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 _LINUX_BONDING_H
16 #define _LINUX_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>
30 #include "bond_options.h"
32 #define DRV_VERSION "3.7.1"
33 #define DRV_RELDATE "April 27, 2011"
34 #define DRV_NAME "bonding"
35 #define DRV_DESCRIPTION "Ethernet Channel Bonding Driver"
37 #define bond_version DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n"
39 #define BOND_MAX_ARP_TARGETS 16
41 #define BOND_DEFAULT_MIIMON 100
44 ((((dev)->flags & IFF_UP) == IFF_UP) && \
45 netif_running(dev) && \
46 netif_carrier_ok(dev))
49 * Checks whether slave is ready for transmit.
51 #define SLAVE_IS_OK(slave) \
52 (((slave)->dev->flags & IFF_UP) && \
53 netif_running((slave)->dev) && \
54 ((slave)->link == BOND_LINK_UP) && \
55 bond_is_active_slave(slave))
58 #define USES_PRIMARY(mode) \
59 (((mode) == BOND_MODE_ACTIVEBACKUP) || \
60 ((mode) == BOND_MODE_TLB) || \
61 ((mode) == BOND_MODE_ALB))
63 #define BOND_NO_USES_ARP(mode) \
64 (((mode) == BOND_MODE_8023AD) || \
65 ((mode) == BOND_MODE_TLB) || \
66 ((mode) == BOND_MODE_ALB))
68 #define TX_QUEUE_OVERRIDE(mode) \
69 (((mode) == BOND_MODE_ACTIVEBACKUP) || \
70 ((mode) == BOND_MODE_ROUNDROBIN))
72 #define BOND_MODE_IS_LB(mode) \
73 (((mode) == BOND_MODE_TLB) || \
74 ((mode) == BOND_MODE_ALB))
76 #define IS_IP_TARGET_UNUSABLE_ADDRESS(a) \
77 ((htonl(INADDR_BROADCAST) == a) || \
80 * Less bad way to call ioctl from within the kernel; this needs to be
81 * done some other way to get the call out of interrupt context.
82 * Needs "ioctl" variable to be supplied by calling context.
84 #define IOCTL(dev, arg, cmd) ({ \
86 mm_segment_t fs = get_fs(); \
88 res = ioctl(dev, arg, cmd); \
92 /* slave list primitives */
93 #define bond_slave_list(bond) (&(bond)->dev->adj_list.lower)
95 #define bond_has_slaves(bond) !list_empty(bond_slave_list(bond))
97 /* IMPORTANT: bond_first/last_slave can return NULL in case of an empty list */
98 #define bond_first_slave(bond) \
99 (bond_has_slaves(bond) ? \
100 netdev_adjacent_get_private(bond_slave_list(bond)->next) : \
102 #define bond_last_slave(bond) \
103 (bond_has_slaves(bond) ? \
104 netdev_adjacent_get_private(bond_slave_list(bond)->prev) : \
107 /* Caller must have rcu_read_lock */
108 #define bond_first_slave_rcu(bond) \
109 netdev_lower_get_first_private_rcu(bond->dev)
111 #define bond_is_first_slave(bond, pos) (pos == bond_first_slave(bond))
112 #define bond_is_last_slave(bond, pos) (pos == bond_last_slave(bond))
115 * bond_for_each_slave - iterate over all slaves
116 * @bond: the bond holding this list
117 * @pos: current slave
118 * @iter: list_head * iterator
120 * Caller must hold bond->lock
122 #define bond_for_each_slave(bond, pos, iter) \
123 netdev_for_each_lower_private((bond)->dev, pos, iter)
125 /* Caller must have rcu_read_lock */
126 #define bond_for_each_slave_rcu(bond, pos, iter) \
127 netdev_for_each_lower_private_rcu((bond)->dev, pos, iter)
129 #ifdef CONFIG_NET_POLL_CONTROLLER
130 extern atomic_t netpoll_block_tx
;
132 static inline void block_netpoll_tx(void)
134 atomic_inc(&netpoll_block_tx
);
137 static inline void unblock_netpoll_tx(void)
139 atomic_dec(&netpoll_block_tx
);
142 static inline int is_netpoll_tx_blocked(struct net_device
*dev
)
144 if (unlikely(netpoll_tx_running(dev
)))
145 return atomic_read(&netpoll_block_tx
);
149 #define block_netpoll_tx()
150 #define unblock_netpoll_tx()
151 #define is_netpoll_tx_blocked(dev) (0)
167 unsigned int min_links
;
169 char primary
[IFNAMSIZ
];
170 int primary_reselect
;
171 __be32 arp_targets
[BOND_MAX_ARP_TARGETS
];
173 int all_slaves_active
;
176 int packets_per_slave
;
177 struct reciprocal_value reciprocal_packets_per_slave
;
180 struct bond_parm_tbl
{
185 #define BOND_MAX_MODENAME_LEN 20
188 struct net_device
*dev
; /* first - useful for panic debug */
189 struct bonding
*bond
; /* our master */
191 unsigned long jiffies
;
192 unsigned long last_arp_rx
;
193 unsigned long target_last_arp_rx
[BOND_MAX_ARP_TARGETS
];
194 s8 link
; /* one of BOND_LINK_XXXX */
196 u8 backup
:1, /* indicates backup slave. Value corresponds with
197 BOND_STATE_ACTIVE and BOND_STATE_BACKUP */
198 inactive
:1, /* indicates inactive slave */
199 should_notify
:1; /* indicateds whether the state changed */
202 u32 link_failure_count
;
205 u8 perm_hwaddr
[ETH_ALEN
];
206 struct ad_slave_info ad_info
; /* HUGE - better to dynamically alloc */
207 struct tlb_slave_info tlb_info
;
208 #ifdef CONFIG_NET_POLL_CONTROLLER
215 * Link pseudo-state only used internally by monitors
217 #define BOND_LINK_NOCHANGE -1
220 * Here are the locking policies for the two bonding locks:
222 * 1) Get bond->lock when reading/writing slave list.
223 * 2) Get bond->curr_slave_lock when reading/writing bond->curr_active_slave.
224 * (It is unnecessary when the write-lock is put with bond->lock.)
225 * 3) When we lock with bond->curr_slave_lock, we must lock with bond->lock
229 struct net_device
*dev
; /* first - useful for panic debug */
230 struct slave
*curr_active_slave
;
231 struct slave
*current_arp_slave
;
232 struct slave
*primary_slave
;
234 s32 slave_cnt
; /* never change this value outside the attach/detach wrappers */
235 int (*recv_probe
)(const struct sk_buff
*, struct bonding
*,
238 rwlock_t curr_slave_lock
;
241 #ifdef CONFIG_PROC_FS
242 struct proc_dir_entry
*proc_entry
;
243 char proc_file_name
[IFNAMSIZ
];
244 #endif /* CONFIG_PROC_FS */
245 struct list_head bond_list
;
247 struct ad_bond_info ad_info
;
248 struct alb_bond_info alb_info
;
249 struct bond_params params
;
250 struct workqueue_struct
*wq
;
251 struct delayed_work mii_work
;
252 struct delayed_work arp_work
;
253 struct delayed_work alb_work
;
254 struct delayed_work ad_work
;
255 struct delayed_work mcast_work
;
256 #ifdef CONFIG_DEBUG_FS
257 /* debugging support via debugfs */
258 struct dentry
*debug_dir
;
259 #endif /* CONFIG_DEBUG_FS */
262 #define bond_slave_get_rcu(dev) \
263 ((struct slave *) rcu_dereference(dev->rx_handler_data))
265 #define bond_slave_get_rtnl(dev) \
266 ((struct slave *) rtnl_dereference(dev->rx_handler_data))
269 * Returns NULL if the net_device does not belong to any of the bond's slaves
271 * Caller must hold bond lock for read
273 static inline struct slave
*bond_get_slave_by_dev(struct bonding
*bond
,
274 struct net_device
*slave_dev
)
276 return netdev_lower_dev_get_private(bond
->dev
, slave_dev
);
279 static inline struct bonding
*bond_get_bond_by_slave(struct slave
*slave
)
281 if (!slave
|| !slave
->bond
)
286 static inline bool bond_is_lb(const struct bonding
*bond
)
288 return BOND_MODE_IS_LB(bond
->params
.mode
);
291 static inline void bond_set_active_slave(struct slave
*slave
)
295 rtmsg_ifinfo(RTM_NEWLINK
, slave
->dev
, 0, GFP_KERNEL
);
299 static inline void bond_set_backup_slave(struct slave
*slave
)
301 if (!slave
->backup
) {
303 rtmsg_ifinfo(RTM_NEWLINK
, slave
->dev
, 0, GFP_KERNEL
);
307 static inline void bond_set_slave_state(struct slave
*slave
,
308 int slave_state
, bool notify
)
310 if (slave
->backup
== slave_state
)
313 slave
->backup
= slave_state
;
315 rtmsg_ifinfo(RTM_NEWLINK
, slave
->dev
, 0, GFP_KERNEL
);
316 slave
->should_notify
= 0;
318 if (slave
->should_notify
)
319 slave
->should_notify
= 0;
321 slave
->should_notify
= 1;
325 static inline void bond_slave_state_change(struct bonding
*bond
)
327 struct list_head
*iter
;
330 bond_for_each_slave(bond
, tmp
, iter
) {
331 if (tmp
->link
== BOND_LINK_UP
)
332 bond_set_active_slave(tmp
);
333 else if (tmp
->link
== BOND_LINK_DOWN
)
334 bond_set_backup_slave(tmp
);
338 static inline void bond_slave_state_notify(struct bonding
*bond
)
340 struct list_head
*iter
;
343 bond_for_each_slave(bond
, tmp
, iter
) {
344 if (tmp
->should_notify
) {
345 rtmsg_ifinfo(RTM_NEWLINK
, tmp
->dev
, 0, GFP_KERNEL
);
346 tmp
->should_notify
= 0;
351 static inline int bond_slave_state(struct slave
*slave
)
353 return slave
->backup
;
356 static inline bool bond_is_active_slave(struct slave
*slave
)
358 return !bond_slave_state(slave
);
361 #define BOND_PRI_RESELECT_ALWAYS 0
362 #define BOND_PRI_RESELECT_BETTER 1
363 #define BOND_PRI_RESELECT_FAILURE 2
365 #define BOND_FOM_NONE 0
366 #define BOND_FOM_ACTIVE 1
367 #define BOND_FOM_FOLLOW 2
369 #define BOND_ARP_TARGETS_ANY 0
370 #define BOND_ARP_TARGETS_ALL 1
372 #define BOND_ARP_VALIDATE_NONE 0
373 #define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE)
374 #define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP)
375 #define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \
376 BOND_ARP_VALIDATE_BACKUP)
378 #define BOND_SLAVE_NOTIFY_NOW true
379 #define BOND_SLAVE_NOTIFY_LATER false
381 static inline int slave_do_arp_validate(struct bonding
*bond
,
384 return bond
->params
.arp_validate
& (1 << bond_slave_state(slave
));
387 /* Get the oldest arp which we've received on this slave for bond's
390 static inline unsigned long slave_oldest_target_arp_rx(struct bonding
*bond
,
394 unsigned long ret
= slave
->target_last_arp_rx
[0];
396 for (; (i
< BOND_MAX_ARP_TARGETS
) && bond
->params
.arp_targets
[i
]; i
++)
397 if (time_before(slave
->target_last_arp_rx
[i
], ret
))
398 ret
= slave
->target_last_arp_rx
[i
];
403 static inline unsigned long slave_last_rx(struct bonding
*bond
,
406 if (slave_do_arp_validate(bond
, slave
)) {
407 if (bond
->params
.arp_all_targets
== BOND_ARP_TARGETS_ALL
)
408 return slave_oldest_target_arp_rx(bond
, slave
);
410 return slave
->last_arp_rx
;
413 return slave
->dev
->last_rx
;
416 #ifdef CONFIG_NET_POLL_CONTROLLER
417 static inline void bond_netpoll_send_skb(const struct slave
*slave
,
420 struct netpoll
*np
= slave
->np
;
423 netpoll_send_skb(np
, skb
);
426 static inline void bond_netpoll_send_skb(const struct slave
*slave
,
432 static inline void bond_set_slave_inactive_flags(struct slave
*slave
,
435 if (!bond_is_lb(slave
->bond
))
436 bond_set_slave_state(slave
, BOND_STATE_BACKUP
, notify
);
437 if (!slave
->bond
->params
.all_slaves_active
)
441 static inline void bond_set_slave_active_flags(struct slave
*slave
,
444 bond_set_slave_state(slave
, BOND_STATE_ACTIVE
, notify
);
448 static inline bool bond_is_slave_inactive(struct slave
*slave
)
450 return slave
->inactive
;
453 static inline __be32
bond_confirm_addr(struct net_device
*dev
, __be32 dst
, __be32 local
)
455 struct in_device
*in_dev
;
459 in_dev
= __in_dev_get_rcu(dev
);
462 addr
= inet_confirm_addr(dev_net(dev
), in_dev
, dst
, local
,
468 static inline bool slave_can_tx(struct slave
*slave
)
470 if (IS_UP(slave
->dev
) && slave
->link
== BOND_LINK_UP
&&
471 bond_is_active_slave(slave
))
479 int bond_arp_rcv(const struct sk_buff
*skb
, struct bonding
*bond
, struct slave
*slave
);
480 void bond_dev_queue_xmit(struct bonding
*bond
, struct sk_buff
*skb
, struct net_device
*slave_dev
);
481 int bond_create(struct net
*net
, const char *name
);
482 int bond_create_sysfs(struct bond_net
*net
);
483 void bond_destroy_sysfs(struct bond_net
*net
);
484 void bond_prepare_sysfs_group(struct bonding
*bond
);
485 int bond_sysfs_slave_add(struct slave
*slave
);
486 void bond_sysfs_slave_del(struct slave
*slave
);
487 int bond_enslave(struct net_device
*bond_dev
, struct net_device
*slave_dev
);
488 int bond_release(struct net_device
*bond_dev
, struct net_device
*slave_dev
);
489 int bond_xmit_hash(struct bonding
*bond
, struct sk_buff
*skb
, int count
);
490 int bond_parse_parm(const char *mode_arg
, const struct bond_parm_tbl
*tbl
);
491 int bond_parm_tbl_lookup(int mode
, const struct bond_parm_tbl
*tbl
);
492 void bond_select_active_slave(struct bonding
*bond
);
493 void bond_change_active_slave(struct bonding
*bond
, struct slave
*new_active
);
494 void bond_create_debugfs(void);
495 void bond_destroy_debugfs(void);
496 void bond_debug_register(struct bonding
*bond
);
497 void bond_debug_unregister(struct bonding
*bond
);
498 void bond_debug_reregister(struct bonding
*bond
);
499 const char *bond_mode_name(int mode
);
500 void bond_setup(struct net_device
*bond_dev
);
501 unsigned int bond_get_num_tx_queues(void);
502 int bond_netlink_init(void);
503 void bond_netlink_fini(void);
504 int bond_option_arp_ip_target_add(struct bonding
*bond
, __be32 target
);
505 int bond_option_arp_ip_target_rem(struct bonding
*bond
, __be32 target
);
506 struct net_device
*bond_option_active_slave_get_rcu(struct bonding
*bond
);
507 struct net_device
*bond_option_active_slave_get(struct bonding
*bond
);
508 const char *bond_slave_link_status(s8 link
);
511 struct net
* net
; /* Associated network namespace */
512 struct list_head dev_list
;
513 #ifdef CONFIG_PROC_FS
514 struct proc_dir_entry
* proc_dir
;
516 struct class_attribute class_attr_bonding_masters
;
519 #ifdef CONFIG_PROC_FS
520 void bond_create_proc_entry(struct bonding
*bond
);
521 void bond_remove_proc_entry(struct bonding
*bond
);
522 void bond_create_proc_dir(struct bond_net
*bn
);
523 void bond_destroy_proc_dir(struct bond_net
*bn
);
525 static inline void bond_create_proc_entry(struct bonding
*bond
)
529 static inline void bond_remove_proc_entry(struct bonding
*bond
)
533 static inline void bond_create_proc_dir(struct bond_net
*bn
)
537 static inline void bond_destroy_proc_dir(struct bond_net
*bn
)
542 static inline struct slave
*bond_slave_has_mac(struct bonding
*bond
,
545 struct list_head
*iter
;
548 bond_for_each_slave(bond
, tmp
, iter
)
549 if (ether_addr_equal_64bits(mac
, tmp
->dev
->dev_addr
))
555 /* Caller must hold rcu_read_lock() for read */
556 static inline struct slave
*bond_slave_has_mac_rcu(struct bonding
*bond
,
559 struct list_head
*iter
;
562 bond_for_each_slave_rcu(bond
, tmp
, iter
)
563 if (ether_addr_equal_64bits(mac
, tmp
->dev
->dev_addr
))
569 /* Check if the ip is present in arp ip list, or first free slot if ip == 0
570 * Returns -1 if not found, index if found
572 static inline int bond_get_targets_ip(__be32
*targets
, __be32 ip
)
576 for (i
= 0; i
< BOND_MAX_ARP_TARGETS
; i
++)
577 if (targets
[i
] == ip
)
579 else if (targets
[i
] == 0)
585 /* exported from bond_main.c */
586 extern int bond_net_id
;
587 extern const struct bond_parm_tbl bond_lacp_tbl
[];
588 extern const struct bond_parm_tbl xmit_hashtype_tbl
[];
589 extern const struct bond_parm_tbl arp_validate_tbl
[];
590 extern const struct bond_parm_tbl arp_all_targets_tbl
[];
591 extern const struct bond_parm_tbl fail_over_mac_tbl
[];
592 extern const struct bond_parm_tbl pri_reselect_tbl
[];
593 extern struct bond_parm_tbl ad_select_tbl
[];
595 /* exported from bond_netlink.c */
596 extern struct rtnl_link_ops bond_link_ops
;
598 #endif /* _LINUX_BONDING_H */