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 BOND_MODE(bond) ((bond)->params.mode)
43 /* slave list primitives */
44 #define bond_slave_list(bond) (&(bond)->dev->adj_list.lower)
46 #define bond_has_slaves(bond) !list_empty(bond_slave_list(bond))
48 /* IMPORTANT: bond_first/last_slave can return NULL in case of an empty list */
49 #define bond_first_slave(bond) \
50 (bond_has_slaves(bond) ? \
51 netdev_adjacent_get_private(bond_slave_list(bond)->next) : \
53 #define bond_last_slave(bond) \
54 (bond_has_slaves(bond) ? \
55 netdev_adjacent_get_private(bond_slave_list(bond)->prev) : \
58 /* Caller must have rcu_read_lock */
59 #define bond_first_slave_rcu(bond) \
60 netdev_lower_get_first_private_rcu(bond->dev)
62 #define bond_is_first_slave(bond, pos) (pos == bond_first_slave(bond))
63 #define bond_is_last_slave(bond, pos) (pos == bond_last_slave(bond))
66 * bond_for_each_slave - iterate over all slaves
67 * @bond: the bond holding this list
69 * @iter: list_head * iterator
71 * Caller must hold RTNL
73 #define bond_for_each_slave(bond, pos, iter) \
74 netdev_for_each_lower_private((bond)->dev, pos, iter)
76 /* Caller must have rcu_read_lock */
77 #define bond_for_each_slave_rcu(bond, pos, iter) \
78 netdev_for_each_lower_private_rcu((bond)->dev, pos, iter)
80 #ifdef CONFIG_NET_POLL_CONTROLLER
81 extern atomic_t netpoll_block_tx
;
83 static inline void block_netpoll_tx(void)
85 atomic_inc(&netpoll_block_tx
);
88 static inline void unblock_netpoll_tx(void)
90 atomic_dec(&netpoll_block_tx
);
93 static inline int is_netpoll_tx_blocked(struct net_device
*dev
)
95 if (unlikely(netpoll_tx_running(dev
)))
96 return atomic_read(&netpoll_block_tx
);
100 #define block_netpoll_tx()
101 #define unblock_netpoll_tx()
102 #define is_netpoll_tx_blocked(dev) (0)
118 unsigned int min_links
;
120 char primary
[IFNAMSIZ
];
121 int primary_reselect
;
122 __be32 arp_targets
[BOND_MAX_ARP_TARGETS
];
124 int all_slaves_active
;
127 int packets_per_slave
;
129 struct reciprocal_value reciprocal_packets_per_slave
;
130 u16 ad_actor_sys_prio
;
131 u16 ad_user_port_key
;
133 /* 2 bytes of padding : see ether_addr_equal_64bits() */
134 u8 ad_actor_system
[ETH_ALEN
+ 2];
137 struct bond_parm_tbl
{
142 struct netdev_notify_work
{
143 struct delayed_work work
;
144 struct net_device
*dev
;
145 struct netdev_bonding_info bonding_info
;
149 struct net_device
*dev
; /* first - useful for panic debug */
150 struct bonding
*bond
; /* our master */
152 /* all three in jiffies */
153 unsigned long last_link_up
;
154 unsigned long last_rx
;
155 unsigned long target_last_arp_rx
[BOND_MAX_ARP_TARGETS
];
156 s8 link
; /* one of BOND_LINK_XXXX */
157 s8 link_new_state
; /* one of BOND_LINK_XXXX */
159 u8 backup
:1, /* indicates backup slave. Value corresponds with
160 BOND_STATE_ACTIVE and BOND_STATE_BACKUP */
161 inactive
:1, /* indicates inactive slave */
162 should_notify
:1, /* indicates whether the state changed */
163 should_notify_link
:1; /* indicates whether the link changed */
166 u32 link_failure_count
;
169 u8 perm_hwaddr
[MAX_ADDR_LEN
];
170 struct ad_slave_info
*ad_info
;
171 struct tlb_slave_info tlb_info
;
172 #ifdef CONFIG_NET_POLL_CONTROLLER
176 struct rtnl_link_stats64 slave_stats
;
179 struct bond_up_slave
{
182 struct slave
*arr
[0];
186 * Link pseudo-state only used internally by monitors
188 #define BOND_LINK_NOCHANGE -1
191 * Here are the locking policies for the two bonding locks:
192 * Get rcu_read_lock when reading or RTNL when writing slave list.
195 struct net_device
*dev
; /* first - useful for panic debug */
196 struct slave __rcu
*curr_active_slave
;
197 struct slave __rcu
*current_arp_slave
;
198 struct slave __rcu
*primary_slave
;
199 struct bond_up_slave __rcu
*slave_arr
; /* Array of usable slaves */
201 s32 slave_cnt
; /* never change this value outside the attach/detach wrappers */
202 int (*recv_probe
)(const struct sk_buff
*, struct bonding
*,
204 /* mode_lock is used for mode-specific locking needs, currently used by:
205 * 3ad mode (4) - protect against running bond_3ad_unbind_slave() and
206 * bond_3ad_state_machine_handler() concurrently and also
207 * the access to the state machine shared variables.
208 * TLB mode (5) - to sync the use and modifications of its hash table
209 * ALB mode (6) - to sync the use and modifications of its hash table
211 spinlock_t mode_lock
;
212 spinlock_t stats_lock
;
215 #ifdef CONFIG_PROC_FS
216 struct proc_dir_entry
*proc_entry
;
217 char proc_file_name
[IFNAMSIZ
];
218 #endif /* CONFIG_PROC_FS */
219 struct list_head bond_list
;
221 struct ad_bond_info ad_info
;
222 struct alb_bond_info alb_info
;
223 struct bond_params params
;
224 struct workqueue_struct
*wq
;
225 struct delayed_work mii_work
;
226 struct delayed_work arp_work
;
227 struct delayed_work alb_work
;
228 struct delayed_work ad_work
;
229 struct delayed_work mcast_work
;
230 struct delayed_work slave_arr_work
;
231 #ifdef CONFIG_DEBUG_FS
232 /* debugging support via debugfs */
233 struct dentry
*debug_dir
;
234 #endif /* CONFIG_DEBUG_FS */
235 struct rtnl_link_stats64 bond_stats
;
238 #define bond_slave_get_rcu(dev) \
239 ((struct slave *) rcu_dereference(dev->rx_handler_data))
241 #define bond_slave_get_rtnl(dev) \
242 ((struct slave *) rtnl_dereference(dev->rx_handler_data))
244 void bond_queue_slave_event(struct slave
*slave
);
245 void bond_lower_state_changed(struct slave
*slave
);
247 struct bond_vlan_tag
{
249 unsigned short vlan_id
;
253 * Returns NULL if the net_device does not belong to any of the bond's slaves
255 * Caller must hold bond lock for read
257 static inline struct slave
*bond_get_slave_by_dev(struct bonding
*bond
,
258 struct net_device
*slave_dev
)
260 return netdev_lower_dev_get_private(bond
->dev
, slave_dev
);
263 static inline struct bonding
*bond_get_bond_by_slave(struct slave
*slave
)
268 static inline bool bond_should_override_tx_queue(struct bonding
*bond
)
270 return BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
||
271 BOND_MODE(bond
) == BOND_MODE_ROUNDROBIN
;
274 static inline bool bond_is_lb(const struct bonding
*bond
)
276 return BOND_MODE(bond
) == BOND_MODE_TLB
||
277 BOND_MODE(bond
) == BOND_MODE_ALB
;
280 static inline bool bond_is_nondyn_tlb(const struct bonding
*bond
)
282 return (BOND_MODE(bond
) == BOND_MODE_TLB
) &&
283 (bond
->params
.tlb_dynamic_lb
== 0);
286 static inline bool bond_mode_uses_xmit_hash(const struct bonding
*bond
)
288 return (BOND_MODE(bond
) == BOND_MODE_8023AD
||
289 BOND_MODE(bond
) == BOND_MODE_XOR
||
290 bond_is_nondyn_tlb(bond
));
293 static inline bool bond_mode_uses_arp(int mode
)
295 return mode
!= BOND_MODE_8023AD
&& mode
!= BOND_MODE_TLB
&&
296 mode
!= BOND_MODE_ALB
;
299 static inline bool bond_mode_uses_primary(int mode
)
301 return mode
== BOND_MODE_ACTIVEBACKUP
|| mode
== BOND_MODE_TLB
||
302 mode
== BOND_MODE_ALB
;
305 static inline bool bond_uses_primary(struct bonding
*bond
)
307 return bond_mode_uses_primary(BOND_MODE(bond
));
310 static inline struct net_device
*bond_option_active_slave_get_rcu(struct bonding
*bond
)
312 struct slave
*slave
= rcu_dereference(bond
->curr_active_slave
);
314 return bond_uses_primary(bond
) && slave
? slave
->dev
: NULL
;
317 static inline bool bond_slave_is_up(struct slave
*slave
)
319 return netif_running(slave
->dev
) && netif_carrier_ok(slave
->dev
);
322 static inline void bond_set_active_slave(struct slave
*slave
)
326 bond_queue_slave_event(slave
);
327 bond_lower_state_changed(slave
);
328 rtmsg_ifinfo(RTM_NEWLINK
, slave
->dev
, 0, GFP_ATOMIC
);
332 static inline void bond_set_backup_slave(struct slave
*slave
)
334 if (!slave
->backup
) {
336 bond_queue_slave_event(slave
);
337 bond_lower_state_changed(slave
);
338 rtmsg_ifinfo(RTM_NEWLINK
, slave
->dev
, 0, GFP_ATOMIC
);
342 static inline void bond_set_slave_state(struct slave
*slave
,
343 int slave_state
, bool notify
)
345 if (slave
->backup
== slave_state
)
348 slave
->backup
= slave_state
;
350 bond_lower_state_changed(slave
);
351 rtmsg_ifinfo(RTM_NEWLINK
, slave
->dev
, 0, GFP_ATOMIC
);
352 bond_queue_slave_event(slave
);
353 slave
->should_notify
= 0;
355 if (slave
->should_notify
)
356 slave
->should_notify
= 0;
358 slave
->should_notify
= 1;
362 static inline void bond_slave_state_change(struct bonding
*bond
)
364 struct list_head
*iter
;
367 bond_for_each_slave(bond
, tmp
, iter
) {
368 if (tmp
->link
== BOND_LINK_UP
)
369 bond_set_active_slave(tmp
);
370 else if (tmp
->link
== BOND_LINK_DOWN
)
371 bond_set_backup_slave(tmp
);
375 static inline void bond_slave_state_notify(struct bonding
*bond
)
377 struct list_head
*iter
;
380 bond_for_each_slave(bond
, tmp
, iter
) {
381 if (tmp
->should_notify
) {
382 bond_lower_state_changed(tmp
);
383 rtmsg_ifinfo(RTM_NEWLINK
, tmp
->dev
, 0, GFP_ATOMIC
);
384 tmp
->should_notify
= 0;
389 static inline int bond_slave_state(struct slave
*slave
)
391 return slave
->backup
;
394 static inline bool bond_is_active_slave(struct slave
*slave
)
396 return !bond_slave_state(slave
);
399 static inline bool bond_slave_can_tx(struct slave
*slave
)
401 return bond_slave_is_up(slave
) && slave
->link
== BOND_LINK_UP
&&
402 bond_is_active_slave(slave
);
405 static inline void bond_hw_addr_copy(u8
*dst
, const u8
*src
, unsigned int len
)
407 if (len
== ETH_ALEN
) {
408 ether_addr_copy(dst
, src
);
412 memcpy(dst
, src
, len
);
415 #define BOND_PRI_RESELECT_ALWAYS 0
416 #define BOND_PRI_RESELECT_BETTER 1
417 #define BOND_PRI_RESELECT_FAILURE 2
419 #define BOND_FOM_NONE 0
420 #define BOND_FOM_ACTIVE 1
421 #define BOND_FOM_FOLLOW 2
423 #define BOND_ARP_TARGETS_ANY 0
424 #define BOND_ARP_TARGETS_ALL 1
426 #define BOND_ARP_VALIDATE_NONE 0
427 #define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE)
428 #define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP)
429 #define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \
430 BOND_ARP_VALIDATE_BACKUP)
431 #define BOND_ARP_FILTER (BOND_ARP_VALIDATE_ALL + 1)
432 #define BOND_ARP_FILTER_ACTIVE (BOND_ARP_VALIDATE_ACTIVE | \
434 #define BOND_ARP_FILTER_BACKUP (BOND_ARP_VALIDATE_BACKUP | \
437 #define BOND_SLAVE_NOTIFY_NOW true
438 #define BOND_SLAVE_NOTIFY_LATER false
440 static inline int slave_do_arp_validate(struct bonding
*bond
,
443 return bond
->params
.arp_validate
& (1 << bond_slave_state(slave
));
446 static inline int slave_do_arp_validate_only(struct bonding
*bond
)
448 return bond
->params
.arp_validate
& BOND_ARP_FILTER
;
451 static inline int bond_is_ip_target_ok(__be32 addr
)
453 return !ipv4_is_lbcast(addr
) && !ipv4_is_zeronet(addr
);
456 /* Get the oldest arp which we've received on this slave for bond's
459 static inline unsigned long slave_oldest_target_arp_rx(struct bonding
*bond
,
463 unsigned long ret
= slave
->target_last_arp_rx
[0];
465 for (; (i
< BOND_MAX_ARP_TARGETS
) && bond
->params
.arp_targets
[i
]; i
++)
466 if (time_before(slave
->target_last_arp_rx
[i
], ret
))
467 ret
= slave
->target_last_arp_rx
[i
];
472 static inline unsigned long slave_last_rx(struct bonding
*bond
,
475 if (bond
->params
.arp_all_targets
== BOND_ARP_TARGETS_ALL
)
476 return slave_oldest_target_arp_rx(bond
, slave
);
478 return slave
->last_rx
;
481 #ifdef CONFIG_NET_POLL_CONTROLLER
482 static inline void bond_netpoll_send_skb(const struct slave
*slave
,
485 struct netpoll
*np
= slave
->np
;
488 netpoll_send_skb(np
, skb
);
491 static inline void bond_netpoll_send_skb(const struct slave
*slave
,
497 static inline void bond_set_slave_inactive_flags(struct slave
*slave
,
500 if (!bond_is_lb(slave
->bond
))
501 bond_set_slave_state(slave
, BOND_STATE_BACKUP
, notify
);
502 if (!slave
->bond
->params
.all_slaves_active
)
506 static inline void bond_set_slave_active_flags(struct slave
*slave
,
509 bond_set_slave_state(slave
, BOND_STATE_ACTIVE
, notify
);
513 static inline bool bond_is_slave_inactive(struct slave
*slave
)
515 return slave
->inactive
;
518 static inline void bond_propose_link_state(struct slave
*slave
, int state
)
520 slave
->link_new_state
= state
;
523 static inline void bond_commit_link_state(struct slave
*slave
, bool notify
)
525 if (slave
->link
== slave
->link_new_state
)
528 slave
->link
= slave
->link_new_state
;
530 bond_queue_slave_event(slave
);
531 bond_lower_state_changed(slave
);
532 slave
->should_notify_link
= 0;
534 if (slave
->should_notify_link
)
535 slave
->should_notify_link
= 0;
537 slave
->should_notify_link
= 1;
541 static inline void bond_set_slave_link_state(struct slave
*slave
, int state
,
544 bond_propose_link_state(slave
, state
);
545 bond_commit_link_state(slave
, notify
);
548 static inline void bond_slave_link_notify(struct bonding
*bond
)
550 struct list_head
*iter
;
553 bond_for_each_slave(bond
, tmp
, iter
) {
554 if (tmp
->should_notify_link
) {
555 bond_queue_slave_event(tmp
);
556 bond_lower_state_changed(tmp
);
557 tmp
->should_notify_link
= 0;
562 static inline __be32
bond_confirm_addr(struct net_device
*dev
, __be32 dst
, __be32 local
)
564 struct in_device
*in_dev
;
568 in_dev
= __in_dev_get_rcu(dev
);
571 addr
= inet_confirm_addr(dev_net(dev
), in_dev
, dst
, local
,
578 struct net
*net
; /* Associated network namespace */
579 struct list_head dev_list
;
580 #ifdef CONFIG_PROC_FS
581 struct proc_dir_entry
*proc_dir
;
583 struct class_attribute class_attr_bonding_masters
;
586 int bond_arp_rcv(const struct sk_buff
*skb
, struct bonding
*bond
, struct slave
*slave
);
587 void bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
, struct net_device
*slave_dev
);
588 int bond_create(struct net
*net
, const char *name
);
589 int bond_create_sysfs(struct bond_net
*net
);
590 void bond_destroy_sysfs(struct bond_net
*net
);
591 void bond_prepare_sysfs_group(struct bonding
*bond
);
592 int bond_sysfs_slave_add(struct slave
*slave
);
593 void bond_sysfs_slave_del(struct slave
*slave
);
594 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
);
595 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
);
596 u32
bond_xmit_hash(struct bonding
*bond
, struct sk_buff
*skb
);
597 int bond_set_carrier(struct bonding
*bond
);
598 void bond_select_active_slave(struct bonding
*bond
);
599 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
);
600 void bond_create_debugfs(void);
601 void bond_destroy_debugfs(void);
602 void bond_debug_register(struct bonding
*bond
);
603 void bond_debug_unregister(struct bonding
*bond
);
604 void bond_debug_reregister(struct bonding
*bond
);
605 const char *bond_mode_name(int mode
);
606 void bond_setup(struct net_device
*bond_dev
);
607 unsigned int bond_get_num_tx_queues(void);
608 int bond_netlink_init(void);
609 void bond_netlink_fini(void);
610 struct net_device
*bond_option_active_slave_get_rcu(struct bonding
*bond
);
611 const char *bond_slave_link_status(s8 link
);
612 struct bond_vlan_tag
*bond_verify_device_path(struct net_device
*start_dev
,
613 struct net_device
*end_dev
,
615 int bond_update_slave_arr(struct bonding
*bond
, struct slave
*skipslave
);
616 void bond_slave_arr_work_rearm(struct bonding
*bond
, unsigned long delay
);
618 #ifdef CONFIG_PROC_FS
619 void bond_create_proc_entry(struct bonding
*bond
);
620 void bond_remove_proc_entry(struct bonding
*bond
);
621 void bond_create_proc_dir(struct bond_net
*bn
);
622 void bond_destroy_proc_dir(struct bond_net
*bn
);
624 static inline void bond_create_proc_entry(struct bonding
*bond
)
628 static inline void bond_remove_proc_entry(struct bonding
*bond
)
632 static inline void bond_create_proc_dir(struct bond_net
*bn
)
636 static inline void bond_destroy_proc_dir(struct bond_net
*bn
)
641 static inline struct slave
*bond_slave_has_mac(struct bonding
*bond
,
644 struct list_head
*iter
;
647 bond_for_each_slave(bond
, tmp
, iter
)
648 if (ether_addr_equal_64bits(mac
, tmp
->dev
->dev_addr
))
654 /* Caller must hold rcu_read_lock() for read */
655 static inline struct slave
*bond_slave_has_mac_rcu(struct bonding
*bond
,
658 struct list_head
*iter
;
661 bond_for_each_slave_rcu(bond
, tmp
, iter
)
662 if (ether_addr_equal_64bits(mac
, tmp
->dev
->dev_addr
))
668 /* Caller must hold rcu_read_lock() for read */
669 static inline bool bond_slave_has_mac_rx(struct bonding
*bond
, const u8
*mac
)
671 struct list_head
*iter
;
673 struct netdev_hw_addr
*ha
;
675 bond_for_each_slave_rcu(bond
, tmp
, iter
)
676 if (ether_addr_equal_64bits(mac
, tmp
->dev
->dev_addr
))
679 if (netdev_uc_empty(bond
->dev
))
682 netdev_for_each_uc_addr(ha
, bond
->dev
)
683 if (ether_addr_equal_64bits(mac
, ha
->addr
))
689 /* Check if the ip is present in arp ip list, or first free slot if ip == 0
690 * Returns -1 if not found, index if found
692 static inline int bond_get_targets_ip(__be32
*targets
, __be32 ip
)
696 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
; i
++)
697 if (targets
[i
] == ip
)
699 else if (targets
[i
] == 0)
705 /* exported from bond_main.c */
706 extern unsigned int bond_net_id
;
707 extern const struct bond_parm_tbl bond_lacp_tbl
[];
708 extern const struct bond_parm_tbl xmit_hashtype_tbl
[];
709 extern const struct bond_parm_tbl arp_validate_tbl
[];
710 extern const struct bond_parm_tbl arp_all_targets_tbl
[];
711 extern const struct bond_parm_tbl fail_over_mac_tbl
[];
712 extern const struct bond_parm_tbl pri_reselect_tbl
[];
713 extern struct bond_parm_tbl ad_select_tbl
[];
715 /* exported from bond_netlink.c */
716 extern struct rtnl_link_ops bond_link_ops
;
718 static inline void bond_tx_drop(struct net_device
*dev
, struct sk_buff
*skb
)
720 atomic_long_inc(&dev
->tx_dropped
);
721 dev_kfree_skb_any(skb
);
724 #endif /* _NET_BONDING_H */