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 #ifdef CONFIG_NET_POLL_CONTROLLER
93 extern atomic_t netpoll_block_tx
;
95 static inline void block_netpoll_tx(void)
97 atomic_inc(&netpoll_block_tx
);
100 static inline void unblock_netpoll_tx(void)
102 atomic_dec(&netpoll_block_tx
);
105 static inline int is_netpoll_tx_blocked(struct net_device
*dev
)
107 if (unlikely(netpoll_tx_running(dev
)))
108 return atomic_read(&netpoll_block_tx
);
112 #define block_netpoll_tx()
113 #define unblock_netpoll_tx()
114 #define is_netpoll_tx_blocked(dev) (0)
129 int peer_notif_delay
;
131 unsigned int min_links
;
133 char primary
[IFNAMSIZ
];
134 int primary_reselect
;
135 __be32 arp_targets
[BOND_MAX_ARP_TARGETS
];
137 int all_slaves_active
;
140 int packets_per_slave
;
142 struct reciprocal_value reciprocal_packets_per_slave
;
143 u16 ad_actor_sys_prio
;
144 u16 ad_user_port_key
;
146 /* 2 bytes of padding : see ether_addr_equal_64bits() */
147 u8 ad_actor_system
[ETH_ALEN
+ 2];
150 struct bond_parm_tbl
{
156 struct net_device
*dev
; /* first - useful for panic debug */
157 struct bonding
*bond
; /* our master */
159 /* all three in jiffies */
160 unsigned long last_link_up
;
161 unsigned long last_rx
;
162 unsigned long target_last_arp_rx
[BOND_MAX_ARP_TARGETS
];
163 s8 link
; /* one of BOND_LINK_XXXX */
164 s8 link_new_state
; /* one of BOND_LINK_XXXX */
165 u8 backup
:1, /* indicates backup slave. Value corresponds with
166 BOND_STATE_ACTIVE and BOND_STATE_BACKUP */
167 inactive
:1, /* indicates inactive slave */
168 should_notify
:1, /* indicates whether the state changed */
169 should_notify_link
:1; /* indicates whether the link changed */
172 u32 link_failure_count
;
175 u8 perm_hwaddr
[MAX_ADDR_LEN
];
176 struct ad_slave_info
*ad_info
;
177 struct tlb_slave_info tlb_info
;
178 #ifdef CONFIG_NET_POLL_CONTROLLER
181 struct delayed_work notify_work
;
183 struct rtnl_link_stats64 slave_stats
;
186 static inline struct slave
*to_slave(struct kobject
*kobj
)
188 return container_of(kobj
, struct slave
, kobj
);
191 struct bond_up_slave
{
198 * Link pseudo-state only used internally by monitors
200 #define BOND_LINK_NOCHANGE -1
203 * Here are the locking policies for the two bonding locks:
204 * Get rcu_read_lock when reading or RTNL when writing slave list.
207 struct net_device
*dev
; /* first - useful for panic debug */
208 struct slave __rcu
*curr_active_slave
;
209 struct slave __rcu
*current_arp_slave
;
210 struct slave __rcu
*primary_slave
;
211 struct bond_up_slave __rcu
*usable_slaves
;
212 struct bond_up_slave __rcu
*all_slaves
;
214 s32 slave_cnt
; /* never change this value outside the attach/detach wrappers */
215 int (*recv_probe
)(const struct sk_buff
*, struct bonding
*,
217 /* mode_lock is used for mode-specific locking needs, currently used by:
218 * 3ad mode (4) - protect against running bond_3ad_unbind_slave() and
219 * bond_3ad_state_machine_handler() concurrently and also
220 * the access to the state machine shared variables.
221 * TLB mode (5) - to sync the use and modifications of its hash table
222 * ALB mode (6) - to sync the use and modifications of its hash table
224 spinlock_t mode_lock
;
225 spinlock_t stats_lock
;
228 #ifdef CONFIG_PROC_FS
229 struct proc_dir_entry
*proc_entry
;
230 char proc_file_name
[IFNAMSIZ
];
231 #endif /* CONFIG_PROC_FS */
232 struct list_head bond_list
;
234 struct ad_bond_info ad_info
;
235 struct alb_bond_info alb_info
;
236 struct bond_params params
;
237 struct workqueue_struct
*wq
;
238 struct delayed_work mii_work
;
239 struct delayed_work arp_work
;
240 struct delayed_work alb_work
;
241 struct delayed_work ad_work
;
242 struct delayed_work mcast_work
;
243 struct delayed_work slave_arr_work
;
244 #ifdef CONFIG_DEBUG_FS
245 /* debugging support via debugfs */
246 struct dentry
*debug_dir
;
247 #endif /* CONFIG_DEBUG_FS */
248 struct rtnl_link_stats64 bond_stats
;
249 #ifdef CONFIG_XFRM_OFFLOAD
250 struct xfrm_state
*xs
;
251 #endif /* CONFIG_XFRM_OFFLOAD */
254 #define bond_slave_get_rcu(dev) \
255 ((struct slave *) rcu_dereference(dev->rx_handler_data))
257 #define bond_slave_get_rtnl(dev) \
258 ((struct slave *) rtnl_dereference(dev->rx_handler_data))
260 void bond_queue_slave_event(struct slave
*slave
);
261 void bond_lower_state_changed(struct slave
*slave
);
263 struct bond_vlan_tag
{
265 unsigned short vlan_id
;
269 * Returns NULL if the net_device does not belong to any of the bond's slaves
271 * Caller must hold bond lock for read
273 static inline struct slave
*bond_get_slave_by_dev(struct bonding
*bond
,
274 struct net_device
*slave_dev
)
276 return netdev_lower_dev_get_private(bond
->dev
, slave_dev
);
279 static inline struct bonding
*bond_get_bond_by_slave(struct slave
*slave
)
284 static inline bool bond_should_override_tx_queue(struct bonding
*bond
)
286 return BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
||
287 BOND_MODE(bond
) == BOND_MODE_ROUNDROBIN
;
290 static inline bool bond_is_lb(const struct bonding
*bond
)
292 return BOND_MODE(bond
) == BOND_MODE_TLB
||
293 BOND_MODE(bond
) == BOND_MODE_ALB
;
296 static inline bool bond_needs_speed_duplex(const struct bonding
*bond
)
298 return BOND_MODE(bond
) == BOND_MODE_8023AD
|| bond_is_lb(bond
);
301 static inline bool bond_is_nondyn_tlb(const struct bonding
*bond
)
303 return (bond_is_lb(bond
) && bond
->params
.tlb_dynamic_lb
== 0);
306 static inline bool bond_mode_can_use_xmit_hash(const struct bonding
*bond
)
308 return (BOND_MODE(bond
) == BOND_MODE_8023AD
||
309 BOND_MODE(bond
) == BOND_MODE_XOR
||
310 BOND_MODE(bond
) == BOND_MODE_TLB
||
311 BOND_MODE(bond
) == BOND_MODE_ALB
);
314 static inline bool bond_mode_uses_xmit_hash(const struct bonding
*bond
)
316 return (BOND_MODE(bond
) == BOND_MODE_8023AD
||
317 BOND_MODE(bond
) == BOND_MODE_XOR
||
318 bond_is_nondyn_tlb(bond
));
321 static inline bool bond_mode_uses_arp(int mode
)
323 return mode
!= BOND_MODE_8023AD
&& mode
!= BOND_MODE_TLB
&&
324 mode
!= BOND_MODE_ALB
;
327 static inline bool bond_mode_uses_primary(int mode
)
329 return mode
== BOND_MODE_ACTIVEBACKUP
|| mode
== BOND_MODE_TLB
||
330 mode
== BOND_MODE_ALB
;
333 static inline bool bond_uses_primary(struct bonding
*bond
)
335 return bond_mode_uses_primary(BOND_MODE(bond
));
338 static inline struct net_device
*bond_option_active_slave_get_rcu(struct bonding
*bond
)
340 struct slave
*slave
= rcu_dereference(bond
->curr_active_slave
);
342 return bond_uses_primary(bond
) && slave
? slave
->dev
: NULL
;
345 static inline bool bond_slave_is_up(struct slave
*slave
)
347 return netif_running(slave
->dev
) && netif_carrier_ok(slave
->dev
);
350 static inline void bond_set_active_slave(struct slave
*slave
)
354 bond_queue_slave_event(slave
);
355 bond_lower_state_changed(slave
);
359 static inline void bond_set_backup_slave(struct slave
*slave
)
361 if (!slave
->backup
) {
363 bond_queue_slave_event(slave
);
364 bond_lower_state_changed(slave
);
368 static inline void bond_set_slave_state(struct slave
*slave
,
369 int slave_state
, bool notify
)
371 if (slave
->backup
== slave_state
)
374 slave
->backup
= slave_state
;
376 bond_lower_state_changed(slave
);
377 bond_queue_slave_event(slave
);
378 slave
->should_notify
= 0;
380 if (slave
->should_notify
)
381 slave
->should_notify
= 0;
383 slave
->should_notify
= 1;
387 static inline void bond_slave_state_change(struct bonding
*bond
)
389 struct list_head
*iter
;
392 bond_for_each_slave(bond
, tmp
, iter
) {
393 if (tmp
->link
== BOND_LINK_UP
)
394 bond_set_active_slave(tmp
);
395 else if (tmp
->link
== BOND_LINK_DOWN
)
396 bond_set_backup_slave(tmp
);
400 static inline void bond_slave_state_notify(struct bonding
*bond
)
402 struct list_head
*iter
;
405 bond_for_each_slave(bond
, tmp
, iter
) {
406 if (tmp
->should_notify
) {
407 bond_lower_state_changed(tmp
);
408 tmp
->should_notify
= 0;
413 static inline int bond_slave_state(struct slave
*slave
)
415 return slave
->backup
;
418 static inline bool bond_is_active_slave(struct slave
*slave
)
420 return !bond_slave_state(slave
);
423 static inline bool bond_slave_can_tx(struct slave
*slave
)
425 return bond_slave_is_up(slave
) && slave
->link
== BOND_LINK_UP
&&
426 bond_is_active_slave(slave
);
429 static inline bool bond_is_active_slave_dev(const struct net_device
*slave_dev
)
435 slave
= bond_slave_get_rcu(slave_dev
);
436 active
= bond_is_active_slave(slave
);
442 static inline void bond_hw_addr_copy(u8
*dst
, const u8
*src
, unsigned int len
)
444 if (len
== ETH_ALEN
) {
445 ether_addr_copy(dst
, src
);
449 memcpy(dst
, src
, len
);
452 #define BOND_PRI_RESELECT_ALWAYS 0
453 #define BOND_PRI_RESELECT_BETTER 1
454 #define BOND_PRI_RESELECT_FAILURE 2
456 #define BOND_FOM_NONE 0
457 #define BOND_FOM_ACTIVE 1
458 #define BOND_FOM_FOLLOW 2
460 #define BOND_ARP_TARGETS_ANY 0
461 #define BOND_ARP_TARGETS_ALL 1
463 #define BOND_ARP_VALIDATE_NONE 0
464 #define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE)
465 #define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP)
466 #define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \
467 BOND_ARP_VALIDATE_BACKUP)
468 #define BOND_ARP_FILTER (BOND_ARP_VALIDATE_ALL + 1)
469 #define BOND_ARP_FILTER_ACTIVE (BOND_ARP_VALIDATE_ACTIVE | \
471 #define BOND_ARP_FILTER_BACKUP (BOND_ARP_VALIDATE_BACKUP | \
474 #define BOND_SLAVE_NOTIFY_NOW true
475 #define BOND_SLAVE_NOTIFY_LATER false
477 static inline int slave_do_arp_validate(struct bonding
*bond
,
480 return bond
->params
.arp_validate
& (1 << bond_slave_state(slave
));
483 static inline int slave_do_arp_validate_only(struct bonding
*bond
)
485 return bond
->params
.arp_validate
& BOND_ARP_FILTER
;
488 static inline int bond_is_ip_target_ok(__be32 addr
)
490 return !ipv4_is_lbcast(addr
) && !ipv4_is_zeronet(addr
);
493 /* Get the oldest arp which we've received on this slave for bond's
496 static inline unsigned long slave_oldest_target_arp_rx(struct bonding
*bond
,
500 unsigned long ret
= slave
->target_last_arp_rx
[0];
502 for (; (i
< BOND_MAX_ARP_TARGETS
) && bond
->params
.arp_targets
[i
]; i
++)
503 if (time_before(slave
->target_last_arp_rx
[i
], ret
))
504 ret
= slave
->target_last_arp_rx
[i
];
509 static inline unsigned long slave_last_rx(struct bonding
*bond
,
512 if (bond
->params
.arp_all_targets
== BOND_ARP_TARGETS_ALL
)
513 return slave_oldest_target_arp_rx(bond
, slave
);
515 return slave
->last_rx
;
518 #ifdef CONFIG_NET_POLL_CONTROLLER
519 static inline netdev_tx_t
bond_netpoll_send_skb(const struct slave
*slave
,
522 return netpoll_send_skb(slave
->np
, skb
);
525 static inline netdev_tx_t
bond_netpoll_send_skb(const struct slave
*slave
,
533 static inline void bond_set_slave_inactive_flags(struct slave
*slave
,
536 if (!bond_is_lb(slave
->bond
))
537 bond_set_slave_state(slave
, BOND_STATE_BACKUP
, notify
);
538 if (!slave
->bond
->params
.all_slaves_active
)
542 static inline void bond_set_slave_active_flags(struct slave
*slave
,
545 bond_set_slave_state(slave
, BOND_STATE_ACTIVE
, notify
);
549 static inline bool bond_is_slave_inactive(struct slave
*slave
)
551 return slave
->inactive
;
554 static inline void bond_propose_link_state(struct slave
*slave
, int state
)
556 slave
->link_new_state
= state
;
559 static inline void bond_commit_link_state(struct slave
*slave
, bool notify
)
561 if (slave
->link_new_state
== BOND_LINK_NOCHANGE
)
564 slave
->link
= slave
->link_new_state
;
566 bond_queue_slave_event(slave
);
567 bond_lower_state_changed(slave
);
568 slave
->should_notify_link
= 0;
570 if (slave
->should_notify_link
)
571 slave
->should_notify_link
= 0;
573 slave
->should_notify_link
= 1;
577 static inline void bond_set_slave_link_state(struct slave
*slave
, int state
,
580 bond_propose_link_state(slave
, state
);
581 bond_commit_link_state(slave
, notify
);
584 static inline void bond_slave_link_notify(struct bonding
*bond
)
586 struct list_head
*iter
;
589 bond_for_each_slave(bond
, tmp
, iter
) {
590 if (tmp
->should_notify_link
) {
591 bond_queue_slave_event(tmp
);
592 bond_lower_state_changed(tmp
);
593 tmp
->should_notify_link
= 0;
598 static inline __be32
bond_confirm_addr(struct net_device
*dev
, __be32 dst
, __be32 local
)
600 struct in_device
*in_dev
;
604 in_dev
= __in_dev_get_rcu(dev
);
607 addr
= inet_confirm_addr(dev_net(dev
), in_dev
, dst
, local
,
614 struct net
*net
; /* Associated network namespace */
615 struct list_head dev_list
;
616 #ifdef CONFIG_PROC_FS
617 struct proc_dir_entry
*proc_dir
;
619 struct class_attribute class_attr_bonding_masters
;
622 int bond_arp_rcv(const struct sk_buff
*skb
, struct bonding
*bond
, struct slave
*slave
);
623 netdev_tx_t
bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
, struct net_device
*slave_dev
);
624 int bond_create(struct net
*net
, const char *name
);
625 int bond_create_sysfs(struct bond_net
*net
);
626 void bond_destroy_sysfs(struct bond_net
*net
);
627 void bond_prepare_sysfs_group(struct bonding
*bond
);
628 int bond_sysfs_slave_add(struct slave
*slave
);
629 void bond_sysfs_slave_del(struct slave
*slave
);
630 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
,
631 struct netlink_ext_ack
*extack
);
632 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
);
633 u32
bond_xmit_hash(struct bonding
*bond
, struct sk_buff
*skb
);
634 int bond_set_carrier(struct bonding
*bond
);
635 void bond_select_active_slave(struct bonding
*bond
);
636 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
);
637 void bond_create_debugfs(void);
638 void bond_destroy_debugfs(void);
639 void bond_debug_register(struct bonding
*bond
);
640 void bond_debug_unregister(struct bonding
*bond
);
641 void bond_debug_reregister(struct bonding
*bond
);
642 const char *bond_mode_name(int mode
);
643 void bond_setup(struct net_device
*bond_dev
);
644 unsigned int bond_get_num_tx_queues(void);
645 int bond_netlink_init(void);
646 void bond_netlink_fini(void);
647 struct net_device
*bond_option_active_slave_get_rcu(struct bonding
*bond
);
648 const char *bond_slave_link_status(s8 link
);
649 struct bond_vlan_tag
*bond_verify_device_path(struct net_device
*start_dev
,
650 struct net_device
*end_dev
,
652 int bond_update_slave_arr(struct bonding
*bond
, struct slave
*skipslave
);
653 void bond_slave_arr_work_rearm(struct bonding
*bond
, unsigned long delay
);
654 void bond_work_init_all(struct bonding
*bond
);
656 #ifdef CONFIG_PROC_FS
657 void bond_create_proc_entry(struct bonding
*bond
);
658 void bond_remove_proc_entry(struct bonding
*bond
);
659 void bond_create_proc_dir(struct bond_net
*bn
);
660 void bond_destroy_proc_dir(struct bond_net
*bn
);
662 static inline void bond_create_proc_entry(struct bonding
*bond
)
666 static inline void bond_remove_proc_entry(struct bonding
*bond
)
670 static inline void bond_create_proc_dir(struct bond_net
*bn
)
674 static inline void bond_destroy_proc_dir(struct bond_net
*bn
)
679 static inline struct slave
*bond_slave_has_mac(struct bonding
*bond
,
682 struct list_head
*iter
;
685 bond_for_each_slave(bond
, tmp
, iter
)
686 if (ether_addr_equal_64bits(mac
, tmp
->dev
->dev_addr
))
692 /* Caller must hold rcu_read_lock() for read */
693 static inline struct slave
*bond_slave_has_mac_rcu(struct bonding
*bond
,
696 struct list_head
*iter
;
699 bond_for_each_slave_rcu(bond
, tmp
, iter
)
700 if (ether_addr_equal_64bits(mac
, tmp
->dev
->dev_addr
))
706 /* Caller must hold rcu_read_lock() for read */
707 static inline bool bond_slave_has_mac_rx(struct bonding
*bond
, const u8
*mac
)
709 struct list_head
*iter
;
711 struct netdev_hw_addr
*ha
;
713 bond_for_each_slave_rcu(bond
, tmp
, iter
)
714 if (ether_addr_equal_64bits(mac
, tmp
->dev
->dev_addr
))
717 if (netdev_uc_empty(bond
->dev
))
720 netdev_for_each_uc_addr(ha
, bond
->dev
)
721 if (ether_addr_equal_64bits(mac
, ha
->addr
))
727 /* Check if the ip is present in arp ip list, or first free slot if ip == 0
728 * Returns -1 if not found, index if found
730 static inline int bond_get_targets_ip(__be32
*targets
, __be32 ip
)
734 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
; i
++)
735 if (targets
[i
] == ip
)
737 else if (targets
[i
] == 0)
743 /* exported from bond_main.c */
744 extern unsigned int bond_net_id
;
745 extern const struct bond_parm_tbl bond_lacp_tbl
[];
746 extern const struct bond_parm_tbl xmit_hashtype_tbl
[];
747 extern const struct bond_parm_tbl arp_validate_tbl
[];
748 extern const struct bond_parm_tbl arp_all_targets_tbl
[];
749 extern const struct bond_parm_tbl fail_over_mac_tbl
[];
750 extern const struct bond_parm_tbl pri_reselect_tbl
[];
751 extern struct bond_parm_tbl ad_select_tbl
[];
753 /* exported from bond_netlink.c */
754 extern struct rtnl_link_ops bond_link_ops
;
756 /* exported from bond_sysfs_slave.c */
757 extern const struct sysfs_ops slave_sysfs_ops
;
759 static inline netdev_tx_t
bond_tx_drop(struct net_device
*dev
, struct sk_buff
*skb
)
761 atomic_long_inc(&dev
->tx_dropped
);
762 dev_kfree_skb_any(skb
);
763 return NET_XMIT_DROP
;
766 #endif /* _NET_BONDING_H */