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 #ifdef CONFIG_XFRM_OFFLOAD
90 #define BOND_XFRM_FEATURES (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM | \
92 #endif /* CONFIG_XFRM_OFFLOAD */
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 struct bond_up_slave
{
195 * Link pseudo-state only used internally by monitors
197 #define BOND_LINK_NOCHANGE -1
200 * Here are the locking policies for the two bonding locks:
201 * Get rcu_read_lock when reading or RTNL when writing slave list.
204 struct net_device
*dev
; /* first - useful for panic debug */
205 struct slave __rcu
*curr_active_slave
;
206 struct slave __rcu
*current_arp_slave
;
207 struct slave __rcu
*primary_slave
;
208 struct bond_up_slave __rcu
*usable_slaves
;
209 struct bond_up_slave __rcu
*all_slaves
;
211 s32 slave_cnt
; /* never change this value outside the attach/detach wrappers */
212 int (*recv_probe
)(const struct sk_buff
*, struct bonding
*,
214 /* mode_lock is used for mode-specific locking needs, currently used by:
215 * 3ad mode (4) - protect against running bond_3ad_unbind_slave() and
216 * bond_3ad_state_machine_handler() concurrently and also
217 * the access to the state machine shared variables.
218 * TLB mode (5) - to sync the use and modifications of its hash table
219 * ALB mode (6) - to sync the use and modifications of its hash table
221 spinlock_t mode_lock
;
222 spinlock_t stats_lock
;
225 #ifdef CONFIG_PROC_FS
226 struct proc_dir_entry
*proc_entry
;
227 char proc_file_name
[IFNAMSIZ
];
228 #endif /* CONFIG_PROC_FS */
229 struct list_head bond_list
;
231 struct ad_bond_info ad_info
;
232 struct alb_bond_info alb_info
;
233 struct bond_params params
;
234 struct workqueue_struct
*wq
;
235 struct delayed_work mii_work
;
236 struct delayed_work arp_work
;
237 struct delayed_work alb_work
;
238 struct delayed_work ad_work
;
239 struct delayed_work mcast_work
;
240 struct delayed_work slave_arr_work
;
241 #ifdef CONFIG_DEBUG_FS
242 /* debugging support via debugfs */
243 struct dentry
*debug_dir
;
244 #endif /* CONFIG_DEBUG_FS */
245 struct rtnl_link_stats64 bond_stats
;
246 #ifdef CONFIG_XFRM_OFFLOAD
247 struct xfrm_state
*xs
;
248 #endif /* CONFIG_XFRM_OFFLOAD */
251 #define bond_slave_get_rcu(dev) \
252 ((struct slave *) rcu_dereference(dev->rx_handler_data))
254 #define bond_slave_get_rtnl(dev) \
255 ((struct slave *) rtnl_dereference(dev->rx_handler_data))
257 void bond_queue_slave_event(struct slave
*slave
);
258 void bond_lower_state_changed(struct slave
*slave
);
260 struct bond_vlan_tag
{
262 unsigned short vlan_id
;
266 * Returns NULL if the net_device does not belong to any of the bond's slaves
268 * Caller must hold bond lock for read
270 static inline struct slave
*bond_get_slave_by_dev(struct bonding
*bond
,
271 struct net_device
*slave_dev
)
273 return netdev_lower_dev_get_private(bond
->dev
, slave_dev
);
276 static inline struct bonding
*bond_get_bond_by_slave(struct slave
*slave
)
281 static inline bool bond_should_override_tx_queue(struct bonding
*bond
)
283 return BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
||
284 BOND_MODE(bond
) == BOND_MODE_ROUNDROBIN
;
287 static inline bool bond_is_lb(const struct bonding
*bond
)
289 return BOND_MODE(bond
) == BOND_MODE_TLB
||
290 BOND_MODE(bond
) == BOND_MODE_ALB
;
293 static inline bool bond_needs_speed_duplex(const struct bonding
*bond
)
295 return BOND_MODE(bond
) == BOND_MODE_8023AD
|| bond_is_lb(bond
);
298 static inline bool bond_is_nondyn_tlb(const struct bonding
*bond
)
300 return (bond_is_lb(bond
) && bond
->params
.tlb_dynamic_lb
== 0);
303 static inline bool bond_mode_can_use_xmit_hash(const struct bonding
*bond
)
305 return (BOND_MODE(bond
) == BOND_MODE_8023AD
||
306 BOND_MODE(bond
) == BOND_MODE_XOR
||
307 BOND_MODE(bond
) == BOND_MODE_TLB
||
308 BOND_MODE(bond
) == BOND_MODE_ALB
);
311 static inline bool bond_mode_uses_xmit_hash(const struct bonding
*bond
)
313 return (BOND_MODE(bond
) == BOND_MODE_8023AD
||
314 BOND_MODE(bond
) == BOND_MODE_XOR
||
315 bond_is_nondyn_tlb(bond
));
318 static inline bool bond_mode_uses_arp(int mode
)
320 return mode
!= BOND_MODE_8023AD
&& mode
!= BOND_MODE_TLB
&&
321 mode
!= BOND_MODE_ALB
;
324 static inline bool bond_mode_uses_primary(int mode
)
326 return mode
== BOND_MODE_ACTIVEBACKUP
|| mode
== BOND_MODE_TLB
||
327 mode
== BOND_MODE_ALB
;
330 static inline bool bond_uses_primary(struct bonding
*bond
)
332 return bond_mode_uses_primary(BOND_MODE(bond
));
335 static inline struct net_device
*bond_option_active_slave_get_rcu(struct bonding
*bond
)
337 struct slave
*slave
= rcu_dereference(bond
->curr_active_slave
);
339 return bond_uses_primary(bond
) && slave
? slave
->dev
: NULL
;
342 static inline bool bond_slave_is_up(struct slave
*slave
)
344 return netif_running(slave
->dev
) && netif_carrier_ok(slave
->dev
);
347 static inline void bond_set_active_slave(struct slave
*slave
)
351 bond_queue_slave_event(slave
);
352 bond_lower_state_changed(slave
);
356 static inline void bond_set_backup_slave(struct slave
*slave
)
358 if (!slave
->backup
) {
360 bond_queue_slave_event(slave
);
361 bond_lower_state_changed(slave
);
365 static inline void bond_set_slave_state(struct slave
*slave
,
366 int slave_state
, bool notify
)
368 if (slave
->backup
== slave_state
)
371 slave
->backup
= slave_state
;
373 bond_lower_state_changed(slave
);
374 bond_queue_slave_event(slave
);
375 slave
->should_notify
= 0;
377 if (slave
->should_notify
)
378 slave
->should_notify
= 0;
380 slave
->should_notify
= 1;
384 static inline void bond_slave_state_change(struct bonding
*bond
)
386 struct list_head
*iter
;
389 bond_for_each_slave(bond
, tmp
, iter
) {
390 if (tmp
->link
== BOND_LINK_UP
)
391 bond_set_active_slave(tmp
);
392 else if (tmp
->link
== BOND_LINK_DOWN
)
393 bond_set_backup_slave(tmp
);
397 static inline void bond_slave_state_notify(struct bonding
*bond
)
399 struct list_head
*iter
;
402 bond_for_each_slave(bond
, tmp
, iter
) {
403 if (tmp
->should_notify
) {
404 bond_lower_state_changed(tmp
);
405 tmp
->should_notify
= 0;
410 static inline int bond_slave_state(struct slave
*slave
)
412 return slave
->backup
;
415 static inline bool bond_is_active_slave(struct slave
*slave
)
417 return !bond_slave_state(slave
);
420 static inline bool bond_slave_can_tx(struct slave
*slave
)
422 return bond_slave_is_up(slave
) && slave
->link
== BOND_LINK_UP
&&
423 bond_is_active_slave(slave
);
426 static inline bool bond_is_active_slave_dev(const struct net_device
*slave_dev
)
432 slave
= bond_slave_get_rcu(slave_dev
);
433 active
= bond_is_active_slave(slave
);
439 static inline void bond_hw_addr_copy(u8
*dst
, const u8
*src
, unsigned int len
)
441 if (len
== ETH_ALEN
) {
442 ether_addr_copy(dst
, src
);
446 memcpy(dst
, src
, len
);
449 #define BOND_PRI_RESELECT_ALWAYS 0
450 #define BOND_PRI_RESELECT_BETTER 1
451 #define BOND_PRI_RESELECT_FAILURE 2
453 #define BOND_FOM_NONE 0
454 #define BOND_FOM_ACTIVE 1
455 #define BOND_FOM_FOLLOW 2
457 #define BOND_ARP_TARGETS_ANY 0
458 #define BOND_ARP_TARGETS_ALL 1
460 #define BOND_ARP_VALIDATE_NONE 0
461 #define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE)
462 #define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP)
463 #define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \
464 BOND_ARP_VALIDATE_BACKUP)
465 #define BOND_ARP_FILTER (BOND_ARP_VALIDATE_ALL + 1)
466 #define BOND_ARP_FILTER_ACTIVE (BOND_ARP_VALIDATE_ACTIVE | \
468 #define BOND_ARP_FILTER_BACKUP (BOND_ARP_VALIDATE_BACKUP | \
471 #define BOND_SLAVE_NOTIFY_NOW true
472 #define BOND_SLAVE_NOTIFY_LATER false
474 static inline int slave_do_arp_validate(struct bonding
*bond
,
477 return bond
->params
.arp_validate
& (1 << bond_slave_state(slave
));
480 static inline int slave_do_arp_validate_only(struct bonding
*bond
)
482 return bond
->params
.arp_validate
& BOND_ARP_FILTER
;
485 static inline int bond_is_ip_target_ok(__be32 addr
)
487 return !ipv4_is_lbcast(addr
) && !ipv4_is_zeronet(addr
);
490 /* Get the oldest arp which we've received on this slave for bond's
493 static inline unsigned long slave_oldest_target_arp_rx(struct bonding
*bond
,
497 unsigned long ret
= slave
->target_last_arp_rx
[0];
499 for (; (i
< BOND_MAX_ARP_TARGETS
) && bond
->params
.arp_targets
[i
]; i
++)
500 if (time_before(slave
->target_last_arp_rx
[i
], ret
))
501 ret
= slave
->target_last_arp_rx
[i
];
506 static inline unsigned long slave_last_rx(struct bonding
*bond
,
509 if (bond
->params
.arp_all_targets
== BOND_ARP_TARGETS_ALL
)
510 return slave_oldest_target_arp_rx(bond
, slave
);
512 return slave
->last_rx
;
515 #ifdef CONFIG_NET_POLL_CONTROLLER
516 static inline netdev_tx_t
bond_netpoll_send_skb(const struct slave
*slave
,
519 return netpoll_send_skb(slave
->np
, skb
);
522 static inline netdev_tx_t
bond_netpoll_send_skb(const struct slave
*slave
,
530 static inline void bond_set_slave_inactive_flags(struct slave
*slave
,
533 if (!bond_is_lb(slave
->bond
))
534 bond_set_slave_state(slave
, BOND_STATE_BACKUP
, notify
);
535 if (!slave
->bond
->params
.all_slaves_active
)
539 static inline void bond_set_slave_active_flags(struct slave
*slave
,
542 bond_set_slave_state(slave
, BOND_STATE_ACTIVE
, notify
);
546 static inline bool bond_is_slave_inactive(struct slave
*slave
)
548 return slave
->inactive
;
551 static inline void bond_propose_link_state(struct slave
*slave
, int state
)
553 slave
->link_new_state
= state
;
556 static inline void bond_commit_link_state(struct slave
*slave
, bool notify
)
558 if (slave
->link_new_state
== BOND_LINK_NOCHANGE
)
561 slave
->link
= slave
->link_new_state
;
563 bond_queue_slave_event(slave
);
564 bond_lower_state_changed(slave
);
565 slave
->should_notify_link
= 0;
567 if (slave
->should_notify_link
)
568 slave
->should_notify_link
= 0;
570 slave
->should_notify_link
= 1;
574 static inline void bond_set_slave_link_state(struct slave
*slave
, int state
,
577 bond_propose_link_state(slave
, state
);
578 bond_commit_link_state(slave
, notify
);
581 static inline void bond_slave_link_notify(struct bonding
*bond
)
583 struct list_head
*iter
;
586 bond_for_each_slave(bond
, tmp
, iter
) {
587 if (tmp
->should_notify_link
) {
588 bond_queue_slave_event(tmp
);
589 bond_lower_state_changed(tmp
);
590 tmp
->should_notify_link
= 0;
595 static inline __be32
bond_confirm_addr(struct net_device
*dev
, __be32 dst
, __be32 local
)
597 struct in_device
*in_dev
;
601 in_dev
= __in_dev_get_rcu(dev
);
604 addr
= inet_confirm_addr(dev_net(dev
), in_dev
, dst
, local
,
611 struct net
*net
; /* Associated network namespace */
612 struct list_head dev_list
;
613 #ifdef CONFIG_PROC_FS
614 struct proc_dir_entry
*proc_dir
;
616 struct class_attribute class_attr_bonding_masters
;
619 int bond_arp_rcv(const struct sk_buff
*skb
, struct bonding
*bond
, struct slave
*slave
);
620 netdev_tx_t
bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
, struct net_device
*slave_dev
);
621 int bond_create(struct net
*net
, const char *name
);
622 int bond_create_sysfs(struct bond_net
*net
);
623 void bond_destroy_sysfs(struct bond_net
*net
);
624 void bond_prepare_sysfs_group(struct bonding
*bond
);
625 int bond_sysfs_slave_add(struct slave
*slave
);
626 void bond_sysfs_slave_del(struct slave
*slave
);
627 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
,
628 struct netlink_ext_ack
*extack
);
629 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
);
630 u32
bond_xmit_hash(struct bonding
*bond
, struct sk_buff
*skb
);
631 int bond_set_carrier(struct bonding
*bond
);
632 void bond_select_active_slave(struct bonding
*bond
);
633 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
);
634 void bond_create_debugfs(void);
635 void bond_destroy_debugfs(void);
636 void bond_debug_register(struct bonding
*bond
);
637 void bond_debug_unregister(struct bonding
*bond
);
638 void bond_debug_reregister(struct bonding
*bond
);
639 const char *bond_mode_name(int mode
);
640 void bond_setup(struct net_device
*bond_dev
);
641 unsigned int bond_get_num_tx_queues(void);
642 int bond_netlink_init(void);
643 void bond_netlink_fini(void);
644 struct net_device
*bond_option_active_slave_get_rcu(struct bonding
*bond
);
645 const char *bond_slave_link_status(s8 link
);
646 struct bond_vlan_tag
*bond_verify_device_path(struct net_device
*start_dev
,
647 struct net_device
*end_dev
,
649 int bond_update_slave_arr(struct bonding
*bond
, struct slave
*skipslave
);
650 void bond_slave_arr_work_rearm(struct bonding
*bond
, unsigned long delay
);
651 void bond_work_init_all(struct bonding
*bond
);
653 #ifdef CONFIG_PROC_FS
654 void bond_create_proc_entry(struct bonding
*bond
);
655 void bond_remove_proc_entry(struct bonding
*bond
);
656 void bond_create_proc_dir(struct bond_net
*bn
);
657 void bond_destroy_proc_dir(struct bond_net
*bn
);
659 static inline void bond_create_proc_entry(struct bonding
*bond
)
663 static inline void bond_remove_proc_entry(struct bonding
*bond
)
667 static inline void bond_create_proc_dir(struct bond_net
*bn
)
671 static inline void bond_destroy_proc_dir(struct bond_net
*bn
)
676 static inline struct slave
*bond_slave_has_mac(struct bonding
*bond
,
679 struct list_head
*iter
;
682 bond_for_each_slave(bond
, tmp
, iter
)
683 if (ether_addr_equal_64bits(mac
, tmp
->dev
->dev_addr
))
689 /* Caller must hold rcu_read_lock() for read */
690 static inline struct slave
*bond_slave_has_mac_rcu(struct bonding
*bond
,
693 struct list_head
*iter
;
696 bond_for_each_slave_rcu(bond
, tmp
, iter
)
697 if (ether_addr_equal_64bits(mac
, tmp
->dev
->dev_addr
))
703 /* Caller must hold rcu_read_lock() for read */
704 static inline bool bond_slave_has_mac_rx(struct bonding
*bond
, const u8
*mac
)
706 struct list_head
*iter
;
708 struct netdev_hw_addr
*ha
;
710 bond_for_each_slave_rcu(bond
, tmp
, iter
)
711 if (ether_addr_equal_64bits(mac
, tmp
->dev
->dev_addr
))
714 if (netdev_uc_empty(bond
->dev
))
717 netdev_for_each_uc_addr(ha
, bond
->dev
)
718 if (ether_addr_equal_64bits(mac
, ha
->addr
))
724 /* Check if the ip is present in arp ip list, or first free slot if ip == 0
725 * Returns -1 if not found, index if found
727 static inline int bond_get_targets_ip(__be32
*targets
, __be32 ip
)
731 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
; i
++)
732 if (targets
[i
] == ip
)
734 else if (targets
[i
] == 0)
740 /* exported from bond_main.c */
741 extern unsigned int bond_net_id
;
742 extern const struct bond_parm_tbl bond_lacp_tbl
[];
743 extern const struct bond_parm_tbl xmit_hashtype_tbl
[];
744 extern const struct bond_parm_tbl arp_validate_tbl
[];
745 extern const struct bond_parm_tbl arp_all_targets_tbl
[];
746 extern const struct bond_parm_tbl fail_over_mac_tbl
[];
747 extern const struct bond_parm_tbl pri_reselect_tbl
[];
748 extern struct bond_parm_tbl ad_select_tbl
[];
750 /* exported from bond_netlink.c */
751 extern struct rtnl_link_ops bond_link_ops
;
753 static inline netdev_tx_t
bond_tx_drop(struct net_device
*dev
, struct sk_buff
*skb
)
755 atomic_long_inc(&dev
->tx_dropped
);
756 dev_kfree_skb_any(skb
);
757 return NET_XMIT_DROP
;
760 #endif /* _NET_BONDING_H */