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
38 * Less bad way to call ioctl from within the kernel; this needs to be
39 * done some other way to get the call out of interrupt context.
40 * Needs "ioctl" variable to be supplied by calling context.
42 #define IOCTL(dev, arg, cmd) ({ \
44 mm_segment_t fs = get_fs(); \
46 res = ioctl(dev, arg, cmd); \
50 #define BOND_MODE(bond) ((bond)->params.mode)
52 /* slave list primitives */
53 #define bond_slave_list(bond) (&(bond)->dev->adj_list.lower)
55 #define bond_has_slaves(bond) !list_empty(bond_slave_list(bond))
57 /* IMPORTANT: bond_first/last_slave can return NULL in case of an empty list */
58 #define bond_first_slave(bond) \
59 (bond_has_slaves(bond) ? \
60 netdev_adjacent_get_private(bond_slave_list(bond)->next) : \
62 #define bond_last_slave(bond) \
63 (bond_has_slaves(bond) ? \
64 netdev_adjacent_get_private(bond_slave_list(bond)->prev) : \
67 /* Caller must have rcu_read_lock */
68 #define bond_first_slave_rcu(bond) \
69 netdev_lower_get_first_private_rcu(bond->dev)
71 #define bond_is_first_slave(bond, pos) (pos == bond_first_slave(bond))
72 #define bond_is_last_slave(bond, pos) (pos == bond_last_slave(bond))
75 * bond_for_each_slave - iterate over all slaves
76 * @bond: the bond holding this list
78 * @iter: list_head * iterator
80 * Caller must hold RTNL
82 #define bond_for_each_slave(bond, pos, iter) \
83 netdev_for_each_lower_private((bond)->dev, pos, iter)
85 /* Caller must have rcu_read_lock */
86 #define bond_for_each_slave_rcu(bond, pos, iter) \
87 netdev_for_each_lower_private_rcu((bond)->dev, pos, iter)
89 #ifdef CONFIG_NET_POLL_CONTROLLER
90 extern atomic_t netpoll_block_tx;
92 static inline void block_netpoll_tx(void)
94 atomic_inc(&netpoll_block_tx);
97 static inline void unblock_netpoll_tx(void)
99 atomic_dec(&netpoll_block_tx);
102 static inline int is_netpoll_tx_blocked(struct net_device *dev)
104 if (unlikely(netpoll_tx_running(dev)))
105 return atomic_read(&netpoll_block_tx);
109 #define block_netpoll_tx()
110 #define unblock_netpoll_tx()
111 #define is_netpoll_tx_blocked(dev) (0)
127 unsigned int min_links;
129 char primary[IFNAMSIZ];
130 int primary_reselect;
131 __be32 arp_targets[BOND_MAX_ARP_TARGETS];
133 int all_slaves_active;
136 int packets_per_slave;
138 struct reciprocal_value reciprocal_packets_per_slave;
139 u16 ad_actor_sys_prio;
140 u16 ad_user_port_key;
141 u8 ad_actor_system[ETH_ALEN];
144 struct bond_parm_tbl {
150 struct net_device *dev; /* first - useful for panic debug */
151 struct bonding *bond; /* our master */
153 /* all three in jiffies */
154 unsigned long last_link_up;
155 unsigned long last_rx;
156 unsigned long target_last_arp_rx[BOND_MAX_ARP_TARGETS];
157 s8 link; /* one of BOND_LINK_XXXX */
159 u8 backup:1, /* indicates backup slave. Value corresponds with
160 BOND_STATE_ACTIVE and BOND_STATE_BACKUP */
161 inactive:1, /* indicates inactive slave */
162 should_notify:1; /* indicateds whether the state 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
174 struct delayed_work notify_work;
176 struct rtnl_link_stats64 slave_stats;
179 struct bond_up_slave {
182 struct slave *arr[0];
186 * Link pseudo-state only used internally by monitors
188 #define BOND_LINK_NOCHANGE -1
191 * Here are the locking policies for the two bonding locks:
192 * Get rcu_read_lock when reading or RTNL when writing slave list.
195 struct net_device *dev; /* first - useful for panic debug */
196 struct slave __rcu *curr_active_slave;
197 struct slave __rcu *current_arp_slave;
198 struct slave __rcu *primary_slave;
199 struct bond_up_slave __rcu *slave_arr; /* Array of usable slaves */
201 s32 slave_cnt; /* never change this value outside the attach/detach wrappers */
202 int (*recv_probe)(const struct sk_buff *, struct bonding *,
204 /* mode_lock is used for mode-specific locking needs, currently used by:
205 * 3ad mode (4) - protect against running bond_3ad_unbind_slave() and
206 * bond_3ad_state_machine_handler() concurrently and also
207 * the access to the state machine shared variables.
208 * TLB mode (5) - to sync the use and modifications of its hash table
209 * ALB mode (6) - to sync the use and modifications of its hash table
211 spinlock_t mode_lock;
212 spinlock_t stats_lock;
215 #ifdef CONFIG_PROC_FS
216 struct proc_dir_entry *proc_entry;
217 char proc_file_name[IFNAMSIZ];
218 #endif /* CONFIG_PROC_FS */
219 struct list_head bond_list;
221 struct ad_bond_info ad_info;
222 struct alb_bond_info alb_info;
223 struct bond_params params;
224 struct workqueue_struct *wq;
225 struct delayed_work mii_work;
226 struct delayed_work arp_work;
227 struct delayed_work alb_work;
228 struct delayed_work ad_work;
229 struct delayed_work mcast_work;
230 struct delayed_work slave_arr_work;
231 #ifdef CONFIG_DEBUG_FS
232 /* debugging support via debugfs */
233 struct dentry *debug_dir;
234 #endif /* CONFIG_DEBUG_FS */
235 struct rtnl_link_stats64 bond_stats;
238 #define bond_slave_get_rcu(dev) \
239 ((struct slave *) rcu_dereference(dev->rx_handler_data))
241 #define bond_slave_get_rtnl(dev) \
242 ((struct slave *) rtnl_dereference(dev->rx_handler_data))
244 void bond_queue_slave_event(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 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
330 static inline void bond_set_backup_slave(struct slave *slave)
332 if (!slave->backup) {
334 bond_queue_slave_event(slave);
335 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
339 static inline void bond_set_slave_state(struct slave *slave,
340 int slave_state, bool notify)
342 if (slave->backup == slave_state)
345 slave->backup = slave_state;
347 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
348 bond_queue_slave_event(slave);
349 slave->should_notify = 0;
351 if (slave->should_notify)
352 slave->should_notify = 0;
354 slave->should_notify = 1;
358 static inline void bond_slave_state_change(struct bonding *bond)
360 struct list_head *iter;
363 bond_for_each_slave(bond, tmp, iter) {
364 if (tmp->link == BOND_LINK_UP)
365 bond_set_active_slave(tmp);
366 else if (tmp->link == BOND_LINK_DOWN)
367 bond_set_backup_slave(tmp);
371 static inline void bond_slave_state_notify(struct bonding *bond)
373 struct list_head *iter;
376 bond_for_each_slave(bond, tmp, iter) {
377 if (tmp->should_notify) {
378 rtmsg_ifinfo(RTM_NEWLINK, tmp->dev, 0, GFP_ATOMIC);
379 tmp->should_notify = 0;
384 static inline int bond_slave_state(struct slave *slave)
386 return slave->backup;
389 static inline bool bond_is_active_slave(struct slave *slave)
391 return !bond_slave_state(slave);
394 static inline bool bond_slave_can_tx(struct slave *slave)
396 return bond_slave_is_up(slave) && slave->link == BOND_LINK_UP &&
397 bond_is_active_slave(slave);
400 #define BOND_PRI_RESELECT_ALWAYS 0
401 #define BOND_PRI_RESELECT_BETTER 1
402 #define BOND_PRI_RESELECT_FAILURE 2
404 #define BOND_FOM_NONE 0
405 #define BOND_FOM_ACTIVE 1
406 #define BOND_FOM_FOLLOW 2
408 #define BOND_ARP_TARGETS_ANY 0
409 #define BOND_ARP_TARGETS_ALL 1
411 #define BOND_ARP_VALIDATE_NONE 0
412 #define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE)
413 #define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP)
414 #define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \
415 BOND_ARP_VALIDATE_BACKUP)
416 #define BOND_ARP_FILTER (BOND_ARP_VALIDATE_ALL + 1)
417 #define BOND_ARP_FILTER_ACTIVE (BOND_ARP_VALIDATE_ACTIVE | \
419 #define BOND_ARP_FILTER_BACKUP (BOND_ARP_VALIDATE_BACKUP | \
422 #define BOND_SLAVE_NOTIFY_NOW true
423 #define BOND_SLAVE_NOTIFY_LATER false
425 static inline int slave_do_arp_validate(struct bonding *bond,
428 return bond->params.arp_validate & (1 << bond_slave_state(slave));
431 static inline int slave_do_arp_validate_only(struct bonding *bond)
433 return bond->params.arp_validate & BOND_ARP_FILTER;
436 static inline int bond_is_ip_target_ok(__be32 addr)
438 return !ipv4_is_lbcast(addr) && !ipv4_is_zeronet(addr);
441 /* Get the oldest arp which we've received on this slave for bond's
444 static inline unsigned long slave_oldest_target_arp_rx(struct bonding *bond,
448 unsigned long ret = slave->target_last_arp_rx[0];
450 for (; (i < BOND_MAX_ARP_TARGETS) && bond->params.arp_targets[i]; i++)
451 if (time_before(slave->target_last_arp_rx[i], ret))
452 ret = slave->target_last_arp_rx[i];
457 static inline unsigned long slave_last_rx(struct bonding *bond,
460 if (bond->params.arp_all_targets == BOND_ARP_TARGETS_ALL)
461 return slave_oldest_target_arp_rx(bond, slave);
463 return slave->last_rx;
466 #ifdef CONFIG_NET_POLL_CONTROLLER
467 static inline void bond_netpoll_send_skb(const struct slave *slave,
470 struct netpoll *np = slave->np;
473 netpoll_send_skb(np, skb);
476 static inline void bond_netpoll_send_skb(const struct slave *slave,
482 static inline void bond_set_slave_inactive_flags(struct slave *slave,
485 if (!bond_is_lb(slave->bond))
486 bond_set_slave_state(slave, BOND_STATE_BACKUP, notify);
487 if (!slave->bond->params.all_slaves_active)
491 static inline void bond_set_slave_active_flags(struct slave *slave,
494 bond_set_slave_state(slave, BOND_STATE_ACTIVE, notify);
498 static inline bool bond_is_slave_inactive(struct slave *slave)
500 return slave->inactive;
503 static inline void bond_set_slave_link_state(struct slave *slave, int state)
506 bond_queue_slave_event(slave);
509 static inline __be32 bond_confirm_addr(struct net_device *dev, __be32 dst, __be32 local)
511 struct in_device *in_dev;
515 in_dev = __in_dev_get_rcu(dev);
518 addr = inet_confirm_addr(dev_net(dev), in_dev, dst, local,
525 struct net *net; /* Associated network namespace */
526 struct list_head dev_list;
527 #ifdef CONFIG_PROC_FS
528 struct proc_dir_entry *proc_dir;
530 struct class_attribute class_attr_bonding_masters;
533 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, struct slave *slave);
534 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev);
535 int bond_create(struct net *net, const char *name);
536 int bond_create_sysfs(struct bond_net *net);
537 void bond_destroy_sysfs(struct bond_net *net);
538 void bond_prepare_sysfs_group(struct bonding *bond);
539 int bond_sysfs_slave_add(struct slave *slave);
540 void bond_sysfs_slave_del(struct slave *slave);
541 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev);
542 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev);
543 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb);
544 int bond_set_carrier(struct bonding *bond);
545 void bond_select_active_slave(struct bonding *bond);
546 void bond_change_active_slave(struct bonding *bond, struct slave *new_active);
547 void bond_create_debugfs(void);
548 void bond_destroy_debugfs(void);
549 void bond_debug_register(struct bonding *bond);
550 void bond_debug_unregister(struct bonding *bond);
551 void bond_debug_reregister(struct bonding *bond);
552 const char *bond_mode_name(int mode);
553 void bond_setup(struct net_device *bond_dev);
554 unsigned int bond_get_num_tx_queues(void);
555 int bond_netlink_init(void);
556 void bond_netlink_fini(void);
557 struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond);
558 const char *bond_slave_link_status(s8 link);
559 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
560 struct net_device *end_dev,
562 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave);
563 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay);
565 #ifdef CONFIG_PROC_FS
566 void bond_create_proc_entry(struct bonding *bond);
567 void bond_remove_proc_entry(struct bonding *bond);
568 void bond_create_proc_dir(struct bond_net *bn);
569 void bond_destroy_proc_dir(struct bond_net *bn);
571 static inline void bond_create_proc_entry(struct bonding *bond)
575 static inline void bond_remove_proc_entry(struct bonding *bond)
579 static inline void bond_create_proc_dir(struct bond_net *bn)
583 static inline void bond_destroy_proc_dir(struct bond_net *bn)
588 static inline struct slave *bond_slave_has_mac(struct bonding *bond,
591 struct list_head *iter;
594 bond_for_each_slave(bond, tmp, iter)
595 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
601 /* Caller must hold rcu_read_lock() for read */
602 static inline struct slave *bond_slave_has_mac_rcu(struct bonding *bond,
605 struct list_head *iter;
608 bond_for_each_slave_rcu(bond, tmp, iter)
609 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
615 /* Caller must hold rcu_read_lock() for read */
616 static inline bool bond_slave_has_mac_rx(struct bonding *bond, const u8 *mac)
618 struct list_head *iter;
620 struct netdev_hw_addr *ha;
622 bond_for_each_slave_rcu(bond, tmp, iter)
623 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
626 if (netdev_uc_empty(bond->dev))
629 netdev_for_each_uc_addr(ha, bond->dev)
630 if (ether_addr_equal_64bits(mac, ha->addr))
636 /* Check if the ip is present in arp ip list, or first free slot if ip == 0
637 * Returns -1 if not found, index if found
639 static inline int bond_get_targets_ip(__be32 *targets, __be32 ip)
643 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
644 if (targets[i] == ip)
646 else if (targets[i] == 0)
652 /* exported from bond_main.c */
653 extern int bond_net_id;
654 extern const struct bond_parm_tbl bond_lacp_tbl[];
655 extern const struct bond_parm_tbl xmit_hashtype_tbl[];
656 extern const struct bond_parm_tbl arp_validate_tbl[];
657 extern const struct bond_parm_tbl arp_all_targets_tbl[];
658 extern const struct bond_parm_tbl fail_over_mac_tbl[];
659 extern const struct bond_parm_tbl pri_reselect_tbl[];
660 extern struct bond_parm_tbl ad_select_tbl[];
662 /* exported from bond_netlink.c */
663 extern struct rtnl_link_ops bond_link_ops;
665 static inline void bond_tx_drop(struct net_device *dev, struct sk_buff *skb)
667 atomic_long_inc(&dev->tx_dropped);
668 dev_kfree_skb_any(skb);
671 #endif /* _NET_BONDING_H */