perf tools: Don't clone maps from parent when synthesizing forks
[linux/fpc-iii.git] / drivers / net / team / team.c
blobdb633ae9f784a7cf0128e922a046160d25b898e1
1 /*
2 * drivers/net/team/team.c - Network team device driver
3 * Copyright (c) 2011 Jiri Pirko <jpirko@redhat.com>
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 */
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 <net/switchdev.h>
32 #include <generated/utsrelease.h>
33 #include <linux/if_team.h>
35 #define DRV_NAME "team"
38 /**********
39 * Helpers
40 **********/
42 #define team_port_exists(dev) (dev->priv_flags & IFF_TEAM_PORT)
44 static struct team_port *team_port_get_rtnl(const struct net_device *dev)
46 struct team_port *port = rtnl_dereference(dev->rx_handler_data);
48 return team_port_exists(dev) ? port : NULL;
52 * Since the ability to change device address for open port device is tested in
53 * team_port_add, this function can be called without control of return value
55 static int __set_port_dev_addr(struct net_device *port_dev,
56 const unsigned char *dev_addr)
58 struct sockaddr_storage addr;
60 memcpy(addr.__data, dev_addr, port_dev->addr_len);
61 addr.ss_family = port_dev->type;
62 return dev_set_mac_address(port_dev, (struct sockaddr *)&addr);
65 static int team_port_set_orig_dev_addr(struct team_port *port)
67 return __set_port_dev_addr(port->dev, port->orig.dev_addr);
70 static int team_port_set_team_dev_addr(struct team *team,
71 struct team_port *port)
73 return __set_port_dev_addr(port->dev, team->dev->dev_addr);
76 int team_modeop_port_enter(struct team *team, struct team_port *port)
78 return team_port_set_team_dev_addr(team, port);
80 EXPORT_SYMBOL(team_modeop_port_enter);
82 void team_modeop_port_change_dev_addr(struct team *team,
83 struct team_port *port)
85 team_port_set_team_dev_addr(team, port);
87 EXPORT_SYMBOL(team_modeop_port_change_dev_addr);
89 static void team_lower_state_changed(struct team_port *port)
91 struct netdev_lag_lower_state_info info;
93 info.link_up = port->linkup;
94 info.tx_enabled = team_port_enabled(port);
95 netdev_lower_state_changed(port->dev, &info);
98 static void team_refresh_port_linkup(struct team_port *port)
100 bool new_linkup = port->user.linkup_enabled ? port->user.linkup :
101 port->state.linkup;
103 if (port->linkup != new_linkup) {
104 port->linkup = new_linkup;
105 team_lower_state_changed(port);
110 /*******************
111 * Options handling
112 *******************/
114 struct team_option_inst { /* One for each option instance */
115 struct list_head list;
116 struct list_head tmp_list;
117 struct team_option *option;
118 struct team_option_inst_info info;
119 bool changed;
120 bool removed;
123 static struct team_option *__team_find_option(struct team *team,
124 const char *opt_name)
126 struct team_option *option;
128 list_for_each_entry(option, &team->option_list, list) {
129 if (strcmp(option->name, opt_name) == 0)
130 return option;
132 return NULL;
135 static void __team_option_inst_del(struct team_option_inst *opt_inst)
137 list_del(&opt_inst->list);
138 kfree(opt_inst);
141 static void __team_option_inst_del_option(struct team *team,
142 struct team_option *option)
144 struct team_option_inst *opt_inst, *tmp;
146 list_for_each_entry_safe(opt_inst, tmp, &team->option_inst_list, list) {
147 if (opt_inst->option == option)
148 __team_option_inst_del(opt_inst);
152 static int __team_option_inst_add(struct team *team, struct team_option *option,
153 struct team_port *port)
155 struct team_option_inst *opt_inst;
156 unsigned int array_size;
157 unsigned int i;
158 int err;
160 array_size = option->array_size;
161 if (!array_size)
162 array_size = 1; /* No array but still need one instance */
164 for (i = 0; i < array_size; i++) {
165 opt_inst = kmalloc(sizeof(*opt_inst), GFP_KERNEL);
166 if (!opt_inst)
167 return -ENOMEM;
168 opt_inst->option = option;
169 opt_inst->info.port = port;
170 opt_inst->info.array_index = i;
171 opt_inst->changed = true;
172 opt_inst->removed = false;
173 list_add_tail(&opt_inst->list, &team->option_inst_list);
174 if (option->init) {
175 err = option->init(team, &opt_inst->info);
176 if (err)
177 return err;
181 return 0;
184 static int __team_option_inst_add_option(struct team *team,
185 struct team_option *option)
187 int err;
189 if (!option->per_port) {
190 err = __team_option_inst_add(team, option, NULL);
191 if (err)
192 goto inst_del_option;
194 return 0;
196 inst_del_option:
197 __team_option_inst_del_option(team, option);
198 return err;
201 static void __team_option_inst_mark_removed_option(struct team *team,
202 struct team_option *option)
204 struct team_option_inst *opt_inst;
206 list_for_each_entry(opt_inst, &team->option_inst_list, list) {
207 if (opt_inst->option == option) {
208 opt_inst->changed = true;
209 opt_inst->removed = true;
214 static void __team_option_inst_del_port(struct team *team,
215 struct team_port *port)
217 struct team_option_inst *opt_inst, *tmp;
219 list_for_each_entry_safe(opt_inst, tmp, &team->option_inst_list, list) {
220 if (opt_inst->option->per_port &&
221 opt_inst->info.port == port)
222 __team_option_inst_del(opt_inst);
226 static int __team_option_inst_add_port(struct team *team,
227 struct team_port *port)
229 struct team_option *option;
230 int err;
232 list_for_each_entry(option, &team->option_list, list) {
233 if (!option->per_port)
234 continue;
235 err = __team_option_inst_add(team, option, port);
236 if (err)
237 goto inst_del_port;
239 return 0;
241 inst_del_port:
242 __team_option_inst_del_port(team, port);
243 return err;
246 static void __team_option_inst_mark_removed_port(struct team *team,
247 struct team_port *port)
249 struct team_option_inst *opt_inst;
251 list_for_each_entry(opt_inst, &team->option_inst_list, list) {
252 if (opt_inst->info.port == port) {
253 opt_inst->changed = true;
254 opt_inst->removed = true;
259 static bool __team_option_inst_tmp_find(const struct list_head *opts,
260 const struct team_option_inst *needle)
262 struct team_option_inst *opt_inst;
264 list_for_each_entry(opt_inst, opts, tmp_list)
265 if (opt_inst == needle)
266 return true;
267 return false;
270 static int __team_options_register(struct team *team,
271 const struct team_option *option,
272 size_t option_count)
274 int i;
275 struct team_option **dst_opts;
276 int err;
278 dst_opts = kcalloc(option_count, sizeof(struct team_option *),
279 GFP_KERNEL);
280 if (!dst_opts)
281 return -ENOMEM;
282 for (i = 0; i < option_count; i++, option++) {
283 if (__team_find_option(team, option->name)) {
284 err = -EEXIST;
285 goto alloc_rollback;
287 dst_opts[i] = kmemdup(option, sizeof(*option), GFP_KERNEL);
288 if (!dst_opts[i]) {
289 err = -ENOMEM;
290 goto alloc_rollback;
294 for (i = 0; i < option_count; i++) {
295 err = __team_option_inst_add_option(team, dst_opts[i]);
296 if (err)
297 goto inst_rollback;
298 list_add_tail(&dst_opts[i]->list, &team->option_list);
301 kfree(dst_opts);
302 return 0;
304 inst_rollback:
305 for (i--; i >= 0; i--)
306 __team_option_inst_del_option(team, dst_opts[i]);
308 i = option_count - 1;
309 alloc_rollback:
310 for (i--; i >= 0; i--)
311 kfree(dst_opts[i]);
313 kfree(dst_opts);
314 return err;
317 static void __team_options_mark_removed(struct team *team,
318 const struct team_option *option,
319 size_t option_count)
321 int i;
323 for (i = 0; i < option_count; i++, option++) {
324 struct team_option *del_opt;
326 del_opt = __team_find_option(team, option->name);
327 if (del_opt)
328 __team_option_inst_mark_removed_option(team, del_opt);
332 static void __team_options_unregister(struct team *team,
333 const struct team_option *option,
334 size_t option_count)
336 int i;
338 for (i = 0; i < option_count; i++, option++) {
339 struct team_option *del_opt;
341 del_opt = __team_find_option(team, option->name);
342 if (del_opt) {
343 __team_option_inst_del_option(team, del_opt);
344 list_del(&del_opt->list);
345 kfree(del_opt);
350 static void __team_options_change_check(struct team *team);
352 int team_options_register(struct team *team,
353 const struct team_option *option,
354 size_t option_count)
356 int err;
358 err = __team_options_register(team, option, option_count);
359 if (err)
360 return err;
361 __team_options_change_check(team);
362 return 0;
364 EXPORT_SYMBOL(team_options_register);
366 void team_options_unregister(struct team *team,
367 const struct team_option *option,
368 size_t option_count)
370 __team_options_mark_removed(team, option, option_count);
371 __team_options_change_check(team);
372 __team_options_unregister(team, option, option_count);
374 EXPORT_SYMBOL(team_options_unregister);
376 static int team_option_get(struct team *team,
377 struct team_option_inst *opt_inst,
378 struct team_gsetter_ctx *ctx)
380 if (!opt_inst->option->getter)
381 return -EOPNOTSUPP;
382 return opt_inst->option->getter(team, ctx);
385 static int team_option_set(struct team *team,
386 struct team_option_inst *opt_inst,
387 struct team_gsetter_ctx *ctx)
389 if (!opt_inst->option->setter)
390 return -EOPNOTSUPP;
391 return opt_inst->option->setter(team, ctx);
394 void team_option_inst_set_change(struct team_option_inst_info *opt_inst_info)
396 struct team_option_inst *opt_inst;
398 opt_inst = container_of(opt_inst_info, struct team_option_inst, info);
399 opt_inst->changed = true;
401 EXPORT_SYMBOL(team_option_inst_set_change);
403 void team_options_change_check(struct team *team)
405 __team_options_change_check(team);
407 EXPORT_SYMBOL(team_options_change_check);
410 /****************
411 * Mode handling
412 ****************/
414 static LIST_HEAD(mode_list);
415 static DEFINE_SPINLOCK(mode_list_lock);
417 struct team_mode_item {
418 struct list_head list;
419 const struct team_mode *mode;
422 static struct team_mode_item *__find_mode(const char *kind)
424 struct team_mode_item *mitem;
426 list_for_each_entry(mitem, &mode_list, list) {
427 if (strcmp(mitem->mode->kind, kind) == 0)
428 return mitem;
430 return NULL;
433 static bool is_good_mode_name(const char *name)
435 while (*name != '\0') {
436 if (!isalpha(*name) && !isdigit(*name) && *name != '_')
437 return false;
438 name++;
440 return true;
443 int team_mode_register(const struct team_mode *mode)
445 int err = 0;
446 struct team_mode_item *mitem;
448 if (!is_good_mode_name(mode->kind) ||
449 mode->priv_size > TEAM_MODE_PRIV_SIZE)
450 return -EINVAL;
452 mitem = kmalloc(sizeof(*mitem), GFP_KERNEL);
453 if (!mitem)
454 return -ENOMEM;
456 spin_lock(&mode_list_lock);
457 if (__find_mode(mode->kind)) {
458 err = -EEXIST;
459 kfree(mitem);
460 goto unlock;
462 mitem->mode = mode;
463 list_add_tail(&mitem->list, &mode_list);
464 unlock:
465 spin_unlock(&mode_list_lock);
466 return err;
468 EXPORT_SYMBOL(team_mode_register);
470 void team_mode_unregister(const struct team_mode *mode)
472 struct team_mode_item *mitem;
474 spin_lock(&mode_list_lock);
475 mitem = __find_mode(mode->kind);
476 if (mitem) {
477 list_del_init(&mitem->list);
478 kfree(mitem);
480 spin_unlock(&mode_list_lock);
482 EXPORT_SYMBOL(team_mode_unregister);
484 static const struct team_mode *team_mode_get(const char *kind)
486 struct team_mode_item *mitem;
487 const struct team_mode *mode = NULL;
489 spin_lock(&mode_list_lock);
490 mitem = __find_mode(kind);
491 if (!mitem) {
492 spin_unlock(&mode_list_lock);
493 request_module("team-mode-%s", kind);
494 spin_lock(&mode_list_lock);
495 mitem = __find_mode(kind);
497 if (mitem) {
498 mode = mitem->mode;
499 if (!try_module_get(mode->owner))
500 mode = NULL;
503 spin_unlock(&mode_list_lock);
504 return mode;
507 static void team_mode_put(const struct team_mode *mode)
509 module_put(mode->owner);
512 static bool team_dummy_transmit(struct team *team, struct sk_buff *skb)
514 dev_kfree_skb_any(skb);
515 return false;
518 static rx_handler_result_t team_dummy_receive(struct team *team,
519 struct team_port *port,
520 struct sk_buff *skb)
522 return RX_HANDLER_ANOTHER;
525 static const struct team_mode __team_no_mode = {
526 .kind = "*NOMODE*",
529 static bool team_is_mode_set(struct team *team)
531 return team->mode != &__team_no_mode;
534 static void team_set_no_mode(struct team *team)
536 team->user_carrier_enabled = false;
537 team->mode = &__team_no_mode;
540 static void team_adjust_ops(struct team *team)
543 * To avoid checks in rx/tx skb paths, ensure here that non-null and
544 * correct ops are always set.
547 if (!team->en_port_count || !team_is_mode_set(team) ||
548 !team->mode->ops->transmit)
549 team->ops.transmit = team_dummy_transmit;
550 else
551 team->ops.transmit = team->mode->ops->transmit;
553 if (!team->en_port_count || !team_is_mode_set(team) ||
554 !team->mode->ops->receive)
555 team->ops.receive = team_dummy_receive;
556 else
557 team->ops.receive = team->mode->ops->receive;
561 * We can benefit from the fact that it's ensured no port is present
562 * at the time of mode change. Therefore no packets are in fly so there's no
563 * need to set mode operations in any special way.
565 static int __team_change_mode(struct team *team,
566 const struct team_mode *new_mode)
568 /* Check if mode was previously set and do cleanup if so */
569 if (team_is_mode_set(team)) {
570 void (*exit_op)(struct team *team) = team->ops.exit;
572 /* Clear ops area so no callback is called any longer */
573 memset(&team->ops, 0, sizeof(struct team_mode_ops));
574 team_adjust_ops(team);
576 if (exit_op)
577 exit_op(team);
578 team_mode_put(team->mode);
579 team_set_no_mode(team);
580 /* zero private data area */
581 memset(&team->mode_priv, 0,
582 sizeof(struct team) - offsetof(struct team, mode_priv));
585 if (!new_mode)
586 return 0;
588 if (new_mode->ops->init) {
589 int err;
591 err = new_mode->ops->init(team);
592 if (err)
593 return err;
596 team->mode = new_mode;
597 memcpy(&team->ops, new_mode->ops, sizeof(struct team_mode_ops));
598 team_adjust_ops(team);
600 return 0;
603 static int team_change_mode(struct team *team, const char *kind)
605 const struct team_mode *new_mode;
606 struct net_device *dev = team->dev;
607 int err;
609 if (!list_empty(&team->port_list)) {
610 netdev_err(dev, "No ports can be present during mode change\n");
611 return -EBUSY;
614 if (team_is_mode_set(team) && strcmp(team->mode->kind, kind) == 0) {
615 netdev_err(dev, "Unable to change to the same mode the team is in\n");
616 return -EINVAL;
619 new_mode = team_mode_get(kind);
620 if (!new_mode) {
621 netdev_err(dev, "Mode \"%s\" not found\n", kind);
622 return -EINVAL;
625 err = __team_change_mode(team, new_mode);
626 if (err) {
627 netdev_err(dev, "Failed to change to mode \"%s\"\n", kind);
628 team_mode_put(new_mode);
629 return err;
632 netdev_info(dev, "Mode changed to \"%s\"\n", kind);
633 return 0;
637 /*********************
638 * Peers notification
639 *********************/
641 static void team_notify_peers_work(struct work_struct *work)
643 struct team *team;
644 int val;
646 team = container_of(work, struct team, notify_peers.dw.work);
648 if (!rtnl_trylock()) {
649 schedule_delayed_work(&team->notify_peers.dw, 0);
650 return;
652 val = atomic_dec_if_positive(&team->notify_peers.count_pending);
653 if (val < 0) {
654 rtnl_unlock();
655 return;
657 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, team->dev);
658 rtnl_unlock();
659 if (val)
660 schedule_delayed_work(&team->notify_peers.dw,
661 msecs_to_jiffies(team->notify_peers.interval));
664 static void team_notify_peers(struct team *team)
666 if (!team->notify_peers.count || !netif_running(team->dev))
667 return;
668 atomic_add(team->notify_peers.count, &team->notify_peers.count_pending);
669 schedule_delayed_work(&team->notify_peers.dw, 0);
672 static void team_notify_peers_init(struct team *team)
674 INIT_DELAYED_WORK(&team->notify_peers.dw, team_notify_peers_work);
677 static void team_notify_peers_fini(struct team *team)
679 cancel_delayed_work_sync(&team->notify_peers.dw);
683 /*******************************
684 * Send multicast group rejoins
685 *******************************/
687 static void team_mcast_rejoin_work(struct work_struct *work)
689 struct team *team;
690 int val;
692 team = container_of(work, struct team, mcast_rejoin.dw.work);
694 if (!rtnl_trylock()) {
695 schedule_delayed_work(&team->mcast_rejoin.dw, 0);
696 return;
698 val = atomic_dec_if_positive(&team->mcast_rejoin.count_pending);
699 if (val < 0) {
700 rtnl_unlock();
701 return;
703 call_netdevice_notifiers(NETDEV_RESEND_IGMP, team->dev);
704 rtnl_unlock();
705 if (val)
706 schedule_delayed_work(&team->mcast_rejoin.dw,
707 msecs_to_jiffies(team->mcast_rejoin.interval));
710 static void team_mcast_rejoin(struct team *team)
712 if (!team->mcast_rejoin.count || !netif_running(team->dev))
713 return;
714 atomic_add(team->mcast_rejoin.count, &team->mcast_rejoin.count_pending);
715 schedule_delayed_work(&team->mcast_rejoin.dw, 0);
718 static void team_mcast_rejoin_init(struct team *team)
720 INIT_DELAYED_WORK(&team->mcast_rejoin.dw, team_mcast_rejoin_work);
723 static void team_mcast_rejoin_fini(struct team *team)
725 cancel_delayed_work_sync(&team->mcast_rejoin.dw);
729 /************************
730 * Rx path frame handler
731 ************************/
733 /* note: already called with rcu_read_lock */
734 static rx_handler_result_t team_handle_frame(struct sk_buff **pskb)
736 struct sk_buff *skb = *pskb;
737 struct team_port *port;
738 struct team *team;
739 rx_handler_result_t res;
741 skb = skb_share_check(skb, GFP_ATOMIC);
742 if (!skb)
743 return RX_HANDLER_CONSUMED;
745 *pskb = skb;
747 port = team_port_get_rcu(skb->dev);
748 team = port->team;
749 if (!team_port_enabled(port)) {
750 /* allow exact match delivery for disabled ports */
751 res = RX_HANDLER_EXACT;
752 } else {
753 res = team->ops.receive(team, port, skb);
755 if (res == RX_HANDLER_ANOTHER) {
756 struct team_pcpu_stats *pcpu_stats;
758 pcpu_stats = this_cpu_ptr(team->pcpu_stats);
759 u64_stats_update_begin(&pcpu_stats->syncp);
760 pcpu_stats->rx_packets++;
761 pcpu_stats->rx_bytes += skb->len;
762 if (skb->pkt_type == PACKET_MULTICAST)
763 pcpu_stats->rx_multicast++;
764 u64_stats_update_end(&pcpu_stats->syncp);
766 skb->dev = team->dev;
767 } else if (res == RX_HANDLER_EXACT) {
768 this_cpu_inc(team->pcpu_stats->rx_nohandler);
769 } else {
770 this_cpu_inc(team->pcpu_stats->rx_dropped);
773 return res;
777 /*************************************
778 * Multiqueue Tx port select override
779 *************************************/
781 static int team_queue_override_init(struct team *team)
783 struct list_head *listarr;
784 unsigned int queue_cnt = team->dev->num_tx_queues - 1;
785 unsigned int i;
787 if (!queue_cnt)
788 return 0;
789 listarr = kmalloc_array(queue_cnt, sizeof(struct list_head),
790 GFP_KERNEL);
791 if (!listarr)
792 return -ENOMEM;
793 team->qom_lists = listarr;
794 for (i = 0; i < queue_cnt; i++)
795 INIT_LIST_HEAD(listarr++);
796 return 0;
799 static void team_queue_override_fini(struct team *team)
801 kfree(team->qom_lists);
804 static struct list_head *__team_get_qom_list(struct team *team, u16 queue_id)
806 return &team->qom_lists[queue_id - 1];
810 * note: already called with rcu_read_lock
812 static bool team_queue_override_transmit(struct team *team, struct sk_buff *skb)
814 struct list_head *qom_list;
815 struct team_port *port;
817 if (!team->queue_override_enabled || !skb->queue_mapping)
818 return false;
819 qom_list = __team_get_qom_list(team, skb->queue_mapping);
820 list_for_each_entry_rcu(port, qom_list, qom_list) {
821 if (!team_dev_queue_xmit(team, port, skb))
822 return true;
824 return false;
827 static void __team_queue_override_port_del(struct team *team,
828 struct team_port *port)
830 if (!port->queue_id)
831 return;
832 list_del_rcu(&port->qom_list);
835 static bool team_queue_override_port_has_gt_prio_than(struct team_port *port,
836 struct team_port *cur)
838 if (port->priority < cur->priority)
839 return true;
840 if (port->priority > cur->priority)
841 return false;
842 if (port->index < cur->index)
843 return true;
844 return false;
847 static void __team_queue_override_port_add(struct team *team,
848 struct team_port *port)
850 struct team_port *cur;
851 struct list_head *qom_list;
852 struct list_head *node;
854 if (!port->queue_id)
855 return;
856 qom_list = __team_get_qom_list(team, port->queue_id);
857 node = qom_list;
858 list_for_each_entry(cur, qom_list, qom_list) {
859 if (team_queue_override_port_has_gt_prio_than(port, cur))
860 break;
861 node = &cur->qom_list;
863 list_add_tail_rcu(&port->qom_list, node);
866 static void __team_queue_override_enabled_check(struct team *team)
868 struct team_port *port;
869 bool enabled = false;
871 list_for_each_entry(port, &team->port_list, list) {
872 if (port->queue_id) {
873 enabled = true;
874 break;
877 if (enabled == team->queue_override_enabled)
878 return;
879 netdev_dbg(team->dev, "%s queue override\n",
880 enabled ? "Enabling" : "Disabling");
881 team->queue_override_enabled = enabled;
884 static void team_queue_override_port_prio_changed(struct team *team,
885 struct team_port *port)
887 if (!port->queue_id || team_port_enabled(port))
888 return;
889 __team_queue_override_port_del(team, port);
890 __team_queue_override_port_add(team, port);
891 __team_queue_override_enabled_check(team);
894 static void team_queue_override_port_change_queue_id(struct team *team,
895 struct team_port *port,
896 u16 new_queue_id)
898 if (team_port_enabled(port)) {
899 __team_queue_override_port_del(team, port);
900 port->queue_id = new_queue_id;
901 __team_queue_override_port_add(team, port);
902 __team_queue_override_enabled_check(team);
903 } else {
904 port->queue_id = new_queue_id;
908 static void team_queue_override_port_add(struct team *team,
909 struct team_port *port)
911 __team_queue_override_port_add(team, port);
912 __team_queue_override_enabled_check(team);
915 static void team_queue_override_port_del(struct team *team,
916 struct team_port *port)
918 __team_queue_override_port_del(team, port);
919 __team_queue_override_enabled_check(team);
923 /****************
924 * Port handling
925 ****************/
927 static bool team_port_find(const struct team *team,
928 const struct team_port *port)
930 struct team_port *cur;
932 list_for_each_entry(cur, &team->port_list, list)
933 if (cur == port)
934 return true;
935 return false;
939 * Enable/disable port by adding to enabled port hashlist and setting
940 * port->index (Might be racy so reader could see incorrect ifindex when
941 * processing a flying packet, but that is not a problem). Write guarded
942 * by team->lock.
944 static void team_port_enable(struct team *team,
945 struct team_port *port)
947 if (team_port_enabled(port))
948 return;
949 port->index = team->en_port_count++;
950 hlist_add_head_rcu(&port->hlist,
951 team_port_index_hash(team, port->index));
952 team_adjust_ops(team);
953 team_queue_override_port_add(team, port);
954 if (team->ops.port_enabled)
955 team->ops.port_enabled(team, port);
956 team_notify_peers(team);
957 team_mcast_rejoin(team);
958 team_lower_state_changed(port);
961 static void __reconstruct_port_hlist(struct team *team, int rm_index)
963 int i;
964 struct team_port *port;
966 for (i = rm_index + 1; i < team->en_port_count; i++) {
967 port = team_get_port_by_index(team, i);
968 hlist_del_rcu(&port->hlist);
969 port->index--;
970 hlist_add_head_rcu(&port->hlist,
971 team_port_index_hash(team, port->index));
975 static void team_port_disable(struct team *team,
976 struct team_port *port)
978 if (!team_port_enabled(port))
979 return;
980 if (team->ops.port_disabled)
981 team->ops.port_disabled(team, port);
982 hlist_del_rcu(&port->hlist);
983 __reconstruct_port_hlist(team, port->index);
984 port->index = -1;
985 team->en_port_count--;
986 team_queue_override_port_del(team, port);
987 team_adjust_ops(team);
988 team_notify_peers(team);
989 team_mcast_rejoin(team);
990 team_lower_state_changed(port);
993 #define TEAM_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
994 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
995 NETIF_F_HIGHDMA | NETIF_F_LRO)
997 #define TEAM_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
998 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1000 static void __team_compute_features(struct team *team)
1002 struct team_port *port;
1003 netdev_features_t vlan_features = TEAM_VLAN_FEATURES &
1004 NETIF_F_ALL_FOR_ALL;
1005 netdev_features_t enc_features = TEAM_ENC_FEATURES;
1006 unsigned short max_hard_header_len = ETH_HLEN;
1007 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1008 IFF_XMIT_DST_RELEASE_PERM;
1010 list_for_each_entry(port, &team->port_list, list) {
1011 vlan_features = netdev_increment_features(vlan_features,
1012 port->dev->vlan_features,
1013 TEAM_VLAN_FEATURES);
1014 enc_features =
1015 netdev_increment_features(enc_features,
1016 port->dev->hw_enc_features,
1017 TEAM_ENC_FEATURES);
1020 dst_release_flag &= port->dev->priv_flags;
1021 if (port->dev->hard_header_len > max_hard_header_len)
1022 max_hard_header_len = port->dev->hard_header_len;
1025 team->dev->vlan_features = vlan_features;
1026 team->dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1027 NETIF_F_GSO_UDP_L4;
1028 team->dev->hard_header_len = max_hard_header_len;
1030 team->dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1031 if (dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1032 team->dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1035 static void team_compute_features(struct team *team)
1037 mutex_lock(&team->lock);
1038 __team_compute_features(team);
1039 mutex_unlock(&team->lock);
1040 netdev_change_features(team->dev);
1043 static int team_port_enter(struct team *team, struct team_port *port)
1045 int err = 0;
1047 dev_hold(team->dev);
1048 if (team->ops.port_enter) {
1049 err = team->ops.port_enter(team, port);
1050 if (err) {
1051 netdev_err(team->dev, "Device %s failed to enter team mode\n",
1052 port->dev->name);
1053 goto err_port_enter;
1057 return 0;
1059 err_port_enter:
1060 dev_put(team->dev);
1062 return err;
1065 static void team_port_leave(struct team *team, struct team_port *port)
1067 if (team->ops.port_leave)
1068 team->ops.port_leave(team, port);
1069 dev_put(team->dev);
1072 #ifdef CONFIG_NET_POLL_CONTROLLER
1073 static int __team_port_enable_netpoll(struct team_port *port)
1075 struct netpoll *np;
1076 int err;
1078 np = kzalloc(sizeof(*np), GFP_KERNEL);
1079 if (!np)
1080 return -ENOMEM;
1082 err = __netpoll_setup(np, port->dev);
1083 if (err) {
1084 kfree(np);
1085 return err;
1087 port->np = np;
1088 return err;
1091 static int team_port_enable_netpoll(struct team_port *port)
1093 if (!port->team->dev->npinfo)
1094 return 0;
1096 return __team_port_enable_netpoll(port);
1099 static void team_port_disable_netpoll(struct team_port *port)
1101 struct netpoll *np = port->np;
1103 if (!np)
1104 return;
1105 port->np = NULL;
1107 __netpoll_free(np);
1109 #else
1110 static int team_port_enable_netpoll(struct team_port *port)
1112 return 0;
1114 static void team_port_disable_netpoll(struct team_port *port)
1117 #endif
1119 static int team_upper_dev_link(struct team *team, struct team_port *port,
1120 struct netlink_ext_ack *extack)
1122 struct netdev_lag_upper_info lag_upper_info;
1123 int err;
1125 lag_upper_info.tx_type = team->mode->lag_tx_type;
1126 lag_upper_info.hash_type = NETDEV_LAG_HASH_UNKNOWN;
1127 err = netdev_master_upper_dev_link(port->dev, team->dev, NULL,
1128 &lag_upper_info, extack);
1129 if (err)
1130 return err;
1131 port->dev->priv_flags |= IFF_TEAM_PORT;
1132 return 0;
1135 static void team_upper_dev_unlink(struct team *team, struct team_port *port)
1137 netdev_upper_dev_unlink(port->dev, team->dev);
1138 port->dev->priv_flags &= ~IFF_TEAM_PORT;
1141 static void __team_port_change_port_added(struct team_port *port, bool linkup);
1142 static int team_dev_type_check_change(struct net_device *dev,
1143 struct net_device *port_dev);
1145 static int team_port_add(struct team *team, struct net_device *port_dev,
1146 struct netlink_ext_ack *extack)
1148 struct net_device *dev = team->dev;
1149 struct team_port *port;
1150 char *portname = port_dev->name;
1151 int err;
1153 if (port_dev->flags & IFF_LOOPBACK) {
1154 NL_SET_ERR_MSG(extack, "Loopback device can't be added as a team port");
1155 netdev_err(dev, "Device %s is loopback device. Loopback devices can't be added as a team port\n",
1156 portname);
1157 return -EINVAL;
1160 if (team_port_exists(port_dev)) {
1161 NL_SET_ERR_MSG(extack, "Device is already a port of a team device");
1162 netdev_err(dev, "Device %s is already a port "
1163 "of a team device\n", portname);
1164 return -EBUSY;
1167 if (dev == port_dev) {
1168 NL_SET_ERR_MSG(extack, "Cannot enslave team device to itself");
1169 netdev_err(dev, "Cannot enslave team device to itself\n");
1170 return -EINVAL;
1173 if (port_dev->features & NETIF_F_VLAN_CHALLENGED &&
1174 vlan_uses_dev(dev)) {
1175 NL_SET_ERR_MSG(extack, "Device is VLAN challenged and team device has VLAN set up");
1176 netdev_err(dev, "Device %s is VLAN challenged and team device has VLAN set up\n",
1177 portname);
1178 return -EPERM;
1181 err = team_dev_type_check_change(dev, port_dev);
1182 if (err)
1183 return err;
1185 if (port_dev->flags & IFF_UP) {
1186 NL_SET_ERR_MSG(extack, "Device is up. Set it down before adding it as a team port");
1187 netdev_err(dev, "Device %s is up. Set it down before adding it as a team port\n",
1188 portname);
1189 return -EBUSY;
1192 port = kzalloc(sizeof(struct team_port) + team->mode->port_priv_size,
1193 GFP_KERNEL);
1194 if (!port)
1195 return -ENOMEM;
1197 port->dev = port_dev;
1198 port->team = team;
1199 INIT_LIST_HEAD(&port->qom_list);
1201 port->orig.mtu = port_dev->mtu;
1202 err = dev_set_mtu(port_dev, dev->mtu);
1203 if (err) {
1204 netdev_dbg(dev, "Error %d calling dev_set_mtu\n", err);
1205 goto err_set_mtu;
1208 memcpy(port->orig.dev_addr, port_dev->dev_addr, port_dev->addr_len);
1210 err = team_port_enter(team, port);
1211 if (err) {
1212 netdev_err(dev, "Device %s failed to enter team mode\n",
1213 portname);
1214 goto err_port_enter;
1217 err = dev_open(port_dev);
1218 if (err) {
1219 netdev_dbg(dev, "Device %s opening failed\n",
1220 portname);
1221 goto err_dev_open;
1224 err = vlan_vids_add_by_dev(port_dev, dev);
1225 if (err) {
1226 netdev_err(dev, "Failed to add vlan ids to device %s\n",
1227 portname);
1228 goto err_vids_add;
1231 err = team_port_enable_netpoll(port);
1232 if (err) {
1233 netdev_err(dev, "Failed to enable netpoll on device %s\n",
1234 portname);
1235 goto err_enable_netpoll;
1238 if (!(dev->features & NETIF_F_LRO))
1239 dev_disable_lro(port_dev);
1241 err = netdev_rx_handler_register(port_dev, team_handle_frame,
1242 port);
1243 if (err) {
1244 netdev_err(dev, "Device %s failed to register rx_handler\n",
1245 portname);
1246 goto err_handler_register;
1249 err = team_upper_dev_link(team, port, extack);
1250 if (err) {
1251 netdev_err(dev, "Device %s failed to set upper link\n",
1252 portname);
1253 goto err_set_upper_link;
1256 err = __team_option_inst_add_port(team, port);
1257 if (err) {
1258 netdev_err(dev, "Device %s failed to add per-port options\n",
1259 portname);
1260 goto err_option_port_add;
1263 netif_addr_lock_bh(dev);
1264 dev_uc_sync_multiple(port_dev, dev);
1265 dev_mc_sync_multiple(port_dev, dev);
1266 netif_addr_unlock_bh(dev);
1268 port->index = -1;
1269 list_add_tail_rcu(&port->list, &team->port_list);
1270 team_port_enable(team, port);
1271 __team_compute_features(team);
1272 __team_port_change_port_added(port, !!netif_carrier_ok(port_dev));
1273 __team_options_change_check(team);
1275 netdev_info(dev, "Port device %s added\n", portname);
1277 return 0;
1279 err_option_port_add:
1280 team_upper_dev_unlink(team, port);
1282 err_set_upper_link:
1283 netdev_rx_handler_unregister(port_dev);
1285 err_handler_register:
1286 team_port_disable_netpoll(port);
1288 err_enable_netpoll:
1289 vlan_vids_del_by_dev(port_dev, dev);
1291 err_vids_add:
1292 dev_close(port_dev);
1294 err_dev_open:
1295 team_port_leave(team, port);
1296 team_port_set_orig_dev_addr(port);
1298 err_port_enter:
1299 dev_set_mtu(port_dev, port->orig.mtu);
1301 err_set_mtu:
1302 kfree(port);
1304 return err;
1307 static void __team_port_change_port_removed(struct team_port *port);
1309 static int team_port_del(struct team *team, struct net_device *port_dev)
1311 struct net_device *dev = team->dev;
1312 struct team_port *port;
1313 char *portname = port_dev->name;
1315 port = team_port_get_rtnl(port_dev);
1316 if (!port || !team_port_find(team, port)) {
1317 netdev_err(dev, "Device %s does not act as a port of this team\n",
1318 portname);
1319 return -ENOENT;
1322 team_port_disable(team, port);
1323 list_del_rcu(&port->list);
1324 team_upper_dev_unlink(team, port);
1325 netdev_rx_handler_unregister(port_dev);
1326 team_port_disable_netpoll(port);
1327 vlan_vids_del_by_dev(port_dev, dev);
1328 dev_uc_unsync(port_dev, dev);
1329 dev_mc_unsync(port_dev, dev);
1330 dev_close(port_dev);
1331 team_port_leave(team, port);
1333 __team_option_inst_mark_removed_port(team, port);
1334 __team_options_change_check(team);
1335 __team_option_inst_del_port(team, port);
1336 __team_port_change_port_removed(port);
1338 team_port_set_orig_dev_addr(port);
1339 dev_set_mtu(port_dev, port->orig.mtu);
1340 kfree_rcu(port, rcu);
1341 netdev_info(dev, "Port device %s removed\n", portname);
1342 __team_compute_features(team);
1344 return 0;
1348 /*****************
1349 * Net device ops
1350 *****************/
1352 static int team_mode_option_get(struct team *team, struct team_gsetter_ctx *ctx)
1354 ctx->data.str_val = team->mode->kind;
1355 return 0;
1358 static int team_mode_option_set(struct team *team, struct team_gsetter_ctx *ctx)
1360 return team_change_mode(team, ctx->data.str_val);
1363 static int team_notify_peers_count_get(struct team *team,
1364 struct team_gsetter_ctx *ctx)
1366 ctx->data.u32_val = team->notify_peers.count;
1367 return 0;
1370 static int team_notify_peers_count_set(struct team *team,
1371 struct team_gsetter_ctx *ctx)
1373 team->notify_peers.count = ctx->data.u32_val;
1374 return 0;
1377 static int team_notify_peers_interval_get(struct team *team,
1378 struct team_gsetter_ctx *ctx)
1380 ctx->data.u32_val = team->notify_peers.interval;
1381 return 0;
1384 static int team_notify_peers_interval_set(struct team *team,
1385 struct team_gsetter_ctx *ctx)
1387 team->notify_peers.interval = ctx->data.u32_val;
1388 return 0;
1391 static int team_mcast_rejoin_count_get(struct team *team,
1392 struct team_gsetter_ctx *ctx)
1394 ctx->data.u32_val = team->mcast_rejoin.count;
1395 return 0;
1398 static int team_mcast_rejoin_count_set(struct team *team,
1399 struct team_gsetter_ctx *ctx)
1401 team->mcast_rejoin.count = ctx->data.u32_val;
1402 return 0;
1405 static int team_mcast_rejoin_interval_get(struct team *team,
1406 struct team_gsetter_ctx *ctx)
1408 ctx->data.u32_val = team->mcast_rejoin.interval;
1409 return 0;
1412 static int team_mcast_rejoin_interval_set(struct team *team,
1413 struct team_gsetter_ctx *ctx)
1415 team->mcast_rejoin.interval = ctx->data.u32_val;
1416 return 0;
1419 static int team_port_en_option_get(struct team *team,
1420 struct team_gsetter_ctx *ctx)
1422 struct team_port *port = ctx->info->port;
1424 ctx->data.bool_val = team_port_enabled(port);
1425 return 0;
1428 static int team_port_en_option_set(struct team *team,
1429 struct team_gsetter_ctx *ctx)
1431 struct team_port *port = ctx->info->port;
1433 if (ctx->data.bool_val)
1434 team_port_enable(team, port);
1435 else
1436 team_port_disable(team, port);
1437 return 0;
1440 static int team_user_linkup_option_get(struct team *team,
1441 struct team_gsetter_ctx *ctx)
1443 struct team_port *port = ctx->info->port;
1445 ctx->data.bool_val = port->user.linkup;
1446 return 0;
1449 static void __team_carrier_check(struct team *team);
1451 static int team_user_linkup_option_set(struct team *team,
1452 struct team_gsetter_ctx *ctx)
1454 struct team_port *port = ctx->info->port;
1456 port->user.linkup = ctx->data.bool_val;
1457 team_refresh_port_linkup(port);
1458 __team_carrier_check(port->team);
1459 return 0;
1462 static int team_user_linkup_en_option_get(struct team *team,
1463 struct team_gsetter_ctx *ctx)
1465 struct team_port *port = ctx->info->port;
1467 ctx->data.bool_val = port->user.linkup_enabled;
1468 return 0;
1471 static int team_user_linkup_en_option_set(struct team *team,
1472 struct team_gsetter_ctx *ctx)
1474 struct team_port *port = ctx->info->port;
1476 port->user.linkup_enabled = ctx->data.bool_val;
1477 team_refresh_port_linkup(port);
1478 __team_carrier_check(port->team);
1479 return 0;
1482 static int team_priority_option_get(struct team *team,
1483 struct team_gsetter_ctx *ctx)
1485 struct team_port *port = ctx->info->port;
1487 ctx->data.s32_val = port->priority;
1488 return 0;
1491 static int team_priority_option_set(struct team *team,
1492 struct team_gsetter_ctx *ctx)
1494 struct team_port *port = ctx->info->port;
1495 s32 priority = ctx->data.s32_val;
1497 if (port->priority == priority)
1498 return 0;
1499 port->priority = priority;
1500 team_queue_override_port_prio_changed(team, port);
1501 return 0;
1504 static int team_queue_id_option_get(struct team *team,
1505 struct team_gsetter_ctx *ctx)
1507 struct team_port *port = ctx->info->port;
1509 ctx->data.u32_val = port->queue_id;
1510 return 0;
1513 static int team_queue_id_option_set(struct team *team,
1514 struct team_gsetter_ctx *ctx)
1516 struct team_port *port = ctx->info->port;
1517 u16 new_queue_id = ctx->data.u32_val;
1519 if (port->queue_id == new_queue_id)
1520 return 0;
1521 if (new_queue_id >= team->dev->real_num_tx_queues)
1522 return -EINVAL;
1523 team_queue_override_port_change_queue_id(team, port, new_queue_id);
1524 return 0;
1527 static const struct team_option team_options[] = {
1529 .name = "mode",
1530 .type = TEAM_OPTION_TYPE_STRING,
1531 .getter = team_mode_option_get,
1532 .setter = team_mode_option_set,
1535 .name = "notify_peers_count",
1536 .type = TEAM_OPTION_TYPE_U32,
1537 .getter = team_notify_peers_count_get,
1538 .setter = team_notify_peers_count_set,
1541 .name = "notify_peers_interval",
1542 .type = TEAM_OPTION_TYPE_U32,
1543 .getter = team_notify_peers_interval_get,
1544 .setter = team_notify_peers_interval_set,
1547 .name = "mcast_rejoin_count",
1548 .type = TEAM_OPTION_TYPE_U32,
1549 .getter = team_mcast_rejoin_count_get,
1550 .setter = team_mcast_rejoin_count_set,
1553 .name = "mcast_rejoin_interval",
1554 .type = TEAM_OPTION_TYPE_U32,
1555 .getter = team_mcast_rejoin_interval_get,
1556 .setter = team_mcast_rejoin_interval_set,
1559 .name = "enabled",
1560 .type = TEAM_OPTION_TYPE_BOOL,
1561 .per_port = true,
1562 .getter = team_port_en_option_get,
1563 .setter = team_port_en_option_set,
1566 .name = "user_linkup",
1567 .type = TEAM_OPTION_TYPE_BOOL,
1568 .per_port = true,
1569 .getter = team_user_linkup_option_get,
1570 .setter = team_user_linkup_option_set,
1573 .name = "user_linkup_enabled",
1574 .type = TEAM_OPTION_TYPE_BOOL,
1575 .per_port = true,
1576 .getter = team_user_linkup_en_option_get,
1577 .setter = team_user_linkup_en_option_set,
1580 .name = "priority",
1581 .type = TEAM_OPTION_TYPE_S32,
1582 .per_port = true,
1583 .getter = team_priority_option_get,
1584 .setter = team_priority_option_set,
1587 .name = "queue_id",
1588 .type = TEAM_OPTION_TYPE_U32,
1589 .per_port = true,
1590 .getter = team_queue_id_option_get,
1591 .setter = team_queue_id_option_set,
1596 static int team_init(struct net_device *dev)
1598 struct team *team = netdev_priv(dev);
1599 int i;
1600 int err;
1602 team->dev = dev;
1603 mutex_init(&team->lock);
1604 team_set_no_mode(team);
1606 team->pcpu_stats = netdev_alloc_pcpu_stats(struct team_pcpu_stats);
1607 if (!team->pcpu_stats)
1608 return -ENOMEM;
1610 for (i = 0; i < TEAM_PORT_HASHENTRIES; i++)
1611 INIT_HLIST_HEAD(&team->en_port_hlist[i]);
1612 INIT_LIST_HEAD(&team->port_list);
1613 err = team_queue_override_init(team);
1614 if (err)
1615 goto err_team_queue_override_init;
1617 team_adjust_ops(team);
1619 INIT_LIST_HEAD(&team->option_list);
1620 INIT_LIST_HEAD(&team->option_inst_list);
1622 team_notify_peers_init(team);
1623 team_mcast_rejoin_init(team);
1625 err = team_options_register(team, team_options, ARRAY_SIZE(team_options));
1626 if (err)
1627 goto err_options_register;
1628 netif_carrier_off(dev);
1630 netdev_lockdep_set_classes(dev);
1632 return 0;
1634 err_options_register:
1635 team_mcast_rejoin_fini(team);
1636 team_notify_peers_fini(team);
1637 team_queue_override_fini(team);
1638 err_team_queue_override_init:
1639 free_percpu(team->pcpu_stats);
1641 return err;
1644 static void team_uninit(struct net_device *dev)
1646 struct team *team = netdev_priv(dev);
1647 struct team_port *port;
1648 struct team_port *tmp;
1650 mutex_lock(&team->lock);
1651 list_for_each_entry_safe(port, tmp, &team->port_list, list)
1652 team_port_del(team, port->dev);
1654 __team_change_mode(team, NULL); /* cleanup */
1655 __team_options_unregister(team, team_options, ARRAY_SIZE(team_options));
1656 team_mcast_rejoin_fini(team);
1657 team_notify_peers_fini(team);
1658 team_queue_override_fini(team);
1659 mutex_unlock(&team->lock);
1660 netdev_change_features(dev);
1663 static void team_destructor(struct net_device *dev)
1665 struct team *team = netdev_priv(dev);
1667 free_percpu(team->pcpu_stats);
1670 static int team_open(struct net_device *dev)
1672 return 0;
1675 static int team_close(struct net_device *dev)
1677 return 0;
1681 * note: already called with rcu_read_lock
1683 static netdev_tx_t team_xmit(struct sk_buff *skb, struct net_device *dev)
1685 struct team *team = netdev_priv(dev);
1686 bool tx_success;
1687 unsigned int len = skb->len;
1689 tx_success = team_queue_override_transmit(team, skb);
1690 if (!tx_success)
1691 tx_success = team->ops.transmit(team, skb);
1692 if (tx_success) {
1693 struct team_pcpu_stats *pcpu_stats;
1695 pcpu_stats = this_cpu_ptr(team->pcpu_stats);
1696 u64_stats_update_begin(&pcpu_stats->syncp);
1697 pcpu_stats->tx_packets++;
1698 pcpu_stats->tx_bytes += len;
1699 u64_stats_update_end(&pcpu_stats->syncp);
1700 } else {
1701 this_cpu_inc(team->pcpu_stats->tx_dropped);
1704 return NETDEV_TX_OK;
1707 static u16 team_select_queue(struct net_device *dev, struct sk_buff *skb,
1708 struct net_device *sb_dev,
1709 select_queue_fallback_t fallback)
1712 * This helper function exists to help dev_pick_tx get the correct
1713 * destination queue. Using a helper function skips a call to
1714 * skb_tx_hash and will put the skbs in the queue we expect on their
1715 * way down to the team driver.
1717 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
1720 * Save the original txq to restore before passing to the driver
1722 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
1724 if (unlikely(txq >= dev->real_num_tx_queues)) {
1725 do {
1726 txq -= dev->real_num_tx_queues;
1727 } while (txq >= dev->real_num_tx_queues);
1729 return txq;
1732 static void team_change_rx_flags(struct net_device *dev, int change)
1734 struct team *team = netdev_priv(dev);
1735 struct team_port *port;
1736 int inc;
1738 rcu_read_lock();
1739 list_for_each_entry_rcu(port, &team->port_list, list) {
1740 if (change & IFF_PROMISC) {
1741 inc = dev->flags & IFF_PROMISC ? 1 : -1;
1742 dev_set_promiscuity(port->dev, inc);
1744 if (change & IFF_ALLMULTI) {
1745 inc = dev->flags & IFF_ALLMULTI ? 1 : -1;
1746 dev_set_allmulti(port->dev, inc);
1749 rcu_read_unlock();
1752 static void team_set_rx_mode(struct net_device *dev)
1754 struct team *team = netdev_priv(dev);
1755 struct team_port *port;
1757 rcu_read_lock();
1758 list_for_each_entry_rcu(port, &team->port_list, list) {
1759 dev_uc_sync_multiple(port->dev, dev);
1760 dev_mc_sync_multiple(port->dev, dev);
1762 rcu_read_unlock();
1765 static int team_set_mac_address(struct net_device *dev, void *p)
1767 struct sockaddr *addr = p;
1768 struct team *team = netdev_priv(dev);
1769 struct team_port *port;
1771 if (dev->type == ARPHRD_ETHER && !is_valid_ether_addr(addr->sa_data))
1772 return -EADDRNOTAVAIL;
1773 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
1774 mutex_lock(&team->lock);
1775 list_for_each_entry(port, &team->port_list, list)
1776 if (team->ops.port_change_dev_addr)
1777 team->ops.port_change_dev_addr(team, port);
1778 mutex_unlock(&team->lock);
1779 return 0;
1782 static int team_change_mtu(struct net_device *dev, int new_mtu)
1784 struct team *team = netdev_priv(dev);
1785 struct team_port *port;
1786 int err;
1789 * Alhough this is reader, it's guarded by team lock. It's not possible
1790 * to traverse list in reverse under rcu_read_lock
1792 mutex_lock(&team->lock);
1793 team->port_mtu_change_allowed = true;
1794 list_for_each_entry(port, &team->port_list, list) {
1795 err = dev_set_mtu(port->dev, new_mtu);
1796 if (err) {
1797 netdev_err(dev, "Device %s failed to change mtu",
1798 port->dev->name);
1799 goto unwind;
1802 team->port_mtu_change_allowed = false;
1803 mutex_unlock(&team->lock);
1805 dev->mtu = new_mtu;
1807 return 0;
1809 unwind:
1810 list_for_each_entry_continue_reverse(port, &team->port_list, list)
1811 dev_set_mtu(port->dev, dev->mtu);
1812 team->port_mtu_change_allowed = false;
1813 mutex_unlock(&team->lock);
1815 return err;
1818 static void
1819 team_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1821 struct team *team = netdev_priv(dev);
1822 struct team_pcpu_stats *p;
1823 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
1824 u32 rx_dropped = 0, tx_dropped = 0, rx_nohandler = 0;
1825 unsigned int start;
1826 int i;
1828 for_each_possible_cpu(i) {
1829 p = per_cpu_ptr(team->pcpu_stats, i);
1830 do {
1831 start = u64_stats_fetch_begin_irq(&p->syncp);
1832 rx_packets = p->rx_packets;
1833 rx_bytes = p->rx_bytes;
1834 rx_multicast = p->rx_multicast;
1835 tx_packets = p->tx_packets;
1836 tx_bytes = p->tx_bytes;
1837 } while (u64_stats_fetch_retry_irq(&p->syncp, start));
1839 stats->rx_packets += rx_packets;
1840 stats->rx_bytes += rx_bytes;
1841 stats->multicast += rx_multicast;
1842 stats->tx_packets += tx_packets;
1843 stats->tx_bytes += tx_bytes;
1845 * rx_dropped, tx_dropped & rx_nohandler are u32,
1846 * updated without syncp protection.
1848 rx_dropped += p->rx_dropped;
1849 tx_dropped += p->tx_dropped;
1850 rx_nohandler += p->rx_nohandler;
1852 stats->rx_dropped = rx_dropped;
1853 stats->tx_dropped = tx_dropped;
1854 stats->rx_nohandler = rx_nohandler;
1857 static int team_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid)
1859 struct team *team = netdev_priv(dev);
1860 struct team_port *port;
1861 int err;
1864 * Alhough this is reader, it's guarded by team lock. It's not possible
1865 * to traverse list in reverse under rcu_read_lock
1867 mutex_lock(&team->lock);
1868 list_for_each_entry(port, &team->port_list, list) {
1869 err = vlan_vid_add(port->dev, proto, vid);
1870 if (err)
1871 goto unwind;
1873 mutex_unlock(&team->lock);
1875 return 0;
1877 unwind:
1878 list_for_each_entry_continue_reverse(port, &team->port_list, list)
1879 vlan_vid_del(port->dev, proto, vid);
1880 mutex_unlock(&team->lock);
1882 return err;
1885 static int team_vlan_rx_kill_vid(struct net_device *dev, __be16 proto, u16 vid)
1887 struct team *team = netdev_priv(dev);
1888 struct team_port *port;
1890 mutex_lock(&team->lock);
1891 list_for_each_entry(port, &team->port_list, list)
1892 vlan_vid_del(port->dev, proto, vid);
1893 mutex_unlock(&team->lock);
1895 return 0;
1898 #ifdef CONFIG_NET_POLL_CONTROLLER
1899 static void team_poll_controller(struct net_device *dev)
1903 static void __team_netpoll_cleanup(struct team *team)
1905 struct team_port *port;
1907 list_for_each_entry(port, &team->port_list, list)
1908 team_port_disable_netpoll(port);
1911 static void team_netpoll_cleanup(struct net_device *dev)
1913 struct team *team = netdev_priv(dev);
1915 mutex_lock(&team->lock);
1916 __team_netpoll_cleanup(team);
1917 mutex_unlock(&team->lock);
1920 static int team_netpoll_setup(struct net_device *dev,
1921 struct netpoll_info *npifo)
1923 struct team *team = netdev_priv(dev);
1924 struct team_port *port;
1925 int err = 0;
1927 mutex_lock(&team->lock);
1928 list_for_each_entry(port, &team->port_list, list) {
1929 err = __team_port_enable_netpoll(port);
1930 if (err) {
1931 __team_netpoll_cleanup(team);
1932 break;
1935 mutex_unlock(&team->lock);
1936 return err;
1938 #endif
1940 static int team_add_slave(struct net_device *dev, struct net_device *port_dev,
1941 struct netlink_ext_ack *extack)
1943 struct team *team = netdev_priv(dev);
1944 int err;
1946 mutex_lock(&team->lock);
1947 err = team_port_add(team, port_dev, extack);
1948 mutex_unlock(&team->lock);
1950 if (!err)
1951 netdev_change_features(dev);
1953 return err;
1956 static int team_del_slave(struct net_device *dev, struct net_device *port_dev)
1958 struct team *team = netdev_priv(dev);
1959 int err;
1961 mutex_lock(&team->lock);
1962 err = team_port_del(team, port_dev);
1963 mutex_unlock(&team->lock);
1965 if (!err)
1966 netdev_change_features(dev);
1968 return err;
1971 static netdev_features_t team_fix_features(struct net_device *dev,
1972 netdev_features_t features)
1974 struct team_port *port;
1975 struct team *team = netdev_priv(dev);
1976 netdev_features_t mask;
1978 mask = features;
1979 features &= ~NETIF_F_ONE_FOR_ALL;
1980 features |= NETIF_F_ALL_FOR_ALL;
1982 rcu_read_lock();
1983 list_for_each_entry_rcu(port, &team->port_list, list) {
1984 features = netdev_increment_features(features,
1985 port->dev->features,
1986 mask);
1988 rcu_read_unlock();
1990 features = netdev_add_tso_features(features, mask);
1992 return features;
1995 static int team_change_carrier(struct net_device *dev, bool new_carrier)
1997 struct team *team = netdev_priv(dev);
1999 team->user_carrier_enabled = true;
2001 if (new_carrier)
2002 netif_carrier_on(dev);
2003 else
2004 netif_carrier_off(dev);
2005 return 0;
2008 static const struct net_device_ops team_netdev_ops = {
2009 .ndo_init = team_init,
2010 .ndo_uninit = team_uninit,
2011 .ndo_open = team_open,
2012 .ndo_stop = team_close,
2013 .ndo_start_xmit = team_xmit,
2014 .ndo_select_queue = team_select_queue,
2015 .ndo_change_rx_flags = team_change_rx_flags,
2016 .ndo_set_rx_mode = team_set_rx_mode,
2017 .ndo_set_mac_address = team_set_mac_address,
2018 .ndo_change_mtu = team_change_mtu,
2019 .ndo_get_stats64 = team_get_stats64,
2020 .ndo_vlan_rx_add_vid = team_vlan_rx_add_vid,
2021 .ndo_vlan_rx_kill_vid = team_vlan_rx_kill_vid,
2022 #ifdef CONFIG_NET_POLL_CONTROLLER
2023 .ndo_poll_controller = team_poll_controller,
2024 .ndo_netpoll_setup = team_netpoll_setup,
2025 .ndo_netpoll_cleanup = team_netpoll_cleanup,
2026 #endif
2027 .ndo_add_slave = team_add_slave,
2028 .ndo_del_slave = team_del_slave,
2029 .ndo_fix_features = team_fix_features,
2030 .ndo_change_carrier = team_change_carrier,
2031 .ndo_features_check = passthru_features_check,
2034 /***********************
2035 * ethtool interface
2036 ***********************/
2038 static void team_ethtool_get_drvinfo(struct net_device *dev,
2039 struct ethtool_drvinfo *drvinfo)
2041 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
2042 strlcpy(drvinfo->version, UTS_RELEASE, sizeof(drvinfo->version));
2045 static const struct ethtool_ops team_ethtool_ops = {
2046 .get_drvinfo = team_ethtool_get_drvinfo,
2047 .get_link = ethtool_op_get_link,
2050 /***********************
2051 * rt netlink interface
2052 ***********************/
2054 static void team_setup_by_port(struct net_device *dev,
2055 struct net_device *port_dev)
2057 dev->header_ops = port_dev->header_ops;
2058 dev->type = port_dev->type;
2059 dev->hard_header_len = port_dev->hard_header_len;
2060 dev->addr_len = port_dev->addr_len;
2061 dev->mtu = port_dev->mtu;
2062 memcpy(dev->broadcast, port_dev->broadcast, port_dev->addr_len);
2063 eth_hw_addr_inherit(dev, port_dev);
2066 static int team_dev_type_check_change(struct net_device *dev,
2067 struct net_device *port_dev)
2069 struct team *team = netdev_priv(dev);
2070 char *portname = port_dev->name;
2071 int err;
2073 if (dev->type == port_dev->type)
2074 return 0;
2075 if (!list_empty(&team->port_list)) {
2076 netdev_err(dev, "Device %s is of different type\n", portname);
2077 return -EBUSY;
2079 err = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE, dev);
2080 err = notifier_to_errno(err);
2081 if (err) {
2082 netdev_err(dev, "Refused to change device type\n");
2083 return err;
2085 dev_uc_flush(dev);
2086 dev_mc_flush(dev);
2087 team_setup_by_port(dev, port_dev);
2088 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE, dev);
2089 return 0;
2092 static void team_setup(struct net_device *dev)
2094 ether_setup(dev);
2095 dev->max_mtu = ETH_MAX_MTU;
2097 dev->netdev_ops = &team_netdev_ops;
2098 dev->ethtool_ops = &team_ethtool_ops;
2099 dev->needs_free_netdev = true;
2100 dev->priv_destructor = team_destructor;
2101 dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
2102 dev->priv_flags |= IFF_NO_QUEUE;
2103 dev->priv_flags |= IFF_TEAM;
2106 * Indicate we support unicast address filtering. That way core won't
2107 * bring us to promisc mode in case a unicast addr is added.
2108 * Let this up to underlay drivers.
2110 dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
2112 dev->features |= NETIF_F_LLTX;
2113 dev->features |= NETIF_F_GRO;
2115 /* Don't allow team devices to change network namespaces. */
2116 dev->features |= NETIF_F_NETNS_LOCAL;
2118 dev->hw_features = TEAM_VLAN_FEATURES |
2119 NETIF_F_HW_VLAN_CTAG_TX |
2120 NETIF_F_HW_VLAN_CTAG_RX |
2121 NETIF_F_HW_VLAN_CTAG_FILTER;
2123 dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4;
2124 dev->features |= dev->hw_features;
2127 static int team_newlink(struct net *src_net, struct net_device *dev,
2128 struct nlattr *tb[], struct nlattr *data[],
2129 struct netlink_ext_ack *extack)
2131 if (tb[IFLA_ADDRESS] == NULL)
2132 eth_hw_addr_random(dev);
2134 return register_netdevice(dev);
2137 static int team_validate(struct nlattr *tb[], struct nlattr *data[],
2138 struct netlink_ext_ack *extack)
2140 if (tb[IFLA_ADDRESS]) {
2141 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
2142 return -EINVAL;
2143 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
2144 return -EADDRNOTAVAIL;
2146 return 0;
2149 static unsigned int team_get_num_tx_queues(void)
2151 return TEAM_DEFAULT_NUM_TX_QUEUES;
2154 static unsigned int team_get_num_rx_queues(void)
2156 return TEAM_DEFAULT_NUM_RX_QUEUES;
2159 static struct rtnl_link_ops team_link_ops __read_mostly = {
2160 .kind = DRV_NAME,
2161 .priv_size = sizeof(struct team),
2162 .setup = team_setup,
2163 .newlink = team_newlink,
2164 .validate = team_validate,
2165 .get_num_tx_queues = team_get_num_tx_queues,
2166 .get_num_rx_queues = team_get_num_rx_queues,
2170 /***********************************
2171 * Generic netlink custom interface
2172 ***********************************/
2174 static struct genl_family team_nl_family;
2176 static const struct nla_policy team_nl_policy[TEAM_ATTR_MAX + 1] = {
2177 [TEAM_ATTR_UNSPEC] = { .type = NLA_UNSPEC, },
2178 [TEAM_ATTR_TEAM_IFINDEX] = { .type = NLA_U32 },
2179 [TEAM_ATTR_LIST_OPTION] = { .type = NLA_NESTED },
2180 [TEAM_ATTR_LIST_PORT] = { .type = NLA_NESTED },
2183 static const struct nla_policy
2184 team_nl_option_policy[TEAM_ATTR_OPTION_MAX + 1] = {
2185 [TEAM_ATTR_OPTION_UNSPEC] = { .type = NLA_UNSPEC, },
2186 [TEAM_ATTR_OPTION_NAME] = {
2187 .type = NLA_STRING,
2188 .len = TEAM_STRING_MAX_LEN,
2190 [TEAM_ATTR_OPTION_CHANGED] = { .type = NLA_FLAG },
2191 [TEAM_ATTR_OPTION_TYPE] = { .type = NLA_U8 },
2192 [TEAM_ATTR_OPTION_DATA] = { .type = NLA_BINARY },
2195 static int team_nl_cmd_noop(struct sk_buff *skb, struct genl_info *info)
2197 struct sk_buff *msg;
2198 void *hdr;
2199 int err;
2201 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2202 if (!msg)
2203 return -ENOMEM;
2205 hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
2206 &team_nl_family, 0, TEAM_CMD_NOOP);
2207 if (!hdr) {
2208 err = -EMSGSIZE;
2209 goto err_msg_put;
2212 genlmsg_end(msg, hdr);
2214 return genlmsg_unicast(genl_info_net(info), msg, info->snd_portid);
2216 err_msg_put:
2217 nlmsg_free(msg);
2219 return err;
2223 * Netlink cmd functions should be locked by following two functions.
2224 * Since dev gets held here, that ensures dev won't disappear in between.
2226 static struct team *team_nl_team_get(struct genl_info *info)
2228 struct net *net = genl_info_net(info);
2229 int ifindex;
2230 struct net_device *dev;
2231 struct team *team;
2233 if (!info->attrs[TEAM_ATTR_TEAM_IFINDEX])
2234 return NULL;
2236 ifindex = nla_get_u32(info->attrs[TEAM_ATTR_TEAM_IFINDEX]);
2237 dev = dev_get_by_index(net, ifindex);
2238 if (!dev || dev->netdev_ops != &team_netdev_ops) {
2239 if (dev)
2240 dev_put(dev);
2241 return NULL;
2244 team = netdev_priv(dev);
2245 mutex_lock(&team->lock);
2246 return team;
2249 static void team_nl_team_put(struct team *team)
2251 mutex_unlock(&team->lock);
2252 dev_put(team->dev);
2255 typedef int team_nl_send_func_t(struct sk_buff *skb,
2256 struct team *team, u32 portid);
2258 static int team_nl_send_unicast(struct sk_buff *skb, struct team *team, u32 portid)
2260 return genlmsg_unicast(dev_net(team->dev), skb, portid);
2263 static int team_nl_fill_one_option_get(struct sk_buff *skb, struct team *team,
2264 struct team_option_inst *opt_inst)
2266 struct nlattr *option_item;
2267 struct team_option *option = opt_inst->option;
2268 struct team_option_inst_info *opt_inst_info = &opt_inst->info;
2269 struct team_gsetter_ctx ctx;
2270 int err;
2272 ctx.info = opt_inst_info;
2273 err = team_option_get(team, opt_inst, &ctx);
2274 if (err)
2275 return err;
2277 option_item = nla_nest_start(skb, TEAM_ATTR_ITEM_OPTION);
2278 if (!option_item)
2279 return -EMSGSIZE;
2281 if (nla_put_string(skb, TEAM_ATTR_OPTION_NAME, option->name))
2282 goto nest_cancel;
2283 if (opt_inst_info->port &&
2284 nla_put_u32(skb, TEAM_ATTR_OPTION_PORT_IFINDEX,
2285 opt_inst_info->port->dev->ifindex))
2286 goto nest_cancel;
2287 if (opt_inst->option->array_size &&
2288 nla_put_u32(skb, TEAM_ATTR_OPTION_ARRAY_INDEX,
2289 opt_inst_info->array_index))
2290 goto nest_cancel;
2292 switch (option->type) {
2293 case TEAM_OPTION_TYPE_U32:
2294 if (nla_put_u8(skb, TEAM_ATTR_OPTION_TYPE, NLA_U32))
2295 goto nest_cancel;
2296 if (nla_put_u32(skb, TEAM_ATTR_OPTION_DATA, ctx.data.u32_val))
2297 goto nest_cancel;
2298 break;
2299 case TEAM_OPTION_TYPE_STRING:
2300 if (nla_put_u8(skb, TEAM_ATTR_OPTION_TYPE, NLA_STRING))
2301 goto nest_cancel;
2302 if (nla_put_string(skb, TEAM_ATTR_OPTION_DATA,
2303 ctx.data.str_val))
2304 goto nest_cancel;
2305 break;
2306 case TEAM_OPTION_TYPE_BINARY:
2307 if (nla_put_u8(skb, TEAM_ATTR_OPTION_TYPE, NLA_BINARY))
2308 goto nest_cancel;
2309 if (nla_put(skb, TEAM_ATTR_OPTION_DATA, ctx.data.bin_val.len,
2310 ctx.data.bin_val.ptr))
2311 goto nest_cancel;
2312 break;
2313 case TEAM_OPTION_TYPE_BOOL:
2314 if (nla_put_u8(skb, TEAM_ATTR_OPTION_TYPE, NLA_FLAG))
2315 goto nest_cancel;
2316 if (ctx.data.bool_val &&
2317 nla_put_flag(skb, TEAM_ATTR_OPTION_DATA))
2318 goto nest_cancel;
2319 break;
2320 case TEAM_OPTION_TYPE_S32:
2321 if (nla_put_u8(skb, TEAM_ATTR_OPTION_TYPE, NLA_S32))
2322 goto nest_cancel;
2323 if (nla_put_s32(skb, TEAM_ATTR_OPTION_DATA, ctx.data.s32_val))
2324 goto nest_cancel;
2325 break;
2326 default:
2327 BUG();
2329 if (opt_inst->removed && nla_put_flag(skb, TEAM_ATTR_OPTION_REMOVED))
2330 goto nest_cancel;
2331 if (opt_inst->changed) {
2332 if (nla_put_flag(skb, TEAM_ATTR_OPTION_CHANGED))
2333 goto nest_cancel;
2334 opt_inst->changed = false;
2336 nla_nest_end(skb, option_item);
2337 return 0;
2339 nest_cancel:
2340 nla_nest_cancel(skb, option_item);
2341 return -EMSGSIZE;
2344 static int __send_and_alloc_skb(struct sk_buff **pskb,
2345 struct team *team, u32 portid,
2346 team_nl_send_func_t *send_func)
2348 int err;
2350 if (*pskb) {
2351 err = send_func(*pskb, team, portid);
2352 if (err)
2353 return err;
2355 *pskb = genlmsg_new(GENLMSG_DEFAULT_SIZE, GFP_KERNEL);
2356 if (!*pskb)
2357 return -ENOMEM;
2358 return 0;
2361 static int team_nl_send_options_get(struct team *team, u32 portid, u32 seq,
2362 int flags, team_nl_send_func_t *send_func,
2363 struct list_head *sel_opt_inst_list)
2365 struct nlattr *option_list;
2366 struct nlmsghdr *nlh;
2367 void *hdr;
2368 struct team_option_inst *opt_inst;
2369 int err;
2370 struct sk_buff *skb = NULL;
2371 bool incomplete;
2372 int i;
2374 opt_inst = list_first_entry(sel_opt_inst_list,
2375 struct team_option_inst, tmp_list);
2377 start_again:
2378 err = __send_and_alloc_skb(&skb, team, portid, send_func);
2379 if (err)
2380 return err;
2382 hdr = genlmsg_put(skb, portid, seq, &team_nl_family, flags | NLM_F_MULTI,
2383 TEAM_CMD_OPTIONS_GET);
2384 if (!hdr) {
2385 nlmsg_free(skb);
2386 return -EMSGSIZE;
2389 if (nla_put_u32(skb, TEAM_ATTR_TEAM_IFINDEX, team->dev->ifindex))
2390 goto nla_put_failure;
2391 option_list = nla_nest_start(skb, TEAM_ATTR_LIST_OPTION);
2392 if (!option_list)
2393 goto nla_put_failure;
2395 i = 0;
2396 incomplete = false;
2397 list_for_each_entry_from(opt_inst, sel_opt_inst_list, tmp_list) {
2398 err = team_nl_fill_one_option_get(skb, team, opt_inst);
2399 if (err) {
2400 if (err == -EMSGSIZE) {
2401 if (!i)
2402 goto errout;
2403 incomplete = true;
2404 break;
2406 goto errout;
2408 i++;
2411 nla_nest_end(skb, option_list);
2412 genlmsg_end(skb, hdr);
2413 if (incomplete)
2414 goto start_again;
2416 send_done:
2417 nlh = nlmsg_put(skb, portid, seq, NLMSG_DONE, 0, flags | NLM_F_MULTI);
2418 if (!nlh) {
2419 err = __send_and_alloc_skb(&skb, team, portid, send_func);
2420 if (err)
2421 return err;
2422 goto send_done;
2425 return send_func(skb, team, portid);
2427 nla_put_failure:
2428 err = -EMSGSIZE;
2429 errout:
2430 nlmsg_free(skb);
2431 return err;
2434 static int team_nl_cmd_options_get(struct sk_buff *skb, struct genl_info *info)
2436 struct team *team;
2437 struct team_option_inst *opt_inst;
2438 int err;
2439 LIST_HEAD(sel_opt_inst_list);
2441 team = team_nl_team_get(info);
2442 if (!team)
2443 return -EINVAL;
2445 list_for_each_entry(opt_inst, &team->option_inst_list, list)
2446 list_add_tail(&opt_inst->tmp_list, &sel_opt_inst_list);
2447 err = team_nl_send_options_get(team, info->snd_portid, info->snd_seq,
2448 NLM_F_ACK, team_nl_send_unicast,
2449 &sel_opt_inst_list);
2451 team_nl_team_put(team);
2453 return err;
2456 static int team_nl_send_event_options_get(struct team *team,
2457 struct list_head *sel_opt_inst_list);
2459 static int team_nl_cmd_options_set(struct sk_buff *skb, struct genl_info *info)
2461 struct team *team;
2462 int err = 0;
2463 int i;
2464 struct nlattr *nl_option;
2465 LIST_HEAD(opt_inst_list);
2467 rtnl_lock();
2469 team = team_nl_team_get(info);
2470 if (!team) {
2471 err = -EINVAL;
2472 goto rtnl_unlock;
2475 err = -EINVAL;
2476 if (!info->attrs[TEAM_ATTR_LIST_OPTION]) {
2477 err = -EINVAL;
2478 goto team_put;
2481 nla_for_each_nested(nl_option, info->attrs[TEAM_ATTR_LIST_OPTION], i) {
2482 struct nlattr *opt_attrs[TEAM_ATTR_OPTION_MAX + 1];
2483 struct nlattr *attr;
2484 struct nlattr *attr_data;
2485 enum team_option_type opt_type;
2486 int opt_port_ifindex = 0; /* != 0 for per-port options */
2487 u32 opt_array_index = 0;
2488 bool opt_is_array = false;
2489 struct team_option_inst *opt_inst;
2490 char *opt_name;
2491 bool opt_found = false;
2493 if (nla_type(nl_option) != TEAM_ATTR_ITEM_OPTION) {
2494 err = -EINVAL;
2495 goto team_put;
2497 err = nla_parse_nested(opt_attrs, TEAM_ATTR_OPTION_MAX,
2498 nl_option, team_nl_option_policy,
2499 info->extack);
2500 if (err)
2501 goto team_put;
2502 if (!opt_attrs[TEAM_ATTR_OPTION_NAME] ||
2503 !opt_attrs[TEAM_ATTR_OPTION_TYPE]) {
2504 err = -EINVAL;
2505 goto team_put;
2507 switch (nla_get_u8(opt_attrs[TEAM_ATTR_OPTION_TYPE])) {
2508 case NLA_U32:
2509 opt_type = TEAM_OPTION_TYPE_U32;
2510 break;
2511 case NLA_STRING:
2512 opt_type = TEAM_OPTION_TYPE_STRING;
2513 break;
2514 case NLA_BINARY:
2515 opt_type = TEAM_OPTION_TYPE_BINARY;
2516 break;
2517 case NLA_FLAG:
2518 opt_type = TEAM_OPTION_TYPE_BOOL;
2519 break;
2520 case NLA_S32:
2521 opt_type = TEAM_OPTION_TYPE_S32;
2522 break;
2523 default:
2524 goto team_put;
2527 attr_data = opt_attrs[TEAM_ATTR_OPTION_DATA];
2528 if (opt_type != TEAM_OPTION_TYPE_BOOL && !attr_data) {
2529 err = -EINVAL;
2530 goto team_put;
2533 opt_name = nla_data(opt_attrs[TEAM_ATTR_OPTION_NAME]);
2534 attr = opt_attrs[TEAM_ATTR_OPTION_PORT_IFINDEX];
2535 if (attr)
2536 opt_port_ifindex = nla_get_u32(attr);
2538 attr = opt_attrs[TEAM_ATTR_OPTION_ARRAY_INDEX];
2539 if (attr) {
2540 opt_is_array = true;
2541 opt_array_index = nla_get_u32(attr);
2544 list_for_each_entry(opt_inst, &team->option_inst_list, list) {
2545 struct team_option *option = opt_inst->option;
2546 struct team_gsetter_ctx ctx;
2547 struct team_option_inst_info *opt_inst_info;
2548 int tmp_ifindex;
2550 opt_inst_info = &opt_inst->info;
2551 tmp_ifindex = opt_inst_info->port ?
2552 opt_inst_info->port->dev->ifindex : 0;
2553 if (option->type != opt_type ||
2554 strcmp(option->name, opt_name) ||
2555 tmp_ifindex != opt_port_ifindex ||
2556 (option->array_size && !opt_is_array) ||
2557 opt_inst_info->array_index != opt_array_index)
2558 continue;
2559 opt_found = true;
2560 ctx.info = opt_inst_info;
2561 switch (opt_type) {
2562 case TEAM_OPTION_TYPE_U32:
2563 ctx.data.u32_val = nla_get_u32(attr_data);
2564 break;
2565 case TEAM_OPTION_TYPE_STRING:
2566 if (nla_len(attr_data) > TEAM_STRING_MAX_LEN) {
2567 err = -EINVAL;
2568 goto team_put;
2570 ctx.data.str_val = nla_data(attr_data);
2571 break;
2572 case TEAM_OPTION_TYPE_BINARY:
2573 ctx.data.bin_val.len = nla_len(attr_data);
2574 ctx.data.bin_val.ptr = nla_data(attr_data);
2575 break;
2576 case TEAM_OPTION_TYPE_BOOL:
2577 ctx.data.bool_val = attr_data ? true : false;
2578 break;
2579 case TEAM_OPTION_TYPE_S32:
2580 ctx.data.s32_val = nla_get_s32(attr_data);
2581 break;
2582 default:
2583 BUG();
2585 err = team_option_set(team, opt_inst, &ctx);
2586 if (err)
2587 goto team_put;
2588 opt_inst->changed = true;
2590 /* dumb/evil user-space can send us duplicate opt,
2591 * keep only the last one
2593 if (__team_option_inst_tmp_find(&opt_inst_list,
2594 opt_inst))
2595 continue;
2597 list_add(&opt_inst->tmp_list, &opt_inst_list);
2599 if (!opt_found) {
2600 err = -ENOENT;
2601 goto team_put;
2605 err = team_nl_send_event_options_get(team, &opt_inst_list);
2607 team_put:
2608 team_nl_team_put(team);
2609 rtnl_unlock:
2610 rtnl_unlock();
2611 return err;
2614 static int team_nl_fill_one_port_get(struct sk_buff *skb,
2615 struct team_port *port)
2617 struct nlattr *port_item;
2619 port_item = nla_nest_start(skb, TEAM_ATTR_ITEM_PORT);
2620 if (!port_item)
2621 goto nest_cancel;
2622 if (nla_put_u32(skb, TEAM_ATTR_PORT_IFINDEX, port->dev->ifindex))
2623 goto nest_cancel;
2624 if (port->changed) {
2625 if (nla_put_flag(skb, TEAM_ATTR_PORT_CHANGED))
2626 goto nest_cancel;
2627 port->changed = false;
2629 if ((port->removed &&
2630 nla_put_flag(skb, TEAM_ATTR_PORT_REMOVED)) ||
2631 (port->state.linkup &&
2632 nla_put_flag(skb, TEAM_ATTR_PORT_LINKUP)) ||
2633 nla_put_u32(skb, TEAM_ATTR_PORT_SPEED, port->state.speed) ||
2634 nla_put_u8(skb, TEAM_ATTR_PORT_DUPLEX, port->state.duplex))
2635 goto nest_cancel;
2636 nla_nest_end(skb, port_item);
2637 return 0;
2639 nest_cancel:
2640 nla_nest_cancel(skb, port_item);
2641 return -EMSGSIZE;
2644 static int team_nl_send_port_list_get(struct team *team, u32 portid, u32 seq,
2645 int flags, team_nl_send_func_t *send_func,
2646 struct team_port *one_port)
2648 struct nlattr *port_list;
2649 struct nlmsghdr *nlh;
2650 void *hdr;
2651 struct team_port *port;
2652 int err;
2653 struct sk_buff *skb = NULL;
2654 bool incomplete;
2655 int i;
2657 port = list_first_entry_or_null(&team->port_list,
2658 struct team_port, list);
2660 start_again:
2661 err = __send_and_alloc_skb(&skb, team, portid, send_func);
2662 if (err)
2663 return err;
2665 hdr = genlmsg_put(skb, portid, seq, &team_nl_family, flags | NLM_F_MULTI,
2666 TEAM_CMD_PORT_LIST_GET);
2667 if (!hdr) {
2668 nlmsg_free(skb);
2669 return -EMSGSIZE;
2672 if (nla_put_u32(skb, TEAM_ATTR_TEAM_IFINDEX, team->dev->ifindex))
2673 goto nla_put_failure;
2674 port_list = nla_nest_start(skb, TEAM_ATTR_LIST_PORT);
2675 if (!port_list)
2676 goto nla_put_failure;
2678 i = 0;
2679 incomplete = false;
2681 /* If one port is selected, called wants to send port list containing
2682 * only this port. Otherwise go through all listed ports and send all
2684 if (one_port) {
2685 err = team_nl_fill_one_port_get(skb, one_port);
2686 if (err)
2687 goto errout;
2688 } else if (port) {
2689 list_for_each_entry_from(port, &team->port_list, list) {
2690 err = team_nl_fill_one_port_get(skb, port);
2691 if (err) {
2692 if (err == -EMSGSIZE) {
2693 if (!i)
2694 goto errout;
2695 incomplete = true;
2696 break;
2698 goto errout;
2700 i++;
2704 nla_nest_end(skb, port_list);
2705 genlmsg_end(skb, hdr);
2706 if (incomplete)
2707 goto start_again;
2709 send_done:
2710 nlh = nlmsg_put(skb, portid, seq, NLMSG_DONE, 0, flags | NLM_F_MULTI);
2711 if (!nlh) {
2712 err = __send_and_alloc_skb(&skb, team, portid, send_func);
2713 if (err)
2714 return err;
2715 goto send_done;
2718 return send_func(skb, team, portid);
2720 nla_put_failure:
2721 err = -EMSGSIZE;
2722 errout:
2723 nlmsg_free(skb);
2724 return err;
2727 static int team_nl_cmd_port_list_get(struct sk_buff *skb,
2728 struct genl_info *info)
2730 struct team *team;
2731 int err;
2733 team = team_nl_team_get(info);
2734 if (!team)
2735 return -EINVAL;
2737 err = team_nl_send_port_list_get(team, info->snd_portid, info->snd_seq,
2738 NLM_F_ACK, team_nl_send_unicast, NULL);
2740 team_nl_team_put(team);
2742 return err;
2745 static const struct genl_ops team_nl_ops[] = {
2747 .cmd = TEAM_CMD_NOOP,
2748 .doit = team_nl_cmd_noop,
2749 .policy = team_nl_policy,
2752 .cmd = TEAM_CMD_OPTIONS_SET,
2753 .doit = team_nl_cmd_options_set,
2754 .policy = team_nl_policy,
2755 .flags = GENL_ADMIN_PERM,
2758 .cmd = TEAM_CMD_OPTIONS_GET,
2759 .doit = team_nl_cmd_options_get,
2760 .policy = team_nl_policy,
2761 .flags = GENL_ADMIN_PERM,
2764 .cmd = TEAM_CMD_PORT_LIST_GET,
2765 .doit = team_nl_cmd_port_list_get,
2766 .policy = team_nl_policy,
2767 .flags = GENL_ADMIN_PERM,
2771 static const struct genl_multicast_group team_nl_mcgrps[] = {
2772 { .name = TEAM_GENL_CHANGE_EVENT_MC_GRP_NAME, },
2775 static struct genl_family team_nl_family __ro_after_init = {
2776 .name = TEAM_GENL_NAME,
2777 .version = TEAM_GENL_VERSION,
2778 .maxattr = TEAM_ATTR_MAX,
2779 .netnsok = true,
2780 .module = THIS_MODULE,
2781 .ops = team_nl_ops,
2782 .n_ops = ARRAY_SIZE(team_nl_ops),
2783 .mcgrps = team_nl_mcgrps,
2784 .n_mcgrps = ARRAY_SIZE(team_nl_mcgrps),
2787 static int team_nl_send_multicast(struct sk_buff *skb,
2788 struct team *team, u32 portid)
2790 return genlmsg_multicast_netns(&team_nl_family, dev_net(team->dev),
2791 skb, 0, 0, GFP_KERNEL);
2794 static int team_nl_send_event_options_get(struct team *team,
2795 struct list_head *sel_opt_inst_list)
2797 return team_nl_send_options_get(team, 0, 0, 0, team_nl_send_multicast,
2798 sel_opt_inst_list);
2801 static int team_nl_send_event_port_get(struct team *team,
2802 struct team_port *port)
2804 return team_nl_send_port_list_get(team, 0, 0, 0, team_nl_send_multicast,
2805 port);
2808 static int __init team_nl_init(void)
2810 return genl_register_family(&team_nl_family);
2813 static void team_nl_fini(void)
2815 genl_unregister_family(&team_nl_family);
2819 /******************
2820 * Change checkers
2821 ******************/
2823 static void __team_options_change_check(struct team *team)
2825 int err;
2826 struct team_option_inst *opt_inst;
2827 LIST_HEAD(sel_opt_inst_list);
2829 list_for_each_entry(opt_inst, &team->option_inst_list, list) {
2830 if (opt_inst->changed)
2831 list_add_tail(&opt_inst->tmp_list, &sel_opt_inst_list);
2833 err = team_nl_send_event_options_get(team, &sel_opt_inst_list);
2834 if (err && err != -ESRCH)
2835 netdev_warn(team->dev, "Failed to send options change via netlink (err %d)\n",
2836 err);
2839 /* rtnl lock is held */
2841 static void __team_port_change_send(struct team_port *port, bool linkup)
2843 int err;
2845 port->changed = true;
2846 port->state.linkup = linkup;
2847 team_refresh_port_linkup(port);
2848 if (linkup) {
2849 struct ethtool_link_ksettings ecmd;
2851 err = __ethtool_get_link_ksettings(port->dev, &ecmd);
2852 if (!err) {
2853 port->state.speed = ecmd.base.speed;
2854 port->state.duplex = ecmd.base.duplex;
2855 goto send_event;
2858 port->state.speed = 0;
2859 port->state.duplex = 0;
2861 send_event:
2862 err = team_nl_send_event_port_get(port->team, port);
2863 if (err && err != -ESRCH)
2864 netdev_warn(port->team->dev, "Failed to send port change of device %s via netlink (err %d)\n",
2865 port->dev->name, err);
2869 static void __team_carrier_check(struct team *team)
2871 struct team_port *port;
2872 bool team_linkup;
2874 if (team->user_carrier_enabled)
2875 return;
2877 team_linkup = false;
2878 list_for_each_entry(port, &team->port_list, list) {
2879 if (port->linkup) {
2880 team_linkup = true;
2881 break;
2885 if (team_linkup)
2886 netif_carrier_on(team->dev);
2887 else
2888 netif_carrier_off(team->dev);
2891 static void __team_port_change_check(struct team_port *port, bool linkup)
2893 if (port->state.linkup != linkup)
2894 __team_port_change_send(port, linkup);
2895 __team_carrier_check(port->team);
2898 static void __team_port_change_port_added(struct team_port *port, bool linkup)
2900 __team_port_change_send(port, linkup);
2901 __team_carrier_check(port->team);
2904 static void __team_port_change_port_removed(struct team_port *port)
2906 port->removed = true;
2907 __team_port_change_send(port, false);
2908 __team_carrier_check(port->team);
2911 static void team_port_change_check(struct team_port *port, bool linkup)
2913 struct team *team = port->team;
2915 mutex_lock(&team->lock);
2916 __team_port_change_check(port, linkup);
2917 mutex_unlock(&team->lock);
2921 /************************************
2922 * Net device notifier event handler
2923 ************************************/
2925 static int team_device_event(struct notifier_block *unused,
2926 unsigned long event, void *ptr)
2928 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2929 struct team_port *port;
2931 port = team_port_get_rtnl(dev);
2932 if (!port)
2933 return NOTIFY_DONE;
2935 switch (event) {
2936 case NETDEV_UP:
2937 if (netif_carrier_ok(dev))
2938 team_port_change_check(port, true);
2939 break;
2940 case NETDEV_DOWN:
2941 team_port_change_check(port, false);
2942 break;
2943 case NETDEV_CHANGE:
2944 if (netif_running(port->dev))
2945 team_port_change_check(port,
2946 !!netif_oper_up(port->dev));
2947 break;
2948 case NETDEV_UNREGISTER:
2949 team_del_slave(port->team->dev, dev);
2950 break;
2951 case NETDEV_FEAT_CHANGE:
2952 team_compute_features(port->team);
2953 break;
2954 case NETDEV_PRECHANGEMTU:
2955 /* Forbid to change mtu of underlaying device */
2956 if (!port->team->port_mtu_change_allowed)
2957 return NOTIFY_BAD;
2958 break;
2959 case NETDEV_PRE_TYPE_CHANGE:
2960 /* Forbid to change type of underlaying device */
2961 return NOTIFY_BAD;
2962 case NETDEV_RESEND_IGMP:
2963 /* Propagate to master device */
2964 call_netdevice_notifiers(event, port->team->dev);
2965 break;
2967 return NOTIFY_DONE;
2970 static struct notifier_block team_notifier_block __read_mostly = {
2971 .notifier_call = team_device_event,
2975 /***********************
2976 * Module init and exit
2977 ***********************/
2979 static int __init team_module_init(void)
2981 int err;
2983 register_netdevice_notifier(&team_notifier_block);
2985 err = rtnl_link_register(&team_link_ops);
2986 if (err)
2987 goto err_rtnl_reg;
2989 err = team_nl_init();
2990 if (err)
2991 goto err_nl_init;
2993 return 0;
2995 err_nl_init:
2996 rtnl_link_unregister(&team_link_ops);
2998 err_rtnl_reg:
2999 unregister_netdevice_notifier(&team_notifier_block);
3001 return err;
3004 static void __exit team_module_exit(void)
3006 team_nl_fini();
3007 rtnl_link_unregister(&team_link_ops);
3008 unregister_netdevice_notifier(&team_notifier_block);
3011 module_init(team_module_init);
3012 module_exit(team_module_exit);
3014 MODULE_LICENSE("GPL v2");
3015 MODULE_AUTHOR("Jiri Pirko <jpirko@redhat.com>");
3016 MODULE_DESCRIPTION("Ethernet team device driver");
3017 MODULE_ALIAS_RTNL_LINK(DRV_NAME);