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
{
143 struct net_device
*dev
; /* first - useful for panic debug */
144 struct bonding
*bond
; /* our master */
146 /* all three in jiffies */
147 unsigned long last_link_up
;
148 unsigned long last_rx
;
149 unsigned long target_last_arp_rx
[BOND_MAX_ARP_TARGETS
];
150 s8 link
; /* one of BOND_LINK_XXXX */
151 s8 link_new_state
; /* one of BOND_LINK_XXXX */
153 u8 backup
:1, /* indicates backup slave. Value corresponds with
154 BOND_STATE_ACTIVE and BOND_STATE_BACKUP */
155 inactive
:1, /* indicates inactive slave */
156 should_notify
:1, /* indicates whether the state changed */
157 should_notify_link
:1; /* indicates whether the link changed */
160 u32 link_failure_count
;
163 u8 perm_hwaddr
[MAX_ADDR_LEN
];
164 struct ad_slave_info
*ad_info
;
165 struct tlb_slave_info tlb_info
;
166 #ifdef CONFIG_NET_POLL_CONTROLLER
169 struct delayed_work notify_work
;
171 struct rtnl_link_stats64 slave_stats
;
174 struct bond_up_slave
{
177 struct slave
*arr
[0];
181 * Link pseudo-state only used internally by monitors
183 #define BOND_LINK_NOCHANGE -1
186 * Here are the locking policies for the two bonding locks:
187 * Get rcu_read_lock when reading or RTNL when writing slave list.
190 struct net_device
*dev
; /* first - useful for panic debug */
191 struct slave __rcu
*curr_active_slave
;
192 struct slave __rcu
*current_arp_slave
;
193 struct slave __rcu
*primary_slave
;
194 struct bond_up_slave __rcu
*slave_arr
; /* Array of usable slaves */
197 s32 slave_cnt
; /* never change this value outside the attach/detach wrappers */
198 int (*recv_probe
)(const struct sk_buff
*, struct bonding
*,
200 /* mode_lock is used for mode-specific locking needs, currently used by:
201 * 3ad mode (4) - protect against running bond_3ad_unbind_slave() and
202 * bond_3ad_state_machine_handler() concurrently and also
203 * the access to the state machine shared variables.
204 * TLB mode (5) - to sync the use and modifications of its hash table
205 * ALB mode (6) - to sync the use and modifications of its hash table
207 spinlock_t mode_lock
;
208 spinlock_t stats_lock
;
211 #ifdef CONFIG_PROC_FS
212 struct proc_dir_entry
*proc_entry
;
213 char proc_file_name
[IFNAMSIZ
];
214 #endif /* CONFIG_PROC_FS */
215 struct list_head bond_list
;
217 struct ad_bond_info ad_info
;
218 struct alb_bond_info alb_info
;
219 struct bond_params params
;
220 struct workqueue_struct
*wq
;
221 struct delayed_work mii_work
;
222 struct delayed_work arp_work
;
223 struct delayed_work alb_work
;
224 struct delayed_work ad_work
;
225 struct delayed_work mcast_work
;
226 struct delayed_work slave_arr_work
;
227 #ifdef CONFIG_DEBUG_FS
228 /* debugging support via debugfs */
229 struct dentry
*debug_dir
;
230 #endif /* CONFIG_DEBUG_FS */
231 struct rtnl_link_stats64 bond_stats
;
234 #define bond_slave_get_rcu(dev) \
235 ((struct slave *) rcu_dereference(dev->rx_handler_data))
237 #define bond_slave_get_rtnl(dev) \
238 ((struct slave *) rtnl_dereference(dev->rx_handler_data))
240 void bond_queue_slave_event(struct slave
*slave
);
241 void bond_lower_state_changed(struct slave
*slave
);
243 struct bond_vlan_tag
{
245 unsigned short vlan_id
;
249 * Returns NULL if the net_device does not belong to any of the bond's slaves
251 * Caller must hold bond lock for read
253 static inline struct slave
*bond_get_slave_by_dev(struct bonding
*bond
,
254 struct net_device
*slave_dev
)
256 return netdev_lower_dev_get_private(bond
->dev
, slave_dev
);
259 static inline struct bonding
*bond_get_bond_by_slave(struct slave
*slave
)
264 static inline bool bond_should_override_tx_queue(struct bonding
*bond
)
266 return BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
||
267 BOND_MODE(bond
) == BOND_MODE_ROUNDROBIN
;
270 static inline bool bond_is_lb(const struct bonding
*bond
)
272 return BOND_MODE(bond
) == BOND_MODE_TLB
||
273 BOND_MODE(bond
) == BOND_MODE_ALB
;
276 static inline bool bond_needs_speed_duplex(const struct bonding
*bond
)
278 return BOND_MODE(bond
) == BOND_MODE_8023AD
|| bond_is_lb(bond
);
281 static inline bool bond_is_nondyn_tlb(const struct bonding
*bond
)
283 return (BOND_MODE(bond
) == BOND_MODE_TLB
) &&
284 (bond
->params
.tlb_dynamic_lb
== 0);
287 static inline bool bond_mode_uses_xmit_hash(const struct bonding
*bond
)
289 return (BOND_MODE(bond
) == BOND_MODE_8023AD
||
290 BOND_MODE(bond
) == BOND_MODE_XOR
||
291 bond_is_nondyn_tlb(bond
));
294 static inline bool bond_mode_uses_arp(int mode
)
296 return mode
!= BOND_MODE_8023AD
&& mode
!= BOND_MODE_TLB
&&
297 mode
!= BOND_MODE_ALB
;
300 static inline bool bond_mode_uses_primary(int mode
)
302 return mode
== BOND_MODE_ACTIVEBACKUP
|| mode
== BOND_MODE_TLB
||
303 mode
== BOND_MODE_ALB
;
306 static inline bool bond_uses_primary(struct bonding
*bond
)
308 return bond_mode_uses_primary(BOND_MODE(bond
));
311 static inline struct net_device
*bond_option_active_slave_get_rcu(struct bonding
*bond
)
313 struct slave
*slave
= rcu_dereference(bond
->curr_active_slave
);
315 return bond_uses_primary(bond
) && slave
? slave
->dev
: NULL
;
318 static inline bool bond_slave_is_up(struct slave
*slave
)
320 return netif_running(slave
->dev
) && netif_carrier_ok(slave
->dev
);
323 static inline void bond_set_active_slave(struct slave
*slave
)
327 bond_queue_slave_event(slave
);
328 bond_lower_state_changed(slave
);
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
);
341 static inline void bond_set_slave_state(struct slave
*slave
,
342 int slave_state
, bool notify
)
344 if (slave
->backup
== slave_state
)
347 slave
->backup
= slave_state
;
349 bond_lower_state_changed(slave
);
350 bond_queue_slave_event(slave
);
351 slave
->should_notify
= 0;
353 if (slave
->should_notify
)
354 slave
->should_notify
= 0;
356 slave
->should_notify
= 1;
360 static inline void bond_slave_state_change(struct bonding
*bond
)
362 struct list_head
*iter
;
365 bond_for_each_slave(bond
, tmp
, iter
) {
366 if (tmp
->link
== BOND_LINK_UP
)
367 bond_set_active_slave(tmp
);
368 else if (tmp
->link
== BOND_LINK_DOWN
)
369 bond_set_backup_slave(tmp
);
373 static inline void bond_slave_state_notify(struct bonding
*bond
)
375 struct list_head
*iter
;
378 bond_for_each_slave(bond
, tmp
, iter
) {
379 if (tmp
->should_notify
) {
380 bond_lower_state_changed(tmp
);
381 tmp
->should_notify
= 0;
386 static inline int bond_slave_state(struct slave
*slave
)
388 return slave
->backup
;
391 static inline bool bond_is_active_slave(struct slave
*slave
)
393 return !bond_slave_state(slave
);
396 static inline bool bond_slave_can_tx(struct slave
*slave
)
398 return bond_slave_is_up(slave
) && slave
->link
== BOND_LINK_UP
&&
399 bond_is_active_slave(slave
);
402 static inline void bond_hw_addr_copy(u8
*dst
, const u8
*src
, unsigned int len
)
404 if (len
== ETH_ALEN
) {
405 ether_addr_copy(dst
, src
);
409 memcpy(dst
, src
, len
);
412 #define BOND_PRI_RESELECT_ALWAYS 0
413 #define BOND_PRI_RESELECT_BETTER 1
414 #define BOND_PRI_RESELECT_FAILURE 2
416 #define BOND_FOM_NONE 0
417 #define BOND_FOM_ACTIVE 1
418 #define BOND_FOM_FOLLOW 2
420 #define BOND_ARP_TARGETS_ANY 0
421 #define BOND_ARP_TARGETS_ALL 1
423 #define BOND_ARP_VALIDATE_NONE 0
424 #define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE)
425 #define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP)
426 #define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \
427 BOND_ARP_VALIDATE_BACKUP)
428 #define BOND_ARP_FILTER (BOND_ARP_VALIDATE_ALL + 1)
429 #define BOND_ARP_FILTER_ACTIVE (BOND_ARP_VALIDATE_ACTIVE | \
431 #define BOND_ARP_FILTER_BACKUP (BOND_ARP_VALIDATE_BACKUP | \
434 #define BOND_SLAVE_NOTIFY_NOW true
435 #define BOND_SLAVE_NOTIFY_LATER false
437 static inline int slave_do_arp_validate(struct bonding
*bond
,
440 return bond
->params
.arp_validate
& (1 << bond_slave_state(slave
));
443 static inline int slave_do_arp_validate_only(struct bonding
*bond
)
445 return bond
->params
.arp_validate
& BOND_ARP_FILTER
;
448 static inline int bond_is_ip_target_ok(__be32 addr
)
450 return !ipv4_is_lbcast(addr
) && !ipv4_is_zeronet(addr
);
453 /* Get the oldest arp which we've received on this slave for bond's
456 static inline unsigned long slave_oldest_target_arp_rx(struct bonding
*bond
,
460 unsigned long ret
= slave
->target_last_arp_rx
[0];
462 for (; (i
< BOND_MAX_ARP_TARGETS
) && bond
->params
.arp_targets
[i
]; i
++)
463 if (time_before(slave
->target_last_arp_rx
[i
], ret
))
464 ret
= slave
->target_last_arp_rx
[i
];
469 static inline unsigned long slave_last_rx(struct bonding
*bond
,
472 if (bond
->params
.arp_all_targets
== BOND_ARP_TARGETS_ALL
)
473 return slave_oldest_target_arp_rx(bond
, slave
);
475 return slave
->last_rx
;
478 #ifdef CONFIG_NET_POLL_CONTROLLER
479 static inline void bond_netpoll_send_skb(const struct slave
*slave
,
482 struct netpoll
*np
= slave
->np
;
485 netpoll_send_skb(np
, skb
);
488 static inline void bond_netpoll_send_skb(const struct slave
*slave
,
494 static inline void bond_set_slave_inactive_flags(struct slave
*slave
,
497 if (!bond_is_lb(slave
->bond
))
498 bond_set_slave_state(slave
, BOND_STATE_BACKUP
, notify
);
499 if (!slave
->bond
->params
.all_slaves_active
)
503 static inline void bond_set_slave_active_flags(struct slave
*slave
,
506 bond_set_slave_state(slave
, BOND_STATE_ACTIVE
, notify
);
510 static inline bool bond_is_slave_inactive(struct slave
*slave
)
512 return slave
->inactive
;
515 static inline void bond_propose_link_state(struct slave
*slave
, int state
)
517 slave
->link_new_state
= state
;
520 static inline void bond_commit_link_state(struct slave
*slave
, bool notify
)
522 if (slave
->link
== slave
->link_new_state
)
525 slave
->link
= slave
->link_new_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_set_slave_link_state(struct slave
*slave
, int state
,
541 bond_propose_link_state(slave
, state
);
542 bond_commit_link_state(slave
, notify
);
545 static inline void bond_slave_link_notify(struct bonding
*bond
)
547 struct list_head
*iter
;
550 bond_for_each_slave(bond
, tmp
, iter
) {
551 if (tmp
->should_notify_link
) {
552 bond_queue_slave_event(tmp
);
553 bond_lower_state_changed(tmp
);
554 tmp
->should_notify_link
= 0;
559 static inline __be32
bond_confirm_addr(struct net_device
*dev
, __be32 dst
, __be32 local
)
561 struct in_device
*in_dev
;
565 in_dev
= __in_dev_get_rcu(dev
);
568 addr
= inet_confirm_addr(dev_net(dev
), in_dev
, dst
, local
,
575 struct net
*net
; /* Associated network namespace */
576 struct list_head dev_list
;
577 #ifdef CONFIG_PROC_FS
578 struct proc_dir_entry
*proc_dir
;
580 struct class_attribute class_attr_bonding_masters
;
583 int bond_arp_rcv(const struct sk_buff
*skb
, struct bonding
*bond
, struct slave
*slave
);
584 void bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
, struct net_device
*slave_dev
);
585 int bond_create(struct net
*net
, const char *name
);
586 int bond_create_sysfs(struct bond_net
*net
);
587 void bond_destroy_sysfs(struct bond_net
*net
);
588 void bond_prepare_sysfs_group(struct bonding
*bond
);
589 int bond_sysfs_slave_add(struct slave
*slave
);
590 void bond_sysfs_slave_del(struct slave
*slave
);
591 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
,
592 struct netlink_ext_ack
*extack
);
593 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
);
594 u32
bond_xmit_hash(struct bonding
*bond
, struct sk_buff
*skb
);
595 int bond_set_carrier(struct bonding
*bond
);
596 void bond_select_active_slave(struct bonding
*bond
);
597 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
);
598 void bond_create_debugfs(void);
599 void bond_destroy_debugfs(void);
600 void bond_debug_register(struct bonding
*bond
);
601 void bond_debug_unregister(struct bonding
*bond
);
602 void bond_debug_reregister(struct bonding
*bond
);
603 const char *bond_mode_name(int mode
);
604 void bond_setup(struct net_device
*bond_dev
);
605 unsigned int bond_get_num_tx_queues(void);
606 int bond_netlink_init(void);
607 void bond_netlink_fini(void);
608 struct net_device
*bond_option_active_slave_get_rcu(struct bonding
*bond
);
609 const char *bond_slave_link_status(s8 link
);
610 struct bond_vlan_tag
*bond_verify_device_path(struct net_device
*start_dev
,
611 struct net_device
*end_dev
,
613 int bond_update_slave_arr(struct bonding
*bond
, struct slave
*skipslave
);
614 void bond_slave_arr_work_rearm(struct bonding
*bond
, unsigned long delay
);
615 void bond_work_init_all(struct bonding
*bond
);
617 #ifdef CONFIG_PROC_FS
618 void bond_create_proc_entry(struct bonding
*bond
);
619 void bond_remove_proc_entry(struct bonding
*bond
);
620 void bond_create_proc_dir(struct bond_net
*bn
);
621 void bond_destroy_proc_dir(struct bond_net
*bn
);
623 static inline void bond_create_proc_entry(struct bonding
*bond
)
627 static inline void bond_remove_proc_entry(struct bonding
*bond
)
631 static inline void bond_create_proc_dir(struct bond_net
*bn
)
635 static inline void bond_destroy_proc_dir(struct bond_net
*bn
)
640 static inline struct slave
*bond_slave_has_mac(struct bonding
*bond
,
643 struct list_head
*iter
;
646 bond_for_each_slave(bond
, tmp
, iter
)
647 if (ether_addr_equal_64bits(mac
, tmp
->dev
->dev_addr
))
653 /* Caller must hold rcu_read_lock() for read */
654 static inline struct slave
*bond_slave_has_mac_rcu(struct bonding
*bond
,
657 struct list_head
*iter
;
660 bond_for_each_slave_rcu(bond
, tmp
, iter
)
661 if (ether_addr_equal_64bits(mac
, tmp
->dev
->dev_addr
))
667 /* Caller must hold rcu_read_lock() for read */
668 static inline bool bond_slave_has_mac_rx(struct bonding
*bond
, const u8
*mac
)
670 struct list_head
*iter
;
672 struct netdev_hw_addr
*ha
;
674 bond_for_each_slave_rcu(bond
, tmp
, iter
)
675 if (ether_addr_equal_64bits(mac
, tmp
->dev
->dev_addr
))
678 if (netdev_uc_empty(bond
->dev
))
681 netdev_for_each_uc_addr(ha
, bond
->dev
)
682 if (ether_addr_equal_64bits(mac
, ha
->addr
))
688 /* Check if the ip is present in arp ip list, or first free slot if ip == 0
689 * Returns -1 if not found, index if found
691 static inline int bond_get_targets_ip(__be32
*targets
, __be32 ip
)
695 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
; i
++)
696 if (targets
[i
] == ip
)
698 else if (targets
[i
] == 0)
704 /* exported from bond_main.c */
705 extern unsigned int bond_net_id
;
706 extern const struct bond_parm_tbl bond_lacp_tbl
[];
707 extern const struct bond_parm_tbl xmit_hashtype_tbl
[];
708 extern const struct bond_parm_tbl arp_validate_tbl
[];
709 extern const struct bond_parm_tbl arp_all_targets_tbl
[];
710 extern const struct bond_parm_tbl fail_over_mac_tbl
[];
711 extern const struct bond_parm_tbl pri_reselect_tbl
[];
712 extern struct bond_parm_tbl ad_select_tbl
[];
714 /* exported from bond_netlink.c */
715 extern struct rtnl_link_ops bond_link_ops
;
717 static inline void bond_tx_drop(struct net_device
*dev
, struct sk_buff
*skb
)
719 atomic_long_inc(&dev
->tx_dropped
);
720 dev_kfree_skb_any(skb
);
723 #endif /* _NET_BONDING_H */