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 */
158 u8 backup:1, /* indicates backup slave. Value corresponds with
159 BOND_STATE_ACTIVE and BOND_STATE_BACKUP */
160 inactive:1, /* indicates inactive slave */
161 should_notify:1, /* indicates whether the state changed */
162 should_notify_link:1; /* indicates whether the link changed */
165 u32 link_failure_count;
168 u8 perm_hwaddr[ETH_ALEN];
169 struct ad_slave_info *ad_info;
170 struct tlb_slave_info tlb_info;
171 #ifdef CONFIG_NET_POLL_CONTROLLER
175 struct rtnl_link_stats64 slave_stats;
178 struct bond_up_slave {
181 struct slave *arr[0];
185 * Link pseudo-state only used internally by monitors
187 #define BOND_LINK_NOCHANGE -1
190 * Here are the locking policies for the two bonding locks:
191 * Get rcu_read_lock when reading or RTNL when writing slave list.
194 struct net_device *dev; /* first - useful for panic debug */
195 struct slave __rcu *curr_active_slave;
196 struct slave __rcu *current_arp_slave;
197 struct slave __rcu *primary_slave;
198 struct bond_up_slave __rcu *slave_arr; /* Array of usable slaves */
200 s32 slave_cnt; /* never change this value outside the attach/detach wrappers */
201 int (*recv_probe)(const struct sk_buff *, struct bonding *,
203 /* mode_lock is used for mode-specific locking needs, currently used by:
204 * 3ad mode (4) - protect against running bond_3ad_unbind_slave() and
205 * bond_3ad_state_machine_handler() concurrently and also
206 * the access to the state machine shared variables.
207 * TLB mode (5) - to sync the use and modifications of its hash table
208 * ALB mode (6) - to sync the use and modifications of its hash table
210 spinlock_t mode_lock;
211 spinlock_t stats_lock;
214 #ifdef CONFIG_PROC_FS
215 struct proc_dir_entry *proc_entry;
216 char proc_file_name[IFNAMSIZ];
217 #endif /* CONFIG_PROC_FS */
218 struct list_head bond_list;
220 struct ad_bond_info ad_info;
221 struct alb_bond_info alb_info;
222 struct bond_params params;
223 struct workqueue_struct *wq;
224 struct delayed_work mii_work;
225 struct delayed_work arp_work;
226 struct delayed_work alb_work;
227 struct delayed_work ad_work;
228 struct delayed_work mcast_work;
229 struct delayed_work slave_arr_work;
230 #ifdef CONFIG_DEBUG_FS
231 /* debugging support via debugfs */
232 struct dentry *debug_dir;
233 #endif /* CONFIG_DEBUG_FS */
234 struct rtnl_link_stats64 bond_stats;
237 #define bond_slave_get_rcu(dev) \
238 ((struct slave *) rcu_dereference(dev->rx_handler_data))
240 #define bond_slave_get_rtnl(dev) \
241 ((struct slave *) rtnl_dereference(dev->rx_handler_data))
243 void bond_queue_slave_event(struct slave *slave);
244 void bond_lower_state_changed(struct slave *slave);
246 struct bond_vlan_tag {
248 unsigned short vlan_id;
252 * Returns NULL if the net_device does not belong to any of the bond's slaves
254 * Caller must hold bond lock for read
256 static inline struct slave *bond_get_slave_by_dev(struct bonding *bond,
257 struct net_device *slave_dev)
259 return netdev_lower_dev_get_private(bond->dev, slave_dev);
262 static inline struct bonding *bond_get_bond_by_slave(struct slave *slave)
267 static inline bool bond_should_override_tx_queue(struct bonding *bond)
269 return BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
270 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN;
273 static inline bool bond_is_lb(const struct bonding *bond)
275 return BOND_MODE(bond) == BOND_MODE_TLB ||
276 BOND_MODE(bond) == BOND_MODE_ALB;
279 static inline bool bond_is_nondyn_tlb(const struct bonding *bond)
281 return (BOND_MODE(bond) == BOND_MODE_TLB) &&
282 (bond->params.tlb_dynamic_lb == 0);
285 static inline bool bond_mode_uses_xmit_hash(const struct bonding *bond)
287 return (BOND_MODE(bond) == BOND_MODE_8023AD ||
288 BOND_MODE(bond) == BOND_MODE_XOR ||
289 bond_is_nondyn_tlb(bond));
292 static inline bool bond_mode_uses_arp(int mode)
294 return mode != BOND_MODE_8023AD && mode != BOND_MODE_TLB &&
295 mode != BOND_MODE_ALB;
298 static inline bool bond_mode_uses_primary(int mode)
300 return mode == BOND_MODE_ACTIVEBACKUP || mode == BOND_MODE_TLB ||
301 mode == BOND_MODE_ALB;
304 static inline bool bond_uses_primary(struct bonding *bond)
306 return bond_mode_uses_primary(BOND_MODE(bond));
309 static inline struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond)
311 struct slave *slave = rcu_dereference(bond->curr_active_slave);
313 return bond_uses_primary(bond) && slave ? slave->dev : NULL;
316 static inline bool bond_slave_is_up(struct slave *slave)
318 return netif_running(slave->dev) && netif_carrier_ok(slave->dev);
321 static inline void bond_set_active_slave(struct slave *slave)
325 bond_queue_slave_event(slave);
326 bond_lower_state_changed(slave);
327 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
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);
337 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
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 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
351 bond_queue_slave_event(slave);
352 slave->should_notify = 0;
354 if (slave->should_notify)
355 slave->should_notify = 0;
357 slave->should_notify = 1;
361 static inline void bond_slave_state_change(struct bonding *bond)
363 struct list_head *iter;
366 bond_for_each_slave(bond, tmp, iter) {
367 if (tmp->link == BOND_LINK_UP)
368 bond_set_active_slave(tmp);
369 else if (tmp->link == BOND_LINK_DOWN)
370 bond_set_backup_slave(tmp);
374 static inline void bond_slave_state_notify(struct bonding *bond)
376 struct list_head *iter;
379 bond_for_each_slave(bond, tmp, iter) {
380 if (tmp->should_notify) {
381 bond_lower_state_changed(tmp);
382 rtmsg_ifinfo(RTM_NEWLINK, tmp->dev, 0, GFP_ATOMIC);
383 tmp->should_notify = 0;
388 static inline int bond_slave_state(struct slave *slave)
390 return slave->backup;
393 static inline bool bond_is_active_slave(struct slave *slave)
395 return !bond_slave_state(slave);
398 static inline bool bond_slave_can_tx(struct slave *slave)
400 return bond_slave_is_up(slave) && slave->link == BOND_LINK_UP &&
401 bond_is_active_slave(slave);
404 #define BOND_PRI_RESELECT_ALWAYS 0
405 #define BOND_PRI_RESELECT_BETTER 1
406 #define BOND_PRI_RESELECT_FAILURE 2
408 #define BOND_FOM_NONE 0
409 #define BOND_FOM_ACTIVE 1
410 #define BOND_FOM_FOLLOW 2
412 #define BOND_ARP_TARGETS_ANY 0
413 #define BOND_ARP_TARGETS_ALL 1
415 #define BOND_ARP_VALIDATE_NONE 0
416 #define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE)
417 #define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP)
418 #define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \
419 BOND_ARP_VALIDATE_BACKUP)
420 #define BOND_ARP_FILTER (BOND_ARP_VALIDATE_ALL + 1)
421 #define BOND_ARP_FILTER_ACTIVE (BOND_ARP_VALIDATE_ACTIVE | \
423 #define BOND_ARP_FILTER_BACKUP (BOND_ARP_VALIDATE_BACKUP | \
426 #define BOND_SLAVE_NOTIFY_NOW true
427 #define BOND_SLAVE_NOTIFY_LATER false
429 static inline int slave_do_arp_validate(struct bonding *bond,
432 return bond->params.arp_validate & (1 << bond_slave_state(slave));
435 static inline int slave_do_arp_validate_only(struct bonding *bond)
437 return bond->params.arp_validate & BOND_ARP_FILTER;
440 static inline int bond_is_ip_target_ok(__be32 addr)
442 return !ipv4_is_lbcast(addr) && !ipv4_is_zeronet(addr);
445 /* Get the oldest arp which we've received on this slave for bond's
448 static inline unsigned long slave_oldest_target_arp_rx(struct bonding *bond,
452 unsigned long ret = slave->target_last_arp_rx[0];
454 for (; (i < BOND_MAX_ARP_TARGETS) && bond->params.arp_targets[i]; i++)
455 if (time_before(slave->target_last_arp_rx[i], ret))
456 ret = slave->target_last_arp_rx[i];
461 static inline unsigned long slave_last_rx(struct bonding *bond,
464 if (bond->params.arp_all_targets == BOND_ARP_TARGETS_ALL)
465 return slave_oldest_target_arp_rx(bond, slave);
467 return slave->last_rx;
470 #ifdef CONFIG_NET_POLL_CONTROLLER
471 static inline void bond_netpoll_send_skb(const struct slave *slave,
474 struct netpoll *np = slave->np;
477 netpoll_send_skb(np, skb);
480 static inline void bond_netpoll_send_skb(const struct slave *slave,
486 static inline void bond_set_slave_inactive_flags(struct slave *slave,
489 if (!bond_is_lb(slave->bond))
490 bond_set_slave_state(slave, BOND_STATE_BACKUP, notify);
491 if (!slave->bond->params.all_slaves_active)
495 static inline void bond_set_slave_active_flags(struct slave *slave,
498 bond_set_slave_state(slave, BOND_STATE_ACTIVE, notify);
502 static inline bool bond_is_slave_inactive(struct slave *slave)
504 return slave->inactive;
507 static inline void bond_set_slave_link_state(struct slave *slave, int state,
510 if (slave->link == state)
515 bond_queue_slave_event(slave);
516 bond_lower_state_changed(slave);
517 slave->should_notify_link = 0;
519 if (slave->should_notify_link)
520 slave->should_notify_link = 0;
522 slave->should_notify_link = 1;
526 static inline void bond_slave_link_notify(struct bonding *bond)
528 struct list_head *iter;
531 bond_for_each_slave(bond, tmp, iter) {
532 if (tmp->should_notify_link) {
533 bond_queue_slave_event(tmp);
534 bond_lower_state_changed(tmp);
535 tmp->should_notify_link = 0;
540 static inline __be32 bond_confirm_addr(struct net_device *dev, __be32 dst, __be32 local)
542 struct in_device *in_dev;
546 in_dev = __in_dev_get_rcu(dev);
549 addr = inet_confirm_addr(dev_net(dev), in_dev, dst, local,
556 struct net *net; /* Associated network namespace */
557 struct list_head dev_list;
558 #ifdef CONFIG_PROC_FS
559 struct proc_dir_entry *proc_dir;
561 struct class_attribute class_attr_bonding_masters;
564 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, struct slave *slave);
565 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev);
566 int bond_create(struct net *net, const char *name);
567 int bond_create_sysfs(struct bond_net *net);
568 void bond_destroy_sysfs(struct bond_net *net);
569 void bond_prepare_sysfs_group(struct bonding *bond);
570 int bond_sysfs_slave_add(struct slave *slave);
571 void bond_sysfs_slave_del(struct slave *slave);
572 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev);
573 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev);
574 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb);
575 int bond_set_carrier(struct bonding *bond);
576 void bond_select_active_slave(struct bonding *bond);
577 void bond_change_active_slave(struct bonding *bond, struct slave *new_active);
578 void bond_create_debugfs(void);
579 void bond_destroy_debugfs(void);
580 void bond_debug_register(struct bonding *bond);
581 void bond_debug_unregister(struct bonding *bond);
582 void bond_debug_reregister(struct bonding *bond);
583 const char *bond_mode_name(int mode);
584 void bond_setup(struct net_device *bond_dev);
585 unsigned int bond_get_num_tx_queues(void);
586 int bond_netlink_init(void);
587 void bond_netlink_fini(void);
588 struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond);
589 const char *bond_slave_link_status(s8 link);
590 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
591 struct net_device *end_dev,
593 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave);
594 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay);
596 #ifdef CONFIG_PROC_FS
597 void bond_create_proc_entry(struct bonding *bond);
598 void bond_remove_proc_entry(struct bonding *bond);
599 void bond_create_proc_dir(struct bond_net *bn);
600 void bond_destroy_proc_dir(struct bond_net *bn);
602 static inline void bond_create_proc_entry(struct bonding *bond)
606 static inline void bond_remove_proc_entry(struct bonding *bond)
610 static inline void bond_create_proc_dir(struct bond_net *bn)
614 static inline void bond_destroy_proc_dir(struct bond_net *bn)
619 static inline struct slave *bond_slave_has_mac(struct bonding *bond,
622 struct list_head *iter;
625 bond_for_each_slave(bond, tmp, iter)
626 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
632 /* Caller must hold rcu_read_lock() for read */
633 static inline struct slave *bond_slave_has_mac_rcu(struct bonding *bond,
636 struct list_head *iter;
639 bond_for_each_slave_rcu(bond, tmp, iter)
640 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
646 /* Caller must hold rcu_read_lock() for read */
647 static inline bool bond_slave_has_mac_rx(struct bonding *bond, const u8 *mac)
649 struct list_head *iter;
651 struct netdev_hw_addr *ha;
653 bond_for_each_slave_rcu(bond, tmp, iter)
654 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
657 if (netdev_uc_empty(bond->dev))
660 netdev_for_each_uc_addr(ha, bond->dev)
661 if (ether_addr_equal_64bits(mac, ha->addr))
667 /* Check if the ip is present in arp ip list, or first free slot if ip == 0
668 * Returns -1 if not found, index if found
670 static inline int bond_get_targets_ip(__be32 *targets, __be32 ip)
674 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
675 if (targets[i] == ip)
677 else if (targets[i] == 0)
683 /* exported from bond_main.c */
684 extern unsigned int bond_net_id;
685 extern const struct bond_parm_tbl bond_lacp_tbl[];
686 extern const struct bond_parm_tbl xmit_hashtype_tbl[];
687 extern const struct bond_parm_tbl arp_validate_tbl[];
688 extern const struct bond_parm_tbl arp_all_targets_tbl[];
689 extern const struct bond_parm_tbl fail_over_mac_tbl[];
690 extern const struct bond_parm_tbl pri_reselect_tbl[];
691 extern struct bond_parm_tbl ad_select_tbl[];
693 /* exported from bond_netlink.c */
694 extern struct rtnl_link_ops bond_link_ops;
696 static inline void bond_tx_drop(struct net_device *dev, struct sk_buff *skb)
698 atomic_long_inc(&dev->tx_dropped);
699 dev_kfree_skb_any(skb);
702 #endif /* _NET_BONDING_H */