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 three in jiffies */
165 unsigned long last_link_up;
166 unsigned long last_rx;
167 unsigned long target_last_arp_rx[BOND_MAX_ARP_TARGETS];
168 s8 link; /* one of BOND_LINK_XXXX */
169 s8 link_new_state; /* one of BOND_LINK_XXXX */
170 u8 backup:1, /* indicates backup slave. Value corresponds with
171 BOND_STATE_ACTIVE and BOND_STATE_BACKUP */
172 inactive:1, /* indicates inactive slave */
173 should_notify:1, /* indicates whether the state changed */
174 should_notify_link:1; /* indicates whether the link changed */
177 u32 link_failure_count;
180 u8 perm_hwaddr[MAX_ADDR_LEN];
181 struct ad_slave_info *ad_info;
182 struct tlb_slave_info tlb_info;
183 #ifdef CONFIG_NET_POLL_CONTROLLER
186 struct delayed_work notify_work;
188 struct rtnl_link_stats64 slave_stats;
191 static inline struct slave *to_slave(struct kobject *kobj)
193 return container_of(kobj, struct slave, kobj);
196 struct bond_up_slave {
203 * Link pseudo-state only used internally by monitors
205 #define BOND_LINK_NOCHANGE -1
208 struct list_head list;
209 struct xfrm_state *xs;
213 * Here are the locking policies for the two bonding locks:
214 * Get rcu_read_lock when reading or RTNL when writing slave list.
217 struct net_device *dev; /* first - useful for panic debug */
218 struct slave __rcu *curr_active_slave;
219 struct slave __rcu *current_arp_slave;
220 struct slave __rcu *primary_slave;
221 struct bond_up_slave __rcu *usable_slaves;
222 struct bond_up_slave __rcu *all_slaves;
224 s32 slave_cnt; /* never change this value outside the attach/detach wrappers */
225 int (*recv_probe)(const struct sk_buff *, struct bonding *,
227 /* mode_lock is used for mode-specific locking needs, currently used by:
228 * 3ad mode (4) - protect against running bond_3ad_unbind_slave() and
229 * bond_3ad_state_machine_handler() concurrently and also
230 * the access to the state machine shared variables.
231 * TLB mode (5) - to sync the use and modifications of its hash table
232 * ALB mode (6) - to sync the use and modifications of its hash table
234 spinlock_t mode_lock;
235 spinlock_t stats_lock;
238 #ifdef CONFIG_PROC_FS
239 struct proc_dir_entry *proc_entry;
240 char proc_file_name[IFNAMSIZ];
241 #endif /* CONFIG_PROC_FS */
242 struct list_head bond_list;
243 u32 __percpu *rr_tx_counter;
244 struct ad_bond_info ad_info;
245 struct alb_bond_info alb_info;
246 struct bond_params params;
247 struct workqueue_struct *wq;
248 struct delayed_work mii_work;
249 struct delayed_work arp_work;
250 struct delayed_work alb_work;
251 struct delayed_work ad_work;
252 struct delayed_work mcast_work;
253 struct delayed_work slave_arr_work;
254 #ifdef CONFIG_DEBUG_FS
255 /* debugging support via debugfs */
256 struct dentry *debug_dir;
257 #endif /* CONFIG_DEBUG_FS */
258 struct rtnl_link_stats64 bond_stats;
259 #ifdef CONFIG_XFRM_OFFLOAD
260 struct list_head ipsec_list;
261 /* protecting ipsec_list */
262 spinlock_t ipsec_lock;
263 #endif /* CONFIG_XFRM_OFFLOAD */
264 struct bpf_prog *xdp_prog;
267 #define bond_slave_get_rcu(dev) \
268 ((struct slave *) rcu_dereference(dev->rx_handler_data))
270 #define bond_slave_get_rtnl(dev) \
271 ((struct slave *) rtnl_dereference(dev->rx_handler_data))
273 void bond_queue_slave_event(struct slave *slave);
274 void bond_lower_state_changed(struct slave *slave);
276 struct bond_vlan_tag {
278 unsigned short vlan_id;
281 bool bond_sk_check(struct bonding *bond);
284 * Returns NULL if the net_device does not belong to any of the bond's slaves
286 * Caller must hold bond lock for read
288 static inline struct slave *bond_get_slave_by_dev(struct bonding *bond,
289 struct net_device *slave_dev)
291 return netdev_lower_dev_get_private(bond->dev, slave_dev);
294 static inline struct bonding *bond_get_bond_by_slave(struct slave *slave)
299 static inline bool bond_should_override_tx_queue(struct bonding *bond)
301 return BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
302 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN;
305 static inline bool bond_is_lb(const struct bonding *bond)
307 return BOND_MODE(bond) == BOND_MODE_TLB ||
308 BOND_MODE(bond) == BOND_MODE_ALB;
311 static inline bool bond_needs_speed_duplex(const struct bonding *bond)
313 return BOND_MODE(bond) == BOND_MODE_8023AD || bond_is_lb(bond);
316 static inline bool bond_is_nondyn_tlb(const struct bonding *bond)
318 return (bond_is_lb(bond) && bond->params.tlb_dynamic_lb == 0);
321 static inline bool bond_mode_can_use_xmit_hash(const struct bonding *bond)
323 return (BOND_MODE(bond) == BOND_MODE_8023AD ||
324 BOND_MODE(bond) == BOND_MODE_XOR ||
325 BOND_MODE(bond) == BOND_MODE_TLB ||
326 BOND_MODE(bond) == BOND_MODE_ALB);
329 static inline bool bond_mode_uses_xmit_hash(const struct bonding *bond)
331 return (BOND_MODE(bond) == BOND_MODE_8023AD ||
332 BOND_MODE(bond) == BOND_MODE_XOR ||
333 bond_is_nondyn_tlb(bond));
336 static inline bool bond_mode_uses_arp(int mode)
338 return mode != BOND_MODE_8023AD && mode != BOND_MODE_TLB &&
339 mode != BOND_MODE_ALB;
342 static inline bool bond_mode_uses_primary(int mode)
344 return mode == BOND_MODE_ACTIVEBACKUP || mode == BOND_MODE_TLB ||
345 mode == BOND_MODE_ALB;
348 static inline bool bond_uses_primary(struct bonding *bond)
350 return bond_mode_uses_primary(BOND_MODE(bond));
353 static inline struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond)
355 struct slave *slave = rcu_dereference_rtnl(bond->curr_active_slave);
357 return bond_uses_primary(bond) && slave ? slave->dev : NULL;
360 static inline bool bond_slave_is_up(struct slave *slave)
362 return netif_running(slave->dev) && netif_carrier_ok(slave->dev);
365 static inline void bond_set_active_slave(struct slave *slave)
369 bond_queue_slave_event(slave);
370 bond_lower_state_changed(slave);
374 static inline void bond_set_backup_slave(struct slave *slave)
376 if (!slave->backup) {
378 bond_queue_slave_event(slave);
379 bond_lower_state_changed(slave);
383 static inline void bond_set_slave_state(struct slave *slave,
384 int slave_state, bool notify)
386 if (slave->backup == slave_state)
389 slave->backup = slave_state;
391 bond_lower_state_changed(slave);
392 bond_queue_slave_event(slave);
393 slave->should_notify = 0;
395 if (slave->should_notify)
396 slave->should_notify = 0;
398 slave->should_notify = 1;
402 static inline void bond_slave_state_change(struct bonding *bond)
404 struct list_head *iter;
407 bond_for_each_slave(bond, tmp, iter) {
408 if (tmp->link == BOND_LINK_UP)
409 bond_set_active_slave(tmp);
410 else if (tmp->link == BOND_LINK_DOWN)
411 bond_set_backup_slave(tmp);
415 static inline void bond_slave_state_notify(struct bonding *bond)
417 struct list_head *iter;
420 bond_for_each_slave(bond, tmp, iter) {
421 if (tmp->should_notify) {
422 bond_lower_state_changed(tmp);
423 tmp->should_notify = 0;
428 static inline int bond_slave_state(struct slave *slave)
430 return slave->backup;
433 static inline bool bond_is_active_slave(struct slave *slave)
435 return !bond_slave_state(slave);
438 static inline bool bond_slave_can_tx(struct slave *slave)
440 return bond_slave_is_up(slave) && slave->link == BOND_LINK_UP &&
441 bond_is_active_slave(slave);
444 static inline bool bond_is_active_slave_dev(const struct net_device *slave_dev)
450 slave = bond_slave_get_rcu(slave_dev);
451 active = bond_is_active_slave(slave);
457 static inline void bond_hw_addr_copy(u8 *dst, const u8 *src, unsigned int len)
459 if (len == ETH_ALEN) {
460 ether_addr_copy(dst, src);
464 memcpy(dst, src, len);
467 #define BOND_PRI_RESELECT_ALWAYS 0
468 #define BOND_PRI_RESELECT_BETTER 1
469 #define BOND_PRI_RESELECT_FAILURE 2
471 #define BOND_FOM_NONE 0
472 #define BOND_FOM_ACTIVE 1
473 #define BOND_FOM_FOLLOW 2
475 #define BOND_ARP_TARGETS_ANY 0
476 #define BOND_ARP_TARGETS_ALL 1
478 #define BOND_ARP_VALIDATE_NONE 0
479 #define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE)
480 #define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP)
481 #define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \
482 BOND_ARP_VALIDATE_BACKUP)
483 #define BOND_ARP_FILTER (BOND_ARP_VALIDATE_ALL + 1)
484 #define BOND_ARP_FILTER_ACTIVE (BOND_ARP_VALIDATE_ACTIVE | \
486 #define BOND_ARP_FILTER_BACKUP (BOND_ARP_VALIDATE_BACKUP | \
489 #define BOND_SLAVE_NOTIFY_NOW true
490 #define BOND_SLAVE_NOTIFY_LATER false
492 static inline int slave_do_arp_validate(struct bonding *bond,
495 return bond->params.arp_validate & (1 << bond_slave_state(slave));
498 static inline int slave_do_arp_validate_only(struct bonding *bond)
500 return bond->params.arp_validate & BOND_ARP_FILTER;
503 static inline int bond_is_ip_target_ok(__be32 addr)
505 return !ipv4_is_lbcast(addr) && !ipv4_is_zeronet(addr);
508 #if IS_ENABLED(CONFIG_IPV6)
509 static inline int bond_is_ip6_target_ok(struct in6_addr *addr)
511 return !ipv6_addr_any(addr) &&
512 !ipv6_addr_loopback(addr) &&
513 !ipv6_addr_is_multicast(addr);
517 /* Get the oldest arp which we've received on this slave for bond's
520 static inline unsigned long slave_oldest_target_arp_rx(struct bonding *bond,
524 unsigned long ret = slave->target_last_arp_rx[0];
526 for (; (i < BOND_MAX_ARP_TARGETS) && bond->params.arp_targets[i]; i++)
527 if (time_before(slave->target_last_arp_rx[i], ret))
528 ret = slave->target_last_arp_rx[i];
533 static inline unsigned long slave_last_rx(struct bonding *bond,
536 if (bond->params.arp_all_targets == BOND_ARP_TARGETS_ALL)
537 return slave_oldest_target_arp_rx(bond, slave);
539 return slave->last_rx;
542 #ifdef CONFIG_NET_POLL_CONTROLLER
543 static inline netdev_tx_t bond_netpoll_send_skb(const struct slave *slave,
546 return netpoll_send_skb(slave->np, skb);
549 static inline netdev_tx_t bond_netpoll_send_skb(const struct slave *slave,
557 static inline void bond_set_slave_inactive_flags(struct slave *slave,
560 if (!bond_is_lb(slave->bond))
561 bond_set_slave_state(slave, BOND_STATE_BACKUP, notify);
562 if (!slave->bond->params.all_slaves_active)
566 static inline void bond_set_slave_active_flags(struct slave *slave,
569 bond_set_slave_state(slave, BOND_STATE_ACTIVE, notify);
573 static inline bool bond_is_slave_inactive(struct slave *slave)
575 return slave->inactive;
578 static inline void bond_propose_link_state(struct slave *slave, int state)
580 slave->link_new_state = state;
583 static inline void bond_commit_link_state(struct slave *slave, bool notify)
585 if (slave->link_new_state == BOND_LINK_NOCHANGE)
588 slave->link = slave->link_new_state;
590 bond_queue_slave_event(slave);
591 bond_lower_state_changed(slave);
592 slave->should_notify_link = 0;
594 if (slave->should_notify_link)
595 slave->should_notify_link = 0;
597 slave->should_notify_link = 1;
601 static inline void bond_set_slave_link_state(struct slave *slave, int state,
604 bond_propose_link_state(slave, state);
605 bond_commit_link_state(slave, notify);
608 static inline void bond_slave_link_notify(struct bonding *bond)
610 struct list_head *iter;
613 bond_for_each_slave(bond, tmp, iter) {
614 if (tmp->should_notify_link) {
615 bond_queue_slave_event(tmp);
616 bond_lower_state_changed(tmp);
617 tmp->should_notify_link = 0;
622 static inline __be32 bond_confirm_addr(struct net_device *dev, __be32 dst, __be32 local)
624 struct in_device *in_dev;
628 in_dev = __in_dev_get_rcu(dev);
631 addr = inet_confirm_addr(dev_net(dev), in_dev, dst, local,
638 struct net *net; /* Associated network namespace */
639 struct list_head dev_list;
640 #ifdef CONFIG_PROC_FS
641 struct proc_dir_entry *proc_dir;
643 struct class_attribute class_attr_bonding_masters;
646 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond, struct slave *slave);
647 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev);
648 int bond_create(struct net *net, const char *name);
649 int bond_create_sysfs(struct bond_net *net);
650 void bond_destroy_sysfs(struct bond_net *net);
651 void bond_prepare_sysfs_group(struct bonding *bond);
652 int bond_sysfs_slave_add(struct slave *slave);
653 void bond_sysfs_slave_del(struct slave *slave);
654 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
655 struct netlink_ext_ack *extack);
656 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev);
657 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb);
658 int bond_set_carrier(struct bonding *bond);
659 void bond_select_active_slave(struct bonding *bond);
660 void bond_change_active_slave(struct bonding *bond, struct slave *new_active);
661 void bond_create_debugfs(void);
662 void bond_destroy_debugfs(void);
663 void bond_debug_register(struct bonding *bond);
664 void bond_debug_unregister(struct bonding *bond);
665 void bond_debug_reregister(struct bonding *bond);
666 const char *bond_mode_name(int mode);
667 void bond_setup(struct net_device *bond_dev);
668 unsigned int bond_get_num_tx_queues(void);
669 int bond_netlink_init(void);
670 void bond_netlink_fini(void);
671 struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond);
672 const char *bond_slave_link_status(s8 link);
673 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
674 struct net_device *end_dev,
676 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave);
677 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay);
678 void bond_work_init_all(struct bonding *bond);
680 #ifdef CONFIG_PROC_FS
681 void bond_create_proc_entry(struct bonding *bond);
682 void bond_remove_proc_entry(struct bonding *bond);
683 void bond_create_proc_dir(struct bond_net *bn);
684 void bond_destroy_proc_dir(struct bond_net *bn);
686 static inline void bond_create_proc_entry(struct bonding *bond)
690 static inline void bond_remove_proc_entry(struct bonding *bond)
694 static inline void bond_create_proc_dir(struct bond_net *bn)
698 static inline void bond_destroy_proc_dir(struct bond_net *bn)
703 static inline struct slave *bond_slave_has_mac(struct bonding *bond,
706 struct list_head *iter;
709 bond_for_each_slave(bond, tmp, iter)
710 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
716 /* Caller must hold rcu_read_lock() for read */
717 static inline bool bond_slave_has_mac_rx(struct bonding *bond, const u8 *mac)
719 struct list_head *iter;
721 struct netdev_hw_addr *ha;
723 bond_for_each_slave_rcu(bond, tmp, iter)
724 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
727 if (netdev_uc_empty(bond->dev))
730 netdev_for_each_uc_addr(ha, bond->dev)
731 if (ether_addr_equal_64bits(mac, ha->addr))
737 /* Check if the ip is present in arp ip list, or first free slot if ip == 0
738 * Returns -1 if not found, index if found
740 static inline int bond_get_targets_ip(__be32 *targets, __be32 ip)
744 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
745 if (targets[i] == ip)
747 else if (targets[i] == 0)
753 #if IS_ENABLED(CONFIG_IPV6)
754 static inline int bond_get_targets_ip6(struct in6_addr *targets, struct in6_addr *ip)
758 for (i = 0; i < BOND_MAX_NS_TARGETS; i++)
759 if (ipv6_addr_equal(&targets[i], ip))
761 else if (ipv6_addr_any(&targets[i]))
768 /* exported from bond_main.c */
769 extern unsigned int bond_net_id;
771 /* exported from bond_netlink.c */
772 extern struct rtnl_link_ops bond_link_ops;
774 /* exported from bond_sysfs_slave.c */
775 extern const struct sysfs_ops slave_sysfs_ops;
777 static inline netdev_tx_t bond_tx_drop(struct net_device *dev, struct sk_buff *skb)
779 dev_core_stats_tx_dropped_inc(dev);
780 dev_kfree_skb_any(skb);
781 return NET_XMIT_DROP;
784 #endif /* _NET_BONDING_H */