1 /* SPDX-License-Identifier: GPL-1.0+ */
3 * Bond several ethernet interfaces into a Cisco, running 'Etherchannel'.
5 * Portions are (c) Copyright 1995 Simon "Guru Aleph-Null" Janes
6 * NCM: Network and Communications Management, Inc.
8 * BUT, I'm the one who modified it for ethernet, so:
9 * (c) Copyright 1999, Thomas Davis, tadavis@lbl.gov
13 #ifndef _NET_BONDING_H
14 #define _NET_BONDING_H
16 #include <linux/timer.h>
17 #include <linux/proc_fs.h>
18 #include <linux/if_bonding.h>
19 #include <linux/cpumask.h>
20 #include <linux/in6.h>
21 #include <linux/netpoll.h>
22 #include <linux/inetdevice.h>
23 #include <linux/etherdevice.h>
24 #include <linux/reciprocal_div.h>
25 #include <linux/if_link.h>
27 #include <net/bond_3ad.h>
28 #include <net/bond_alb.h>
29 #include <net/bond_options.h>
31 #include <net/addrconf.h>
33 #define BOND_MAX_ARP_TARGETS 16
34 #define BOND_MAX_NS_TARGETS BOND_MAX_ARP_TARGETS
36 #define BOND_DEFAULT_MIIMON 100
38 #ifndef __long_aligned
39 #define __long_aligned __attribute__((aligned((sizeof(long)))))
42 #define slave_info(bond_dev, slave_dev, fmt, ...) \
43 netdev_info(bond_dev, "(slave %s): " fmt, (slave_dev)->name, ##__VA_ARGS__)
44 #define slave_warn(bond_dev, slave_dev, fmt, ...) \
45 netdev_warn(bond_dev, "(slave %s): " fmt, (slave_dev)->name, ##__VA_ARGS__)
46 #define slave_dbg(bond_dev, slave_dev, fmt, ...) \
47 netdev_dbg(bond_dev, "(slave %s): " fmt, (slave_dev)->name, ##__VA_ARGS__)
48 #define slave_err(bond_dev, slave_dev, fmt, ...) \
49 netdev_err(bond_dev, "(slave %s): " fmt, (slave_dev)->name, ##__VA_ARGS__)
51 #define BOND_MODE(bond) ((bond)->params.mode)
53 /* slave list primitives */
54 #define bond_slave_list(bond) (&(bond)->dev->adj_list.lower)
56 #define bond_has_slaves(bond) !list_empty(bond_slave_list(bond))
58 /* IMPORTANT: bond_first/last_slave can return NULL in case of an empty list */
59 #define bond_first_slave(bond) \
60 (bond_has_slaves(bond) ? \
61 netdev_adjacent_get_private(bond_slave_list(bond)->next) : \
63 #define bond_last_slave(bond) \
64 (bond_has_slaves(bond) ? \
65 netdev_adjacent_get_private(bond_slave_list(bond)->prev) : \
68 /* Caller must have rcu_read_lock */
69 #define bond_first_slave_rcu(bond) \
70 netdev_lower_get_first_private_rcu(bond->dev)
72 #define bond_is_first_slave(bond, pos) (pos == bond_first_slave(bond))
73 #define bond_is_last_slave(bond, pos) (pos == bond_last_slave(bond))
76 * bond_for_each_slave - iterate over all slaves
77 * @bond: the bond holding this list
79 * @iter: list_head * iterator
81 * Caller must hold RTNL
83 #define bond_for_each_slave(bond, pos, iter) \
84 netdev_for_each_lower_private((bond)->dev, pos, iter)
86 /* Caller must have rcu_read_lock */
87 #define bond_for_each_slave_rcu(bond, pos, iter) \
88 netdev_for_each_lower_private_rcu((bond)->dev, pos, iter)
90 #define BOND_XFRM_FEATURES (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM | \
93 #ifdef CONFIG_NET_POLL_CONTROLLER
94 extern atomic_t netpoll_block_tx;
96 static inline void block_netpoll_tx(void)
98 atomic_inc(&netpoll_block_tx);
101 static inline void unblock_netpoll_tx(void)
103 atomic_dec(&netpoll_block_tx);
106 static inline int is_netpoll_tx_blocked(struct net_device *dev)
108 if (unlikely(netpoll_tx_running(dev)))
109 return atomic_read(&netpoll_block_tx);
113 #define block_netpoll_tx()
114 #define unblock_netpoll_tx()
115 #define is_netpoll_tx_blocked(dev) (0)
131 int peer_notif_delay;
134 unsigned int min_links;
136 char primary[IFNAMSIZ];
137 int primary_reselect;
138 __be32 arp_targets[BOND_MAX_ARP_TARGETS];
140 int all_slaves_active;
143 int packets_per_slave;
145 struct reciprocal_value reciprocal_packets_per_slave;
146 u16 ad_actor_sys_prio;
147 u16 ad_user_port_key;
148 #if IS_ENABLED(CONFIG_IPV6)
149 struct in6_addr ns_targets[BOND_MAX_NS_TARGETS];
152 /* 2 bytes of padding : see ether_addr_equal_64bits() */
153 u8 ad_actor_system[ETH_ALEN + 2];
157 struct net_device *dev; /* first - useful for panic debug */
158 struct bonding *bond; /* our master */
160 /* all 4 in jiffies */
161 unsigned long last_link_up;
162 unsigned long last_tx;
163 unsigned long last_rx;
164 unsigned long target_last_arp_rx[BOND_MAX_ARP_TARGETS];
165 s8 link; /* one of BOND_LINK_XXXX */
166 s8 link_new_state; /* one of BOND_LINK_XXXX */
167 u8 backup:1, /* indicates backup slave. Value corresponds with
168 BOND_STATE_ACTIVE and BOND_STATE_BACKUP */
169 inactive:1, /* indicates inactive slave */
170 should_notify:1, /* indicates whether the state changed */
171 should_notify_link:1; /* indicates whether the link changed */
174 u32 link_failure_count;
177 u8 perm_hwaddr[MAX_ADDR_LEN];
179 struct ad_slave_info *ad_info;
180 struct tlb_slave_info tlb_info;
181 #ifdef CONFIG_NET_POLL_CONTROLLER
184 struct delayed_work notify_work;
186 struct rtnl_link_stats64 slave_stats;
189 static inline struct slave *to_slave(struct kobject *kobj)
191 return container_of(kobj, struct slave, kobj);
194 struct bond_up_slave {
201 * Link pseudo-state only used internally by monitors
203 #define BOND_LINK_NOCHANGE -1
206 struct list_head list;
207 struct xfrm_state *xs;
211 * Here are the locking policies for the two bonding locks:
212 * Get rcu_read_lock when reading or RTNL when writing slave list.
215 struct net_device *dev; /* first - useful for panic debug */
216 struct slave __rcu *curr_active_slave;
217 struct slave __rcu *current_arp_slave;
218 struct slave __rcu *primary_slave;
219 struct bond_up_slave __rcu *usable_slaves;
220 struct bond_up_slave __rcu *all_slaves;
223 s32 slave_cnt; /* never change this value outside the attach/detach wrappers */
224 int (*recv_probe)(const struct sk_buff *, struct bonding *,
226 /* mode_lock is used for mode-specific locking needs, currently used by:
227 * 3ad mode (4) - protect against running bond_3ad_unbind_slave() and
228 * bond_3ad_state_machine_handler() concurrently and also
229 * the access to the state machine shared variables.
230 * TLB mode (5) - to sync the use and modifications of its hash table
231 * ALB mode (6) - to sync the use and modifications of its hash table
233 spinlock_t mode_lock;
234 spinlock_t stats_lock;
237 #ifdef CONFIG_PROC_FS
238 struct proc_dir_entry *proc_entry;
239 char proc_file_name[IFNAMSIZ];
240 #endif /* CONFIG_PROC_FS */
241 struct list_head bond_list;
242 u32 __percpu *rr_tx_counter;
243 struct ad_bond_info ad_info;
244 struct alb_bond_info alb_info;
245 struct bond_params params;
246 struct workqueue_struct *wq;
247 struct delayed_work mii_work;
248 struct delayed_work arp_work;
249 struct delayed_work alb_work;
250 struct delayed_work ad_work;
251 struct delayed_work mcast_work;
252 struct delayed_work slave_arr_work;
253 #ifdef CONFIG_DEBUG_FS
254 /* debugging support via debugfs */
255 struct dentry *debug_dir;
256 #endif /* CONFIG_DEBUG_FS */
257 struct rtnl_link_stats64 bond_stats;
258 #ifdef CONFIG_XFRM_OFFLOAD
259 struct list_head ipsec_list;
260 /* protecting ipsec_list */
261 spinlock_t ipsec_lock;
262 #endif /* CONFIG_XFRM_OFFLOAD */
263 struct bpf_prog *xdp_prog;
266 #define bond_slave_get_rcu(dev) \
267 ((struct slave *) rcu_dereference(dev->rx_handler_data))
269 #define bond_slave_get_rtnl(dev) \
270 ((struct slave *) rtnl_dereference(dev->rx_handler_data))
272 void bond_queue_slave_event(struct slave *slave);
273 void bond_lower_state_changed(struct slave *slave);
275 struct bond_vlan_tag {
277 unsigned short vlan_id;
281 * Returns NULL if the net_device does not belong to any of the bond's slaves
283 * Caller must hold bond lock for read
285 static inline struct slave *bond_get_slave_by_dev(struct bonding *bond,
286 struct net_device *slave_dev)
288 return netdev_lower_dev_get_private(bond->dev, slave_dev);
291 static inline struct bonding *bond_get_bond_by_slave(struct slave *slave)
296 static inline bool bond_should_override_tx_queue(struct bonding *bond)
298 return BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
299 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN;
302 static inline bool bond_is_lb(const struct bonding *bond)
304 return BOND_MODE(bond) == BOND_MODE_TLB ||
305 BOND_MODE(bond) == BOND_MODE_ALB;
308 static inline bool bond_needs_speed_duplex(const struct bonding *bond)
310 return BOND_MODE(bond) == BOND_MODE_8023AD || bond_is_lb(bond);
313 static inline bool bond_is_nondyn_tlb(const struct bonding *bond)
315 return (bond_is_lb(bond) && bond->params.tlb_dynamic_lb == 0);
318 static inline bool bond_mode_can_use_xmit_hash(const struct bonding *bond)
320 return (BOND_MODE(bond) == BOND_MODE_8023AD ||
321 BOND_MODE(bond) == BOND_MODE_XOR ||
322 BOND_MODE(bond) == BOND_MODE_TLB ||
323 BOND_MODE(bond) == BOND_MODE_ALB);
326 static inline bool bond_mode_uses_xmit_hash(const struct bonding *bond)
328 return (BOND_MODE(bond) == BOND_MODE_8023AD ||
329 BOND_MODE(bond) == BOND_MODE_XOR ||
330 bond_is_nondyn_tlb(bond));
333 static inline bool bond_mode_uses_arp(int mode)
335 return mode != BOND_MODE_8023AD && mode != BOND_MODE_TLB &&
336 mode != BOND_MODE_ALB;
339 static inline bool bond_mode_uses_primary(int mode)
341 return mode == BOND_MODE_ACTIVEBACKUP || mode == BOND_MODE_TLB ||
342 mode == BOND_MODE_ALB;
345 static inline bool bond_uses_primary(struct bonding *bond)
347 return bond_mode_uses_primary(BOND_MODE(bond));
350 static inline struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond)
352 struct slave *slave = rcu_dereference_rtnl(bond->curr_active_slave);
354 return bond_uses_primary(bond) && slave ? slave->dev : NULL;
357 static inline bool bond_slave_is_up(struct slave *slave)
359 return netif_running(slave->dev) && netif_carrier_ok(slave->dev);
362 static inline void bond_set_active_slave(struct slave *slave)
366 bond_queue_slave_event(slave);
367 bond_lower_state_changed(slave);
371 static inline void bond_set_backup_slave(struct slave *slave)
373 if (!slave->backup) {
375 bond_queue_slave_event(slave);
376 bond_lower_state_changed(slave);
380 static inline void bond_set_slave_state(struct slave *slave,
381 int slave_state, bool notify)
383 if (slave->backup == slave_state)
386 slave->backup = slave_state;
388 bond_lower_state_changed(slave);
389 bond_queue_slave_event(slave);
390 slave->should_notify = 0;
392 if (slave->should_notify)
393 slave->should_notify = 0;
395 slave->should_notify = 1;
399 static inline void bond_slave_state_change(struct bonding *bond)
401 struct list_head *iter;
404 bond_for_each_slave(bond, tmp, iter) {
405 if (tmp->link == BOND_LINK_UP)
406 bond_set_active_slave(tmp);
407 else if (tmp->link == BOND_LINK_DOWN)
408 bond_set_backup_slave(tmp);
412 static inline void bond_slave_state_notify(struct bonding *bond)
414 struct list_head *iter;
417 bond_for_each_slave(bond, tmp, iter) {
418 if (tmp->should_notify) {
419 bond_lower_state_changed(tmp);
420 tmp->should_notify = 0;
425 static inline int bond_slave_state(struct slave *slave)
427 return slave->backup;
430 static inline bool bond_is_active_slave(struct slave *slave)
432 return !bond_slave_state(slave);
435 static inline bool bond_slave_can_tx(struct slave *slave)
437 return bond_slave_is_up(slave) && slave->link == BOND_LINK_UP &&
438 bond_is_active_slave(slave);
441 static inline bool bond_is_active_slave_dev(const struct net_device *slave_dev)
447 slave = bond_slave_get_rcu(slave_dev);
448 active = bond_is_active_slave(slave);
454 static inline void bond_hw_addr_copy(u8 *dst, const u8 *src, unsigned int len)
456 if (len == ETH_ALEN) {
457 ether_addr_copy(dst, src);
461 memcpy(dst, src, len);
464 #define BOND_PRI_RESELECT_ALWAYS 0
465 #define BOND_PRI_RESELECT_BETTER 1
466 #define BOND_PRI_RESELECT_FAILURE 2
468 #define BOND_FOM_NONE 0
469 #define BOND_FOM_ACTIVE 1
470 #define BOND_FOM_FOLLOW 2
472 #define BOND_ARP_TARGETS_ANY 0
473 #define BOND_ARP_TARGETS_ALL 1
475 #define BOND_ARP_VALIDATE_NONE 0
476 #define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE)
477 #define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP)
478 #define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \
479 BOND_ARP_VALIDATE_BACKUP)
480 #define BOND_ARP_FILTER (BOND_ARP_VALIDATE_ALL + 1)
481 #define BOND_ARP_FILTER_ACTIVE (BOND_ARP_VALIDATE_ACTIVE | \
483 #define BOND_ARP_FILTER_BACKUP (BOND_ARP_VALIDATE_BACKUP | \
486 #define BOND_SLAVE_NOTIFY_NOW true
487 #define BOND_SLAVE_NOTIFY_LATER false
489 static inline int slave_do_arp_validate(struct bonding *bond,
492 return bond->params.arp_validate & (1 << bond_slave_state(slave));
495 static inline int slave_do_arp_validate_only(struct bonding *bond)
497 return bond->params.arp_validate & BOND_ARP_FILTER;
500 static inline int bond_is_ip_target_ok(__be32 addr)
502 return !ipv4_is_lbcast(addr) && !ipv4_is_zeronet(addr);
505 #if IS_ENABLED(CONFIG_IPV6)
506 static inline int bond_is_ip6_target_ok(struct in6_addr *addr)
508 return !ipv6_addr_any(addr) &&
509 !ipv6_addr_loopback(addr) &&
510 !ipv6_addr_is_multicast(addr);
514 /* Get the oldest arp which we've received on this slave for bond's
517 static inline unsigned long slave_oldest_target_arp_rx(struct bonding *bond,
521 unsigned long ret = slave->target_last_arp_rx[0];
523 for (; (i < BOND_MAX_ARP_TARGETS) && bond->params.arp_targets[i]; i++)
524 if (time_before(slave->target_last_arp_rx[i], ret))
525 ret = slave->target_last_arp_rx[i];
530 static inline unsigned long slave_last_rx(struct bonding *bond,
533 if (bond->params.arp_all_targets == BOND_ARP_TARGETS_ALL)
534 return slave_oldest_target_arp_rx(bond, slave);
536 return slave->last_rx;
539 static inline void slave_update_last_tx(struct slave *slave)
541 WRITE_ONCE(slave->last_tx, jiffies);
544 static inline unsigned long slave_last_tx(struct slave *slave)
546 return READ_ONCE(slave->last_tx);
549 #ifdef CONFIG_NET_POLL_CONTROLLER
550 static inline netdev_tx_t bond_netpoll_send_skb(const struct slave *slave,
553 return netpoll_send_skb(slave->np, skb);
556 static inline netdev_tx_t bond_netpoll_send_skb(const struct slave *slave,
564 static inline void bond_set_slave_inactive_flags(struct slave *slave,
567 if (!bond_is_lb(slave->bond))
568 bond_set_slave_state(slave, BOND_STATE_BACKUP, notify);
569 if (!slave->bond->params.all_slaves_active)
573 static inline void bond_set_slave_active_flags(struct slave *slave,
576 bond_set_slave_state(slave, BOND_STATE_ACTIVE, notify);
580 static inline bool bond_is_slave_inactive(struct slave *slave)
582 return slave->inactive;
585 static inline void bond_propose_link_state(struct slave *slave, int state)
587 slave->link_new_state = state;
590 static inline void bond_commit_link_state(struct slave *slave, bool notify)
592 if (slave->link_new_state == BOND_LINK_NOCHANGE)
595 slave->link = slave->link_new_state;
597 bond_queue_slave_event(slave);
598 bond_lower_state_changed(slave);
599 slave->should_notify_link = 0;
601 if (slave->should_notify_link)
602 slave->should_notify_link = 0;
604 slave->should_notify_link = 1;
608 static inline void bond_set_slave_link_state(struct slave *slave, int state,
611 bond_propose_link_state(slave, state);
612 bond_commit_link_state(slave, notify);
615 static inline void bond_slave_link_notify(struct bonding *bond)
617 struct list_head *iter;
620 bond_for_each_slave(bond, tmp, iter) {
621 if (tmp->should_notify_link) {
622 bond_queue_slave_event(tmp);
623 bond_lower_state_changed(tmp);
624 tmp->should_notify_link = 0;
629 static inline __be32 bond_confirm_addr(struct net_device *dev, __be32 dst, __be32 local)
631 struct in_device *in_dev;
635 in_dev = __in_dev_get_rcu(dev);
638 addr = inet_confirm_addr(dev_net(dev), in_dev, dst, local,
645 struct net *net; /* Associated network namespace */
646 struct list_head dev_list;
647 #ifdef CONFIG_PROC_FS
648 struct proc_dir_entry *proc_dir;
650 struct class_attribute class_attr_bonding_masters;
653 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond, struct slave *slave);
654 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev);
655 int bond_create(struct net *net, const char *name);
656 int bond_create_sysfs(struct bond_net *net);
657 void bond_destroy_sysfs(struct bond_net *net);
658 void bond_prepare_sysfs_group(struct bonding *bond);
659 int bond_sysfs_slave_add(struct slave *slave);
660 void bond_sysfs_slave_del(struct slave *slave);
661 void bond_xdp_set_features(struct net_device *bond_dev);
662 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
663 struct netlink_ext_ack *extack);
664 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev);
665 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb);
666 int bond_set_carrier(struct bonding *bond);
667 void bond_select_active_slave(struct bonding *bond);
668 void bond_change_active_slave(struct bonding *bond, struct slave *new_active);
669 void bond_create_debugfs(void);
670 void bond_destroy_debugfs(void);
671 void bond_debug_register(struct bonding *bond);
672 void bond_debug_unregister(struct bonding *bond);
673 void bond_debug_reregister(struct bonding *bond);
674 const char *bond_mode_name(int mode);
675 void bond_setup(struct net_device *bond_dev);
676 unsigned int bond_get_num_tx_queues(void);
677 int bond_netlink_init(void);
678 void bond_netlink_fini(void);
679 struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond);
680 const char *bond_slave_link_status(s8 link);
681 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
682 struct net_device *end_dev,
684 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave);
685 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay);
686 void bond_work_init_all(struct bonding *bond);
688 #ifdef CONFIG_PROC_FS
689 void bond_create_proc_entry(struct bonding *bond);
690 void bond_remove_proc_entry(struct bonding *bond);
691 void bond_create_proc_dir(struct bond_net *bn);
692 void bond_destroy_proc_dir(struct bond_net *bn);
694 static inline void bond_create_proc_entry(struct bonding *bond)
698 static inline void bond_remove_proc_entry(struct bonding *bond)
702 static inline void bond_create_proc_dir(struct bond_net *bn)
706 static inline void bond_destroy_proc_dir(struct bond_net *bn)
711 static inline struct slave *bond_slave_has_mac(struct bonding *bond,
714 struct list_head *iter;
717 bond_for_each_slave(bond, tmp, iter)
718 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
724 /* Caller must hold rcu_read_lock() for read */
725 static inline bool bond_slave_has_mac_rx(struct bonding *bond, const u8 *mac)
727 struct list_head *iter;
729 struct netdev_hw_addr *ha;
731 bond_for_each_slave_rcu(bond, tmp, iter)
732 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
735 if (netdev_uc_empty(bond->dev))
738 netdev_for_each_uc_addr(ha, bond->dev)
739 if (ether_addr_equal_64bits(mac, ha->addr))
745 /* Check if the ip is present in arp ip list, or first free slot if ip == 0
746 * Returns -1 if not found, index if found
748 static inline int bond_get_targets_ip(__be32 *targets, __be32 ip)
752 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
753 if (targets[i] == ip)
755 else if (targets[i] == 0)
761 #if IS_ENABLED(CONFIG_IPV6)
762 static inline int bond_get_targets_ip6(struct in6_addr *targets, struct in6_addr *ip)
764 struct in6_addr mcaddr;
767 for (i = 0; i < BOND_MAX_NS_TARGETS; i++) {
768 addrconf_addr_solict_mult(&targets[i], &mcaddr);
769 if ((ipv6_addr_equal(&targets[i], ip)) ||
770 (ipv6_addr_equal(&mcaddr, ip)))
772 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 /* exported from bond_3ad.c */
790 extern const u8 lacpdu_mcast_addr[];
792 static inline netdev_tx_t bond_tx_drop(struct net_device *dev, struct sk_buff *skb)
794 dev_core_stats_tx_dropped_inc(dev);
795 dev_kfree_skb_any(skb);
796 return NET_XMIT_DROP;
799 #endif /* _NET_BONDING_H */