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 #define BOND_MAX_ARP_TARGETS 16
35 #define BOND_DEFAULT_MIIMON 100
37 #ifndef __long_aligned
38 #define __long_aligned __attribute__((aligned((sizeof(long)))))
41 #define slave_info(bond_dev, slave_dev, fmt, ...) \
42 netdev_info(bond_dev, "(slave %s): " fmt, (slave_dev)->name, ##__VA_ARGS__)
43 #define slave_warn(bond_dev, slave_dev, fmt, ...) \
44 netdev_warn(bond_dev, "(slave %s): " fmt, (slave_dev)->name, ##__VA_ARGS__)
45 #define slave_dbg(bond_dev, slave_dev, fmt, ...) \
46 netdev_dbg(bond_dev, "(slave %s): " fmt, (slave_dev)->name, ##__VA_ARGS__)
47 #define slave_err(bond_dev, slave_dev, fmt, ...) \
48 netdev_err(bond_dev, "(slave %s): " fmt, (slave_dev)->name, ##__VA_ARGS__)
50 #define BOND_MODE(bond) ((bond)->params.mode)
52 /* slave list primitives */
53 #define bond_slave_list(bond) (&(bond)->dev->adj_list.lower)
55 #define bond_has_slaves(bond) !list_empty(bond_slave_list(bond))
57 /* IMPORTANT: bond_first/last_slave can return NULL in case of an empty list */
58 #define bond_first_slave(bond) \
59 (bond_has_slaves(bond) ? \
60 netdev_adjacent_get_private(bond_slave_list(bond)->next) : \
62 #define bond_last_slave(bond) \
63 (bond_has_slaves(bond) ? \
64 netdev_adjacent_get_private(bond_slave_list(bond)->prev) : \
67 /* Caller must have rcu_read_lock */
68 #define bond_first_slave_rcu(bond) \
69 netdev_lower_get_first_private_rcu(bond->dev)
71 #define bond_is_first_slave(bond, pos) (pos == bond_first_slave(bond))
72 #define bond_is_last_slave(bond, pos) (pos == bond_last_slave(bond))
75 * bond_for_each_slave - iterate over all slaves
76 * @bond: the bond holding this list
78 * @iter: list_head * iterator
80 * Caller must hold RTNL
82 #define bond_for_each_slave(bond, pos, iter) \
83 netdev_for_each_lower_private((bond)->dev, pos, iter)
85 /* Caller must have rcu_read_lock */
86 #define bond_for_each_slave_rcu(bond, pos, iter) \
87 netdev_for_each_lower_private_rcu((bond)->dev, pos, iter)
89 #define BOND_XFRM_FEATURES (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM | \
92 #define BOND_TLS_FEATURES (NETIF_F_HW_TLS_TX | NETIF_F_HW_TLS_RX)
94 #ifdef CONFIG_NET_POLL_CONTROLLER
95 extern atomic_t netpoll_block_tx;
97 static inline void block_netpoll_tx(void)
99 atomic_inc(&netpoll_block_tx);
102 static inline void unblock_netpoll_tx(void)
104 atomic_dec(&netpoll_block_tx);
107 static inline int is_netpoll_tx_blocked(struct net_device *dev)
109 if (unlikely(netpoll_tx_running(dev)))
110 return atomic_read(&netpoll_block_tx);
114 #define block_netpoll_tx()
115 #define unblock_netpoll_tx()
116 #define is_netpoll_tx_blocked(dev) (0)
131 int peer_notif_delay;
133 unsigned int min_links;
135 char primary[IFNAMSIZ];
136 int primary_reselect;
137 __be32 arp_targets[BOND_MAX_ARP_TARGETS];
139 int all_slaves_active;
142 int packets_per_slave;
144 struct reciprocal_value reciprocal_packets_per_slave;
145 u16 ad_actor_sys_prio;
146 u16 ad_user_port_key;
148 /* 2 bytes of padding : see ether_addr_equal_64bits() */
149 u8 ad_actor_system[ETH_ALEN + 2];
152 struct bond_parm_tbl {
158 struct net_device *dev; /* first - useful for panic debug */
159 struct bonding *bond; /* our master */
161 /* all three in jiffies */
162 unsigned long last_link_up;
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];
178 struct ad_slave_info *ad_info;
179 struct tlb_slave_info tlb_info;
180 #ifdef CONFIG_NET_POLL_CONTROLLER
183 struct delayed_work notify_work;
185 struct rtnl_link_stats64 slave_stats;
188 static inline struct slave *to_slave(struct kobject *kobj)
190 return container_of(kobj, struct slave, kobj);
193 struct bond_up_slave {
200 * Link pseudo-state only used internally by monitors
202 #define BOND_LINK_NOCHANGE -1
205 * Here are the locking policies for the two bonding locks:
206 * Get rcu_read_lock when reading or RTNL when writing slave list.
209 struct net_device *dev; /* first - useful for panic debug */
210 struct slave __rcu *curr_active_slave;
211 struct slave __rcu *current_arp_slave;
212 struct slave __rcu *primary_slave;
213 struct bond_up_slave __rcu *usable_slaves;
214 struct bond_up_slave __rcu *all_slaves;
216 s32 slave_cnt; /* never change this value outside the attach/detach wrappers */
217 int (*recv_probe)(const struct sk_buff *, struct bonding *,
219 /* mode_lock is used for mode-specific locking needs, currently used by:
220 * 3ad mode (4) - protect against running bond_3ad_unbind_slave() and
221 * bond_3ad_state_machine_handler() concurrently and also
222 * the access to the state machine shared variables.
223 * TLB mode (5) - to sync the use and modifications of its hash table
224 * ALB mode (6) - to sync the use and modifications of its hash table
226 spinlock_t mode_lock;
227 spinlock_t stats_lock;
230 #ifdef CONFIG_PROC_FS
231 struct proc_dir_entry *proc_entry;
232 char proc_file_name[IFNAMSIZ];
233 #endif /* CONFIG_PROC_FS */
234 struct list_head bond_list;
235 u32 __percpu *rr_tx_counter;
236 struct ad_bond_info ad_info;
237 struct alb_bond_info alb_info;
238 struct bond_params params;
239 struct workqueue_struct *wq;
240 struct delayed_work mii_work;
241 struct delayed_work arp_work;
242 struct delayed_work alb_work;
243 struct delayed_work ad_work;
244 struct delayed_work mcast_work;
245 struct delayed_work slave_arr_work;
246 #ifdef CONFIG_DEBUG_FS
247 /* debugging support via debugfs */
248 struct dentry *debug_dir;
249 #endif /* CONFIG_DEBUG_FS */
250 struct rtnl_link_stats64 bond_stats;
251 #ifdef CONFIG_XFRM_OFFLOAD
252 struct xfrm_state *xs;
253 #endif /* CONFIG_XFRM_OFFLOAD */
256 #define bond_slave_get_rcu(dev) \
257 ((struct slave *) rcu_dereference(dev->rx_handler_data))
259 #define bond_slave_get_rtnl(dev) \
260 ((struct slave *) rtnl_dereference(dev->rx_handler_data))
262 void bond_queue_slave_event(struct slave *slave);
263 void bond_lower_state_changed(struct slave *slave);
265 struct bond_vlan_tag {
267 unsigned short vlan_id;
270 bool bond_sk_check(struct bonding *bond);
273 * Returns NULL if the net_device does not belong to any of the bond's slaves
275 * Caller must hold bond lock for read
277 static inline struct slave *bond_get_slave_by_dev(struct bonding *bond,
278 struct net_device *slave_dev)
280 return netdev_lower_dev_get_private(bond->dev, slave_dev);
283 static inline struct bonding *bond_get_bond_by_slave(struct slave *slave)
288 static inline bool bond_should_override_tx_queue(struct bonding *bond)
290 return BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
291 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN;
294 static inline bool bond_is_lb(const struct bonding *bond)
296 return BOND_MODE(bond) == BOND_MODE_TLB ||
297 BOND_MODE(bond) == BOND_MODE_ALB;
300 static inline bool bond_needs_speed_duplex(const struct bonding *bond)
302 return BOND_MODE(bond) == BOND_MODE_8023AD || bond_is_lb(bond);
305 static inline bool bond_is_nondyn_tlb(const struct bonding *bond)
307 return (bond_is_lb(bond) && bond->params.tlb_dynamic_lb == 0);
310 static inline bool bond_mode_can_use_xmit_hash(const struct bonding *bond)
312 return (BOND_MODE(bond) == BOND_MODE_8023AD ||
313 BOND_MODE(bond) == BOND_MODE_XOR ||
314 BOND_MODE(bond) == BOND_MODE_TLB ||
315 BOND_MODE(bond) == BOND_MODE_ALB);
318 static inline bool bond_mode_uses_xmit_hash(const struct bonding *bond)
320 return (BOND_MODE(bond) == BOND_MODE_8023AD ||
321 BOND_MODE(bond) == BOND_MODE_XOR ||
322 bond_is_nondyn_tlb(bond));
325 static inline bool bond_mode_uses_arp(int mode)
327 return mode != BOND_MODE_8023AD && mode != BOND_MODE_TLB &&
328 mode != BOND_MODE_ALB;
331 static inline bool bond_mode_uses_primary(int mode)
333 return mode == BOND_MODE_ACTIVEBACKUP || mode == BOND_MODE_TLB ||
334 mode == BOND_MODE_ALB;
337 static inline bool bond_uses_primary(struct bonding *bond)
339 return bond_mode_uses_primary(BOND_MODE(bond));
342 static inline struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond)
344 struct slave *slave = rcu_dereference(bond->curr_active_slave);
346 return bond_uses_primary(bond) && slave ? slave->dev : NULL;
349 static inline bool bond_slave_is_up(struct slave *slave)
351 return netif_running(slave->dev) && netif_carrier_ok(slave->dev);
354 static inline void bond_set_active_slave(struct slave *slave)
358 bond_queue_slave_event(slave);
359 bond_lower_state_changed(slave);
363 static inline void bond_set_backup_slave(struct slave *slave)
365 if (!slave->backup) {
367 bond_queue_slave_event(slave);
368 bond_lower_state_changed(slave);
372 static inline void bond_set_slave_state(struct slave *slave,
373 int slave_state, bool notify)
375 if (slave->backup == slave_state)
378 slave->backup = slave_state;
380 bond_lower_state_changed(slave);
381 bond_queue_slave_event(slave);
382 slave->should_notify = 0;
384 if (slave->should_notify)
385 slave->should_notify = 0;
387 slave->should_notify = 1;
391 static inline void bond_slave_state_change(struct bonding *bond)
393 struct list_head *iter;
396 bond_for_each_slave(bond, tmp, iter) {
397 if (tmp->link == BOND_LINK_UP)
398 bond_set_active_slave(tmp);
399 else if (tmp->link == BOND_LINK_DOWN)
400 bond_set_backup_slave(tmp);
404 static inline void bond_slave_state_notify(struct bonding *bond)
406 struct list_head *iter;
409 bond_for_each_slave(bond, tmp, iter) {
410 if (tmp->should_notify) {
411 bond_lower_state_changed(tmp);
412 tmp->should_notify = 0;
417 static inline int bond_slave_state(struct slave *slave)
419 return slave->backup;
422 static inline bool bond_is_active_slave(struct slave *slave)
424 return !bond_slave_state(slave);
427 static inline bool bond_slave_can_tx(struct slave *slave)
429 return bond_slave_is_up(slave) && slave->link == BOND_LINK_UP &&
430 bond_is_active_slave(slave);
433 static inline bool bond_is_active_slave_dev(const struct net_device *slave_dev)
439 slave = bond_slave_get_rcu(slave_dev);
440 active = bond_is_active_slave(slave);
446 static inline void bond_hw_addr_copy(u8 *dst, const u8 *src, unsigned int len)
448 if (len == ETH_ALEN) {
449 ether_addr_copy(dst, src);
453 memcpy(dst, src, len);
456 #define BOND_PRI_RESELECT_ALWAYS 0
457 #define BOND_PRI_RESELECT_BETTER 1
458 #define BOND_PRI_RESELECT_FAILURE 2
460 #define BOND_FOM_NONE 0
461 #define BOND_FOM_ACTIVE 1
462 #define BOND_FOM_FOLLOW 2
464 #define BOND_ARP_TARGETS_ANY 0
465 #define BOND_ARP_TARGETS_ALL 1
467 #define BOND_ARP_VALIDATE_NONE 0
468 #define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE)
469 #define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP)
470 #define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \
471 BOND_ARP_VALIDATE_BACKUP)
472 #define BOND_ARP_FILTER (BOND_ARP_VALIDATE_ALL + 1)
473 #define BOND_ARP_FILTER_ACTIVE (BOND_ARP_VALIDATE_ACTIVE | \
475 #define BOND_ARP_FILTER_BACKUP (BOND_ARP_VALIDATE_BACKUP | \
478 #define BOND_SLAVE_NOTIFY_NOW true
479 #define BOND_SLAVE_NOTIFY_LATER false
481 static inline int slave_do_arp_validate(struct bonding *bond,
484 return bond->params.arp_validate & (1 << bond_slave_state(slave));
487 static inline int slave_do_arp_validate_only(struct bonding *bond)
489 return bond->params.arp_validate & BOND_ARP_FILTER;
492 static inline int bond_is_ip_target_ok(__be32 addr)
494 return !ipv4_is_lbcast(addr) && !ipv4_is_zeronet(addr);
497 /* Get the oldest arp which we've received on this slave for bond's
500 static inline unsigned long slave_oldest_target_arp_rx(struct bonding *bond,
504 unsigned long ret = slave->target_last_arp_rx[0];
506 for (; (i < BOND_MAX_ARP_TARGETS) && bond->params.arp_targets[i]; i++)
507 if (time_before(slave->target_last_arp_rx[i], ret))
508 ret = slave->target_last_arp_rx[i];
513 static inline unsigned long slave_last_rx(struct bonding *bond,
516 if (bond->params.arp_all_targets == BOND_ARP_TARGETS_ALL)
517 return slave_oldest_target_arp_rx(bond, slave);
519 return slave->last_rx;
522 #ifdef CONFIG_NET_POLL_CONTROLLER
523 static inline netdev_tx_t bond_netpoll_send_skb(const struct slave *slave,
526 return netpoll_send_skb(slave->np, skb);
529 static inline netdev_tx_t bond_netpoll_send_skb(const struct slave *slave,
537 static inline void bond_set_slave_inactive_flags(struct slave *slave,
540 if (!bond_is_lb(slave->bond))
541 bond_set_slave_state(slave, BOND_STATE_BACKUP, notify);
542 if (!slave->bond->params.all_slaves_active)
546 static inline void bond_set_slave_active_flags(struct slave *slave,
549 bond_set_slave_state(slave, BOND_STATE_ACTIVE, notify);
553 static inline bool bond_is_slave_inactive(struct slave *slave)
555 return slave->inactive;
558 static inline void bond_propose_link_state(struct slave *slave, int state)
560 slave->link_new_state = state;
563 static inline void bond_commit_link_state(struct slave *slave, bool notify)
565 if (slave->link_new_state == BOND_LINK_NOCHANGE)
568 slave->link = slave->link_new_state;
570 bond_queue_slave_event(slave);
571 bond_lower_state_changed(slave);
572 slave->should_notify_link = 0;
574 if (slave->should_notify_link)
575 slave->should_notify_link = 0;
577 slave->should_notify_link = 1;
581 static inline void bond_set_slave_link_state(struct slave *slave, int state,
584 bond_propose_link_state(slave, state);
585 bond_commit_link_state(slave, notify);
588 static inline void bond_slave_link_notify(struct bonding *bond)
590 struct list_head *iter;
593 bond_for_each_slave(bond, tmp, iter) {
594 if (tmp->should_notify_link) {
595 bond_queue_slave_event(tmp);
596 bond_lower_state_changed(tmp);
597 tmp->should_notify_link = 0;
602 static inline __be32 bond_confirm_addr(struct net_device *dev, __be32 dst, __be32 local)
604 struct in_device *in_dev;
608 in_dev = __in_dev_get_rcu(dev);
611 addr = inet_confirm_addr(dev_net(dev), in_dev, dst, local,
618 struct net *net; /* Associated network namespace */
619 struct list_head dev_list;
620 #ifdef CONFIG_PROC_FS
621 struct proc_dir_entry *proc_dir;
623 struct class_attribute class_attr_bonding_masters;
626 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, struct slave *slave);
627 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev);
628 int bond_create(struct net *net, const char *name);
629 int bond_create_sysfs(struct bond_net *net);
630 void bond_destroy_sysfs(struct bond_net *net);
631 void bond_prepare_sysfs_group(struct bonding *bond);
632 int bond_sysfs_slave_add(struct slave *slave);
633 void bond_sysfs_slave_del(struct slave *slave);
634 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
635 struct netlink_ext_ack *extack);
636 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev);
637 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb);
638 int bond_set_carrier(struct bonding *bond);
639 void bond_select_active_slave(struct bonding *bond);
640 void bond_change_active_slave(struct bonding *bond, struct slave *new_active);
641 void bond_create_debugfs(void);
642 void bond_destroy_debugfs(void);
643 void bond_debug_register(struct bonding *bond);
644 void bond_debug_unregister(struct bonding *bond);
645 void bond_debug_reregister(struct bonding *bond);
646 const char *bond_mode_name(int mode);
647 void bond_setup(struct net_device *bond_dev);
648 unsigned int bond_get_num_tx_queues(void);
649 int bond_netlink_init(void);
650 void bond_netlink_fini(void);
651 struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond);
652 const char *bond_slave_link_status(s8 link);
653 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
654 struct net_device *end_dev,
656 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave);
657 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay);
658 void bond_work_init_all(struct bonding *bond);
660 #ifdef CONFIG_PROC_FS
661 void bond_create_proc_entry(struct bonding *bond);
662 void bond_remove_proc_entry(struct bonding *bond);
663 void bond_create_proc_dir(struct bond_net *bn);
664 void bond_destroy_proc_dir(struct bond_net *bn);
666 static inline void bond_create_proc_entry(struct bonding *bond)
670 static inline void bond_remove_proc_entry(struct bonding *bond)
674 static inline void bond_create_proc_dir(struct bond_net *bn)
678 static inline void bond_destroy_proc_dir(struct bond_net *bn)
683 static inline struct slave *bond_slave_has_mac(struct bonding *bond,
686 struct list_head *iter;
689 bond_for_each_slave(bond, tmp, iter)
690 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
696 /* Caller must hold rcu_read_lock() for read */
697 static inline struct slave *bond_slave_has_mac_rcu(struct bonding *bond,
700 struct list_head *iter;
703 bond_for_each_slave_rcu(bond, tmp, iter)
704 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
710 /* Caller must hold rcu_read_lock() for read */
711 static inline bool bond_slave_has_mac_rx(struct bonding *bond, const u8 *mac)
713 struct list_head *iter;
715 struct netdev_hw_addr *ha;
717 bond_for_each_slave_rcu(bond, tmp, iter)
718 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
721 if (netdev_uc_empty(bond->dev))
724 netdev_for_each_uc_addr(ha, bond->dev)
725 if (ether_addr_equal_64bits(mac, ha->addr))
731 /* Check if the ip is present in arp ip list, or first free slot if ip == 0
732 * Returns -1 if not found, index if found
734 static inline int bond_get_targets_ip(__be32 *targets, __be32 ip)
738 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
739 if (targets[i] == ip)
741 else if (targets[i] == 0)
747 /* exported from bond_main.c */
748 extern unsigned int bond_net_id;
749 extern const struct bond_parm_tbl bond_lacp_tbl[];
750 extern const struct bond_parm_tbl xmit_hashtype_tbl[];
751 extern const struct bond_parm_tbl arp_validate_tbl[];
752 extern const struct bond_parm_tbl arp_all_targets_tbl[];
753 extern const struct bond_parm_tbl fail_over_mac_tbl[];
754 extern const struct bond_parm_tbl pri_reselect_tbl[];
755 extern struct bond_parm_tbl ad_select_tbl[];
757 /* exported from bond_netlink.c */
758 extern struct rtnl_link_ops bond_link_ops;
760 /* exported from bond_sysfs_slave.c */
761 extern const struct sysfs_ops slave_sysfs_ops;
763 static inline netdev_tx_t bond_tx_drop(struct net_device *dev, struct sk_buff *skb)
765 atomic_long_inc(&dev->tx_dropped);
766 dev_kfree_skb_any(skb);
767 return NET_XMIT_DROP;
770 #endif /* _NET_BONDING_H */