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 * Less bad way to call ioctl from within the kernel; this needs to be
42 * done some other way to get the call out of interrupt context.
43 * Needs "ioctl" variable to be supplied by calling context.
45 #define IOCTL(dev, arg, cmd) ({ \
47 mm_segment_t fs = get_fs(); \
49 res = ioctl(dev, arg, cmd); \
53 #define BOND_MODE(bond) ((bond)->params.mode)
55 /* slave list primitives */
56 #define bond_slave_list(bond) (&(bond)->dev->adj_list.lower)
58 #define bond_has_slaves(bond) !list_empty(bond_slave_list(bond))
60 /* IMPORTANT: bond_first/last_slave can return NULL in case of an empty list */
61 #define bond_first_slave(bond) \
62 (bond_has_slaves(bond) ? \
63 netdev_adjacent_get_private(bond_slave_list(bond)->next) : \
65 #define bond_last_slave(bond) \
66 (bond_has_slaves(bond) ? \
67 netdev_adjacent_get_private(bond_slave_list(bond)->prev) : \
70 /* Caller must have rcu_read_lock */
71 #define bond_first_slave_rcu(bond) \
72 netdev_lower_get_first_private_rcu(bond->dev)
74 #define bond_is_first_slave(bond, pos) (pos == bond_first_slave(bond))
75 #define bond_is_last_slave(bond, pos) (pos == bond_last_slave(bond))
78 * bond_for_each_slave - iterate over all slaves
79 * @bond: the bond holding this list
81 * @iter: list_head * iterator
83 * Caller must hold RTNL
85 #define bond_for_each_slave(bond, pos, iter) \
86 netdev_for_each_lower_private((bond)->dev, pos, iter)
88 /* Caller must have rcu_read_lock */
89 #define bond_for_each_slave_rcu(bond, pos, iter) \
90 netdev_for_each_lower_private_rcu((bond)->dev, pos, iter)
92 #ifdef CONFIG_NET_POLL_CONTROLLER
93 extern atomic_t netpoll_block_tx;
95 static inline void block_netpoll_tx(void)
97 atomic_inc(&netpoll_block_tx);
100 static inline void unblock_netpoll_tx(void)
102 atomic_dec(&netpoll_block_tx);
105 static inline int is_netpoll_tx_blocked(struct net_device *dev)
107 if (unlikely(netpoll_tx_running(dev)))
108 return atomic_read(&netpoll_block_tx);
112 #define block_netpoll_tx()
113 #define unblock_netpoll_tx()
114 #define is_netpoll_tx_blocked(dev) (0)
130 unsigned int min_links;
132 char primary[IFNAMSIZ];
133 int primary_reselect;
134 __be32 arp_targets[BOND_MAX_ARP_TARGETS];
136 int all_slaves_active;
139 int packets_per_slave;
141 struct reciprocal_value reciprocal_packets_per_slave;
142 u16 ad_actor_sys_prio;
143 u16 ad_user_port_key;
145 /* 2 bytes of padding : see ether_addr_equal_64bits() */
146 u8 ad_actor_system[ETH_ALEN + 2];
149 struct bond_parm_tbl {
155 struct net_device *dev; /* first - useful for panic debug */
156 struct bonding *bond; /* our master */
158 /* all three in jiffies */
159 unsigned long last_link_up;
160 unsigned long last_rx;
161 unsigned long target_last_arp_rx[BOND_MAX_ARP_TARGETS];
162 s8 link; /* one of BOND_LINK_XXXX */
164 u8 backup:1, /* indicates backup slave. Value corresponds with
165 BOND_STATE_ACTIVE and BOND_STATE_BACKUP */
166 inactive:1, /* indicates inactive slave */
167 should_notify:1; /* indicateds whether the state changed */
170 u32 link_failure_count;
173 u8 perm_hwaddr[ETH_ALEN];
174 struct ad_slave_info *ad_info;
175 struct tlb_slave_info tlb_info;
176 #ifdef CONFIG_NET_POLL_CONTROLLER
179 struct delayed_work notify_work;
181 struct rtnl_link_stats64 slave_stats;
184 static inline struct slave *to_slave(struct kobject *kobj)
186 return container_of(kobj, struct slave, kobj);
189 struct bond_up_slave {
192 struct slave *arr[0];
196 * Link pseudo-state only used internally by monitors
198 #define BOND_LINK_NOCHANGE -1
201 * Here are the locking policies for the two bonding locks:
202 * Get rcu_read_lock when reading or RTNL when writing slave list.
205 struct net_device *dev; /* first - useful for panic debug */
206 struct slave __rcu *curr_active_slave;
207 struct slave __rcu *current_arp_slave;
208 struct slave __rcu *primary_slave;
209 struct bond_up_slave __rcu *slave_arr; /* Array of usable slaves */
211 s32 slave_cnt; /* never change this value outside the attach/detach wrappers */
212 int (*recv_probe)(const struct sk_buff *, struct bonding *,
214 /* mode_lock is used for mode-specific locking needs, currently used by:
215 * 3ad mode (4) - protect against running bond_3ad_unbind_slave() and
216 * bond_3ad_state_machine_handler() concurrently and also
217 * the access to the state machine shared variables.
218 * TLB mode (5) - to sync the use and modifications of its hash table
219 * ALB mode (6) - to sync the use and modifications of its hash table
221 spinlock_t mode_lock;
222 spinlock_t stats_lock;
225 #ifdef CONFIG_PROC_FS
226 struct proc_dir_entry *proc_entry;
227 char proc_file_name[IFNAMSIZ];
228 #endif /* CONFIG_PROC_FS */
229 struct list_head bond_list;
231 struct ad_bond_info ad_info;
232 struct alb_bond_info alb_info;
233 struct bond_params params;
234 struct workqueue_struct *wq;
235 struct delayed_work mii_work;
236 struct delayed_work arp_work;
237 struct delayed_work alb_work;
238 struct delayed_work ad_work;
239 struct delayed_work mcast_work;
240 struct delayed_work slave_arr_work;
241 #ifdef CONFIG_DEBUG_FS
242 /* debugging support via debugfs */
243 struct dentry *debug_dir;
244 #endif /* CONFIG_DEBUG_FS */
245 struct rtnl_link_stats64 bond_stats;
248 #define bond_slave_get_rcu(dev) \
249 ((struct slave *) rcu_dereference(dev->rx_handler_data))
251 #define bond_slave_get_rtnl(dev) \
252 ((struct slave *) rtnl_dereference(dev->rx_handler_data))
254 void bond_queue_slave_event(struct slave *slave);
256 struct bond_vlan_tag {
258 unsigned short vlan_id;
262 * Returns NULL if the net_device does not belong to any of the bond's slaves
264 * Caller must hold bond lock for read
266 static inline struct slave *bond_get_slave_by_dev(struct bonding *bond,
267 struct net_device *slave_dev)
269 return netdev_lower_dev_get_private(bond->dev, slave_dev);
272 static inline struct bonding *bond_get_bond_by_slave(struct slave *slave)
277 static inline bool bond_should_override_tx_queue(struct bonding *bond)
279 return BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
280 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN;
283 static inline bool bond_is_lb(const struct bonding *bond)
285 return BOND_MODE(bond) == BOND_MODE_TLB ||
286 BOND_MODE(bond) == BOND_MODE_ALB;
289 static inline bool bond_is_nondyn_tlb(const struct bonding *bond)
291 return (BOND_MODE(bond) == BOND_MODE_TLB) &&
292 (bond->params.tlb_dynamic_lb == 0);
295 static inline bool bond_mode_uses_xmit_hash(const struct bonding *bond)
297 return (BOND_MODE(bond) == BOND_MODE_8023AD ||
298 BOND_MODE(bond) == BOND_MODE_XOR ||
299 bond_is_nondyn_tlb(bond));
302 static inline bool bond_mode_uses_arp(int mode)
304 return mode != BOND_MODE_8023AD && mode != BOND_MODE_TLB &&
305 mode != BOND_MODE_ALB;
308 static inline bool bond_mode_uses_primary(int mode)
310 return mode == BOND_MODE_ACTIVEBACKUP || mode == BOND_MODE_TLB ||
311 mode == BOND_MODE_ALB;
314 static inline bool bond_uses_primary(struct bonding *bond)
316 return bond_mode_uses_primary(BOND_MODE(bond));
319 static inline struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond)
321 struct slave *slave = rcu_dereference(bond->curr_active_slave);
323 return bond_uses_primary(bond) && slave ? slave->dev : NULL;
326 static inline bool bond_slave_is_up(struct slave *slave)
328 return netif_running(slave->dev) && netif_carrier_ok(slave->dev);
331 static inline void bond_set_active_slave(struct slave *slave)
335 bond_queue_slave_event(slave);
336 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
340 static inline void bond_set_backup_slave(struct slave *slave)
342 if (!slave->backup) {
344 bond_queue_slave_event(slave);
345 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
349 static inline void bond_set_slave_state(struct slave *slave,
350 int slave_state, bool notify)
352 if (slave->backup == slave_state)
355 slave->backup = slave_state;
357 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
358 bond_queue_slave_event(slave);
359 slave->should_notify = 0;
361 if (slave->should_notify)
362 slave->should_notify = 0;
364 slave->should_notify = 1;
368 static inline void bond_slave_state_change(struct bonding *bond)
370 struct list_head *iter;
373 bond_for_each_slave(bond, tmp, iter) {
374 if (tmp->link == BOND_LINK_UP)
375 bond_set_active_slave(tmp);
376 else if (tmp->link == BOND_LINK_DOWN)
377 bond_set_backup_slave(tmp);
381 static inline void bond_slave_state_notify(struct bonding *bond)
383 struct list_head *iter;
386 bond_for_each_slave(bond, tmp, iter) {
387 if (tmp->should_notify) {
388 rtmsg_ifinfo(RTM_NEWLINK, tmp->dev, 0, GFP_ATOMIC);
389 tmp->should_notify = 0;
394 static inline int bond_slave_state(struct slave *slave)
396 return slave->backup;
399 static inline bool bond_is_active_slave(struct slave *slave)
401 return !bond_slave_state(slave);
404 static inline bool bond_slave_can_tx(struct slave *slave)
406 return bond_slave_is_up(slave) && slave->link == BOND_LINK_UP &&
407 bond_is_active_slave(slave);
410 #define BOND_PRI_RESELECT_ALWAYS 0
411 #define BOND_PRI_RESELECT_BETTER 1
412 #define BOND_PRI_RESELECT_FAILURE 2
414 #define BOND_FOM_NONE 0
415 #define BOND_FOM_ACTIVE 1
416 #define BOND_FOM_FOLLOW 2
418 #define BOND_ARP_TARGETS_ANY 0
419 #define BOND_ARP_TARGETS_ALL 1
421 #define BOND_ARP_VALIDATE_NONE 0
422 #define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE)
423 #define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP)
424 #define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \
425 BOND_ARP_VALIDATE_BACKUP)
426 #define BOND_ARP_FILTER (BOND_ARP_VALIDATE_ALL + 1)
427 #define BOND_ARP_FILTER_ACTIVE (BOND_ARP_VALIDATE_ACTIVE | \
429 #define BOND_ARP_FILTER_BACKUP (BOND_ARP_VALIDATE_BACKUP | \
432 #define BOND_SLAVE_NOTIFY_NOW true
433 #define BOND_SLAVE_NOTIFY_LATER false
435 static inline int slave_do_arp_validate(struct bonding *bond,
438 return bond->params.arp_validate & (1 << bond_slave_state(slave));
441 static inline int slave_do_arp_validate_only(struct bonding *bond)
443 return bond->params.arp_validate & BOND_ARP_FILTER;
446 static inline int bond_is_ip_target_ok(__be32 addr)
448 return !ipv4_is_lbcast(addr) && !ipv4_is_zeronet(addr);
451 /* Get the oldest arp which we've received on this slave for bond's
454 static inline unsigned long slave_oldest_target_arp_rx(struct bonding *bond,
458 unsigned long ret = slave->target_last_arp_rx[0];
460 for (; (i < BOND_MAX_ARP_TARGETS) && bond->params.arp_targets[i]; i++)
461 if (time_before(slave->target_last_arp_rx[i], ret))
462 ret = slave->target_last_arp_rx[i];
467 static inline unsigned long slave_last_rx(struct bonding *bond,
470 if (bond->params.arp_all_targets == BOND_ARP_TARGETS_ALL)
471 return slave_oldest_target_arp_rx(bond, slave);
473 return slave->last_rx;
476 #ifdef CONFIG_NET_POLL_CONTROLLER
477 static inline void bond_netpoll_send_skb(const struct slave *slave,
480 struct netpoll *np = slave->np;
483 netpoll_send_skb(np, skb);
486 static inline void bond_netpoll_send_skb(const struct slave *slave,
492 static inline void bond_set_slave_inactive_flags(struct slave *slave,
495 if (!bond_is_lb(slave->bond))
496 bond_set_slave_state(slave, BOND_STATE_BACKUP, notify);
497 if (!slave->bond->params.all_slaves_active)
501 static inline void bond_set_slave_active_flags(struct slave *slave,
504 bond_set_slave_state(slave, BOND_STATE_ACTIVE, notify);
508 static inline bool bond_is_slave_inactive(struct slave *slave)
510 return slave->inactive;
513 static inline void bond_set_slave_link_state(struct slave *slave, int state)
516 bond_queue_slave_event(slave);
519 static inline __be32 bond_confirm_addr(struct net_device *dev, __be32 dst, __be32 local)
521 struct in_device *in_dev;
525 in_dev = __in_dev_get_rcu(dev);
528 addr = inet_confirm_addr(dev_net(dev), in_dev, dst, local,
535 struct net *net; /* Associated network namespace */
536 struct list_head dev_list;
537 #ifdef CONFIG_PROC_FS
538 struct proc_dir_entry *proc_dir;
540 struct class_attribute class_attr_bonding_masters;
543 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, struct slave *slave);
544 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev);
545 int bond_create(struct net *net, const char *name);
546 int bond_create_sysfs(struct bond_net *net);
547 void bond_destroy_sysfs(struct bond_net *net);
548 void bond_prepare_sysfs_group(struct bonding *bond);
549 int bond_sysfs_slave_add(struct slave *slave);
550 void bond_sysfs_slave_del(struct slave *slave);
551 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev);
552 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev);
553 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb);
554 int bond_set_carrier(struct bonding *bond);
555 void bond_select_active_slave(struct bonding *bond);
556 void bond_change_active_slave(struct bonding *bond, struct slave *new_active);
557 void bond_create_debugfs(void);
558 void bond_destroy_debugfs(void);
559 void bond_debug_register(struct bonding *bond);
560 void bond_debug_unregister(struct bonding *bond);
561 void bond_debug_reregister(struct bonding *bond);
562 const char *bond_mode_name(int mode);
563 void bond_setup(struct net_device *bond_dev);
564 unsigned int bond_get_num_tx_queues(void);
565 int bond_netlink_init(void);
566 void bond_netlink_fini(void);
567 struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond);
568 const char *bond_slave_link_status(s8 link);
569 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
570 struct net_device *end_dev,
572 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave);
573 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay);
575 #ifdef CONFIG_PROC_FS
576 void bond_create_proc_entry(struct bonding *bond);
577 void bond_remove_proc_entry(struct bonding *bond);
578 void bond_create_proc_dir(struct bond_net *bn);
579 void bond_destroy_proc_dir(struct bond_net *bn);
581 static inline void bond_create_proc_entry(struct bonding *bond)
585 static inline void bond_remove_proc_entry(struct bonding *bond)
589 static inline void bond_create_proc_dir(struct bond_net *bn)
593 static inline void bond_destroy_proc_dir(struct bond_net *bn)
598 static inline struct slave *bond_slave_has_mac(struct bonding *bond,
601 struct list_head *iter;
604 bond_for_each_slave(bond, tmp, iter)
605 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
611 /* Caller must hold rcu_read_lock() for read */
612 static inline struct slave *bond_slave_has_mac_rcu(struct bonding *bond,
615 struct list_head *iter;
618 bond_for_each_slave_rcu(bond, tmp, iter)
619 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
625 /* Caller must hold rcu_read_lock() for read */
626 static inline bool bond_slave_has_mac_rx(struct bonding *bond, const u8 *mac)
628 struct list_head *iter;
630 struct netdev_hw_addr *ha;
632 bond_for_each_slave_rcu(bond, tmp, iter)
633 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
636 if (netdev_uc_empty(bond->dev))
639 netdev_for_each_uc_addr(ha, bond->dev)
640 if (ether_addr_equal_64bits(mac, ha->addr))
646 /* Check if the ip is present in arp ip list, or first free slot if ip == 0
647 * Returns -1 if not found, index if found
649 static inline int bond_get_targets_ip(__be32 *targets, __be32 ip)
653 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
654 if (targets[i] == ip)
656 else if (targets[i] == 0)
662 /* exported from bond_main.c */
663 extern int bond_net_id;
664 extern const struct bond_parm_tbl bond_lacp_tbl[];
665 extern const struct bond_parm_tbl xmit_hashtype_tbl[];
666 extern const struct bond_parm_tbl arp_validate_tbl[];
667 extern const struct bond_parm_tbl arp_all_targets_tbl[];
668 extern const struct bond_parm_tbl fail_over_mac_tbl[];
669 extern const struct bond_parm_tbl pri_reselect_tbl[];
670 extern struct bond_parm_tbl ad_select_tbl[];
672 /* exported from bond_netlink.c */
673 extern struct rtnl_link_ops bond_link_ops;
675 /* exported from bond_sysfs_slave.c */
676 extern const struct sysfs_ops slave_sysfs_ops;
678 static inline void bond_tx_drop(struct net_device *dev, struct sk_buff *skb)
680 atomic_long_inc(&dev->tx_dropped);
681 dev_kfree_skb_any(skb);
684 #endif /* _NET_BONDING_H */