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 */
152 u8 backup
:1, /* indicates backup slave. Value corresponds with
153 BOND_STATE_ACTIVE and BOND_STATE_BACKUP */
154 inactive
:1, /* indicates inactive slave */
155 should_notify
:1, /* indicates whether the state changed */
156 should_notify_link
:1; /* indicates whether the link changed */
159 u32 link_failure_count
;
162 u8 perm_hwaddr
[MAX_ADDR_LEN
];
163 struct ad_slave_info
*ad_info
;
164 struct tlb_slave_info tlb_info
;
165 #ifdef CONFIG_NET_POLL_CONTROLLER
168 struct delayed_work notify_work
;
170 struct rtnl_link_stats64 slave_stats
;
173 struct bond_up_slave
{
176 struct slave
*arr
[0];
180 * Link pseudo-state only used internally by monitors
182 #define BOND_LINK_NOCHANGE -1
185 * Here are the locking policies for the two bonding locks:
186 * Get rcu_read_lock when reading or RTNL when writing slave list.
189 struct net_device
*dev
; /* first - useful for panic debug */
190 struct slave __rcu
*curr_active_slave
;
191 struct slave __rcu
*current_arp_slave
;
192 struct slave __rcu
*primary_slave
;
193 struct bond_up_slave __rcu
*slave_arr
; /* Array of usable slaves */
196 s32 slave_cnt
; /* never change this value outside the attach/detach wrappers */
197 int (*recv_probe
)(const struct sk_buff
*, struct bonding
*,
199 /* mode_lock is used for mode-specific locking needs, currently used by:
200 * 3ad mode (4) - protect against running bond_3ad_unbind_slave() and
201 * bond_3ad_state_machine_handler() concurrently and also
202 * the access to the state machine shared variables.
203 * TLB mode (5) - to sync the use and modifications of its hash table
204 * ALB mode (6) - to sync the use and modifications of its hash table
206 spinlock_t mode_lock
;
207 spinlock_t stats_lock
;
210 #ifdef CONFIG_PROC_FS
211 struct proc_dir_entry
*proc_entry
;
212 char proc_file_name
[IFNAMSIZ
];
213 #endif /* CONFIG_PROC_FS */
214 struct list_head bond_list
;
216 struct ad_bond_info ad_info
;
217 struct alb_bond_info alb_info
;
218 struct bond_params params
;
219 struct workqueue_struct
*wq
;
220 struct delayed_work mii_work
;
221 struct delayed_work arp_work
;
222 struct delayed_work alb_work
;
223 struct delayed_work ad_work
;
224 struct delayed_work mcast_work
;
225 struct delayed_work slave_arr_work
;
226 #ifdef CONFIG_DEBUG_FS
227 /* debugging support via debugfs */
228 struct dentry
*debug_dir
;
229 #endif /* CONFIG_DEBUG_FS */
230 struct rtnl_link_stats64 bond_stats
;
233 #define bond_slave_get_rcu(dev) \
234 ((struct slave *) rcu_dereference(dev->rx_handler_data))
236 #define bond_slave_get_rtnl(dev) \
237 ((struct slave *) rtnl_dereference(dev->rx_handler_data))
239 void bond_queue_slave_event(struct slave
*slave
);
240 void bond_lower_state_changed(struct slave
*slave
);
242 struct bond_vlan_tag
{
244 unsigned short vlan_id
;
248 * Returns NULL if the net_device does not belong to any of the bond's slaves
250 * Caller must hold bond lock for read
252 static inline struct slave
*bond_get_slave_by_dev(struct bonding
*bond
,
253 struct net_device
*slave_dev
)
255 return netdev_lower_dev_get_private(bond
->dev
, slave_dev
);
258 static inline struct bonding
*bond_get_bond_by_slave(struct slave
*slave
)
263 static inline bool bond_should_override_tx_queue(struct bonding
*bond
)
265 return BOND_MODE(bond
) == BOND_MODE_ACTIVEBACKUP
||
266 BOND_MODE(bond
) == BOND_MODE_ROUNDROBIN
;
269 static inline bool bond_is_lb(const struct bonding
*bond
)
271 return BOND_MODE(bond
) == BOND_MODE_TLB
||
272 BOND_MODE(bond
) == BOND_MODE_ALB
;
275 static inline bool bond_needs_speed_duplex(const struct bonding
*bond
)
277 return BOND_MODE(bond
) == BOND_MODE_8023AD
|| bond_is_lb(bond
);
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
);
331 static inline void bond_set_backup_slave(struct slave
*slave
)
333 if (!slave
->backup
) {
335 bond_queue_slave_event(slave
);
336 bond_lower_state_changed(slave
);
340 static inline void bond_set_slave_state(struct slave
*slave
,
341 int slave_state
, bool notify
)
343 if (slave
->backup
== slave_state
)
346 slave
->backup
= slave_state
;
348 bond_lower_state_changed(slave
);
349 bond_queue_slave_event(slave
);
350 slave
->should_notify
= 0;
352 if (slave
->should_notify
)
353 slave
->should_notify
= 0;
355 slave
->should_notify
= 1;
359 static inline void bond_slave_state_change(struct bonding
*bond
)
361 struct list_head
*iter
;
364 bond_for_each_slave(bond
, tmp
, iter
) {
365 if (tmp
->link
== BOND_LINK_UP
)
366 bond_set_active_slave(tmp
);
367 else if (tmp
->link
== BOND_LINK_DOWN
)
368 bond_set_backup_slave(tmp
);
372 static inline void bond_slave_state_notify(struct bonding
*bond
)
374 struct list_head
*iter
;
377 bond_for_each_slave(bond
, tmp
, iter
) {
378 if (tmp
->should_notify
) {
379 bond_lower_state_changed(tmp
);
380 tmp
->should_notify
= 0;
385 static inline int bond_slave_state(struct slave
*slave
)
387 return slave
->backup
;
390 static inline bool bond_is_active_slave(struct slave
*slave
)
392 return !bond_slave_state(slave
);
395 static inline bool bond_slave_can_tx(struct slave
*slave
)
397 return bond_slave_is_up(slave
) && slave
->link
== BOND_LINK_UP
&&
398 bond_is_active_slave(slave
);
401 static inline void bond_hw_addr_copy(u8
*dst
, const u8
*src
, unsigned int len
)
403 if (len
== ETH_ALEN
) {
404 ether_addr_copy(dst
, src
);
408 memcpy(dst
, src
, len
);
411 #define BOND_PRI_RESELECT_ALWAYS 0
412 #define BOND_PRI_RESELECT_BETTER 1
413 #define BOND_PRI_RESELECT_FAILURE 2
415 #define BOND_FOM_NONE 0
416 #define BOND_FOM_ACTIVE 1
417 #define BOND_FOM_FOLLOW 2
419 #define BOND_ARP_TARGETS_ANY 0
420 #define BOND_ARP_TARGETS_ALL 1
422 #define BOND_ARP_VALIDATE_NONE 0
423 #define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE)
424 #define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP)
425 #define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \
426 BOND_ARP_VALIDATE_BACKUP)
427 #define BOND_ARP_FILTER (BOND_ARP_VALIDATE_ALL + 1)
428 #define BOND_ARP_FILTER_ACTIVE (BOND_ARP_VALIDATE_ACTIVE | \
430 #define BOND_ARP_FILTER_BACKUP (BOND_ARP_VALIDATE_BACKUP | \
433 #define BOND_SLAVE_NOTIFY_NOW true
434 #define BOND_SLAVE_NOTIFY_LATER false
436 static inline int slave_do_arp_validate(struct bonding
*bond
,
439 return bond
->params
.arp_validate
& (1 << bond_slave_state(slave
));
442 static inline int slave_do_arp_validate_only(struct bonding
*bond
)
444 return bond
->params
.arp_validate
& BOND_ARP_FILTER
;
447 static inline int bond_is_ip_target_ok(__be32 addr
)
449 return !ipv4_is_lbcast(addr
) && !ipv4_is_zeronet(addr
);
452 /* Get the oldest arp which we've received on this slave for bond's
455 static inline unsigned long slave_oldest_target_arp_rx(struct bonding
*bond
,
459 unsigned long ret
= slave
->target_last_arp_rx
[0];
461 for (; (i
< BOND_MAX_ARP_TARGETS
) && bond
->params
.arp_targets
[i
]; i
++)
462 if (time_before(slave
->target_last_arp_rx
[i
], ret
))
463 ret
= slave
->target_last_arp_rx
[i
];
468 static inline unsigned long slave_last_rx(struct bonding
*bond
,
471 if (bond
->params
.arp_all_targets
== BOND_ARP_TARGETS_ALL
)
472 return slave_oldest_target_arp_rx(bond
, slave
);
474 return slave
->last_rx
;
477 #ifdef CONFIG_NET_POLL_CONTROLLER
478 static inline void bond_netpoll_send_skb(const struct slave
*slave
,
481 struct netpoll
*np
= slave
->np
;
484 netpoll_send_skb(np
, skb
);
487 static inline void bond_netpoll_send_skb(const struct slave
*slave
,
493 static inline void bond_set_slave_inactive_flags(struct slave
*slave
,
496 if (!bond_is_lb(slave
->bond
))
497 bond_set_slave_state(slave
, BOND_STATE_BACKUP
, notify
);
498 if (!slave
->bond
->params
.all_slaves_active
)
502 static inline void bond_set_slave_active_flags(struct slave
*slave
,
505 bond_set_slave_state(slave
, BOND_STATE_ACTIVE
, notify
);
509 static inline bool bond_is_slave_inactive(struct slave
*slave
)
511 return slave
->inactive
;
514 static inline void bond_propose_link_state(struct slave
*slave
, int state
)
516 slave
->link_new_state
= state
;
519 static inline void bond_commit_link_state(struct slave
*slave
, bool notify
)
521 if (slave
->link_new_state
== BOND_LINK_NOCHANGE
)
524 slave
->link
= slave
->link_new_state
;
526 bond_queue_slave_event(slave
);
527 bond_lower_state_changed(slave
);
528 slave
->should_notify_link
= 0;
530 if (slave
->should_notify_link
)
531 slave
->should_notify_link
= 0;
533 slave
->should_notify_link
= 1;
537 static inline void bond_set_slave_link_state(struct slave
*slave
, int state
,
540 bond_propose_link_state(slave
, state
);
541 bond_commit_link_state(slave
, notify
);
544 static inline void bond_slave_link_notify(struct bonding
*bond
)
546 struct list_head
*iter
;
549 bond_for_each_slave(bond
, tmp
, iter
) {
550 if (tmp
->should_notify_link
) {
551 bond_queue_slave_event(tmp
);
552 bond_lower_state_changed(tmp
);
553 tmp
->should_notify_link
= 0;
558 static inline __be32
bond_confirm_addr(struct net_device
*dev
, __be32 dst
, __be32 local
)
560 struct in_device
*in_dev
;
564 in_dev
= __in_dev_get_rcu(dev
);
567 addr
= inet_confirm_addr(dev_net(dev
), in_dev
, dst
, local
,
574 struct net
*net
; /* Associated network namespace */
575 struct list_head dev_list
;
576 #ifdef CONFIG_PROC_FS
577 struct proc_dir_entry
*proc_dir
;
579 struct class_attribute class_attr_bonding_masters
;
582 int bond_arp_rcv(const struct sk_buff
*skb
, struct bonding
*bond
, struct slave
*slave
);
583 void bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
, struct net_device
*slave_dev
);
584 int bond_create(struct net
*net
, const char *name
);
585 int bond_create_sysfs(struct bond_net
*net
);
586 void bond_destroy_sysfs(struct bond_net
*net
);
587 void bond_prepare_sysfs_group(struct bonding
*bond
);
588 int bond_sysfs_slave_add(struct slave
*slave
);
589 void bond_sysfs_slave_del(struct slave
*slave
);
590 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
,
591 struct netlink_ext_ack
*extack
);
592 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
);
593 u32
bond_xmit_hash(struct bonding
*bond
, struct sk_buff
*skb
);
594 int bond_set_carrier(struct bonding
*bond
);
595 void bond_select_active_slave(struct bonding
*bond
);
596 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
);
597 void bond_create_debugfs(void);
598 void bond_destroy_debugfs(void);
599 void bond_debug_register(struct bonding
*bond
);
600 void bond_debug_unregister(struct bonding
*bond
);
601 void bond_debug_reregister(struct bonding
*bond
);
602 const char *bond_mode_name(int mode
);
603 void bond_setup(struct net_device
*bond_dev
);
604 unsigned int bond_get_num_tx_queues(void);
605 int bond_netlink_init(void);
606 void bond_netlink_fini(void);
607 struct net_device
*bond_option_active_slave_get_rcu(struct bonding
*bond
);
608 const char *bond_slave_link_status(s8 link
);
609 struct bond_vlan_tag
*bond_verify_device_path(struct net_device
*start_dev
,
610 struct net_device
*end_dev
,
612 int bond_update_slave_arr(struct bonding
*bond
, struct slave
*skipslave
);
613 void bond_slave_arr_work_rearm(struct bonding
*bond
, unsigned long delay
);
614 void bond_work_init_all(struct bonding
*bond
);
616 #ifdef CONFIG_PROC_FS
617 void bond_create_proc_entry(struct bonding
*bond
);
618 void bond_remove_proc_entry(struct bonding
*bond
);
619 void bond_create_proc_dir(struct bond_net
*bn
);
620 void bond_destroy_proc_dir(struct bond_net
*bn
);
622 static inline void bond_create_proc_entry(struct bonding
*bond
)
626 static inline void bond_remove_proc_entry(struct bonding
*bond
)
630 static inline void bond_create_proc_dir(struct bond_net
*bn
)
634 static inline void bond_destroy_proc_dir(struct bond_net
*bn
)
639 static inline struct slave
*bond_slave_has_mac(struct bonding
*bond
,
642 struct list_head
*iter
;
645 bond_for_each_slave(bond
, tmp
, iter
)
646 if (ether_addr_equal_64bits(mac
, tmp
->dev
->dev_addr
))
652 /* Caller must hold rcu_read_lock() for read */
653 static inline struct slave
*bond_slave_has_mac_rcu(struct bonding
*bond
,
656 struct list_head
*iter
;
659 bond_for_each_slave_rcu(bond
, tmp
, iter
)
660 if (ether_addr_equal_64bits(mac
, tmp
->dev
->dev_addr
))
666 /* Caller must hold rcu_read_lock() for read */
667 static inline bool bond_slave_has_mac_rx(struct bonding
*bond
, const u8
*mac
)
669 struct list_head
*iter
;
671 struct netdev_hw_addr
*ha
;
673 bond_for_each_slave_rcu(bond
, tmp
, iter
)
674 if (ether_addr_equal_64bits(mac
, tmp
->dev
->dev_addr
))
677 if (netdev_uc_empty(bond
->dev
))
680 netdev_for_each_uc_addr(ha
, bond
->dev
)
681 if (ether_addr_equal_64bits(mac
, ha
->addr
))
687 /* Check if the ip is present in arp ip list, or first free slot if ip == 0
688 * Returns -1 if not found, index if found
690 static inline int bond_get_targets_ip(__be32
*targets
, __be32 ip
)
694 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
; i
++)
695 if (targets
[i
] == ip
)
697 else if (targets
[i
] == 0)
703 /* exported from bond_main.c */
704 extern unsigned int bond_net_id
;
705 extern const struct bond_parm_tbl bond_lacp_tbl
[];
706 extern const struct bond_parm_tbl xmit_hashtype_tbl
[];
707 extern const struct bond_parm_tbl arp_validate_tbl
[];
708 extern const struct bond_parm_tbl arp_all_targets_tbl
[];
709 extern const struct bond_parm_tbl fail_over_mac_tbl
[];
710 extern const struct bond_parm_tbl pri_reselect_tbl
[];
711 extern struct bond_parm_tbl ad_select_tbl
[];
713 /* exported from bond_netlink.c */
714 extern struct rtnl_link_ops bond_link_ops
;
716 static inline void bond_tx_drop(struct net_device
*dev
, struct sk_buff
*skb
)
718 atomic_long_inc(&dev
->tx_dropped
);
719 dev_kfree_skb_any(skb
);
722 #endif /* _NET_BONDING_H */