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 * Less bad way to call ioctl from within the kernel; this needs to be
42 * done some other way to get the call out of interrupt context.
43 * Needs "ioctl" variable to be supplied by calling context.
45 #define IOCTL(dev, arg, cmd) ({ \
47 mm_segment_t fs = get_fs(); \
49 res = ioctl(dev, arg, cmd); \
53 #define BOND_MODE(bond) ((bond)->params.mode)
55 /* slave list primitives */
56 #define bond_slave_list(bond) (&(bond)->dev->adj_list.lower)
58 #define bond_has_slaves(bond) !list_empty(bond_slave_list(bond))
60 /* IMPORTANT: bond_first/last_slave can return NULL in case of an empty list */
61 #define bond_first_slave(bond) \
62 (bond_has_slaves(bond) ? \
63 netdev_adjacent_get_private(bond_slave_list(bond)->next) : \
65 #define bond_last_slave(bond) \
66 (bond_has_slaves(bond) ? \
67 netdev_adjacent_get_private(bond_slave_list(bond)->prev) : \
70 /* Caller must have rcu_read_lock */
71 #define bond_first_slave_rcu(bond) \
72 netdev_lower_get_first_private_rcu(bond->dev)
74 #define bond_is_first_slave(bond, pos) (pos == bond_first_slave(bond))
75 #define bond_is_last_slave(bond, pos) (pos == bond_last_slave(bond))
78 * bond_for_each_slave - iterate over all slaves
79 * @bond: the bond holding this list
81 * @iter: list_head * iterator
83 * Caller must hold RTNL
85 #define bond_for_each_slave(bond, pos, iter) \
86 netdev_for_each_lower_private((bond)->dev, pos, iter)
88 /* Caller must have rcu_read_lock */
89 #define bond_for_each_slave_rcu(bond, pos, iter) \
90 netdev_for_each_lower_private_rcu((bond)->dev, pos, iter)
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)
130 unsigned int min_links
;
132 char primary
[IFNAMSIZ
];
133 int primary_reselect
;
134 __be32 arp_targets
[BOND_MAX_ARP_TARGETS
];
136 int all_slaves_active
;
139 int packets_per_slave
;
141 struct reciprocal_value reciprocal_packets_per_slave
;
142 u16 ad_actor_sys_prio
;
143 u16 ad_user_port_key
;
145 /* 2 bytes of padding : see ether_addr_equal_64bits() */
146 u8 ad_actor_system
[ETH_ALEN
+ 2];
149 struct bond_parm_tbl
{
154 struct netdev_notify_work
{
155 struct delayed_work work
;
156 struct net_device
*dev
;
157 struct netdev_bonding_info bonding_info
;
161 struct net_device
*dev
; /* first - useful for panic debug */
162 struct bonding
*bond
; /* our master */
164 /* all three in jiffies */
165 unsigned long last_link_up
;
166 unsigned long last_rx
;
167 unsigned long target_last_arp_rx
[BOND_MAX_ARP_TARGETS
];
168 s8 link
; /* one of BOND_LINK_XXXX */
170 u8 backup
:1, /* indicates backup slave. Value corresponds with
171 BOND_STATE_ACTIVE and BOND_STATE_BACKUP */
172 inactive
:1, /* indicates inactive slave */
173 should_notify
:1, /* indicates whether the state changed */
174 should_notify_link
:1; /* indicates whether the link changed */
177 u32 link_failure_count
;
180 u8 perm_hwaddr
[ETH_ALEN
];
181 struct ad_slave_info
*ad_info
;
182 struct tlb_slave_info tlb_info
;
183 #ifdef CONFIG_NET_POLL_CONTROLLER
187 struct rtnl_link_stats64 slave_stats
;
190 struct bond_up_slave
{
193 struct slave
*arr
[0];
197 * Link pseudo-state only used internally by monitors
199 #define BOND_LINK_NOCHANGE -1
202 * Here are the locking policies for the two bonding locks:
203 * Get rcu_read_lock when reading or RTNL when writing slave list.
206 struct net_device
*dev
; /* first - useful for panic debug */
207 struct slave __rcu
*curr_active_slave
;
208 struct slave __rcu
*current_arp_slave
;
209 struct slave __rcu
*primary_slave
;
210 struct bond_up_slave __rcu
*slave_arr
; /* Array of usable slaves */
212 s32 slave_cnt
; /* never change this value outside the attach/detach wrappers */
213 int (*recv_probe
)(const struct sk_buff
*, struct bonding
*,
215 /* mode_lock is used for mode-specific locking needs, currently used by:
216 * 3ad mode (4) - protect against running bond_3ad_unbind_slave() and
217 * bond_3ad_state_machine_handler() concurrently and also
218 * the access to the state machine shared variables.
219 * TLB mode (5) - to sync the use and modifications of its hash table
220 * ALB mode (6) - to sync the use and modifications of its hash table
222 spinlock_t mode_lock
;
223 spinlock_t stats_lock
;
226 #ifdef CONFIG_PROC_FS
227 struct proc_dir_entry
*proc_entry
;
228 char proc_file_name
[IFNAMSIZ
];
229 #endif /* CONFIG_PROC_FS */
230 struct list_head bond_list
;
232 struct ad_bond_info ad_info
;
233 struct alb_bond_info alb_info
;
234 struct bond_params params
;
235 struct workqueue_struct
*wq
;
236 struct delayed_work mii_work
;
237 struct delayed_work arp_work
;
238 struct delayed_work alb_work
;
239 struct delayed_work ad_work
;
240 struct delayed_work mcast_work
;
241 struct delayed_work slave_arr_work
;
242 #ifdef CONFIG_DEBUG_FS
243 /* debugging support via debugfs */
244 struct dentry
*debug_dir
;
245 #endif /* CONFIG_DEBUG_FS */
246 struct rtnl_link_stats64 bond_stats
;
249 #define bond_slave_get_rcu(dev) \
250 ((struct slave *) rcu_dereference(dev->rx_handler_data))
252 #define bond_slave_get_rtnl(dev) \
253 ((struct slave *) rtnl_dereference(dev->rx_handler_data))
255 void bond_queue_slave_event(struct slave
*slave
);
256 void bond_lower_state_changed(struct slave
*slave
);
258 struct bond_vlan_tag
{
260 unsigned short vlan_id
;
264 * Returns NULL if the net_device does not belong to any of the bond's slaves
266 * Caller must hold bond lock for read
268 static inline struct slave
*bond_get_slave_by_dev(struct bonding
*bond
,
269 struct net_device
*slave_dev
)
271 return netdev_lower_dev_get_private(bond
->dev
, slave_dev
);
274 static inline struct bonding
*bond_get_bond_by_slave(struct slave
*slave
)
279 static inline bool bond_should_override_tx_queue(struct bonding
*bond
)
281 return BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
||
282 BOND_MODE(bond
) == BOND_MODE_ROUNDROBIN
;
285 static inline bool bond_is_lb(const struct bonding
*bond
)
287 return BOND_MODE(bond
) == BOND_MODE_TLB
||
288 BOND_MODE(bond
) == BOND_MODE_ALB
;
291 static inline bool bond_is_nondyn_tlb(const struct bonding
*bond
)
293 return (BOND_MODE(bond
) == BOND_MODE_TLB
) &&
294 (bond
->params
.tlb_dynamic_lb
== 0);
297 static inline bool bond_mode_uses_xmit_hash(const struct bonding
*bond
)
299 return (BOND_MODE(bond
) == BOND_MODE_8023AD
||
300 BOND_MODE(bond
) == BOND_MODE_XOR
||
301 bond_is_nondyn_tlb(bond
));
304 static inline bool bond_mode_uses_arp(int mode
)
306 return mode
!= BOND_MODE_8023AD
&& mode
!= BOND_MODE_TLB
&&
307 mode
!= BOND_MODE_ALB
;
310 static inline bool bond_mode_uses_primary(int mode
)
312 return mode
== BOND_MODE_ACTIVEBACKUP
|| mode
== BOND_MODE_TLB
||
313 mode
== BOND_MODE_ALB
;
316 static inline bool bond_uses_primary(struct bonding
*bond
)
318 return bond_mode_uses_primary(BOND_MODE(bond
));
321 static inline struct net_device
*bond_option_active_slave_get_rcu(struct bonding
*bond
)
323 struct slave
*slave
= rcu_dereference(bond
->curr_active_slave
);
325 return bond_uses_primary(bond
) && slave
? slave
->dev
: NULL
;
328 static inline bool bond_slave_is_up(struct slave
*slave
)
330 return netif_running(slave
->dev
) && netif_carrier_ok(slave
->dev
);
333 static inline void bond_set_active_slave(struct slave
*slave
)
337 bond_queue_slave_event(slave
);
338 bond_lower_state_changed(slave
);
339 rtmsg_ifinfo(RTM_NEWLINK
, slave
->dev
, 0, GFP_ATOMIC
);
343 static inline void bond_set_backup_slave(struct slave
*slave
)
345 if (!slave
->backup
) {
347 bond_queue_slave_event(slave
);
348 bond_lower_state_changed(slave
);
349 rtmsg_ifinfo(RTM_NEWLINK
, slave
->dev
, 0, GFP_ATOMIC
);
353 static inline void bond_set_slave_state(struct slave
*slave
,
354 int slave_state
, bool notify
)
356 if (slave
->backup
== slave_state
)
359 slave
->backup
= slave_state
;
361 bond_lower_state_changed(slave
);
362 rtmsg_ifinfo(RTM_NEWLINK
, slave
->dev
, 0, GFP_ATOMIC
);
363 bond_queue_slave_event(slave
);
364 slave
->should_notify
= 0;
366 if (slave
->should_notify
)
367 slave
->should_notify
= 0;
369 slave
->should_notify
= 1;
373 static inline void bond_slave_state_change(struct bonding
*bond
)
375 struct list_head
*iter
;
378 bond_for_each_slave(bond
, tmp
, iter
) {
379 if (tmp
->link
== BOND_LINK_UP
)
380 bond_set_active_slave(tmp
);
381 else if (tmp
->link
== BOND_LINK_DOWN
)
382 bond_set_backup_slave(tmp
);
386 static inline void bond_slave_state_notify(struct bonding
*bond
)
388 struct list_head
*iter
;
391 bond_for_each_slave(bond
, tmp
, iter
) {
392 if (tmp
->should_notify
) {
393 bond_lower_state_changed(tmp
);
394 rtmsg_ifinfo(RTM_NEWLINK
, tmp
->dev
, 0, GFP_ATOMIC
);
395 tmp
->should_notify
= 0;
400 static inline int bond_slave_state(struct slave
*slave
)
402 return slave
->backup
;
405 static inline bool bond_is_active_slave(struct slave
*slave
)
407 return !bond_slave_state(slave
);
410 static inline bool bond_slave_can_tx(struct slave
*slave
)
412 return bond_slave_is_up(slave
) && slave
->link
== BOND_LINK_UP
&&
413 bond_is_active_slave(slave
);
416 #define BOND_PRI_RESELECT_ALWAYS 0
417 #define BOND_PRI_RESELECT_BETTER 1
418 #define BOND_PRI_RESELECT_FAILURE 2
420 #define BOND_FOM_NONE 0
421 #define BOND_FOM_ACTIVE 1
422 #define BOND_FOM_FOLLOW 2
424 #define BOND_ARP_TARGETS_ANY 0
425 #define BOND_ARP_TARGETS_ALL 1
427 #define BOND_ARP_VALIDATE_NONE 0
428 #define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE)
429 #define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP)
430 #define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \
431 BOND_ARP_VALIDATE_BACKUP)
432 #define BOND_ARP_FILTER (BOND_ARP_VALIDATE_ALL + 1)
433 #define BOND_ARP_FILTER_ACTIVE (BOND_ARP_VALIDATE_ACTIVE | \
435 #define BOND_ARP_FILTER_BACKUP (BOND_ARP_VALIDATE_BACKUP | \
438 #define BOND_SLAVE_NOTIFY_NOW true
439 #define BOND_SLAVE_NOTIFY_LATER false
441 static inline int slave_do_arp_validate(struct bonding
*bond
,
444 return bond
->params
.arp_validate
& (1 << bond_slave_state(slave
));
447 static inline int slave_do_arp_validate_only(struct bonding
*bond
)
449 return bond
->params
.arp_validate
& BOND_ARP_FILTER
;
452 static inline int bond_is_ip_target_ok(__be32 addr
)
454 return !ipv4_is_lbcast(addr
) && !ipv4_is_zeronet(addr
);
457 /* Get the oldest arp which we've received on this slave for bond's
460 static inline unsigned long slave_oldest_target_arp_rx(struct bonding
*bond
,
464 unsigned long ret
= slave
->target_last_arp_rx
[0];
466 for (; (i
< BOND_MAX_ARP_TARGETS
) && bond
->params
.arp_targets
[i
]; i
++)
467 if (time_before(slave
->target_last_arp_rx
[i
], ret
))
468 ret
= slave
->target_last_arp_rx
[i
];
473 static inline unsigned long slave_last_rx(struct bonding
*bond
,
476 if (bond
->params
.arp_all_targets
== BOND_ARP_TARGETS_ALL
)
477 return slave_oldest_target_arp_rx(bond
, slave
);
479 return slave
->last_rx
;
482 #ifdef CONFIG_NET_POLL_CONTROLLER
483 static inline void bond_netpoll_send_skb(const struct slave
*slave
,
486 struct netpoll
*np
= slave
->np
;
489 netpoll_send_skb(np
, skb
);
492 static inline void bond_netpoll_send_skb(const struct slave
*slave
,
498 static inline void bond_set_slave_inactive_flags(struct slave
*slave
,
501 if (!bond_is_lb(slave
->bond
))
502 bond_set_slave_state(slave
, BOND_STATE_BACKUP
, notify
);
503 if (!slave
->bond
->params
.all_slaves_active
)
507 static inline void bond_set_slave_active_flags(struct slave
*slave
,
510 bond_set_slave_state(slave
, BOND_STATE_ACTIVE
, notify
);
514 static inline bool bond_is_slave_inactive(struct slave
*slave
)
516 return slave
->inactive
;
519 static inline void bond_set_slave_link_state(struct slave
*slave
, int state
,
522 if (slave
->link
== state
)
527 bond_queue_slave_event(slave
);
528 bond_lower_state_changed(slave
);
529 slave
->should_notify_link
= 0;
531 if (slave
->should_notify_link
)
532 slave
->should_notify_link
= 0;
534 slave
->should_notify_link
= 1;
538 static inline void bond_slave_link_notify(struct bonding
*bond
)
540 struct list_head
*iter
;
543 bond_for_each_slave(bond
, tmp
, iter
) {
544 if (tmp
->should_notify_link
) {
545 bond_queue_slave_event(tmp
);
546 bond_lower_state_changed(tmp
);
547 tmp
->should_notify_link
= 0;
552 static inline __be32
bond_confirm_addr(struct net_device
*dev
, __be32 dst
, __be32 local
)
554 struct in_device
*in_dev
;
558 in_dev
= __in_dev_get_rcu(dev
);
561 addr
= inet_confirm_addr(dev_net(dev
), in_dev
, dst
, local
,
568 struct net
*net
; /* Associated network namespace */
569 struct list_head dev_list
;
570 #ifdef CONFIG_PROC_FS
571 struct proc_dir_entry
*proc_dir
;
573 struct class_attribute class_attr_bonding_masters
;
576 int bond_arp_rcv(const struct sk_buff
*skb
, struct bonding
*bond
, struct slave
*slave
);
577 void bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
, struct net_device
*slave_dev
);
578 int bond_create(struct net
*net
, const char *name
);
579 int bond_create_sysfs(struct bond_net
*net
);
580 void bond_destroy_sysfs(struct bond_net
*net
);
581 void bond_prepare_sysfs_group(struct bonding
*bond
);
582 int bond_sysfs_slave_add(struct slave
*slave
);
583 void bond_sysfs_slave_del(struct slave
*slave
);
584 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
);
585 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
);
586 u32
bond_xmit_hash(struct bonding
*bond
, struct sk_buff
*skb
);
587 int bond_set_carrier(struct bonding
*bond
);
588 void bond_select_active_slave(struct bonding
*bond
);
589 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
);
590 void bond_create_debugfs(void);
591 void bond_destroy_debugfs(void);
592 void bond_debug_register(struct bonding
*bond
);
593 void bond_debug_unregister(struct bonding
*bond
);
594 void bond_debug_reregister(struct bonding
*bond
);
595 const char *bond_mode_name(int mode
);
596 void bond_setup(struct net_device
*bond_dev
);
597 unsigned int bond_get_num_tx_queues(void);
598 int bond_netlink_init(void);
599 void bond_netlink_fini(void);
600 struct net_device
*bond_option_active_slave_get_rcu(struct bonding
*bond
);
601 const char *bond_slave_link_status(s8 link
);
602 struct bond_vlan_tag
*bond_verify_device_path(struct net_device
*start_dev
,
603 struct net_device
*end_dev
,
605 int bond_update_slave_arr(struct bonding
*bond
, struct slave
*skipslave
);
606 void bond_slave_arr_work_rearm(struct bonding
*bond
, unsigned long delay
);
608 #ifdef CONFIG_PROC_FS
609 void bond_create_proc_entry(struct bonding
*bond
);
610 void bond_remove_proc_entry(struct bonding
*bond
);
611 void bond_create_proc_dir(struct bond_net
*bn
);
612 void bond_destroy_proc_dir(struct bond_net
*bn
);
614 static inline void bond_create_proc_entry(struct bonding
*bond
)
618 static inline void bond_remove_proc_entry(struct bonding
*bond
)
622 static inline void bond_create_proc_dir(struct bond_net
*bn
)
626 static inline void bond_destroy_proc_dir(struct bond_net
*bn
)
631 static inline struct slave
*bond_slave_has_mac(struct bonding
*bond
,
634 struct list_head
*iter
;
637 bond_for_each_slave(bond
, tmp
, iter
)
638 if (ether_addr_equal_64bits(mac
, tmp
->dev
->dev_addr
))
644 /* Caller must hold rcu_read_lock() for read */
645 static inline struct slave
*bond_slave_has_mac_rcu(struct bonding
*bond
,
648 struct list_head
*iter
;
651 bond_for_each_slave_rcu(bond
, tmp
, iter
)
652 if (ether_addr_equal_64bits(mac
, tmp
->dev
->dev_addr
))
658 /* Caller must hold rcu_read_lock() for read */
659 static inline bool bond_slave_has_mac_rx(struct bonding
*bond
, const u8
*mac
)
661 struct list_head
*iter
;
663 struct netdev_hw_addr
*ha
;
665 bond_for_each_slave_rcu(bond
, tmp
, iter
)
666 if (ether_addr_equal_64bits(mac
, tmp
->dev
->dev_addr
))
669 if (netdev_uc_empty(bond
->dev
))
672 netdev_for_each_uc_addr(ha
, bond
->dev
)
673 if (ether_addr_equal_64bits(mac
, ha
->addr
))
679 /* Check if the ip is present in arp ip list, or first free slot if ip == 0
680 * Returns -1 if not found, index if found
682 static inline int bond_get_targets_ip(__be32
*targets
, __be32 ip
)
686 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
; i
++)
687 if (targets
[i
] == ip
)
689 else if (targets
[i
] == 0)
695 /* exported from bond_main.c */
696 extern int bond_net_id
;
697 extern const struct bond_parm_tbl bond_lacp_tbl
[];
698 extern const struct bond_parm_tbl xmit_hashtype_tbl
[];
699 extern const struct bond_parm_tbl arp_validate_tbl
[];
700 extern const struct bond_parm_tbl arp_all_targets_tbl
[];
701 extern const struct bond_parm_tbl fail_over_mac_tbl
[];
702 extern const struct bond_parm_tbl pri_reselect_tbl
[];
703 extern struct bond_parm_tbl ad_select_tbl
[];
705 /* exported from bond_netlink.c */
706 extern struct rtnl_link_ops bond_link_ops
;
708 static inline void bond_tx_drop(struct net_device
*dev
, struct sk_buff
*skb
)
710 atomic_long_inc(&dev
->tx_dropped
);
711 dev_kfree_skb_any(skb
);
714 #endif /* _NET_BONDING_H */