue_id)
828		return;
829	qom_list = __team_get_qom_list(team, port->queue_id);
830	node = qom_list;
831	list_for_each_entry(cur, qom_list, qom_list) {
832		if (team_queue_override_port_has_gt_prio_than(port, cur))
833			break;
834		node = &cur->qom_list;
835	}
836	list_add_tail_rcu(&port->qom_list, node);
837}
838
839static void __team_queue_override_enabled_check(struct team *team)
840{
841	struct team_port *port;
842	bool enabled = false;
843
844	list_for_each_entry(port, &team->port_list, list) {
845		if (port->queue_id) {
846			enabled = true;
847			break;
848		}
849	}
850	if (enabled == team->queue_override_enabled)
851		return;
852	netdev_dbg(team->dev, "%s queue override\n",
853		   enabled ? "Enabling" : "Disabling");
854	team->queue_override_enabled = enabled;
855}
856
857static void team_queue_override_port_prio_changed(struct team *team,
858						  struct team_port *port)
859{
860	if (!port->queue_id || team_port_enabled(port))
861		return;
862	__team_queue_override_port_del(team, port);
863	__team_queue_override_port_add(team, port);
864	__team_queue_override_enabled_check(team);
865}
866
867static void team_queue_override_port_change_queue_id(struct team *team,
868						     struct team_port *port,
869						     u16 new_queue_id)
870{
871	if (team_port_enabled(port)) {
872		__team_queue_override_port_del(team, port);
873		port->queue_id = new_queue_id;
874		__team_queue_override_port_add(team, port);
875		__team_queue_override_enabled_check(team);
876	} else {
877		port->queue_id = new_queue_id;
878	}
879}
880
881static void team_queue_override_port_add(struct team *team,
882					 struct team_port *port)
883{
884	__team_queue_override_port_add(team, port);
885	__team_queue_override_enabled_check(team);
886}
887
888static void team_queue_override_port_del(struct team *team,
889					 struct team_port *port)
890{
891	__team_queue_override_port_del(team, port);
892	__team_queue_override_enabled_check(team);
893}
894
895
896/****************
897 * Port handling
898 ****************/
899
900static bool team_port_find(const struct team *team,
901			   const struct team_port *port)
902{
903	struct team_port *cur;
904
905	list_for_each_entry(cur, &team->port_list, list)
906		if (cur == port)
907			return true;
908	return false;
909}
910
911/*
912 * Enable/disable port by adding to enabled port hashlist and setting
913 * port->index (Might be racy so reader could see incorrect ifindex when
914 * processing a flying packet, but that is not a problem). Write guarded
915 * by team->lock.
916 */
917static void team_port_enable(struct team *team,
918			     struct team_port *port)
919{
920	if (team_port_enabled(port))
921		return;
922	port->index = team->en_port_count++;
923	hlist_add_head_rcu(&port->hlist,
924			   team_port_index_hash(team, port->index));
925	team_adjust_ops(team);
926	team_queue_override_port_add(team, port);
927	if (team->ops.port_enabled)
928		team->ops.port_enabled(team, port);
929	team_notify_peers(team);
930	team_mcast_rejoin(team);
931}
932
933static void __reconstruct_port_hlist(struct team *team, int rm_index)
934{
935	int i;
936	struct team_port *port;
937
938	for (i = rm_index + 1; i < team->en_port_count; i++) {
939		port = team_get_port_by_index(team, i);
940		hlist_del_rcu(&port->hlist);
941		port->index--;
942		hlist_add_head_rcu(&port->hlist,
943				   team_port_index_hash(team, port->index));
944	}
945}
946
947static void team_port_disable(struct team *team,
948			      struct team_port *port)
949{
950	if (!team_port_enabled(port))
951		return;
952	if (team->ops.port_disabled)
953		team->ops.port_disabled(team, port);
954	hlist_del_rcu(&port->hlist);
955	__reconstruct_port_hlist(team, port->index);
956	port->index = -1;
957	team->en_port_count--;
958	team_queue_override_port_del(team, port);
959	team_adjust_ops(team);
960	team_notify_peers(team);
961	team_mcast_rejoin(team);
962}
963
964#define TEAM_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
965			    NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
966			    NETIF_F_HIGHDMA | NETIF_F_LRO)
967
968static void __team_compute_features(struct team *team)
969{
970	struct team_port *port;
971	u32 vlan_features = TEAM_VLAN_FEATURES & NETIF_F_ALL_FOR_ALL;
972	unsigned short max_hard_header_len = ETH_HLEN;
973	unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
974					IFF_XMIT_DST_RELEASE_PERM;
975
976	list_for_each_entry(port, &team->port_list, list) {
977		vlan_features = netdev_increment_features(vlan_features,
978					port->dev->vlan_features,
979					TEAM_VLAN_FEATURES);
980
981		dst_release_flag &= port->dev->priv_flags;
982		if (port->dev->hard_header_len > max_hard_header_len)
983			max_hard_header_len = port->dev->hard_header_len;
984	}
985
986	team->dev->vlan_features = vlan_features;
987	team->dev->hard_header_len = max_hard_header_len;
988
989	team->dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
990	if (dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
991		team->dev->priv_flags |= IFF_XMIT_DST_RELEASE;
992
993	netdev_change_features(team->dev);
994}
995
996static void team_compute_features(struct team *team)
997{
998	mutex_lock(&team->lock);
999	__team_compute_features(team);
1000	mutex_unlock(&team->lock);
1001}
1002
1003static int team_port_enter(struct team *team, struct team_port *port)
1004{
1005	int err = 0;
1006
1007	dev_hold(team->dev);
1008	if (team->ops.port_enter) {
1009		err = team->ops.port_enter(team, port);
1010		if (err) {
1011			netdev_err(team->dev, "Device %s failed to enter team mode\n",
1012				   port->dev->name);
1013			goto err_port_enter;
1014		}
1015	}
1016
1017	return 0;
1018
1019err_port_enter:
1020	dev_put(team->dev);
1021
1022	return err;
1023}
1024
1025static void team_port_leave(struct team *team, struct team_port *port)
1026{
1027	if (team->ops.port_leave)
1028		team->ops.port_leave(team, port);
1029	dev_put(team->dev);
1030}
1031
1032#ifdef CONFIG_NET_POLL_CONTROLLER
1033static int team_port_enable_netpoll(struct team *team, struct team_port *port)
1034{
1035	struct netpoll *np;
1036	int err;
1037
1038	if (!team->dev->npinfo)
1039		return 0;
1040
1041	np = kzalloc(sizeof(*np), GFP_KERNEL);
1042	if (!np)
1043		return -ENOMEM;
1044
1045	err = __netpoll_setup(np, port->dev);
1046	if (err) {
1047		kfree(np);
1048		return err;
1049	}
1050	port->np = np;
1051	return err;
1052}
1053
1054static void team_port_disable_netpoll(struct team_port *port)
1055{
1056	struct netpoll *np = port->np;
1057
1058	if (!np)
1059		return;
1060	port->np = NULL;
1061
1062	/* Wait for transmitting packets to finish before freeing. */
1063	synchronize_rcu_bh();
1064	__netpoll_cleanup(np);
1065	kfree(np);
1066}
1067#else
1068static int team_port_enable_netpoll(struct team *team, struct team_port *port)
1069{
1070	return 0;
1071}
1072static void team_port_disable_netpoll(struct team_port *port)
1073{
1074}
1075#endif
1076
1077static int team_upper_dev_link(struct net_device *dev,
1078			       struct net_device *port_dev)
1079{
1080	int err;
1081
1082	err = netdev_master_upper_dev_link(port_dev, dev);
1083	if (err)
1084		return err;
1085	port_dev->priv_flags |= IFF_TEAM_PORT;
1086	return 0;
1087}
1088
1089static void team_upper_dev_unlink(struct net_device *dev,
1090				  struct net_device *port_dev)
1091{
1092	netdev_upper_dev_unlink(port_dev, dev);
1093	port_dev->priv_flags &= ~IFF_TEAM_PORT;
1094}
1095
1096static void __team_port_change_port_added(struct team_port *port, bool linkup);
1097static int team_dev_type_check_change(struct net_device *dev,
1098				      struct net_device *port_dev);
1099
1100static int team_port_add(struct team *team, struct net_device *port_dev)
1101{
1102	struct net_device *dev = team->dev;
1103	struct team_port *port;
1104	char *portname = port_dev->name;
1105	int err;
1106
1107	if (port_dev->flags & IFF_LOOPBACK) {
1108		netdev_err(dev, "Device %s is loopback device. Loopback devices can't be added as a team port\n",
1109			   portname);
1110		return -EINVAL;
1111	}
1112
1113	if (team_port_exists(port_dev)) {
1114		netdev_err(dev, "Device %s is already a port "
1115				"of a team device\n", portname);
1116		return -EBUSY;
1117	}
1118
1119	if (port_dev->features & NETIF_F_VLAN_CHALLENGED &&
1120	    vlan_uses_dev(dev)) {
1121		netdev_err(dev, "Device %s is VLAN challenged and team device has VLAN set up\n",
1122			   portname);
1123		return -EPERM;
1124	}
1125
1126	err = team_dev_type_check_change(dev, port_dev);
1127	if (err)
1128		return err;
1129
1130	if (port_dev->flags & IFF_UP) {
1131		netdev_err(dev, "Device %s is up. Set it down before adding it as a team port\n",
1132			   portname);
1133		return -EBUSY;
1134	}
1135
1136	port = kzalloc(sizeof(struct team_port) + team->mode->port_priv_size,
1137		       GFP_KERNEL);
1138	if (!port)
1139		return -ENOMEM;
1140
1141	port->dev = port_dev;
1142	port->team = team;
1143	INIT_LIST_HEAD(&port->qom_list);
1144
1145	port->orig.mtu = port_dev->mtu;
1146	err = dev_set_mtu(port_dev, dev->mtu);
1147	if (err) {
1148		netdev_dbg(dev, "Error %d calling dev_set_mtu\n", err);
1149		goto err_set_mtu;
1150	}
1151
1152	memcpy(port->orig.dev_addr, port_dev->dev_addr, port_dev->addr_len);
1153
1154	err = team_port_enter(team, port);
1155	if (err) {
1156		netdev_err(dev, "Device %s failed to enter team mode\n",
1157			   portname);
1158		goto err_port_enter;
1159	}
1160
1161	err = dev_open(port_dev);
1162	if (err) {
1163		netdev_dbg(dev, "Device %s opening failed\n",
1164			   portname);
1165		goto err_dev_open;
1166	}
1167
1168	err = vlan_vids_add_by_dev(port_dev, dev);
1169	if (err) {
1170		netdev_err(dev, "Failed to add vlan ids to device %s\n",
1171				portname);
1172		goto err_vids_add;
1173	}
1174
1175	err = team_port_enable_netpoll(team, port);
1176	if (err) {
1177		netdev_err(dev, "Failed to enable netpoll on device %s\n",
1178			   portname);
1179		goto err_enable_netpoll;
1180	}
1181
1182	err = netdev_rx_handler_register(port_dev, team_handle_frame,
1183					 port);
1184	if (err) {
1185		netdev_err(dev, "Device %s failed to register rx_handler\n",
1186			   portname);
1187		goto err_handler_register;
1188	}
1189
1190	err = team_upper_dev_link(dev, port_dev);
1191	if (err) {
1192		netdev_err(dev, "Device %s failed to set upper link\n",
1193			   portname);
1194		goto err_set_upper_link;
1195	}
1196
1197	err = __team_option_inst_add_port(team, port);
1198	if (err) {
1199		netdev_err(dev, "Device %s failed to add per-port options\n",
1200			   portname);
1201		goto err_option_port_add;
1202	}
1203
1204	port->index = -1;
1205	list_add_tail_rcu(&port->list, &team->port_list);
1206	team_port_enable(team, port);
1207	__team_compute_features(team);
1208	__team_port_change_port_added(port, !!netif_carrier_ok(port_dev));
1209	__team_options_change_check(team);
1210
1211	netdev_info(dev, "Port device %s added\n", portname);
1212
1213	return 0;
1214
1215err_option_port_add:
1216	team_upper_dev_unlink(dev, port_dev);
1217
1218err_set_upper_link:
1219	netdev_rx_handler_unregister(port_dev);
1220
1221err_handler_register:
1222	team_port_disable_netpoll(port);
1223
1224err_enable_netpoll:
1225	vlan_vids_del_by_dev(port_dev, dev);
1226
1227err_vids_add:
1228	dev_close(port_dev);
1229
1230err_dev_open:
1231	team_port_leave(team, port);
1232	team_port_set_orig_dev_addr(port);
1233
1234err_port_enter:
1235	dev_set_mtu(port_dev, port->orig.mtu);
1236
1237err_set_mtu:
1238	kfree(port);
1239
1240	return err;
1241}
1242
1243static void __team_port_change_port_removed(struct team_port *port);
1244
1245static int team_port_del(struct team *team, struct net_device *port_dev)
1246{
1247	struct net_device *dev = team->dev;
1248	struct team_port *port;
1249	char *portname = port_dev->name;
1250
1251	port = team_port_get_rtnl(port_dev);
1252	if (!port || !team_port_find(team, port)) {
1253		netdev_err(dev, "Device %s does not act as a port of this team\n",
1254			   portname);
1255		return -ENOENT;
1256	}
1257
1258	team_port_disable(team, port);
1259	list_del_rcu(&port->list);
1260	team_upper_dev_unlink(dev, port_dev);
1261	netdev_rx_handler_unregister(port_dev);
1262	team_port_disable_netpoll(port);
1263	vlan_vids_del_by_dev(port_dev, dev);
1264	dev_uc_unsync(port_dev, dev);
1265	dev_mc_unsync(port_dev, dev);
1266	dev_close(port_dev);
1267	team_port_leave(team, port);
1268
1269	__team_option_inst_mark_removed_port(team, port);
1270	__team_options_change_check(team);
1271	__team_option_inst_del_port(team, port);
1272	__team_port_change_port_removed(port);
1273
1274	team_port_set_orig_dev_addr(port);
1275	dev_set_mtu(port_dev, port->orig.mtu);
1276	kfree_rcu(port, rcu);
1277	netdev_info(dev, "Port device %s removed\n", portname);
1278	__team_compute_features(team);
1279
1280	return 0;
1281}
1282
1283
1284/*****************
1285 * Net device ops
1286 *****************/
1287
1288static int team_mode_option_get(struct team *team, struct team_gsetter_ctx *ctx)
1289{
1290	ctx->data.str_val = team->mode->kind;
1291	return 0;
1292}
1293
1294static int team_mode_option_set(struct team *team, struct team_gsetter_ctx *ctx)
1295{
1296	return team_change_mode(team, ctx->data.str_val);
1297}
1298
1299static int team_notify_peers_count_get(struct team *team,
1300				       struct team_gsetter_ctx *ctx)
1301{
1302	ctx->data.u32_val = team->notify_peers.count;
1303	return 0;
1304}
1305
1306static int team_notify_peers_count_set(struct team *team,
1307				       struct team_gsetter_ctx *ctx)
1308{
1309	team->notify_peers.count = ctx->data.u32_val;
1310	return 0;
1311}
1312
1313static int team_notify_peers_interval_get(struct team *team,
1314					  struct team_gsetter_ctx *ctx)
1315{
1316	ctx->data.u32_val = team->notify_peers.interval;
1317	return 0;
1318}
1319
1320static int team_notify_peers_interval_set(struct team *team,
1321					  struct team_gsetter_ctx *ctx)
1322{
1323	team->notify_peers.interval = ctx->data.u32_val;
1324	return 0;
1325}
1326
1327static int team_mcast_rejoin_count_get(struct team *team,
1328				       struct team_gsetter_ctx *ctx)
1329{
1330	ctx->data.u32_val = team->mcast_rejoin.count;
1331	return 0;
1332}
1333
1334static int team_mcast_rejoin_count_set(struct team *team,
1335				       struct team_gsetter_ctx *ctx)
1336{
1337	team->mcast_rejoin.count = ctx->data.u32_val;
1338	return 0;
1339}
1340
1341static int team_mcast_rejoin_interval_get(struct team *team,
1342					  struct team_gsetter_ctx *ctx)
1343{
1344	ctx->data.u32_val = team->mcast_rejoin.interval;
1345	return 0;
1346}
1347
1348static int team_mcast_rejoin_interval_set(struct team *team,
1349					  struct team_gsetter_ctx *ctx)
1350{
1351	team->mcast_rejoin.interval = ctx->data.u32_val;
1352	return 0;
1353}
1354
1355static int team_port_en_option_get(struct team *team,
1356				   struct team_gsetter_ctx *ctx)
1357{
1358	struct team_port *port = ctx->info->port;
1359
1360	ctx->data.bool_val = team_port_enabled(port);
1361	return 0;
1362}
1363
1364static int team_port_en_option_set(struct team *team,
1365				   struct team_gsetter_ctx *ctx)
1366{
1367	struct team_port *port = ctx->info->port;
1368
1369	if (ctx->data.bool_val)
1370		team_port_enable(team, port);
1371	else
1372		team_port_disable(team, port);
1373	return 0;
1374}
1375
1376static int team_user_linkup_option_get(struct team *team,
1377				       struct team_gsetter_ctx *ctx)
1378{
1379	struct team_port *port = ctx->info->port;
1380
1381	ctx->data.bool_val = port->user.linkup;
1382	return 0;
1383}
1384
1385static void __team_carrier_check(struct team *team);
1386
1387static int team_user_linkup_option_set(struct team *team,
1388				       struct team_gsetter_ctx *ctx)
1389{
1390	struct team_port *port = ctx->info->port;
1391
1392	port->user.linkup = ctx->data.bool_val;
1393	team_refresh_port_linkup(port);
1394	__team_carrier_check(port->team);
1395	return 0;
1396}
1397
1398static int team_user_linkup_en_option_get(struct team *team,
1399					  struct team_gsetter_ctx *ctx)
1400{
1401	struct team_port *port = ctx->info->port;
1402
1403	ctx->data.bool_val = port->user.linkup_enabled;
1404	return 0;
1405}
1406
1407static int team_user_linkup_en_option_set(struct team *team,
1408					  struct team_gsetter_ctx *ctx)
1409{
1410	struct team_port *port = ctx->info->port;
1411
1412	port->user.linkup_enabled = ctx->data.bool_val;
1413	team_refresh_port_linkup(port);
1414	__team_carrier_check(port->team);
1415	return 0;
1416}
1417
1418static int team_priority_option_get(struct team *team,
1419				    struct team_gsetter_ctx *ctx)
1420{
1421	struct team_port *port = ctx->info->port;
1422
1423	ctx->data.s32_val = port->priority;
1424	return 0;
1425}
1426
1427static int team_priority_option_set(struct team *team,
1428				    struct team_gsetter_ctx *ctx)
1429{
1430	struct team_port *port = ctx->info->port;
1431	s32 priority = ctx->data.s32_val;
1432
1433	if (port->priority == priority)
1434		return 0;
1435	port->priority = priority;
1436	team_queue_override_port_prio_changed(team, port);
1437	return 0;
1438}
1439
1440static int team_queue_id_option_get(struct team *team,
1441				    struct team_gsetter_ctx *ctx)
1442{
1443	struct team_port *port = ctx->info->port;
1444
1445	ctx->data.u32_val = port->queue_id;
1446	return 0;
1447}
1448
1449static int team_queue_id_option_set(struct team *team,
1450				    struct team_gsetter_ctx *ctx)
1451{
1452	struct team_port *port = ctx->info->port;
1453	u16 new_queue_id = ctx->data.u32_val;
1454
1455	if (port->queue_id == new_queue_id)
1456		return 0;
1457	if (new_queue_id >= team->dev->real_num_tx_queues)
1458		return -EINVAL;
1459	team_queue_override_port_change_queue_id(team, port, new_queue_id);
1460	return 0;
1461}
1462
1463static const struct team_option team_options[] = {
1464	{
1465		.name = "mode",
1466		.type = TEAM_OPTION_TYPE_STRING,
1467		.getter = team_mode_option_get,
1468		.setter = team_mode_option_set,
1469	},
1470	{
1471		.name = "notify_peers_count",
1472		.type = TEAM_OPTION_TYPE_U32,
1473		.getter = team_notify_peers_count_get,
1474		.setter = team_notify_peers_count_set,
1475	},
1476	{
1477		.name = "notify_peers_interval",
1478		.type = TEAM_OPTION_TYPE_U32,
1479		.getter = team_notify_peers_interval_get,
1480		.setter = team_notify_peers_interval_set,
1481	},
1482	{
1483		.name = "mcast_rejoin_count",
1484		.type = TEAM_OPTION_TYPE_U32,
1485		.getter = team_mcast_rejoin_count_get,
1486		.setter = team_mcast_rejoin_count_set,
1487	},
1488	{
1489		.name = "mcast_rejoin_interval",
1490		.type = TEAM_OPTION_TYPE_U32,
1491		.getter = team_mcast_rejoin_interval_get,
1492		.setter = team_mcast_rejoin_interval_set,
1493	},
1494	{
1495		.name = "enabled",
1496		.type = TEAM_OPTION_TYPE_BOOL,
1497		.per_port = true,
1498		.getter = team_port_en_option_get,
1499		.setter = team_port_en_option_set,
1500	},
1501	{
1502		.name = "user_linkup",
1503		.type = TEAM_OPTION_TYPE_BOOL,
1504		.per_port = true,
1505		.getter = team_user_linkup_option_get,
1506		.setter = team_user_linkup_option_set,
1507	},
1508	{
1509		.name = "user_linkup_enabled",
1510		.type = TEAM_OPTION_TYPE_BOOL,
1511		.per_port = true,
1512		.getter = team_user_linkup_en_option_get,
1513		.setter = team_user_linkup_en_option_set,
1514	},
1515	{
1516		.name = "priority",
1517		.type = TEAM_OPTION_TYPE_S32,
1518		.per_port = true,
1519		.getter = team_priority_option_get,
1520		.setter = team_priority_option_set,
1521	},
1522	{
1523		.name = "queue_id",
1524		.type = TEAM_OPTION_TYPE_U32,
1525		.per_port = true,
1526		.getter = team_queue_id_option_get,
1527		.setter = team_queue_id_option_set,
1528	},
1529};
1530
1531static struct lock_class_key team_netdev_xmit_lock_key;
1532static struct lock_class_key team_netdev_addr_lock_key;
1533static struct lock_class_key team_tx_busylock_key;
1534
1535static void team_set_lockdep_class_one(struct net_device *dev,
1536				       struct netdev_queue *txq,
1537				       void *unused)
1538{
1539	lockdep_set_class(&txq->_xmit_lock, &team_netdev_xmit_lock_key);
1540}
1541
1542static void team_set_lockdep_class(struct net_device *dev)
1543{
1544	lockdep_set_class(&dev->addr_list_lock, &team_netdev_addr_lock_key);
1545	netdev_for_each_tx_queue(dev, team_set_lockdep_class_one, NULL);
1546	dev->qdisc_tx_busylock = &team_tx_busylock_key;
1547}
1548
1549static int team_init(struct net_device *dev)
1550{
1551	struct team *team = netdev_priv(dev);
1552	int i;
1553	int err;
1554
1555	team->dev = dev;
1556	mutex_init(&team->lock);
1557	team_set_no_mode(team);
1558
1559	team->pcpu_stats = netdev_alloc_pcpu_stats(struct team_pcpu_stats);
1560	if (!team->pcpu_stats)
1561		return -ENOMEM;
1562
1563	for (i = 0; i < TEAM_PORT_HASHENTRIES; i++)
1564		INIT_HLIST_HEAD(&team->en_port_hlist[i]);
1565	INIT_LIST_HEAD(&team->port_list);
1566	err = team_queue_override_init(team);
1567	if (err)
1568		goto err_team_queue_override_init;
1569
1570	team_adjust_ops(team);
1571
1572	INIT_LIST_HEAD(&team->option_list);
1573	INIT_LIST_HEAD(&team->option_inst_list);
1574
1575	team_notify_peers_init(team);
1576	team_mcast_rejoin_init(team);
1577
1578	err = team_options_register(team, team_options, ARRAY_SIZE(team_options));
1579	if (err)
1580		goto err_options_register;
1581	netif_carrier_off(dev);
1582
1583	team_set_lockdep_class(dev);
1584
1585	return 0;
1586
1587err_options_register:
1588	team_mcast_rejoin_fini(team);
1589	team_notify_peers_fini(team);
1590	team_queue_override_fini(team);
1591err_team_queue_override_init:
1592	free_percpu(team->pcpu_stats);
1593
1594	return err;
1595}
1596
1597static void team_uninit(struct net_device *dev)
1598{
1599	struct team *team = netdev_priv(dev);
1600	struct team_port *port;
1601	struct team_port *tmp;
1602
1603	mutex_lock(&team->lock);
1604	list_for_each_entry_safe(port, tmp, &team->port_list, list)
1605		team_port_del(team, port->dev);
1606
1607	__team_change_mode(team, NULL); /* cleanup */
1608	__team_options_unregister(team, team_options, ARRAY_SIZE(team_options));
1609	team_mcast_rejoin_fini(team);
1610	team_notify_peers_fini(team);
1611	team_queue_override_fini(team);
1612	mutex_unlock(&team->lock);
1613}
1614
1615static void team_destructor(struct net_device *dev)
1616{
1617	struct team *team = netdev_priv(dev);
1618
1619	free_percpu(team->pcpu_stats);
1620	free_netdev(dev);
1621}
1622
1623static int team_open(struct net_device *dev)
1624{
1625	return 0;
1626}
1627
1628static int team_close(struct net_device *dev)
1629{
1630	return 0;
1631}
1632
1633/*
1634 * note: already called with rcu_read_lock
1635 */
1636static netdev_tx_t team_xmit(struct sk_buff *skb, struct net_device *dev)
1637{
1638	struct team *team = netdev_priv(dev);
1639	bool tx_success;
1640	unsigned int len = skb->len;
1641
1642	tx_success = team_queue_override_transmit(team, skb);
1643	if (!tx_success)
1644		tx_success = team->ops.transmit(team, skb);
1645	if (tx_success) {
1646		struct team_pcpu_stats *pcpu_stats;
1647
1648		pcpu_stats = this_cpu_ptr(team->pcpu_stats);
1649		u64_stats_update_begin(&pcpu_stats->syncp);
1650		pcpu_stats->tx_packets++;
1651		pcpu_stats->tx_bytes += len;
1652		u64_stats_update_end(&pcpu_stats->syncp);
1653	} else {
1654		this_cpu_inc(team->pcpu_stats->tx_dropped);
1655	}
1656
1657	return NETDEV_TX_OK;
1658}
1659
1660static u16 team_select_queue(struct net_device *dev, struct sk_buff *skb,
1661			     void *accel_priv, select_queue_fallback_t fallback)
1662{
1663	/*
1664	 * This helper function exists to help dev_pick_tx get the correct
1665	 * destination queue.  Using a helper function skips a call to
1666	 * skb_tx_hash and will put the skbs in the queue we expect on their
1667	 * way down to the team driver.
1668	 */
1669	u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
1670
1671	/*
1672	 * Save the original txq to restore before passing to the driver
1673	 */
1674	qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
1675
1676	if (unlikely(txq >= dev->real_num_tx_queues)) {
1677		do {
1678			txq -= dev->real_num_tx_queues;
1679		} while (txq >= dev->real_num_tx_queues);
1680	}
1681	return txq;
1682}
1683
1684static void team_change_rx_flags(struct net_device *dev, int change)
1685{
1686	struct team *team = netdev_priv(dev);
1687	struct team_port *port;
1688	int inc;
1689
1690	rcu_read_lock();
1691	list_for_each_entry_rcu(port, &team->port_list, list) {
1692		if (change & IFF_PROMISC) {
1693			inc = dev->flags & IFF_PROMISC ? 1 : -1;
1694			dev_set_promiscuity(port->dev, inc);
1695		}
1696		if (change & IFF_ALLMULTI) {
1697			inc = dev->flags & IFF_ALLMULTI ? 1 : -1;
1698			dev_set_allmulti(port->dev, inc);
1699		}
1700	}
1701	rcu_read_unlock();
1702}
1703
1704static void team_set_rx_mode(struct net_device *dev)
1705{
1706	struct team *team = netdev_priv(dev);
1707	struct team_port *port;
1708
1709	rcu_read_lock();
1710	list_for_each_entry_rcu(port, &team->port_list, list) {
1711		dev_uc_sync_multiple(port->dev, dev);
1712		dev_mc_sync_multiple(port->dev, dev);
1713	}
1714	rcu_read_unlock();
1715}
1716
1717static int team_set_mac_address(struct net_device *dev, void *p)
1718{
1719	struct sockaddr *addr = p;
1720	struct team *team = netdev_priv(dev);
1721	struct team_port *port;
1722
1723	if (dev->type == ARPHRD_ETHER && !is_valid_ether_addr(addr->sa_data))
1724		return -EADDRNOTAVAIL;
1725	memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
1726	rcu_read_lock();
1727	list_for_each_entry_rcu(port, &team->port_list, list)
1728		if (team->ops.port_change_dev_addr)
1729			team->ops.port_change_dev_addr(team, port);
1730	rcu_read_unlock();
1731	return 0;
1732}
1733
1734static int team_change_mtu(struct net_device *dev, int new_mtu)
1735{
1736	struct team *team = netdev_priv(dev);
1737	struct team_port *port;
1738	int err;
1739
1740	/*
1741	 * Alhough this is reader, it's guarded by team lock. It's not possible
1742	 * to traverse list in reverse under rcu_read_lock
1743	 */
1744	mutex_lock(&team->lock);
1745	team->port_mtu_change_allowed = true;
1746	list_for_each_entry(port, &team->port_list, list) {
1747		err = dev_set_mtu(port->dev, new_mtu);
1748		if (err) {
1749			netdev_err(dev, "Device %s failed to change mtu",
1750				   port->dev->name);
1751			goto unwind;
1752		}
1753	}
1754	team->port_mtu_change_allowed = false;
1755	mutex_unlock(&team->lock);
1756
1757	dev->mtu = new_mtu;
1758
1759	return 0;
1760
1761unwind:
1762	list_for_each_entry_continue_reverse(port, &team->port_list, list)
1763		dev_set_mtu(port->dev, dev->mtu);
1764	team->port_mtu_change_allowed = false;
1765	mutex_unlock(&team->lock);
1766
1767	return err;
1768}
1769
1770static struct rtnl_link_stats64 *
1771team_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1772{
1773	struct team *team = netdev_priv(dev);
1774	struct team_pcpu_stats *p;
1775	u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
1776	u32 rx_dropped = 0, tx_dropped = 0;
1777	unsigned int start;
1778	int i;
1779
1780	for_each_possible_cpu(i) {
1781		p = per_cpu_ptr(team->pcpu_stats, i);
1782		do {
1783			start = u64_stats_fetch_begin_irq(&p->syncp);
1784			rx_packets	= p->rx_packets;
1785			rx_bytes	= p->rx_bytes;
1786			rx_multicast	= p->rx_multicast;
1787			tx_packets	= p->tx_packets;
1788			tx_bytes	= p->tx_bytes;
1789		} while (u64_stats_fetch_retry_irq(&p->syncp, start));
1790
1791		stats->rx_packets	+= rx_packets;
1792		stats->rx_bytes		+= rx_bytes;
1793		stats->multicast	+= rx_multicast;
1794		stats->tx_packets	+= tx_packets;
1795		stats->tx_bytes		+= tx_bytes;
1796		/*
1797		 * rx_dropped & tx_dropped are u32, updated
1798		 * without syncp protection.
1799		 */
1800		rx_dropped	+= p->rx_dropped;
1801		tx_dropped	+= p->tx_dropped;
1802	}
1803	stats->rx_dropped	= rx_dropped;
1804	stats->tx_dropped	= tx_dropped;
1805	return stats;
1806}
1807
1808static int team_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
1809{
1810	struct team *team = netdev_priv(dev);
1811	struct team_port *port;
1812	int err;
1813
1814	/*
1815	 * Alhough this is reader, it's guarded by team lock. It's not possible
1816	 * to traverse list in reverse under rcu_read_lock
1817	 */
1818	mutex_lock(&team->lock);
1819	list_for_each_entry(port, &team->port_list, list) {
1820		err = vlan_vid_add(port->dev, proto, vid);
1821		if (err)
1822			goto unwind;
1823	}
1824	mutex_unlock(&team->lock);
1825
1826	return 0;
1827
1828unwind:
1829	list_for_each_entry_continue_reverse(port, &team->port_list, list)
1830		vlan_vid_del(port->dev, proto, vid);
1831	mutex_unlock(&team->lock);
1832
1833	return err;
1834}
1835
1836static int team_vlan_rx_kill_vid(struct net_device *dev, __be16 proto, u16 vid)
1837{
1838	struct team *team = netdev_priv(dev);
1839	struct team_port *port;
1840
1841	rcu_read_lock();
1842	list_for_each_entry_rcu(port, &team->port_list, list)
1843		vlan_vid_del(port->dev, proto, vid);
1844	rcu_read_unlock();
1845
1846	return 0;
1847}
1848
1849#ifdef CONFIG_NET_POLL_CONTROLLER
1850static void team_poll_controller(struct net_device *dev)
1851{
1852}
1853
1854static void __team_netpoll_cleanup(struct team *team)
1855{
1856	struct team_port *port;
1857
1858	list_for_each_entry(port, &team->port_list, list)
1859		team_port_disable_netpoll(port);
1860}
1861
1862static void team_netpoll_cleanup(struct net_device *dev)
1863{
1864	struct team *team = netdev_priv(dev);
1865
1866	mutex_lock(&team->lock);
1867	__team_netpoll_cleanup(team);
1868	mutex_unlock(&team->lock);
1869}
1870
1871static int team_netpoll_setup(struct net_device *dev,
1872			      struct netpoll_info *npifo)
1873{
1874	struct team *team = netdev_priv(dev);
1875	struct team_port *port;
1876	int err = 0;
1877
1878	mutex_lock(&team->lock);
1879	list_for_each_entry(port, &team->port_list, list) {
1880		err = team_port_enable_netpoll(team, port);
1881		if (err) {
1882			__team_netpoll_cleanup(team);
1883			break;
1884		}
1885	}
1886	mutex_unlock(&team->lock);
1887	return err;
1888}
1889#endif
1890
1891static int team_add_slave(struct net_device *dev, struct net_device *port_dev)
1892{
1893	struct team *team = netdev_priv(dev);
1894	int err;
1895
1896	mutex_lock(&team->lock);
1897	err = team_port_add(team, port_dev);
1898	mutex_unlock(&team->lock);
1899	return err;
1900}
1901
1902static int team_del_slave(struct net_device *dev, struct net_device *port_dev)
1903{
1904	struct team *team = netdev_priv(dev);
1905	int err;
1906
1907	mutex_lock(&team->lock);
1908	err = team_port_del(team, port_dev);
1909	mutex_unlock(&team->lock);
1910	return err;
1911}
1912
1913static netdev_features_t team_fix_features(struct net_device *dev,
1914					   netdev_features_t features)
1915{
1916	struct team_port *port;
1917	struct team *team = netdev_priv(dev);
1918	netdev_features_t mask;
1919
1920	mask = features;
1921	features &= ~NETIF_F_ONE_FOR_ALL;
1922	features |= NETIF_F_ALL_FOR_ALL;
1923
1924	rcu_read_lock();
1925	list_for_each_entry_rcu(port, &team->port_list, list) {
1926		features = netdev_increment_features(features,
1927						     port->dev->features,
1928						     mask);
1929	}
1930	rcu_read_unlock();
1931	return features;
1932}
1933
1934static int team_change_carrier(struct net_device *dev, bool new_carrier)
1935{
1936	struct team *team = netdev_priv(dev);
1937
1938	team->user_carrier_enabled = true;
1939
1940	if (new_carrier)
1941		netif_carrier_on(dev);
1942	else
1943		netif_carrier_off(dev);
1944	return 0;
1945}
1946
1947static const struct net_device_ops team_netdev_ops = {
1948	.ndo_init		= team_init,
1949	.ndo_uninit		= team_uninit,
1950	.ndo_open		= team_open,
1951	.ndo_stop		= team_close,
1952	.ndo_start_xmit		= team_xmit,
1953	.ndo_select_queue	= team_select_queue,
1954	.ndo_change_rx_flags	= team_change_rx_flags,
1955	.ndo_set_rx_mode	= team_set_rx_mode,
1956	.ndo_set_mac_address	= team_set_mac_address,
1957	.ndo_change_mtu		= team_change_mtu,
1958	.ndo_get_stats64	= team_get_stats64,
1959	.ndo_vlan_rx_add_vid	= team_vlan_rx_add_vid,
1960	.ndo_vlan_rx_kill_vid	= team_vlan_rx_kill_vid,
1961#ifdef CONFIG_NET_POLL_CONTROLLER
1962	.ndo_poll_controller	= team_poll_controller,
1963	.ndo_netpoll_setup	= team_netpoll_setup,
1964	.ndo_netpoll_cleanup	= team_netpoll_cleanup,
1965#endif
1966	.ndo_add_slave		= team_add_slave,
1967	.ndo_del_slave		= team_del_slave,
1968	.ndo_fix_features	= team_fix_features,
1969	.ndo_change_carrier     = team_change_carrier,
1970};
1971
1972/***********************
1973 * ethtool interface
1974 ***********************/
1975
1976static void team_ethtool_get_drvinfo(struct net_device *dev,
1977				     struct ethtool_drvinfo *drvinfo)
1978{
1979	strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
1980	strlcpy(drvinfo->version, UTS_RELEASE, sizeof(drvinfo->version));
1981}
1982
1983static const struct ethtool_ops team_ethtool_ops = {
1984	.get_drvinfo		= team_ethtool_get_drvinfo,
1985	.get_link		= ethtool_op_get_link,
1986};
1987
1988/***********************
1989 * rt netlink interface
1990 ***********************/
1991
1992static void team_setup_by_port(struct net_device *dev,
1993			       struct net_device *port_dev)
1994{
1995	dev->header_ops	= port_dev->header_ops;
1996	dev->type = port_dev->type;
1997	dev->hard_header_len = port_dev->hard_header_len;
1998	dev->addr_len = port_dev->addr_len;
1999	dev->mtu = port_dev->mtu;
2000	memcpy(dev->broadcast, port_dev->broadcast, port_dev->addr_len);
2001	eth_hw_addr_inherit(dev, port_dev);
2002}
2003
2004static int team_dev_type_check_change(struct net_device *dev,
2005				      struct net_device *port_dev)
2006{
2007	struct team *team = netdev_priv(dev);
2008	char *portname = port_dev->name;
2009	int err;
2010
2011	if (dev->type == port_dev->type)
2012		return 0;
2013	if (!list_empty(&team->port_list)) {
2014		netdev_err(dev, "Device %s is of different type\n", portname);
2015		return -EBUSY;
2016	}
2017	err = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE, dev);
2018	err = notifier_to_errno(err);
2019	if (err) {
2020		netdev_err(dev, "Refused to change device type\n");
2021		return err;
2022	}
2023	dev_uc_flush(dev);
2024	dev_mc_flush(dev);
2025	team_setup_by_port(dev, port_dev);
2026	call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE, dev);
2027	return 0;
2028}
2029
2030static void team_setup(struct net_device *dev)
2031{
2032	ether_setup(dev);
2033
2034	dev->netdev_ops = &team_netdev_ops;
2035	dev->ethtool_ops = &team_ethtool_ops;
2036	dev->destructor	= team_destructor;
2037	dev->tx_queue_len = 0;
2038	dev->flags |= IFF_MULTICAST;
2039	dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
2040
2041	/*
2042	 * Indicate we support unicast address filtering. That way core won't
2043	 * bring us to promisc mode in case a unicast addr is added.
2044	 * Let this up to underlay drivers.
2045	 */
2046	dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
2047
2048	dev->features |= NETIF_F_LLTX;
2049	dev->features |= NETIF_F_GRO;
2050
2051	/* Don't allow team devices to change network namespaces. */
2052	dev->features |= NETIF_F_NETNS_LOCAL;
2053
2054	dev->hw_features = TEAM_VLAN_FEATURES |
2055			   NETIF_F_HW_VLAN_CTAG_TX |
2056			   NETIF_F_HW_VLAN_CTAG_RX |
2057			   NETIF_F_HW_VLAN_CTAG_FILTER;
2058
2059	dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
2060	dev->features |= dev->hw_features;
2061}
2062
2063static int team_newlink(struct net *src_net, struct net_device *dev,
2064			struct nlattr *tb[], struct nlattr *data[])
2065{
2066	if (tb[IFLA_ADDRESS] == NULL)
2067		eth_hw_addr_random(dev);
2068
2069	return register_netdevice(dev);
2070}
2071
2072static int team_validate(struct nlattr *tb[], struct nlattr *data[])
2073{
2074	if (tb[IFLA_ADDRESS]) {
2075		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
2076			return -EINVAL;
2077		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
2078			return -EADDRNOTAVAIL;
2079	}
2080	return 0;
2081}
2082
2083static unsigned int team_get_num_tx_queues(void)
2084{
2085	return TEAM_DEFAULT_NUM_TX_QUEUES;
2086}
2087
2088static unsigned int team_get_num_rx_queues(void)
2089{
2090	return TEAM_DEFAULT_NUM_RX_QUEUES;
2091}
2092
2093static struct rtnl_link_ops team_link_ops __read_mostly = {
2094	.kind			= DRV_NAME,
2095	.priv_size		= sizeof(struct team),
2096	.setup			= team_setup,
2097	.newlink		= team_newlink,
2098	.validate		= team_validate,
2099	.get_num_tx_queues	= team_get_num_tx_queues,
2100	.get_num_rx_queues	= team_get_num_rx_queues,
2101};
2102
2103
2104/***********************************
2105 * Generic netlink custom interface
2106 ***********************************/
2107
2108static struct genl_family team_nl_family = {
2109	.id		= GENL_ID_GENERATE,
2110	.name		= TEAM_GENL_NAME,
2111	.version	= TEAM_GENL_VERSION,
2112	.maxattr	= TEAM_ATTR_MAX,
2113	.netnsok	= true,
2114};
2115
2116static const struct nla_policy team_nl_policy[TEAM_ATTR_MAX + 1] = {
2117	[TEAM_ATTR_UNSPEC]			= { .type = NLA_UNSPEC, },
2118	[TEAM_ATTR_TEAM_IFINDEX]		= { .type = NLA_U32 },
2119	[TEAM_ATTR_LIST_OPTION]			= { .type = NLA_NESTED },
2120	[TEAM_ATTR_LIST_PORT]			= { .type = NLA_NESTED },
2121};
2122
2123static const struct nla_policy
2124team_nl_option_policy[TEAM_ATTR_OPTION_MAX + 1] = {
2125	[TEAM_ATTR_OPTION_UNSPEC]		= { .type = NLA_UNSPEC, },
2126	[TEAM_ATTR_OPTION_NAME] = {
2127		.type = NLA_STRING,
2128		.len = TEAM_STRING_MAX_LEN,
2129	},
2130	[TEAM_ATTR_OPTION_CHANGED]		= { .type = NLA_FLAG },
2131	[TEAM_ATTR_OPTION_TYPE]			= { .type = NLA_U8 },
2132	[TEAM_ATTR_OPTION_DATA]			= { .type = NLA_BINARY },
2133};
2134
2135static int team_nl_cmd_noop(struct sk_buff *skb, struct genl_info *info)
2136{
2137	struct sk_buff *msg;
2138	void *hdr;
2139	int err;
2140
2141	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2142	if (!msg)
2143		return -ENOMEM;
2144
2145	hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
2146			  &team_nl_family, 0, TEAM_CMD_NOOP);
2147	if (!hdr) {
2148		err = -EMSGSIZE;
2149		goto err_msg_put;
2150	}
2151
2152	genlmsg_end(msg, hdr);
2153
2154	return genlmsg_unicast(genl_info_net(info), msg, info->snd_portid);
2155
2156err_msg_put:
2157	nlmsg_free(msg);
2158
2159	return err;
2160}
2161
2162/*
2163 * Netlink cmd functions should be locked by following two functions.
2164 * Since dev gets held here, that ensures dev won't disappear in between.
2165 */
2166static struct team *team_nl_team_get(struct genl_info *info)
2167{
2168	struct net *net = genl_info_net(info);
2169	int ifindex;
2170	struct net_device *dev;
2171	struct team *team;
2172
2173	if (!info->attrs[TEAM_ATTR_TEAM_IFINDEX])
2174		return NULL;
2175
2176	ifindex = nla_get_u32(info->attrs[TEAM_ATTR_TEAM_IFINDEX]);
2177	dev = dev_get_by_index(net, ifindex);
2178	if (!dev || dev->netdev_ops != &team_netdev_ops) {
2179		if (dev)
2180			dev_put(dev);
2181		return NULL;
2182	}
2183
2184	team = netdev_priv(dev);
2185	mutex_lock(&team->lock);
2186	return team;
2187}
2188
2189static void team_nl_team_put(struct team *team)
2190{
2191	mutex_unlock(&team->lock);
2192	dev_put(team->dev);
2193}
2194
2195typedef int team_nl_send_func_t(struct sk_buff *skb,
2196				struct team *team, u32 portid);
2197
2198static int team_nl_send_unicast(struct sk_buff *skb, struct team *team, u32 portid)
2199{
2200	return genlmsg_unicast(dev_net(team->dev), skb, portid);
2201}
2202
2203static int team_nl_fill_one_option_get(struct sk_buff *skb, struct team *team,
2204				       struct team_option_inst *opt_inst)
2205{
2206	struct nlattr *option_item;
2207	struct team_option *option = opt_inst->option;
2208	struct team_option_inst_info *opt_inst_info = &opt_inst->info;
2209	struct team_gsetter_ctx ctx;
2210	int err;
2211
2212	ctx.info = opt_inst_info;
2213	err = team_option_get(team, opt_inst, &ctx);
2214	if (err)
2215		return err;
2216
2217	option_item = nla_nest_start(skb, TEAM_ATTR_ITEM_OPTION);
2218	if (!option_item)
2219		return -EMSGSIZE;
2220
2221	if (nla_put_string(skb, TEAM_ATTR_OPTION_NAME, option->name))
2222		goto nest_cancel;
2223	if (opt_inst_info->port &&
2224	    nla_put_u32(skb, TEAM_ATTR_OPTION_PORT_IFINDEX,
2225			opt_inst_info->port->dev->ifindex))
2226		goto nest_cancel;
2227	if (opt_inst->option->array_size &&
2228	    nla_put_u32(skb, TEAM_ATTR_OPTION_ARRAY_INDEX,
2229			opt_inst_info->array_index))
2230		goto nest_cancel;
2231
2232	switch (option->type) {
2233	case TEAM_OPTION_TYPE_U32:
2234		if (nla_put_u8(skb, TEAM_ATTR_OPTION_TYPE, NLA_U32))
2235			goto nest_cancel;
2236		if (nla_put_u32(skb, TEAM_ATTR_OPTION_DATA, ctx.data.u32_val))
2237			goto nest_cancel;
2238		break;
2239	case TEAM_OPTION_TYPE_STRING:
2240		if (nla_put_u8(skb, TEAM_ATTR_OPTION_TYPE, NLA_STRING))
2241			goto nest_cancel;
2242		if (nla_put_string(skb, TEAM_ATTR_OPTION_DATA,
2243				   ctx.data.str_val))
2244			goto nest_cancel;
2245		break;
2246	case TEAM_OPTION_TYPE_BINARY:
2247		if (nla_put_u8(skb, TEAM_ATTR_OPTION_TYPE, NLA_BINARY))
2248			goto nest_cancel;
2249		if (nla_put(skb, TEAM_ATTR_OPTION_DATA, ctx.data.bin_val.len,
2250			    ctx.data.bin_val.ptr))
2251			goto nest_cancel;
2252		break;
2253	case TEAM_OPTION_TYPE_BOOL:
2254		if (nla_put_u8(skb, TEAM_ATTR_OPTION_TYPE, NLA_FLAG))
2255			goto nest_cancel;
2256		if (ctx.data.bool_val &&
2257		    nla_put_flag(skb, TEAM_ATTR_OPTION_DATA))
2258			goto nest_cancel;
2259		break;
2260	case TEAM_OPTION_TYPE_S32:
2261		if (nla_put_u8(skb, TEAM_ATTR_OPTION_TYPE, NLA_S32))
2262			goto nest_cancel;
2263		if (nla_put_s32(skb, TEAM_ATTR_OPTION_DATA, ctx.data.s32_val))
2264			goto nest_cancel;
2265		break;
2266	default:
2267		BUG();
2268	}
2269	if (opt_inst->removed && nla_put_flag(skb, TEAM_ATTR_OPTION_REMOVED))
2270		goto nest_cancel;
2271	if (opt_inst->changed) {
2272		if (nla_put_flag(skb, TEAM_ATTR_OPTION_CHANGED))
2273			goto nest_cancel;
2274		opt_inst->changed = false;
2275	}
2276	nla_nest_end(skb, option_item);
2277	return 0;
2278
2279nest_cancel:
2280	nla_nest_cancel(skb, option_item);
2281	return -EMSGSIZE;
2282}
2283
2284static int __send_and_alloc_skb(struct sk_buff **pskb,
2285				struct team *team, u32 portid,
2286				team_nl_send_func_t *send_func)
2287{
2288	int err;
2289
2290	if (*pskb) {
2291		err = send_func(*pskb, team, portid);
2292		if (err)
2293			return err;
2294	}
2295	*pskb = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL);
2296	if (!*pskb)
2297		return -ENOMEM;
2298	return 0;
2299}
2300
2301static int team_nl_send_options_get(struct team *team, u32 portid, u32 seq,
2302				    int flags, team_nl_send_func_t *send_func,
2303				    struct list_head *sel_opt_inst_list)
2304{
2305	struct nlattr *option_list;
2306	struct nlmsghdr *nlh;
2307	void *hdr;
2308	struct team_option_inst *opt_inst;
2309	int err;
2310	struct sk_buff *skb = NULL;
2311	bool incomplete;
2312	int i;
2313
2314	opt_inst = list_first_entry(sel_opt_inst_list,
2315				    struct team_option_inst, tmp_list);
2316
2317start_again:
2318	err = __send_and_alloc_skb(&skb, team, portid, send_func);
2319	if (err)
2320		return err;
2321
2322	hdr = genlmsg_put(skb, portid, seq, &team_nl_family, flags | NLM_F_MULTI,
2323			  TEAM_CMD_OPTIONS_GET);
2324	if (!hdr)
2325		return -EMSGSIZE;
2326
2327	if (nla_put_u32(skb, TEAM_ATTR_TEAM_IFINDEX, team->dev->ifindex))
2328		goto nla_put_failure;
2329	option_list = nla_nest_start(skb, TEAM_ATTR_LIST_OPTION);
2330	if (!option_list)
2331		goto nla_put_failure;
2332
2333	i = 0;
2334	incomplete = false;
2335	list_for_each_entry_from(opt_inst, sel_opt_inst_list, tmp_list) {
2336		err = team_nl_fill_one_option_get(skb, team, opt_inst);
2337		if (err) {
2338			if (err == -EMSGSIZE) {
2339				if (!i)
2340					goto errout;
2341				incomplete = true;
2342				break;
2343			}
2344			goto errout;
2345		}
2346		i++;
2347	}
2348
2349	nla_nest_end(skb, option_list);
2350	genlmsg_end(skb, hdr);
2351	if (incomplete)
2352		goto start_again;
2353
2354send_done:
2355	nlh = nlmsg_put(skb, portid, seq, NLMSG_DONE, 0, flags | NLM_F_MULTI);
2356	if (!nlh) {
2357		err = __send_and_alloc_skb(&skb, team, portid, send_func);
2358		if (err)
2359			goto errout;
2360		goto send_done;
2361	}
2362
2363	return send_func(skb, team, portid);
2364
2365nla_put_failure:
2366	err = -EMSGSIZE;
2367errout:
2368	genlmsg_cancel(skb, hdr);
2369	nlmsg_free(skb);
2370	return err;
2371}
2372
2373static int team_nl_cmd_options_get(struct sk_buff *skb, struct genl_info *info)
2374{
2375	struct team *team;
2376	struct team_option_inst *opt_inst;
2377	int err;
2378	LIST_HEAD(sel_opt_inst_list);
2379
2380	team = team_nl_team_get(info);
2381	if (!team)
2382		return -EINVAL;
2383
2384	list_for_each_entry(opt_inst, &team->option_inst_list, list)
2385		list_add_tail(&opt_inst->tmp_list, &sel_opt_inst_list);
2386	err = team_nl_send_options_get(team, info->snd_portid, info->snd_seq,
2387				       NLM_F_ACK, team_nl_send_unicast,
2388				       &sel_opt_inst_list);
2389
2390	team_nl_team_put(team);
2391
2392	return err;
2393}
2394
2395static int team_nl_send_event_options_get(struct team *team,
2396					  struct list_head *sel_opt_inst_list);
2397
2398static int team_nl_cmd_options_set(struct sk_buff *skb, struct genl_info *info)
2399{
2400	struct team *team;
2401	int err = 0;
2402	int i;
2403	struct nlattr *nl_option;
2404	LIST_HEAD(opt_inst_list);
2405
2406	team = team_nl_team_get(info);
2407	if (!team)
2408		return -EINVAL;
2409
2410	err = -EINVAL;
2411	if (!info->attrs[TEAM_ATTR_LIST_OPTION]) {
2412		err = -EINVAL;
2413		goto team_put;
2414	}
2415
2416	nla_for_each_nested(nl_option, info->attrs[TEAM_ATTR_LIST_OPTION], i) {
2417		struct nlattr *opt_attrs[TEAM_ATTR_OPTION_MAX + 1];
2418		struct nlattr *attr;
2419		struct nlattr *attr_data;
2420		enum team_option_type opt_type;
2421		int opt_port_ifindex = 0; /* != 0 for per-port options */
2422		u32 opt_array_index = 0;
2423		bool opt_is_array = false;
2424		struct team_option_inst *opt_inst;
2425		char *opt_name;
2426		bool opt_found = false;
2427
2428		if (nla_type(nl_option) != TEAM_ATTR_ITEM_OPTION) {
2429			err = -EINVAL;
2430			goto team_put;
2431		}
2432		err = nla_parse_nested(opt_attrs, TEAM_ATTR_OPTION_MAX,
2433				       nl_option, team_nl_option_policy);
2434		if (err)
2435			goto team_put;
2436		if (!opt_attrs[TEAM_ATTR_OPTION_NAME] ||
2437		    !opt_attrs[TEAM_ATTR_OPTION_TYPE]) {
2438			err = -EINVAL;
2439			goto team_put;
2440		}
2441		switch (nla_get_u8(opt_attrs[TEAM_ATTR_OPTION_TYPE])) {
2442		case NLA_U32:
2443			opt_type = TEAM_OPTION_TYPE_U32;
2444			break;
2445		case NLA_STRING:
2446			opt_type = TEAM_OPTION_TYPE_STRING;
2447			break;
2448		case NLA_BINARY:
2449			opt_type = TEAM_OPTION_TYPE_BINARY;
2450			break;
2451		case NLA_FLAG:
2452			opt_type = TEAM_OPTION_TYPE_BOOL;
2453			break;
2454		case NLA_S32:
2455			opt_type = TEAM_OPTION_TYPE_S32;
2456			break;
2457		default:
2458			goto team_put;
2459		}
2460
2461		attr_data = opt_attrs[TEAM_ATTR_OPTION_DATA];
2462		if (opt_type != TEAM_OPTION_TYPE_BOOL && !attr_data) {
2463			err = -EINVAL;
2464			goto team_put;
2465		}
2466
2467		opt_name = nla_data(opt_attrs[TEAM_ATTR_OPTION_NAME]);
2468		attr = opt_attrs[TEAM_ATTR_OPTION_PORT_IFINDEX];
2469		if (attr)
2470			opt_port_ifindex = nla_get_u32(attr);
2471
2472		attr = opt_attrs[TEAM_ATTR_OPTION_ARRAY_INDEX];
2473		if (attr) {
2474			opt_is_array = true;
2475			opt_array_index = nla_get_u32(attr);
2476		}
2477
2478		list_for_each_entry(opt_inst, &team->option_inst_list, list) {
2479			struct team_option *option = opt_inst->option;
2480			struct team_gsetter_ctx ctx;
2481			struct team_option_inst_info *opt_inst_info;
2482			int tmp_ifindex;
2483
2484			opt_inst_info = &opt_inst->info;
2485			tmp_ifindex = opt_inst_info->port ?
2486				      opt_inst_info->port->dev->ifindex : 0;
2487			if (option->type != opt_type ||
2488			    strcmp(option->name, opt_name) ||
2489			    tmp_ifindex != opt_port_ifindex ||
2490			    (option->array_size && !opt_is_array) ||
2491			    opt_inst_info->array_index != opt_array_index)
2492				continue;
2493			opt_found = true;
2494			ctx.info = opt_inst_info;
2495			switch (opt_type) {
2496			case TEAM_OPTION_TYPE_U32:
2497				ctx.data.u32_val = nla_get_u32(attr_data);
2498				break;
2499			case TEAM_OPTION_TYPE_STRING:
2500				if (nla_len(attr_data) > TEAM_STRING_MAX_LEN) {
2501					err = -EINVAL;
2502					goto team_put;
2503				}
2504				ctx.data.str_val = nla_data(attr_data);
2505				break;
2506			case TEAM_OPTION_TYPE_BINARY:
2507				ctx.data.bin_val.len = nla_len(attr_data);
2508				ctx.data.bin_val.ptr = nla_data(attr_data);
2509				break;
2510			case TEAM_OPTION_TYPE_BOOL:
2511				ctx.data.bool_val = attr_data ? true : false;
2512				break;
2513			case TEAM_OPTION_TYPE_S32:
2514				ctx.data.s32_val = nla_get_s32(attr_data);
2515				break;
2516			default:
2517				BUG();
2518			}
2519			err = team_option_set(team, opt_inst, &ctx);
2520			if (err)
2521				goto team_put;
2522			opt_inst->changed = true;
2523			list_add(&opt_inst->tmp_list, &opt_inst_list);
2524		}
2525		if (!opt_found) {
2526			err = -ENOENT;
2527			goto team_put;
2528		}
2529	}
2530
2531	err = team_nl_send_event_options_get(team, &opt_inst_list);
2532
2533team_put:
2534	team_nl_team_put(team);
2535
2536	return err;
2537}
2538
2539static int team_nl_fill_one_port_get(struct sk_buff *skb,
2540				     struct team_port *port)
2541{
2542	struct nlattr *port_item;
2543
2544	port_item = nla_nest_start(skb, TEAM_ATTR_ITEM_PORT);
2545	if (!port_item)
2546		goto nest_cancel;
2547	if (nla_put_u32(skb, TEAM_ATTR_PORT_IFINDEX, port->dev->ifindex))
2548		goto nest_cancel;
2549	if (port->changed) {
2550		if (nla_put_flag(skb, TEAM_ATTR_PORT_CHANGED))
2551			goto nest_cancel;
2552		port->changed = false;
2553	}
2554	if ((port->removed &&
2555	     nla_put_flag(skb, TEAM_ATTR_PORT_REMOVED)) ||
2556	    (port->state.linkup &&
2557	     nla_put_flag(skb, TEAM_ATTR_PORT_LINKUP)) ||
2558	    nla_put_u32(skb, TEAM_ATTR_PORT_SPEED, port->state.speed) ||
2559	    nla_put_u8(skb, TEAM_ATTR_PORT_DUPLEX, port->state.duplex))
2560		goto nest_cancel;
2561	nla_nest_end(skb, port_item);
2562	return 0;
2563
2564nest_cancel:
2565	nla_nest_cancel(skb, port_item);
2566	return -EMSGSIZE;
2567}
2568
2569static int team_nl_send_port_list_get(struct team *team, u32 portid, u32 seq,
2570				      int flags, team_nl_send_func_t *send_func,
2571				      struct team_port *one_port)
2572{
2573	struct nlattr *port_list;
2574	struct nlmsghdr *nlh;
2575	void *hdr;
2576	struct team_port *port;
2577	int err;
2578	struct sk_buff *skb = NULL;
2579	bool incomplete;
2580	int i;
2581
2582	port = list_first_entry_or_null(&team->port_list,
2583					struct team_port, list);
2584
2585start_again:
2586	err = __send_and_alloc_skb(&skb, team, portid, send_func);
2587	if (err)
2588		return err;
2589
2590	hdr = genlmsg_put(skb, portid, seq, &team_nl_family, flags | NLM_F_MULTI,
2591			  TEAM_CMD_PORT_LIST_GET);
2592	if (!hdr)
2593		return -EMSGSIZE;
2594
2595	if (nla_put_u32(skb, TEAM_ATTR_TEAM_IFINDEX, team->dev->ifindex))
2596		goto nla_put_failure;
2597	port_list = nla_nest_start(skb, TEAM_ATTR_LIST_PORT);
2598	if (!port_list)
2599		goto nla_put_failure;
2600
2601	i = 0;
2602	incomplete = false;
2603
2604	/* If one port is selected, called wants to send port list containing
2605	 * only this port. Otherwise go through all listed ports and send all
2606	 */
2607	if (one_port) {
2608		err = team_nl_fill_one_port_get(skb, one_port);
2609		if (err)
2610			goto errout;
2611	} else if (port) {
2612		list_for_each_entry_from(port, &team->port_list, list) {
2613			err = team_nl_fill_one_port_get(skb, port);
2614			if (err) {
2615				if (err == -EMSGSIZE) {
2616					if (!i)
2617						goto errout;
2618					incomplete = true;
2619					break;
2620				}
2621				goto errout;
2622			}
2623			i++;
2624		}
2625	}
2626
2627	nla_nest_end(skb, port_list);
2628	genlmsg_end(skb, hdr);
2629	if (incomplete)
2630		goto start_again;
2631
2632send_done:
2633	nlh = nlmsg_put(skb, portid, seq, NLMSG_DONE, 0, flags | NLM_F_MULTI);
2634	if (!nlh) {
2635		err = __send_and_alloc_skb(&skb, team, portid, send_func);
2636		if (err)
2637			goto errout;
2638		goto send_done;
2639	}
2640
2641	return send_func(skb, team, portid);
2642
2643nla_put_failure:
2644	err = -EMSGSIZE;
2645errout:
2646	genlmsg_cancel(skb, hdr);
2647	nlmsg_free(skb);
2648	return err;
2649}
2650
2651static int team_nl_cmd_port_list_get(struct sk_buff *skb,
2652				     struct genl_info *info)
2653{
2654	struct team *team;
2655	int err;
2656
2657	team = team_nl_team_get(info);
2658	if (!team)
2659		return -EINVAL;
2660
2661	err = team_nl_send_port_list_get(team, info->snd_portid, info->snd_seq,
2662					 NLM_F_ACK, team_nl_send_unicast, NULL);
2663
2664	team_nl_team_put(team);
2665
2666	return err;
2667}
2668
2669static const struct genl_ops team_nl_ops[] = {
2670	{
2671		.cmd = TEAM_CMD_NOOP,
2672		.doit = team_nl_cmd_noop,
2673		.policy = team_nl_policy,
2674	},
2675	{
2676		.cmd = TEAM_CMD_OPTIONS_SET,
2677		.doit = team_nl_cmd_options_set,
2678		.policy = team_nl_policy,
2679		.flags = GENL_ADMIN_PERM,
2680	},
2681	{
2682		.cmd = TEAM_CMD_OPTIONS_GET,
2683		.doit = team_nl_cmd_options_get,
2684		.policy = team_nl_policy,
2685		.flags = GENL_ADMIN_PERM,
2686	},
2687	{
2688		.cmd = TEAM_CMD_PORT_LIST_GET,
2689		.doit = team_nl_cmd_port_list_get,
2690		.policy = team_nl_policy,
2691		.flags = GENL_ADMIN_PERM,
2692	},
2693};
2694
2695static const struct genl_multicast_group team_nl_mcgrps[] = {
2696	{ .name = TEAM_GENL_CHANGE_EVENT_MC_GRP_NAME, },
2697};
2698
2699static int team_nl_send_multicast(struct sk_buff *skb,
2700				  struct team *team, u32 portid)
2701{
2702	return genlmsg_multicast_netns(&team_nl_family, dev_net(team->dev),
2703				       skb, 0, 0, GFP_KERNEL);
2704}
2705
2706static int team_nl_send_event_options_get(struct team *team,
2707					  struct list_head *sel_opt_inst_list)
2708{
2709	return team_nl_send_options_get(team, 0, 0, 0, team_nl_send_multicast,
2710					sel_opt_inst_list);
2711}
2712
2713static int team_nl_send_event_port_get(struct team *team,
2714				       struct team_port *port)
2715{
2716	return team_nl_send_port_list_get(team, 0, 0, 0, team_nl_send_multicast,
2717					  port);
2718}
2719
2720static int team_nl_init(void)
2721{
2722	return genl_register_family_with_ops_groups(&team_nl_family, team_nl_ops,
2723						    team_nl_mcgrps);
2724}
2725
2726static void team_nl_fini(void)
2727{
2728	genl_unregister_family(&team_nl_family);
2729}
2730
2731
2732/******************
2733 * Change checkers
2734 ******************/
2735
2736static void __team_options_change_check(struct team *team)
2737{
2738	int err;
2739	struct team_option_inst *opt_inst;
2740	LIST_HEAD(sel_opt_inst_list);
2741
2742	list_for_each_entry(opt_inst, &team->option_inst_list, list) {
2743		if (opt_inst->changed)
2744			list_add_tail(&opt_inst->tmp_list, &sel_opt_inst_list);
2745	}
2746	err = team_nl_send_event_options_get(team, &sel_opt_inst_list);
2747	if (err && err != -ESRCH)
2748		netdev_warn(team->dev, "Failed to send options change via netlink (err %d)\n",
2749			    err);
2750}
2751
2752/* rtnl lock is held */
2753
2754static void __team_port_change_send(struct team_port *port, bool linkup)
2755{
2756	int err;
2757
2758	port->changed = true;
2759	port->state.linkup = linkup;
2760	team_refresh_port_linkup(port);
2761	if (linkup) {
2762		struct ethtool_cmd ecmd;
2763
2764		err = __ethtool_get_settings(port->dev, &ecmd);
2765		if (!err) {
2766			port->state.speed = ethtool_cmd_speed(&ecmd);
2767			port->state.duplex = ecmd.duplex;
2768			goto send_event;
2769		}
2770	}
2771	port->state.speed = 0;
2772	port->state.duplex = 0;
2773
2774send_event:
2775	err = team_nl_send_event_port_get(port->team, port);
2776	if (err && err != -ESRCH)
2777		netdev_warn(port->team->dev, "Failed to send port change of device %s via netlink (err %d)\n",
2778			    port->dev->name, err);
2779
2780}
2781
2782static void __team_carrier_check(struct team *team)
2783{
2784	struct team_port *port;
2785	bool team_linkup;
2786
2787	if (team->user_carrier_enabled)
2788		return;
2789
2790	team_linkup = false;
2791	list_for_each_entry(port, &team->port_list, list) {
2792		if (port->linkup) {
2793			team_linkup = true;
2794			break;
2795		}
2796	}
2797
2798	if (team_linkup)
2799		netif_carrier_on(team->dev);
2800	else
2801		netif_carrier_off(team->dev);
2802}
2803
2804static void __team_port_change_check(struct team_port *port, bool linkup)
2805{
2806	if (port->state.linkup != linkup)
2807		__team_port_change_send(port, linkup);
2808	__team_carrier_check(port->team);
2809}
2810
2811static void __team_port_change_port_added(struct team_port *port, bool linkup)
2812{
2813	__team_port_change_send(port, linkup);
2814	__team_carrier_check(port->team);
2815}
2816
2817static void __team_port_change_port_removed(struct team_port *port)
2818{
2819	port->removed = true;
2820	__team_port_change_send(port, false);
2821	__team_carrier_check(port->team);
2822}
2823
2824static void team_port_change_check(struct team_port *port, bool linkup)
2825{
2826	struct team *team = port->team;
2827
2828	mutex_lock(&team->lock);
2829	__team_port_change_check(port, linkup);
2830	mutex_unlock(&team->lock);
2831}
2832
2833
2834/************************************
2835 * Net device notifier event handler
2836 ************************************/
2837
2838static int team_device_event(struct notifier_block *unused,
2839			     unsigned long event, void *ptr)
2840{
2841	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2842	struct team_port *port;
2843
2844	port = team_port_get_rtnl(dev);
2845	if (!port)
2846		return NOTIFY_DONE;
2847
2848	switch (event) {
2849	case NETDEV_UP:
2850		if (netif_carrier_ok(dev))
2851			team_port_change_check(port, true);
2852		break;
2853	case NETDEV_DOWN:
2854		team_port_change_check(port, false);
2855		break;
2856	case NETDEV_CHANGE:
2857		if (netif_running(port->dev))
2858			team_port_change_check(port,
2859					       !!netif_carrier_ok(port->dev));
2860		break;
2861	case NETDEV_UNREGISTER:
2862		team_del_slave(port->team->dev, dev);
2863		break;
2864	case NETDEV_FEAT_CHANGE:
2865		team_compute_features(port->team);
2866		break;
2867	case NETDEV_PRECHANGEMTU:
2868		/* Forbid to change mtu of underlaying device */
2869		if (!port->team->port_mtu_change_allowed)
2870			return NOTIFY_BAD;
2871		break;
2872	case NETDEV_PRE_TYPE_CHANGE:
2873		/* Forbid to change type of underlaying device */
2874		return NOTIFY_BAD;
2875	case NETDEV_RESEND_IGMP:
2876		/* Propagate to master device */
2877		call_netdevice_notifiers(event, port->team->dev);
2878		break;
2879	}
2880	return NOTIFY_DONE;
2881}
2882
2883static struct notifier_block team_notifier_block __read_mostly = {
2884	.notifier_call = team_device_event,
2885};
2886
2887
2888/***********************
2889 * Module init and exit
2890 ***********************/
2891
2892static int __init team_module_init(void)
2893{
2894	int err;
2895
2896	register_netdevice_notifier(&team_notifier_block);
2897
2898	err = rtnl_link_register(&team_link_ops);
2899	if (err)
2900		goto err_rtnl_reg;
2901
2902	err = team_nl_init();
2903	if (err)
2904		goto err_nl_init;
2905
2906	return 0;
2907
2908err_nl_init:
2909	rtnl_link_unregister(&team_link_ops);
2910
2911err_rtnl_reg:
2912	unregister_netdevice_notifier(&team_notifier_block);
2913
2914	return err;
2915}
2916
2917static void __exit team_module_exit(void)
2918{
2919	team_nl_fini();
2920	rtnl_link_unregister(&team_link_ops);
2921	unregister_netdevice_notifier(&team_notifier_block);
2922}
2923
2924module_init(team_module_init);
2925module_exit(team_module_exit);
2926
2927MODULE_LICENSE("GPL v2");
2928MODULE_AUTHOR("Jiri Pirko <jpirko@redhat.com>");
2929MODULE_DESCRIPTION("Ethernet team device driver");
2930MODULE_ALIAS_RTNL_LINK(DRV_NAME);
2931