2 * Bond several ethernet interfaces into a Cisco, running 'Etherchannel'.
4 * Portions are (c) Copyright 1995 Simon "Guru Aleph-Null" Janes
5 * NCM: Network and Communications Management, Inc.
7 * BUT, I'm the one who modified it for ethernet, so:
8 * (c) Copyright 1999, Thomas Davis, tadavis@lbl.gov
10 * This software may be used and distributed according to the terms
11 * of the GNU Public License, incorporated herein by reference.
15 #ifndef _NET_BONDING_H
16 #define _NET_BONDING_H
18 #include <linux/timer.h>
19 #include <linux/proc_fs.h>
20 #include <linux/if_bonding.h>
21 #include <linux/cpumask.h>
22 #include <linux/in6.h>
23 #include <linux/netpoll.h>
24 #include <linux/inetdevice.h>
25 #include <linux/etherdevice.h>
26 #include <linux/reciprocal_div.h>
27 #include <linux/if_link.h>
29 #include <net/bond_3ad.h>
30 #include <net/bond_alb.h>
31 #include <net/bond_options.h>
33 #include <net/addrconf.h>
35 #define BOND_MAX_ARP_TARGETS 16
36 #define BOND_MAX_NS_TARGETS BOND_MAX_ARP_TARGETS
38 #define BOND_DEFAULT_MIIMON 100
40 #ifndef __long_aligned
41 #define __long_aligned __attribute__((aligned((sizeof(long)))))
44 #define slave_info(bond_dev, slave_dev, fmt, ...) \
45 netdev_info(bond_dev, "(slave %s): " fmt, (slave_dev)->name, ##__VA_ARGS__)
46 #define slave_warn(bond_dev, slave_dev, fmt, ...) \
47 netdev_warn(bond_dev, "(slave %s): " fmt, (slave_dev)->name, ##__VA_ARGS__)
48 #define slave_dbg(bond_dev, slave_dev, fmt, ...) \
49 netdev_dbg(bond_dev, "(slave %s): " fmt, (slave_dev)->name, ##__VA_ARGS__)
50 #define slave_err(bond_dev, slave_dev, fmt, ...) \
51 netdev_err(bond_dev, "(slave %s): " fmt, (slave_dev)->name, ##__VA_ARGS__)
53 #define BOND_MODE(bond) ((bond)->params.mode)
55 /* slave list primitives */
56 #define bond_slave_list(bond) (&(bond)->dev->adj_list.lower)
58 #define bond_has_slaves(bond) !list_empty(bond_slave_list(bond))
60 /* IMPORTANT: bond_first/last_slave can return NULL in case of an empty list */
61 #define bond_first_slave(bond) \
62 (bond_has_slaves(bond) ? \
63 netdev_adjacent_get_private(bond_slave_list(bond)->next) : \
65 #define bond_last_slave(bond) \
66 (bond_has_slaves(bond) ? \
67 netdev_adjacent_get_private(bond_slave_list(bond)->prev) : \
70 /* Caller must have rcu_read_lock */
71 #define bond_first_slave_rcu(bond) \
72 netdev_lower_get_first_private_rcu(bond->dev)
74 #define bond_is_first_slave(bond, pos) (pos == bond_first_slave(bond))
75 #define bond_is_last_slave(bond, pos) (pos == bond_last_slave(bond))
78 * bond_for_each_slave - iterate over all slaves
79 * @bond: the bond holding this list
81 * @iter: list_head * iterator
83 * Caller must hold RTNL
85 #define bond_for_each_slave(bond, pos, iter) \
86 netdev_for_each_lower_private((bond)->dev, pos, iter)
88 /* Caller must have rcu_read_lock */
89 #define bond_for_each_slave_rcu(bond, pos, iter) \
90 netdev_for_each_lower_private_rcu((bond)->dev, pos, iter)
92 #define BOND_XFRM_FEATURES (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM | \
95 #define BOND_TLS_FEATURES (NETIF_F_HW_TLS_TX | NETIF_F_HW_TLS_RX)
97 #ifdef CONFIG_NET_POLL_CONTROLLER
98 extern atomic_t netpoll_block_tx;
100 static inline void block_netpoll_tx(void)
102 atomic_inc(&netpoll_block_tx);
105 static inline void unblock_netpoll_tx(void)
107 atomic_dec(&netpoll_block_tx);
110 static inline int is_netpoll_tx_blocked(struct net_device *dev)
112 if (unlikely(netpoll_tx_running(dev)))
113 return atomic_read(&netpoll_block_tx);
117 #define block_netpoll_tx()
118 #define unblock_netpoll_tx()
119 #define is_netpoll_tx_blocked(dev) (0)
135 int peer_notif_delay;
138 unsigned int min_links;
140 char primary[IFNAMSIZ];
141 int primary_reselect;
142 __be32 arp_targets[BOND_MAX_ARP_TARGETS];
144 int all_slaves_active;
147 int packets_per_slave;
149 struct reciprocal_value reciprocal_packets_per_slave;
150 u16 ad_actor_sys_prio;
151 u16 ad_user_port_key;
152 #if IS_ENABLED(CONFIG_IPV6)
153 struct in6_addr ns_targets[BOND_MAX_NS_TARGETS];
156 /* 2 bytes of padding : see ether_addr_equal_64bits() */
157 u8 ad_actor_system[ETH_ALEN + 2];
161 struct net_device *dev; /* first - useful for panic debug */
162 struct bonding *bond; /* our master */
164 /* all 4 in jiffies */
165 unsigned long last_link_up;
166 unsigned long last_tx;
167 unsigned long last_rx;
168 unsigned long target_last_arp_rx[BOND_MAX_ARP_TARGETS];
169 s8 link; /* one of BOND_LINK_XXXX */
170 s8 link_new_state; /* one of BOND_LINK_XXXX */
171 u8 backup:1, /* indicates backup slave. Value corresponds with
172 BOND_STATE_ACTIVE and BOND_STATE_BACKUP */
173 inactive:1, /* indicates inactive slave */
174 should_notify:1, /* indicates whether the state changed */
175 should_notify_link:1; /* indicates whether the link changed */
178 u32 link_failure_count;
181 u8 perm_hwaddr[MAX_ADDR_LEN];
183 struct ad_slave_info *ad_info;
184 struct tlb_slave_info tlb_info;
185 #ifdef CONFIG_NET_POLL_CONTROLLER
188 struct delayed_work notify_work;
190 struct rtnl_link_stats64 slave_stats;
193 static inline struct slave *to_slave(struct kobject *kobj)
195 return container_of(kobj, struct slave, kobj);
198 struct bond_up_slave {
205 * Link pseudo-state only used internally by monitors
207 #define BOND_LINK_NOCHANGE -1
210 struct list_head list;
211 struct xfrm_state *xs;
215 * Here are the locking policies for the two bonding locks:
216 * Get rcu_read_lock when reading or RTNL when writing slave list.
219 struct net_device *dev; /* first - useful for panic debug */
220 struct slave __rcu *curr_active_slave;
221 struct slave __rcu *current_arp_slave;
222 struct slave __rcu *primary_slave;
223 struct bond_up_slave __rcu *usable_slaves;
224 struct bond_up_slave __rcu *all_slaves;
226 s32 slave_cnt; /* never change this value outside the attach/detach wrappers */
227 int (*recv_probe)(const struct sk_buff *, struct bonding *,
229 /* mode_lock is used for mode-specific locking needs, currently used by:
230 * 3ad mode (4) - protect against running bond_3ad_unbind_slave() and
231 * bond_3ad_state_machine_handler() concurrently and also
232 * the access to the state machine shared variables.
233 * TLB mode (5) - to sync the use and modifications of its hash table
234 * ALB mode (6) - to sync the use and modifications of its hash table
236 spinlock_t mode_lock;
237 spinlock_t stats_lock;
240 #ifdef CONFIG_PROC_FS
241 struct proc_dir_entry *proc_entry;
242 char proc_file_name[IFNAMSIZ];
243 #endif /* CONFIG_PROC_FS */
244 struct list_head bond_list;
245 u32 __percpu *rr_tx_counter;
246 struct ad_bond_info ad_info;
247 struct alb_bond_info alb_info;
248 struct bond_params params;
249 struct workqueue_struct *wq;
250 struct delayed_work mii_work;
251 struct delayed_work arp_work;
252 struct delayed_work alb_work;
253 struct delayed_work ad_work;
254 struct delayed_work mcast_work;
255 struct delayed_work slave_arr_work;
256 #ifdef CONFIG_DEBUG_FS
257 /* debugging support via debugfs */
258 struct dentry *debug_dir;
259 #endif /* CONFIG_DEBUG_FS */
260 struct rtnl_link_stats64 bond_stats;
261 #ifdef CONFIG_XFRM_OFFLOAD
262 struct list_head ipsec_list;
263 /* protecting ipsec_list */
264 spinlock_t ipsec_lock;
265 #endif /* CONFIG_XFRM_OFFLOAD */
266 struct bpf_prog *xdp_prog;
269 #define bond_slave_get_rcu(dev) \
270 ((struct slave *) rcu_dereference(dev->rx_handler_data))
272 #define bond_slave_get_rtnl(dev) \
273 ((struct slave *) rtnl_dereference(dev->rx_handler_data))
275 void bond_queue_slave_event(struct slave *slave);
276 void bond_lower_state_changed(struct slave *slave);
278 struct bond_vlan_tag {
280 unsigned short vlan_id;
283 bool bond_sk_check(struct bonding *bond);
286 * Returns NULL if the net_device does not belong to any of the bond's slaves
288 * Caller must hold bond lock for read
290 static inline struct slave *bond_get_slave_by_dev(struct bonding *bond,
291 struct net_device *slave_dev)
293 return netdev_lower_dev_get_private(bond->dev, slave_dev);
296 static inline struct bonding *bond_get_bond_by_slave(struct slave *slave)
301 static inline bool bond_should_override_tx_queue(struct bonding *bond)
303 return BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
304 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN;
307 static inline bool bond_is_lb(const struct bonding *bond)
309 return BOND_MODE(bond) == BOND_MODE_TLB ||
310 BOND_MODE(bond) == BOND_MODE_ALB;
313 static inline bool bond_needs_speed_duplex(const struct bonding *bond)
315 return BOND_MODE(bond) == BOND_MODE_8023AD || bond_is_lb(bond);
318 static inline bool bond_is_nondyn_tlb(const struct bonding *bond)
320 return (bond_is_lb(bond) && bond->params.tlb_dynamic_lb == 0);
323 static inline bool bond_mode_can_use_xmit_hash(const struct bonding *bond)
325 return (BOND_MODE(bond) == BOND_MODE_8023AD ||
326 BOND_MODE(bond) == BOND_MODE_XOR ||
327 BOND_MODE(bond) == BOND_MODE_TLB ||
328 BOND_MODE(bond) == BOND_MODE_ALB);
331 static inline bool bond_mode_uses_xmit_hash(const struct bonding *bond)
333 return (BOND_MODE(bond) == BOND_MODE_8023AD ||
334 BOND_MODE(bond) == BOND_MODE_XOR ||
335 bond_is_nondyn_tlb(bond));
338 static inline bool bond_mode_uses_arp(int mode)
340 return mode != BOND_MODE_8023AD && mode != BOND_MODE_TLB &&
341 mode != BOND_MODE_ALB;
344 static inline bool bond_mode_uses_primary(int mode)
346 return mode == BOND_MODE_ACTIVEBACKUP || mode == BOND_MODE_TLB ||
347 mode == BOND_MODE_ALB;
350 static inline bool bond_uses_primary(struct bonding *bond)
352 return bond_mode_uses_primary(BOND_MODE(bond));
355 static inline struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond)
357 struct slave *slave = rcu_dereference_rtnl(bond->curr_active_slave);
359 return bond_uses_primary(bond) && slave ? slave->dev : NULL;
362 static inline bool bond_slave_is_up(struct slave *slave)
364 return netif_running(slave->dev) && netif_carrier_ok(slave->dev);
367 static inline void bond_set_active_slave(struct slave *slave)
371 bond_queue_slave_event(slave);
372 bond_lower_state_changed(slave);
376 static inline void bond_set_backup_slave(struct slave *slave)
378 if (!slave->backup) {
380 bond_queue_slave_event(slave);
381 bond_lower_state_changed(slave);
385 static inline void bond_set_slave_state(struct slave *slave,
386 int slave_state, bool notify)
388 if (slave->backup == slave_state)
391 slave->backup = slave_state;
393 bond_lower_state_changed(slave);
394 bond_queue_slave_event(slave);
395 slave->should_notify = 0;
397 if (slave->should_notify)
398 slave->should_notify = 0;
400 slave->should_notify = 1;
404 static inline void bond_slave_state_change(struct bonding *bond)
406 struct list_head *iter;
409 bond_for_each_slave(bond, tmp, iter) {
410 if (tmp->link == BOND_LINK_UP)
411 bond_set_active_slave(tmp);
412 else if (tmp->link == BOND_LINK_DOWN)
413 bond_set_backup_slave(tmp);
417 static inline void bond_slave_state_notify(struct bonding *bond)
419 struct list_head *iter;
422 bond_for_each_slave(bond, tmp, iter) {
423 if (tmp->should_notify) {
424 bond_lower_state_changed(tmp);
425 tmp->should_notify = 0;
430 static inline int bond_slave_state(struct slave *slave)
432 return slave->backup;
435 static inline bool bond_is_active_slave(struct slave *slave)
437 return !bond_slave_state(slave);
440 static inline bool bond_slave_can_tx(struct slave *slave)
442 return bond_slave_is_up(slave) && slave->link == BOND_LINK_UP &&
443 bond_is_active_slave(slave);
446 static inline bool bond_is_active_slave_dev(const struct net_device *slave_dev)
452 slave = bond_slave_get_rcu(slave_dev);
453 active = bond_is_active_slave(slave);
459 static inline void bond_hw_addr_copy(u8 *dst, const u8 *src, unsigned int len)
461 if (len == ETH_ALEN) {
462 ether_addr_copy(dst, src);
466 memcpy(dst, src, len);
469 #define BOND_PRI_RESELECT_ALWAYS 0
470 #define BOND_PRI_RESELECT_BETTER 1
471 #define BOND_PRI_RESELECT_FAILURE 2
473 #define BOND_FOM_NONE 0
474 #define BOND_FOM_ACTIVE 1
475 #define BOND_FOM_FOLLOW 2
477 #define BOND_ARP_TARGETS_ANY 0
478 #define BOND_ARP_TARGETS_ALL 1
480 #define BOND_ARP_VALIDATE_NONE 0
481 #define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE)
482 #define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP)
483 #define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \
484 BOND_ARP_VALIDATE_BACKUP)
485 #define BOND_ARP_FILTER (BOND_ARP_VALIDATE_ALL + 1)
486 #define BOND_ARP_FILTER_ACTIVE (BOND_ARP_VALIDATE_ACTIVE | \
488 #define BOND_ARP_FILTER_BACKUP (BOND_ARP_VALIDATE_BACKUP | \
491 #define BOND_SLAVE_NOTIFY_NOW true
492 #define BOND_SLAVE_NOTIFY_LATER false
494 static inline int slave_do_arp_validate(struct bonding *bond,
497 return bond->params.arp_validate & (1 << bond_slave_state(slave));
500 static inline int slave_do_arp_validate_only(struct bonding *bond)
502 return bond->params.arp_validate & BOND_ARP_FILTER;
505 static inline int bond_is_ip_target_ok(__be32 addr)
507 return !ipv4_is_lbcast(addr) && !ipv4_is_zeronet(addr);
510 #if IS_ENABLED(CONFIG_IPV6)
511 static inline int bond_is_ip6_target_ok(struct in6_addr *addr)
513 return !ipv6_addr_any(addr) &&
514 !ipv6_addr_loopback(addr) &&
515 !ipv6_addr_is_multicast(addr);
519 /* Get the oldest arp which we've received on this slave for bond's
522 static inline unsigned long slave_oldest_target_arp_rx(struct bonding *bond,
526 unsigned long ret = slave->target_last_arp_rx[0];
528 for (; (i < BOND_MAX_ARP_TARGETS) && bond->params.arp_targets[i]; i++)
529 if (time_before(slave->target_last_arp_rx[i], ret))
530 ret = slave->target_last_arp_rx[i];
535 static inline unsigned long slave_last_rx(struct bonding *bond,
538 if (bond->params.arp_all_targets == BOND_ARP_TARGETS_ALL)
539 return slave_oldest_target_arp_rx(bond, slave);
541 return slave->last_rx;
544 static inline void slave_update_last_tx(struct slave *slave)
546 WRITE_ONCE(slave->last_tx, jiffies);
549 static inline unsigned long slave_last_tx(struct slave *slave)
551 return READ_ONCE(slave->last_tx);
554 #ifdef CONFIG_NET_POLL_CONTROLLER
555 static inline netdev_tx_t bond_netpoll_send_skb(const struct slave *slave,
558 return netpoll_send_skb(slave->np, skb);
561 static inline netdev_tx_t bond_netpoll_send_skb(const struct slave *slave,
569 static inline void bond_set_slave_inactive_flags(struct slave *slave,
572 if (!bond_is_lb(slave->bond))
573 bond_set_slave_state(slave, BOND_STATE_BACKUP, notify);
574 if (!slave->bond->params.all_slaves_active)
578 static inline void bond_set_slave_active_flags(struct slave *slave,
581 bond_set_slave_state(slave, BOND_STATE_ACTIVE, notify);
585 static inline bool bond_is_slave_inactive(struct slave *slave)
587 return slave->inactive;
590 static inline void bond_propose_link_state(struct slave *slave, int state)
592 slave->link_new_state = state;
595 static inline void bond_commit_link_state(struct slave *slave, bool notify)
597 if (slave->link_new_state == BOND_LINK_NOCHANGE)
600 slave->link = slave->link_new_state;
602 bond_queue_slave_event(slave);
603 bond_lower_state_changed(slave);
604 slave->should_notify_link = 0;
606 if (slave->should_notify_link)
607 slave->should_notify_link = 0;
609 slave->should_notify_link = 1;
613 static inline void bond_set_slave_link_state(struct slave *slave, int state,
616 bond_propose_link_state(slave, state);
617 bond_commit_link_state(slave, notify);
620 static inline void bond_slave_link_notify(struct bonding *bond)
622 struct list_head *iter;
625 bond_for_each_slave(bond, tmp, iter) {
626 if (tmp->should_notify_link) {
627 bond_queue_slave_event(tmp);
628 bond_lower_state_changed(tmp);
629 tmp->should_notify_link = 0;
634 static inline __be32 bond_confirm_addr(struct net_device *dev, __be32 dst, __be32 local)
636 struct in_device *in_dev;
640 in_dev = __in_dev_get_rcu(dev);
643 addr = inet_confirm_addr(dev_net(dev), in_dev, dst, local,
650 struct net *net; /* Associated network namespace */
651 struct list_head dev_list;
652 #ifdef CONFIG_PROC_FS
653 struct proc_dir_entry *proc_dir;
655 struct class_attribute class_attr_bonding_masters;
658 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond, struct slave *slave);
659 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev);
660 int bond_create(struct net *net, const char *name);
661 int bond_create_sysfs(struct bond_net *net);
662 void bond_destroy_sysfs(struct bond_net *net);
663 void bond_prepare_sysfs_group(struct bonding *bond);
664 int bond_sysfs_slave_add(struct slave *slave);
665 void bond_sysfs_slave_del(struct slave *slave);
666 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
667 struct netlink_ext_ack *extack);
668 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev);
669 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb);
670 int bond_set_carrier(struct bonding *bond);
671 void bond_select_active_slave(struct bonding *bond);
672 void bond_change_active_slave(struct bonding *bond, struct slave *new_active);
673 void bond_create_debugfs(void);
674 void bond_destroy_debugfs(void);
675 void bond_debug_register(struct bonding *bond);
676 void bond_debug_unregister(struct bonding *bond);
677 void bond_debug_reregister(struct bonding *bond);
678 const char *bond_mode_name(int mode);
679 void bond_setup(struct net_device *bond_dev);
680 unsigned int bond_get_num_tx_queues(void);
681 int bond_netlink_init(void);
682 void bond_netlink_fini(void);
683 struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond);
684 const char *bond_slave_link_status(s8 link);
685 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
686 struct net_device *end_dev,
688 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave);
689 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay);
690 void bond_work_init_all(struct bonding *bond);
692 #ifdef CONFIG_PROC_FS
693 void bond_create_proc_entry(struct bonding *bond);
694 void bond_remove_proc_entry(struct bonding *bond);
695 void bond_create_proc_dir(struct bond_net *bn);
696 void bond_destroy_proc_dir(struct bond_net *bn);
698 static inline void bond_create_proc_entry(struct bonding *bond)
702 static inline void bond_remove_proc_entry(struct bonding *bond)
706 static inline void bond_create_proc_dir(struct bond_net *bn)
710 static inline void bond_destroy_proc_dir(struct bond_net *bn)
715 static inline struct slave *bond_slave_has_mac(struct bonding *bond,
718 struct list_head *iter;
721 bond_for_each_slave(bond, tmp, iter)
722 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
728 /* Caller must hold rcu_read_lock() for read */
729 static inline bool bond_slave_has_mac_rx(struct bonding *bond, const u8 *mac)
731 struct list_head *iter;
733 struct netdev_hw_addr *ha;
735 bond_for_each_slave_rcu(bond, tmp, iter)
736 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
739 if (netdev_uc_empty(bond->dev))
742 netdev_for_each_uc_addr(ha, bond->dev)
743 if (ether_addr_equal_64bits(mac, ha->addr))
749 /* Check if the ip is present in arp ip list, or first free slot if ip == 0
750 * Returns -1 if not found, index if found
752 static inline int bond_get_targets_ip(__be32 *targets, __be32 ip)
756 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
757 if (targets[i] == ip)
759 else if (targets[i] == 0)
765 #if IS_ENABLED(CONFIG_IPV6)
766 static inline int bond_get_targets_ip6(struct in6_addr *targets, struct in6_addr *ip)
770 for (i = 0; i < BOND_MAX_NS_TARGETS; i++)
771 if (ipv6_addr_equal(&targets[i], ip))
773 else if (ipv6_addr_any(&targets[i]))
780 /* exported from bond_main.c */
781 extern unsigned int bond_net_id;
783 /* exported from bond_netlink.c */
784 extern struct rtnl_link_ops bond_link_ops;
786 /* exported from bond_sysfs_slave.c */
787 extern const struct sysfs_ops slave_sysfs_ops;
789 static inline netdev_tx_t bond_tx_drop(struct net_device *dev, struct sk_buff *skb)
791 dev_core_stats_tx_dropped_inc(dev);
792 dev_kfree_skb_any(skb);
793 return NET_XMIT_DROP;
796 #endif /* _NET_BONDING_H */