Loading...
1/*
2 * Bond several ethernet interfaces into a Cisco, running 'Etherchannel'.
3 *
4 * Portions are (c) Copyright 1995 Simon "Guru Aleph-Null" Janes
5 * NCM: Network and Communications Management, Inc.
6 *
7 * BUT, I'm the one who modified it for ethernet, so:
8 * (c) Copyright 1999, Thomas Davis, tadavis@lbl.gov
9 *
10 * This software may be used and distributed according to the terms
11 * of the GNU Public License, incorporated herein by reference.
12 *
13 */
14
15#ifndef _NET_BONDING_H
16#define _NET_BONDING_H
17
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>
28
29#include <net/bond_3ad.h>
30#include <net/bond_alb.h>
31#include <net/bond_options.h>
32
33#define BOND_MAX_ARP_TARGETS 16
34
35#define BOND_DEFAULT_MIIMON 100
36
37/*
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.
41 */
42#define IOCTL(dev, arg, cmd) ({ \
43 int res = 0; \
44 mm_segment_t fs = get_fs(); \
45 set_fs(get_ds()); \
46 res = ioctl(dev, arg, cmd); \
47 set_fs(fs); \
48 res; })
49
50#define BOND_MODE(bond) ((bond)->params.mode)
51
52/* slave list primitives */
53#define bond_slave_list(bond) (&(bond)->dev->adj_list.lower)
54
55#define bond_has_slaves(bond) !list_empty(bond_slave_list(bond))
56
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) : \
61 NULL)
62#define bond_last_slave(bond) \
63 (bond_has_slaves(bond) ? \
64 netdev_adjacent_get_private(bond_slave_list(bond)->prev) : \
65 NULL)
66
67/* Caller must have rcu_read_lock */
68#define bond_first_slave_rcu(bond) \
69 netdev_lower_get_first_private_rcu(bond->dev)
70
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))
73
74/**
75 * bond_for_each_slave - iterate over all slaves
76 * @bond: the bond holding this list
77 * @pos: current slave
78 * @iter: list_head * iterator
79 *
80 * Caller must hold RTNL
81 */
82#define bond_for_each_slave(bond, pos, iter) \
83 netdev_for_each_lower_private((bond)->dev, pos, iter)
84
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)
88
89#ifdef CONFIG_NET_POLL_CONTROLLER
90extern atomic_t netpoll_block_tx;
91
92static inline void block_netpoll_tx(void)
93{
94 atomic_inc(&netpoll_block_tx);
95}
96
97static inline void unblock_netpoll_tx(void)
98{
99 atomic_dec(&netpoll_block_tx);
100}
101
102static inline int is_netpoll_tx_blocked(struct net_device *dev)
103{
104 if (unlikely(netpoll_tx_running(dev)))
105 return atomic_read(&netpoll_block_tx);
106 return 0;
107}
108#else
109#define block_netpoll_tx()
110#define unblock_netpoll_tx()
111#define is_netpoll_tx_blocked(dev) (0)
112#endif
113
114struct bond_params {
115 int mode;
116 int xmit_policy;
117 int miimon;
118 u8 num_peer_notif;
119 int arp_interval;
120 int arp_validate;
121 int arp_all_targets;
122 int use_carrier;
123 int fail_over_mac;
124 int updelay;
125 int downdelay;
126 int lacp_fast;
127 unsigned int min_links;
128 int ad_select;
129 char primary[IFNAMSIZ];
130 int primary_reselect;
131 __be32 arp_targets[BOND_MAX_ARP_TARGETS];
132 int tx_queues;
133 int all_slaves_active;
134 int resend_igmp;
135 int lp_interval;
136 int packets_per_slave;
137 int tlb_dynamic_lb;
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];
142};
143
144struct bond_parm_tbl {
145 char *modename;
146 int mode;
147};
148
149struct netdev_notify_work {
150 struct delayed_work work;
151 struct net_device *dev;
152 struct netdev_bonding_info bonding_info;
153};
154
155struct slave {
156 struct net_device *dev; /* first - useful for panic debug */
157 struct bonding *bond; /* our master */
158 int delay;
159 /* all three in jiffies */
160 unsigned long last_link_up;
161 unsigned long last_rx;
162 unsigned long target_last_arp_rx[BOND_MAX_ARP_TARGETS];
163 s8 link; /* one of BOND_LINK_XXXX */
164 s8 new_link;
165 u8 backup:1, /* indicates backup slave. Value corresponds with
166 BOND_STATE_ACTIVE and BOND_STATE_BACKUP */
167 inactive:1, /* indicates inactive slave */
168 should_notify:1, /* indicates whether the state changed */
169 should_notify_link:1; /* indicates whether the link changed */
170 u8 duplex;
171 u32 original_mtu;
172 u32 link_failure_count;
173 u32 speed;
174 u16 queue_id;
175 u8 perm_hwaddr[ETH_ALEN];
176 struct ad_slave_info *ad_info;
177 struct tlb_slave_info tlb_info;
178#ifdef CONFIG_NET_POLL_CONTROLLER
179 struct netpoll *np;
180#endif
181 struct kobject kobj;
182 struct rtnl_link_stats64 slave_stats;
183};
184
185struct bond_up_slave {
186 unsigned int count;
187 struct rcu_head rcu;
188 struct slave *arr[0];
189};
190
191/*
192 * Link pseudo-state only used internally by monitors
193 */
194#define BOND_LINK_NOCHANGE -1
195
196/*
197 * Here are the locking policies for the two bonding locks:
198 * Get rcu_read_lock when reading or RTNL when writing slave list.
199 */
200struct bonding {
201 struct net_device *dev; /* first - useful for panic debug */
202 struct slave __rcu *curr_active_slave;
203 struct slave __rcu *current_arp_slave;
204 struct slave __rcu *primary_slave;
205 struct bond_up_slave __rcu *slave_arr; /* Array of usable slaves */
206 bool force_primary;
207 s32 slave_cnt; /* never change this value outside the attach/detach wrappers */
208 int (*recv_probe)(const struct sk_buff *, struct bonding *,
209 struct slave *);
210 /* mode_lock is used for mode-specific locking needs, currently used by:
211 * 3ad mode (4) - protect against running bond_3ad_unbind_slave() and
212 * bond_3ad_state_machine_handler() concurrently and also
213 * the access to the state machine shared variables.
214 * TLB mode (5) - to sync the use and modifications of its hash table
215 * ALB mode (6) - to sync the use and modifications of its hash table
216 */
217 spinlock_t mode_lock;
218 spinlock_t stats_lock;
219 u8 send_peer_notif;
220 u8 igmp_retrans;
221#ifdef CONFIG_PROC_FS
222 struct proc_dir_entry *proc_entry;
223 char proc_file_name[IFNAMSIZ];
224#endif /* CONFIG_PROC_FS */
225 struct list_head bond_list;
226 u32 rr_tx_counter;
227 struct ad_bond_info ad_info;
228 struct alb_bond_info alb_info;
229 struct bond_params params;
230 struct workqueue_struct *wq;
231 struct delayed_work mii_work;
232 struct delayed_work arp_work;
233 struct delayed_work alb_work;
234 struct delayed_work ad_work;
235 struct delayed_work mcast_work;
236 struct delayed_work slave_arr_work;
237#ifdef CONFIG_DEBUG_FS
238 /* debugging support via debugfs */
239 struct dentry *debug_dir;
240#endif /* CONFIG_DEBUG_FS */
241 struct rtnl_link_stats64 bond_stats;
242};
243
244#define bond_slave_get_rcu(dev) \
245 ((struct slave *) rcu_dereference(dev->rx_handler_data))
246
247#define bond_slave_get_rtnl(dev) \
248 ((struct slave *) rtnl_dereference(dev->rx_handler_data))
249
250void bond_queue_slave_event(struct slave *slave);
251void bond_lower_state_changed(struct slave *slave);
252
253struct bond_vlan_tag {
254 __be16 vlan_proto;
255 unsigned short vlan_id;
256};
257
258/**
259 * Returns NULL if the net_device does not belong to any of the bond's slaves
260 *
261 * Caller must hold bond lock for read
262 */
263static inline struct slave *bond_get_slave_by_dev(struct bonding *bond,
264 struct net_device *slave_dev)
265{
266 return netdev_lower_dev_get_private(bond->dev, slave_dev);
267}
268
269static inline struct bonding *bond_get_bond_by_slave(struct slave *slave)
270{
271 return slave->bond;
272}
273
274static inline bool bond_should_override_tx_queue(struct bonding *bond)
275{
276 return BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
277 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN;
278}
279
280static inline bool bond_is_lb(const struct bonding *bond)
281{
282 return BOND_MODE(bond) == BOND_MODE_TLB ||
283 BOND_MODE(bond) == BOND_MODE_ALB;
284}
285
286static inline bool bond_is_nondyn_tlb(const struct bonding *bond)
287{
288 return (BOND_MODE(bond) == BOND_MODE_TLB) &&
289 (bond->params.tlb_dynamic_lb == 0);
290}
291
292static inline bool bond_mode_uses_xmit_hash(const struct bonding *bond)
293{
294 return (BOND_MODE(bond) == BOND_MODE_8023AD ||
295 BOND_MODE(bond) == BOND_MODE_XOR ||
296 bond_is_nondyn_tlb(bond));
297}
298
299static inline bool bond_mode_uses_arp(int mode)
300{
301 return mode != BOND_MODE_8023AD && mode != BOND_MODE_TLB &&
302 mode != BOND_MODE_ALB;
303}
304
305static inline bool bond_mode_uses_primary(int mode)
306{
307 return mode == BOND_MODE_ACTIVEBACKUP || mode == BOND_MODE_TLB ||
308 mode == BOND_MODE_ALB;
309}
310
311static inline bool bond_uses_primary(struct bonding *bond)
312{
313 return bond_mode_uses_primary(BOND_MODE(bond));
314}
315
316static inline struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond)
317{
318 struct slave *slave = rcu_dereference(bond->curr_active_slave);
319
320 return bond_uses_primary(bond) && slave ? slave->dev : NULL;
321}
322
323static inline bool bond_slave_is_up(struct slave *slave)
324{
325 return netif_running(slave->dev) && netif_carrier_ok(slave->dev);
326}
327
328static inline void bond_set_active_slave(struct slave *slave)
329{
330 if (slave->backup) {
331 slave->backup = 0;
332 bond_queue_slave_event(slave);
333 bond_lower_state_changed(slave);
334 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
335 }
336}
337
338static inline void bond_set_backup_slave(struct slave *slave)
339{
340 if (!slave->backup) {
341 slave->backup = 1;
342 bond_queue_slave_event(slave);
343 bond_lower_state_changed(slave);
344 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
345 }
346}
347
348static inline void bond_set_slave_state(struct slave *slave,
349 int slave_state, bool notify)
350{
351 if (slave->backup == slave_state)
352 return;
353
354 slave->backup = slave_state;
355 if (notify) {
356 bond_lower_state_changed(slave);
357 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
358 bond_queue_slave_event(slave);
359 slave->should_notify = 0;
360 } else {
361 if (slave->should_notify)
362 slave->should_notify = 0;
363 else
364 slave->should_notify = 1;
365 }
366}
367
368static inline void bond_slave_state_change(struct bonding *bond)
369{
370 struct list_head *iter;
371 struct slave *tmp;
372
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);
378 }
379}
380
381static inline void bond_slave_state_notify(struct bonding *bond)
382{
383 struct list_head *iter;
384 struct slave *tmp;
385
386 bond_for_each_slave(bond, tmp, iter) {
387 if (tmp->should_notify) {
388 bond_lower_state_changed(tmp);
389 rtmsg_ifinfo(RTM_NEWLINK, tmp->dev, 0, GFP_ATOMIC);
390 tmp->should_notify = 0;
391 }
392 }
393}
394
395static inline int bond_slave_state(struct slave *slave)
396{
397 return slave->backup;
398}
399
400static inline bool bond_is_active_slave(struct slave *slave)
401{
402 return !bond_slave_state(slave);
403}
404
405static inline bool bond_slave_can_tx(struct slave *slave)
406{
407 return bond_slave_is_up(slave) && slave->link == BOND_LINK_UP &&
408 bond_is_active_slave(slave);
409}
410
411#define BOND_PRI_RESELECT_ALWAYS 0
412#define BOND_PRI_RESELECT_BETTER 1
413#define BOND_PRI_RESELECT_FAILURE 2
414
415#define BOND_FOM_NONE 0
416#define BOND_FOM_ACTIVE 1
417#define BOND_FOM_FOLLOW 2
418
419#define BOND_ARP_TARGETS_ANY 0
420#define BOND_ARP_TARGETS_ALL 1
421
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 | \
429 BOND_ARP_FILTER)
430#define BOND_ARP_FILTER_BACKUP (BOND_ARP_VALIDATE_BACKUP | \
431 BOND_ARP_FILTER)
432
433#define BOND_SLAVE_NOTIFY_NOW true
434#define BOND_SLAVE_NOTIFY_LATER false
435
436static inline int slave_do_arp_validate(struct bonding *bond,
437 struct slave *slave)
438{
439 return bond->params.arp_validate & (1 << bond_slave_state(slave));
440}
441
442static inline int slave_do_arp_validate_only(struct bonding *bond)
443{
444 return bond->params.arp_validate & BOND_ARP_FILTER;
445}
446
447static inline int bond_is_ip_target_ok(__be32 addr)
448{
449 return !ipv4_is_lbcast(addr) && !ipv4_is_zeronet(addr);
450}
451
452/* Get the oldest arp which we've received on this slave for bond's
453 * arp_targets.
454 */
455static inline unsigned long slave_oldest_target_arp_rx(struct bonding *bond,
456 struct slave *slave)
457{
458 int i = 1;
459 unsigned long ret = slave->target_last_arp_rx[0];
460
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];
464
465 return ret;
466}
467
468static inline unsigned long slave_last_rx(struct bonding *bond,
469 struct slave *slave)
470{
471 if (bond->params.arp_all_targets == BOND_ARP_TARGETS_ALL)
472 return slave_oldest_target_arp_rx(bond, slave);
473
474 return slave->last_rx;
475}
476
477#ifdef CONFIG_NET_POLL_CONTROLLER
478static inline void bond_netpoll_send_skb(const struct slave *slave,
479 struct sk_buff *skb)
480{
481 struct netpoll *np = slave->np;
482
483 if (np)
484 netpoll_send_skb(np, skb);
485}
486#else
487static inline void bond_netpoll_send_skb(const struct slave *slave,
488 struct sk_buff *skb)
489{
490}
491#endif
492
493static inline void bond_set_slave_inactive_flags(struct slave *slave,
494 bool notify)
495{
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)
499 slave->inactive = 1;
500}
501
502static inline void bond_set_slave_active_flags(struct slave *slave,
503 bool notify)
504{
505 bond_set_slave_state(slave, BOND_STATE_ACTIVE, notify);
506 slave->inactive = 0;
507}
508
509static inline bool bond_is_slave_inactive(struct slave *slave)
510{
511 return slave->inactive;
512}
513
514static inline void bond_set_slave_link_state(struct slave *slave, int state,
515 bool notify)
516{
517 if (slave->link == state)
518 return;
519
520 slave->link = state;
521 if (notify) {
522 bond_queue_slave_event(slave);
523 bond_lower_state_changed(slave);
524 slave->should_notify_link = 0;
525 } else {
526 if (slave->should_notify_link)
527 slave->should_notify_link = 0;
528 else
529 slave->should_notify_link = 1;
530 }
531}
532
533static inline void bond_slave_link_notify(struct bonding *bond)
534{
535 struct list_head *iter;
536 struct slave *tmp;
537
538 bond_for_each_slave(bond, tmp, iter) {
539 if (tmp->should_notify_link) {
540 bond_queue_slave_event(tmp);
541 bond_lower_state_changed(tmp);
542 tmp->should_notify_link = 0;
543 }
544 }
545}
546
547static inline __be32 bond_confirm_addr(struct net_device *dev, __be32 dst, __be32 local)
548{
549 struct in_device *in_dev;
550 __be32 addr = 0;
551
552 rcu_read_lock();
553 in_dev = __in_dev_get_rcu(dev);
554
555 if (in_dev)
556 addr = inet_confirm_addr(dev_net(dev), in_dev, dst, local,
557 RT_SCOPE_HOST);
558 rcu_read_unlock();
559 return addr;
560}
561
562struct bond_net {
563 struct net *net; /* Associated network namespace */
564 struct list_head dev_list;
565#ifdef CONFIG_PROC_FS
566 struct proc_dir_entry *proc_dir;
567#endif
568 struct class_attribute class_attr_bonding_masters;
569};
570
571int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, struct slave *slave);
572void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev);
573int bond_create(struct net *net, const char *name);
574int bond_create_sysfs(struct bond_net *net);
575void bond_destroy_sysfs(struct bond_net *net);
576void bond_prepare_sysfs_group(struct bonding *bond);
577int bond_sysfs_slave_add(struct slave *slave);
578void bond_sysfs_slave_del(struct slave *slave);
579int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev);
580int bond_release(struct net_device *bond_dev, struct net_device *slave_dev);
581u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb);
582int bond_set_carrier(struct bonding *bond);
583void bond_select_active_slave(struct bonding *bond);
584void bond_change_active_slave(struct bonding *bond, struct slave *new_active);
585void bond_create_debugfs(void);
586void bond_destroy_debugfs(void);
587void bond_debug_register(struct bonding *bond);
588void bond_debug_unregister(struct bonding *bond);
589void bond_debug_reregister(struct bonding *bond);
590const char *bond_mode_name(int mode);
591void bond_setup(struct net_device *bond_dev);
592unsigned int bond_get_num_tx_queues(void);
593int bond_netlink_init(void);
594void bond_netlink_fini(void);
595struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond);
596const char *bond_slave_link_status(s8 link);
597struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
598 struct net_device *end_dev,
599 int level);
600int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave);
601void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay);
602
603#ifdef CONFIG_PROC_FS
604void bond_create_proc_entry(struct bonding *bond);
605void bond_remove_proc_entry(struct bonding *bond);
606void bond_create_proc_dir(struct bond_net *bn);
607void bond_destroy_proc_dir(struct bond_net *bn);
608#else
609static inline void bond_create_proc_entry(struct bonding *bond)
610{
611}
612
613static inline void bond_remove_proc_entry(struct bonding *bond)
614{
615}
616
617static inline void bond_create_proc_dir(struct bond_net *bn)
618{
619}
620
621static inline void bond_destroy_proc_dir(struct bond_net *bn)
622{
623}
624#endif
625
626static inline struct slave *bond_slave_has_mac(struct bonding *bond,
627 const u8 *mac)
628{
629 struct list_head *iter;
630 struct slave *tmp;
631
632 bond_for_each_slave(bond, tmp, iter)
633 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
634 return tmp;
635
636 return NULL;
637}
638
639/* Caller must hold rcu_read_lock() for read */
640static inline struct slave *bond_slave_has_mac_rcu(struct bonding *bond,
641 const u8 *mac)
642{
643 struct list_head *iter;
644 struct slave *tmp;
645
646 bond_for_each_slave_rcu(bond, tmp, iter)
647 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
648 return tmp;
649
650 return NULL;
651}
652
653/* Caller must hold rcu_read_lock() for read */
654static inline bool bond_slave_has_mac_rx(struct bonding *bond, const u8 *mac)
655{
656 struct list_head *iter;
657 struct slave *tmp;
658 struct netdev_hw_addr *ha;
659
660 bond_for_each_slave_rcu(bond, tmp, iter)
661 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
662 return true;
663
664 if (netdev_uc_empty(bond->dev))
665 return false;
666
667 netdev_for_each_uc_addr(ha, bond->dev)
668 if (ether_addr_equal_64bits(mac, ha->addr))
669 return true;
670
671 return false;
672}
673
674/* Check if the ip is present in arp ip list, or first free slot if ip == 0
675 * Returns -1 if not found, index if found
676 */
677static inline int bond_get_targets_ip(__be32 *targets, __be32 ip)
678{
679 int i;
680
681 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
682 if (targets[i] == ip)
683 return i;
684 else if (targets[i] == 0)
685 break;
686
687 return -1;
688}
689
690/* exported from bond_main.c */
691extern int bond_net_id;
692extern const struct bond_parm_tbl bond_lacp_tbl[];
693extern const struct bond_parm_tbl xmit_hashtype_tbl[];
694extern const struct bond_parm_tbl arp_validate_tbl[];
695extern const struct bond_parm_tbl arp_all_targets_tbl[];
696extern const struct bond_parm_tbl fail_over_mac_tbl[];
697extern const struct bond_parm_tbl pri_reselect_tbl[];
698extern struct bond_parm_tbl ad_select_tbl[];
699
700/* exported from bond_netlink.c */
701extern struct rtnl_link_ops bond_link_ops;
702
703static inline void bond_tx_drop(struct net_device *dev, struct sk_buff *skb)
704{
705 atomic_long_inc(&dev->tx_dropped);
706 dev_kfree_skb_any(skb);
707}
708
709#endif /* _NET_BONDING_H */
1/*
2 * Bond several ethernet interfaces into a Cisco, running 'Etherchannel'.
3 *
4 * Portions are (c) Copyright 1995 Simon "Guru Aleph-Null" Janes
5 * NCM: Network and Communications Management, Inc.
6 *
7 * BUT, I'm the one who modified it for ethernet, so:
8 * (c) Copyright 1999, Thomas Davis, tadavis@lbl.gov
9 *
10 * This software may be used and distributed according to the terms
11 * of the GNU Public License, incorporated herein by reference.
12 *
13 */
14
15#ifndef _NET_BONDING_H
16#define _NET_BONDING_H
17
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>
28
29#include <net/bond_3ad.h>
30#include <net/bond_alb.h>
31#include <net/bond_options.h>
32
33#define BOND_MAX_ARP_TARGETS 16
34
35#define BOND_DEFAULT_MIIMON 100
36
37#ifndef __long_aligned
38#define __long_aligned __attribute__((aligned((sizeof(long)))))
39#endif
40
41#define BOND_MODE(bond) ((bond)->params.mode)
42
43/* slave list primitives */
44#define bond_slave_list(bond) (&(bond)->dev->adj_list.lower)
45
46#define bond_has_slaves(bond) !list_empty(bond_slave_list(bond))
47
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) : \
52 NULL)
53#define bond_last_slave(bond) \
54 (bond_has_slaves(bond) ? \
55 netdev_adjacent_get_private(bond_slave_list(bond)->prev) : \
56 NULL)
57
58/* Caller must have rcu_read_lock */
59#define bond_first_slave_rcu(bond) \
60 netdev_lower_get_first_private_rcu(bond->dev)
61
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))
64
65/**
66 * bond_for_each_slave - iterate over all slaves
67 * @bond: the bond holding this list
68 * @pos: current slave
69 * @iter: list_head * iterator
70 *
71 * Caller must hold RTNL
72 */
73#define bond_for_each_slave(bond, pos, iter) \
74 netdev_for_each_lower_private((bond)->dev, pos, iter)
75
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)
79
80#ifdef CONFIG_NET_POLL_CONTROLLER
81extern atomic_t netpoll_block_tx;
82
83static inline void block_netpoll_tx(void)
84{
85 atomic_inc(&netpoll_block_tx);
86}
87
88static inline void unblock_netpoll_tx(void)
89{
90 atomic_dec(&netpoll_block_tx);
91}
92
93static inline int is_netpoll_tx_blocked(struct net_device *dev)
94{
95 if (unlikely(netpoll_tx_running(dev)))
96 return atomic_read(&netpoll_block_tx);
97 return 0;
98}
99#else
100#define block_netpoll_tx()
101#define unblock_netpoll_tx()
102#define is_netpoll_tx_blocked(dev) (0)
103#endif
104
105struct bond_params {
106 int mode;
107 int xmit_policy;
108 int miimon;
109 u8 num_peer_notif;
110 int arp_interval;
111 int arp_validate;
112 int arp_all_targets;
113 int use_carrier;
114 int fail_over_mac;
115 int updelay;
116 int downdelay;
117 int lacp_fast;
118 unsigned int min_links;
119 int ad_select;
120 char primary[IFNAMSIZ];
121 int primary_reselect;
122 __be32 arp_targets[BOND_MAX_ARP_TARGETS];
123 int tx_queues;
124 int all_slaves_active;
125 int resend_igmp;
126 int lp_interval;
127 int packets_per_slave;
128 int tlb_dynamic_lb;
129 struct reciprocal_value reciprocal_packets_per_slave;
130 u16 ad_actor_sys_prio;
131 u16 ad_user_port_key;
132
133 /* 2 bytes of padding : see ether_addr_equal_64bits() */
134 u8 ad_actor_system[ETH_ALEN + 2];
135};
136
137struct bond_parm_tbl {
138 char *modename;
139 int mode;
140};
141
142struct netdev_notify_work {
143 struct delayed_work work;
144 struct net_device *dev;
145 struct netdev_bonding_info bonding_info;
146};
147
148struct slave {
149 struct net_device *dev; /* first - useful for panic debug */
150 struct bonding *bond; /* our master */
151 int delay;
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 */
157 s8 new_link;
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 */
163 u8 duplex;
164 u32 original_mtu;
165 u32 link_failure_count;
166 u32 speed;
167 u16 queue_id;
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
172 struct netpoll *np;
173#endif
174 struct kobject kobj;
175 struct rtnl_link_stats64 slave_stats;
176};
177
178struct bond_up_slave {
179 unsigned int count;
180 struct rcu_head rcu;
181 struct slave *arr[0];
182};
183
184/*
185 * Link pseudo-state only used internally by monitors
186 */
187#define BOND_LINK_NOCHANGE -1
188
189/*
190 * Here are the locking policies for the two bonding locks:
191 * Get rcu_read_lock when reading or RTNL when writing slave list.
192 */
193struct bonding {
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 */
199 bool force_primary;
200 s32 slave_cnt; /* never change this value outside the attach/detach wrappers */
201 int (*recv_probe)(const struct sk_buff *, struct bonding *,
202 struct slave *);
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
209 */
210 spinlock_t mode_lock;
211 spinlock_t stats_lock;
212 u8 send_peer_notif;
213 u8 igmp_retrans;
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;
219 u32 rr_tx_counter;
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;
235};
236
237#define bond_slave_get_rcu(dev) \
238 ((struct slave *) rcu_dereference(dev->rx_handler_data))
239
240#define bond_slave_get_rtnl(dev) \
241 ((struct slave *) rtnl_dereference(dev->rx_handler_data))
242
243void bond_queue_slave_event(struct slave *slave);
244void bond_lower_state_changed(struct slave *slave);
245
246struct bond_vlan_tag {
247 __be16 vlan_proto;
248 unsigned short vlan_id;
249};
250
251/**
252 * Returns NULL if the net_device does not belong to any of the bond's slaves
253 *
254 * Caller must hold bond lock for read
255 */
256static inline struct slave *bond_get_slave_by_dev(struct bonding *bond,
257 struct net_device *slave_dev)
258{
259 return netdev_lower_dev_get_private(bond->dev, slave_dev);
260}
261
262static inline struct bonding *bond_get_bond_by_slave(struct slave *slave)
263{
264 return slave->bond;
265}
266
267static inline bool bond_should_override_tx_queue(struct bonding *bond)
268{
269 return BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
270 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN;
271}
272
273static inline bool bond_is_lb(const struct bonding *bond)
274{
275 return BOND_MODE(bond) == BOND_MODE_TLB ||
276 BOND_MODE(bond) == BOND_MODE_ALB;
277}
278
279static inline bool bond_is_nondyn_tlb(const struct bonding *bond)
280{
281 return (BOND_MODE(bond) == BOND_MODE_TLB) &&
282 (bond->params.tlb_dynamic_lb == 0);
283}
284
285static inline bool bond_mode_uses_xmit_hash(const struct bonding *bond)
286{
287 return (BOND_MODE(bond) == BOND_MODE_8023AD ||
288 BOND_MODE(bond) == BOND_MODE_XOR ||
289 bond_is_nondyn_tlb(bond));
290}
291
292static inline bool bond_mode_uses_arp(int mode)
293{
294 return mode != BOND_MODE_8023AD && mode != BOND_MODE_TLB &&
295 mode != BOND_MODE_ALB;
296}
297
298static inline bool bond_mode_uses_primary(int mode)
299{
300 return mode == BOND_MODE_ACTIVEBACKUP || mode == BOND_MODE_TLB ||
301 mode == BOND_MODE_ALB;
302}
303
304static inline bool bond_uses_primary(struct bonding *bond)
305{
306 return bond_mode_uses_primary(BOND_MODE(bond));
307}
308
309static inline struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond)
310{
311 struct slave *slave = rcu_dereference(bond->curr_active_slave);
312
313 return bond_uses_primary(bond) && slave ? slave->dev : NULL;
314}
315
316static inline bool bond_slave_is_up(struct slave *slave)
317{
318 return netif_running(slave->dev) && netif_carrier_ok(slave->dev);
319}
320
321static inline void bond_set_active_slave(struct slave *slave)
322{
323 if (slave->backup) {
324 slave->backup = 0;
325 bond_queue_slave_event(slave);
326 bond_lower_state_changed(slave);
327 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
328 }
329}
330
331static inline void bond_set_backup_slave(struct slave *slave)
332{
333 if (!slave->backup) {
334 slave->backup = 1;
335 bond_queue_slave_event(slave);
336 bond_lower_state_changed(slave);
337 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
338 }
339}
340
341static inline void bond_set_slave_state(struct slave *slave,
342 int slave_state, bool notify)
343{
344 if (slave->backup == slave_state)
345 return;
346
347 slave->backup = slave_state;
348 if (notify) {
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;
353 } else {
354 if (slave->should_notify)
355 slave->should_notify = 0;
356 else
357 slave->should_notify = 1;
358 }
359}
360
361static inline void bond_slave_state_change(struct bonding *bond)
362{
363 struct list_head *iter;
364 struct slave *tmp;
365
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);
371 }
372}
373
374static inline void bond_slave_state_notify(struct bonding *bond)
375{
376 struct list_head *iter;
377 struct slave *tmp;
378
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;
384 }
385 }
386}
387
388static inline int bond_slave_state(struct slave *slave)
389{
390 return slave->backup;
391}
392
393static inline bool bond_is_active_slave(struct slave *slave)
394{
395 return !bond_slave_state(slave);
396}
397
398static inline bool bond_slave_can_tx(struct slave *slave)
399{
400 return bond_slave_is_up(slave) && slave->link == BOND_LINK_UP &&
401 bond_is_active_slave(slave);
402}
403
404#define BOND_PRI_RESELECT_ALWAYS 0
405#define BOND_PRI_RESELECT_BETTER 1
406#define BOND_PRI_RESELECT_FAILURE 2
407
408#define BOND_FOM_NONE 0
409#define BOND_FOM_ACTIVE 1
410#define BOND_FOM_FOLLOW 2
411
412#define BOND_ARP_TARGETS_ANY 0
413#define BOND_ARP_TARGETS_ALL 1
414
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 | \
422 BOND_ARP_FILTER)
423#define BOND_ARP_FILTER_BACKUP (BOND_ARP_VALIDATE_BACKUP | \
424 BOND_ARP_FILTER)
425
426#define BOND_SLAVE_NOTIFY_NOW true
427#define BOND_SLAVE_NOTIFY_LATER false
428
429static inline int slave_do_arp_validate(struct bonding *bond,
430 struct slave *slave)
431{
432 return bond->params.arp_validate & (1 << bond_slave_state(slave));
433}
434
435static inline int slave_do_arp_validate_only(struct bonding *bond)
436{
437 return bond->params.arp_validate & BOND_ARP_FILTER;
438}
439
440static inline int bond_is_ip_target_ok(__be32 addr)
441{
442 return !ipv4_is_lbcast(addr) && !ipv4_is_zeronet(addr);
443}
444
445/* Get the oldest arp which we've received on this slave for bond's
446 * arp_targets.
447 */
448static inline unsigned long slave_oldest_target_arp_rx(struct bonding *bond,
449 struct slave *slave)
450{
451 int i = 1;
452 unsigned long ret = slave->target_last_arp_rx[0];
453
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];
457
458 return ret;
459}
460
461static inline unsigned long slave_last_rx(struct bonding *bond,
462 struct slave *slave)
463{
464 if (bond->params.arp_all_targets == BOND_ARP_TARGETS_ALL)
465 return slave_oldest_target_arp_rx(bond, slave);
466
467 return slave->last_rx;
468}
469
470#ifdef CONFIG_NET_POLL_CONTROLLER
471static inline void bond_netpoll_send_skb(const struct slave *slave,
472 struct sk_buff *skb)
473{
474 struct netpoll *np = slave->np;
475
476 if (np)
477 netpoll_send_skb(np, skb);
478}
479#else
480static inline void bond_netpoll_send_skb(const struct slave *slave,
481 struct sk_buff *skb)
482{
483}
484#endif
485
486static inline void bond_set_slave_inactive_flags(struct slave *slave,
487 bool notify)
488{
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)
492 slave->inactive = 1;
493}
494
495static inline void bond_set_slave_active_flags(struct slave *slave,
496 bool notify)
497{
498 bond_set_slave_state(slave, BOND_STATE_ACTIVE, notify);
499 slave->inactive = 0;
500}
501
502static inline bool bond_is_slave_inactive(struct slave *slave)
503{
504 return slave->inactive;
505}
506
507static inline void bond_set_slave_link_state(struct slave *slave, int state,
508 bool notify)
509{
510 if (slave->link == state)
511 return;
512
513 slave->link = state;
514 if (notify) {
515 bond_queue_slave_event(slave);
516 bond_lower_state_changed(slave);
517 slave->should_notify_link = 0;
518 } else {
519 if (slave->should_notify_link)
520 slave->should_notify_link = 0;
521 else
522 slave->should_notify_link = 1;
523 }
524}
525
526static inline void bond_slave_link_notify(struct bonding *bond)
527{
528 struct list_head *iter;
529 struct slave *tmp;
530
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;
536 }
537 }
538}
539
540static inline __be32 bond_confirm_addr(struct net_device *dev, __be32 dst, __be32 local)
541{
542 struct in_device *in_dev;
543 __be32 addr = 0;
544
545 rcu_read_lock();
546 in_dev = __in_dev_get_rcu(dev);
547
548 if (in_dev)
549 addr = inet_confirm_addr(dev_net(dev), in_dev, dst, local,
550 RT_SCOPE_HOST);
551 rcu_read_unlock();
552 return addr;
553}
554
555struct bond_net {
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;
560#endif
561 struct class_attribute class_attr_bonding_masters;
562};
563
564int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, struct slave *slave);
565void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev);
566int bond_create(struct net *net, const char *name);
567int bond_create_sysfs(struct bond_net *net);
568void bond_destroy_sysfs(struct bond_net *net);
569void bond_prepare_sysfs_group(struct bonding *bond);
570int bond_sysfs_slave_add(struct slave *slave);
571void bond_sysfs_slave_del(struct slave *slave);
572int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev);
573int bond_release(struct net_device *bond_dev, struct net_device *slave_dev);
574u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb);
575int bond_set_carrier(struct bonding *bond);
576void bond_select_active_slave(struct bonding *bond);
577void bond_change_active_slave(struct bonding *bond, struct slave *new_active);
578void bond_create_debugfs(void);
579void bond_destroy_debugfs(void);
580void bond_debug_register(struct bonding *bond);
581void bond_debug_unregister(struct bonding *bond);
582void bond_debug_reregister(struct bonding *bond);
583const char *bond_mode_name(int mode);
584void bond_setup(struct net_device *bond_dev);
585unsigned int bond_get_num_tx_queues(void);
586int bond_netlink_init(void);
587void bond_netlink_fini(void);
588struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond);
589const char *bond_slave_link_status(s8 link);
590struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
591 struct net_device *end_dev,
592 int level);
593int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave);
594void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay);
595
596#ifdef CONFIG_PROC_FS
597void bond_create_proc_entry(struct bonding *bond);
598void bond_remove_proc_entry(struct bonding *bond);
599void bond_create_proc_dir(struct bond_net *bn);
600void bond_destroy_proc_dir(struct bond_net *bn);
601#else
602static inline void bond_create_proc_entry(struct bonding *bond)
603{
604}
605
606static inline void bond_remove_proc_entry(struct bonding *bond)
607{
608}
609
610static inline void bond_create_proc_dir(struct bond_net *bn)
611{
612}
613
614static inline void bond_destroy_proc_dir(struct bond_net *bn)
615{
616}
617#endif
618
619static inline struct slave *bond_slave_has_mac(struct bonding *bond,
620 const u8 *mac)
621{
622 struct list_head *iter;
623 struct slave *tmp;
624
625 bond_for_each_slave(bond, tmp, iter)
626 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
627 return tmp;
628
629 return NULL;
630}
631
632/* Caller must hold rcu_read_lock() for read */
633static inline struct slave *bond_slave_has_mac_rcu(struct bonding *bond,
634 const u8 *mac)
635{
636 struct list_head *iter;
637 struct slave *tmp;
638
639 bond_for_each_slave_rcu(bond, tmp, iter)
640 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
641 return tmp;
642
643 return NULL;
644}
645
646/* Caller must hold rcu_read_lock() for read */
647static inline bool bond_slave_has_mac_rx(struct bonding *bond, const u8 *mac)
648{
649 struct list_head *iter;
650 struct slave *tmp;
651 struct netdev_hw_addr *ha;
652
653 bond_for_each_slave_rcu(bond, tmp, iter)
654 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
655 return true;
656
657 if (netdev_uc_empty(bond->dev))
658 return false;
659
660 netdev_for_each_uc_addr(ha, bond->dev)
661 if (ether_addr_equal_64bits(mac, ha->addr))
662 return true;
663
664 return false;
665}
666
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
669 */
670static inline int bond_get_targets_ip(__be32 *targets, __be32 ip)
671{
672 int i;
673
674 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
675 if (targets[i] == ip)
676 return i;
677 else if (targets[i] == 0)
678 break;
679
680 return -1;
681}
682
683/* exported from bond_main.c */
684extern unsigned int bond_net_id;
685extern const struct bond_parm_tbl bond_lacp_tbl[];
686extern const struct bond_parm_tbl xmit_hashtype_tbl[];
687extern const struct bond_parm_tbl arp_validate_tbl[];
688extern const struct bond_parm_tbl arp_all_targets_tbl[];
689extern const struct bond_parm_tbl fail_over_mac_tbl[];
690extern const struct bond_parm_tbl pri_reselect_tbl[];
691extern struct bond_parm_tbl ad_select_tbl[];
692
693/* exported from bond_netlink.c */
694extern struct rtnl_link_ops bond_link_ops;
695
696static inline void bond_tx_drop(struct net_device *dev, struct sk_buff *skb)
697{
698 atomic_long_inc(&dev->tx_dropped);
699 dev_kfree_skb_any(skb);
700}
701
702#endif /* _NET_BONDING_H */