bonding: allow notifications for bond_set_slave_link_state
[deliverable/linux.git] / include / net / bonding.h
... / ...
CommitLineData
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 u8 send_peer_notif;
219 u8 igmp_retrans;
220#ifdef CONFIG_PROC_FS
221 struct proc_dir_entry *proc_entry;
222 char proc_file_name[IFNAMSIZ];
223#endif /* CONFIG_PROC_FS */
224 struct list_head bond_list;
225 u32 rr_tx_counter;
226 struct ad_bond_info ad_info;
227 struct alb_bond_info alb_info;
228 struct bond_params params;
229 struct workqueue_struct *wq;
230 struct delayed_work mii_work;
231 struct delayed_work arp_work;
232 struct delayed_work alb_work;
233 struct delayed_work ad_work;
234 struct delayed_work mcast_work;
235 struct delayed_work slave_arr_work;
236#ifdef CONFIG_DEBUG_FS
237 /* debugging support via debugfs */
238 struct dentry *debug_dir;
239#endif /* CONFIG_DEBUG_FS */
240 struct rtnl_link_stats64 bond_stats;
241};
242
243#define bond_slave_get_rcu(dev) \
244 ((struct slave *) rcu_dereference(dev->rx_handler_data))
245
246#define bond_slave_get_rtnl(dev) \
247 ((struct slave *) rtnl_dereference(dev->rx_handler_data))
248
249void bond_queue_slave_event(struct slave *slave);
250
251struct bond_vlan_tag {
252 __be16 vlan_proto;
253 unsigned short vlan_id;
254};
255
256/**
257 * Returns NULL if the net_device does not belong to any of the bond's slaves
258 *
259 * Caller must hold bond lock for read
260 */
261static inline struct slave *bond_get_slave_by_dev(struct bonding *bond,
262 struct net_device *slave_dev)
263{
264 return netdev_lower_dev_get_private(bond->dev, slave_dev);
265}
266
267static inline struct bonding *bond_get_bond_by_slave(struct slave *slave)
268{
269 return slave->bond;
270}
271
272static inline bool bond_should_override_tx_queue(struct bonding *bond)
273{
274 return BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
275 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN;
276}
277
278static inline bool bond_is_lb(const struct bonding *bond)
279{
280 return BOND_MODE(bond) == BOND_MODE_TLB ||
281 BOND_MODE(bond) == BOND_MODE_ALB;
282}
283
284static inline bool bond_is_nondyn_tlb(const struct bonding *bond)
285{
286 return (BOND_MODE(bond) == BOND_MODE_TLB) &&
287 (bond->params.tlb_dynamic_lb == 0);
288}
289
290static inline bool bond_mode_uses_xmit_hash(const struct bonding *bond)
291{
292 return (BOND_MODE(bond) == BOND_MODE_8023AD ||
293 BOND_MODE(bond) == BOND_MODE_XOR ||
294 bond_is_nondyn_tlb(bond));
295}
296
297static inline bool bond_mode_uses_arp(int mode)
298{
299 return mode != BOND_MODE_8023AD && mode != BOND_MODE_TLB &&
300 mode != BOND_MODE_ALB;
301}
302
303static inline bool bond_mode_uses_primary(int mode)
304{
305 return mode == BOND_MODE_ACTIVEBACKUP || mode == BOND_MODE_TLB ||
306 mode == BOND_MODE_ALB;
307}
308
309static inline bool bond_uses_primary(struct bonding *bond)
310{
311 return bond_mode_uses_primary(BOND_MODE(bond));
312}
313
314static inline struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond)
315{
316 struct slave *slave = rcu_dereference(bond->curr_active_slave);
317
318 return bond_uses_primary(bond) && slave ? slave->dev : NULL;
319}
320
321static inline bool bond_slave_is_up(struct slave *slave)
322{
323 return netif_running(slave->dev) && netif_carrier_ok(slave->dev);
324}
325
326static inline void bond_set_active_slave(struct slave *slave)
327{
328 if (slave->backup) {
329 slave->backup = 0;
330 bond_queue_slave_event(slave);
331 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
332 }
333}
334
335static inline void bond_set_backup_slave(struct slave *slave)
336{
337 if (!slave->backup) {
338 slave->backup = 1;
339 bond_queue_slave_event(slave);
340 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
341 }
342}
343
344static inline void bond_set_slave_state(struct slave *slave,
345 int slave_state, bool notify)
346{
347 if (slave->backup == slave_state)
348 return;
349
350 slave->backup = slave_state;
351 if (notify) {
352 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
353 bond_queue_slave_event(slave);
354 slave->should_notify = 0;
355 } else {
356 if (slave->should_notify)
357 slave->should_notify = 0;
358 else
359 slave->should_notify = 1;
360 }
361}
362
363static inline void bond_slave_state_change(struct bonding *bond)
364{
365 struct list_head *iter;
366 struct slave *tmp;
367
368 bond_for_each_slave(bond, tmp, iter) {
369 if (tmp->link == BOND_LINK_UP)
370 bond_set_active_slave(tmp);
371 else if (tmp->link == BOND_LINK_DOWN)
372 bond_set_backup_slave(tmp);
373 }
374}
375
376static inline void bond_slave_state_notify(struct bonding *bond)
377{
378 struct list_head *iter;
379 struct slave *tmp;
380
381 bond_for_each_slave(bond, tmp, iter) {
382 if (tmp->should_notify) {
383 rtmsg_ifinfo(RTM_NEWLINK, tmp->dev, 0, GFP_ATOMIC);
384 tmp->should_notify = 0;
385 }
386 }
387}
388
389static inline int bond_slave_state(struct slave *slave)
390{
391 return slave->backup;
392}
393
394static inline bool bond_is_active_slave(struct slave *slave)
395{
396 return !bond_slave_state(slave);
397}
398
399static inline bool bond_slave_can_tx(struct slave *slave)
400{
401 return bond_slave_is_up(slave) && slave->link == BOND_LINK_UP &&
402 bond_is_active_slave(slave);
403}
404
405#define BOND_PRI_RESELECT_ALWAYS 0
406#define BOND_PRI_RESELECT_BETTER 1
407#define BOND_PRI_RESELECT_FAILURE 2
408
409#define BOND_FOM_NONE 0
410#define BOND_FOM_ACTIVE 1
411#define BOND_FOM_FOLLOW 2
412
413#define BOND_ARP_TARGETS_ANY 0
414#define BOND_ARP_TARGETS_ALL 1
415
416#define BOND_ARP_VALIDATE_NONE 0
417#define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE)
418#define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP)
419#define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \
420 BOND_ARP_VALIDATE_BACKUP)
421#define BOND_ARP_FILTER (BOND_ARP_VALIDATE_ALL + 1)
422#define BOND_ARP_FILTER_ACTIVE (BOND_ARP_VALIDATE_ACTIVE | \
423 BOND_ARP_FILTER)
424#define BOND_ARP_FILTER_BACKUP (BOND_ARP_VALIDATE_BACKUP | \
425 BOND_ARP_FILTER)
426
427#define BOND_SLAVE_NOTIFY_NOW true
428#define BOND_SLAVE_NOTIFY_LATER false
429
430static inline int slave_do_arp_validate(struct bonding *bond,
431 struct slave *slave)
432{
433 return bond->params.arp_validate & (1 << bond_slave_state(slave));
434}
435
436static inline int slave_do_arp_validate_only(struct bonding *bond)
437{
438 return bond->params.arp_validate & BOND_ARP_FILTER;
439}
440
441static inline int bond_is_ip_target_ok(__be32 addr)
442{
443 return !ipv4_is_lbcast(addr) && !ipv4_is_zeronet(addr);
444}
445
446/* Get the oldest arp which we've received on this slave for bond's
447 * arp_targets.
448 */
449static inline unsigned long slave_oldest_target_arp_rx(struct bonding *bond,
450 struct slave *slave)
451{
452 int i = 1;
453 unsigned long ret = slave->target_last_arp_rx[0];
454
455 for (; (i < BOND_MAX_ARP_TARGETS) && bond->params.arp_targets[i]; i++)
456 if (time_before(slave->target_last_arp_rx[i], ret))
457 ret = slave->target_last_arp_rx[i];
458
459 return ret;
460}
461
462static inline unsigned long slave_last_rx(struct bonding *bond,
463 struct slave *slave)
464{
465 if (bond->params.arp_all_targets == BOND_ARP_TARGETS_ALL)
466 return slave_oldest_target_arp_rx(bond, slave);
467
468 return slave->last_rx;
469}
470
471#ifdef CONFIG_NET_POLL_CONTROLLER
472static inline void bond_netpoll_send_skb(const struct slave *slave,
473 struct sk_buff *skb)
474{
475 struct netpoll *np = slave->np;
476
477 if (np)
478 netpoll_send_skb(np, skb);
479}
480#else
481static inline void bond_netpoll_send_skb(const struct slave *slave,
482 struct sk_buff *skb)
483{
484}
485#endif
486
487static inline void bond_set_slave_inactive_flags(struct slave *slave,
488 bool notify)
489{
490 if (!bond_is_lb(slave->bond))
491 bond_set_slave_state(slave, BOND_STATE_BACKUP, notify);
492 if (!slave->bond->params.all_slaves_active)
493 slave->inactive = 1;
494}
495
496static inline void bond_set_slave_active_flags(struct slave *slave,
497 bool notify)
498{
499 bond_set_slave_state(slave, BOND_STATE_ACTIVE, notify);
500 slave->inactive = 0;
501}
502
503static inline bool bond_is_slave_inactive(struct slave *slave)
504{
505 return slave->inactive;
506}
507
508static inline void bond_set_slave_link_state(struct slave *slave, int state,
509 bool notify)
510{
511 if (slave->link == state)
512 return;
513
514 slave->link = state;
515 if (notify) {
516 bond_queue_slave_event(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 tmp->should_notify_link = 0;
535 }
536 }
537}
538
539static inline __be32 bond_confirm_addr(struct net_device *dev, __be32 dst, __be32 local)
540{
541 struct in_device *in_dev;
542 __be32 addr = 0;
543
544 rcu_read_lock();
545 in_dev = __in_dev_get_rcu(dev);
546
547 if (in_dev)
548 addr = inet_confirm_addr(dev_net(dev), in_dev, dst, local,
549 RT_SCOPE_HOST);
550 rcu_read_unlock();
551 return addr;
552}
553
554struct bond_net {
555 struct net *net; /* Associated network namespace */
556 struct list_head dev_list;
557#ifdef CONFIG_PROC_FS
558 struct proc_dir_entry *proc_dir;
559#endif
560 struct class_attribute class_attr_bonding_masters;
561};
562
563int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, struct slave *slave);
564void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev);
565int bond_create(struct net *net, const char *name);
566int bond_create_sysfs(struct bond_net *net);
567void bond_destroy_sysfs(struct bond_net *net);
568void bond_prepare_sysfs_group(struct bonding *bond);
569int bond_sysfs_slave_add(struct slave *slave);
570void bond_sysfs_slave_del(struct slave *slave);
571int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev);
572int bond_release(struct net_device *bond_dev, struct net_device *slave_dev);
573u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb);
574int bond_set_carrier(struct bonding *bond);
575void bond_select_active_slave(struct bonding *bond);
576void bond_change_active_slave(struct bonding *bond, struct slave *new_active);
577void bond_create_debugfs(void);
578void bond_destroy_debugfs(void);
579void bond_debug_register(struct bonding *bond);
580void bond_debug_unregister(struct bonding *bond);
581void bond_debug_reregister(struct bonding *bond);
582const char *bond_mode_name(int mode);
583void bond_setup(struct net_device *bond_dev);
584unsigned int bond_get_num_tx_queues(void);
585int bond_netlink_init(void);
586void bond_netlink_fini(void);
587struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond);
588const char *bond_slave_link_status(s8 link);
589struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
590 struct net_device *end_dev,
591 int level);
592int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave);
593void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay);
594
595#ifdef CONFIG_PROC_FS
596void bond_create_proc_entry(struct bonding *bond);
597void bond_remove_proc_entry(struct bonding *bond);
598void bond_create_proc_dir(struct bond_net *bn);
599void bond_destroy_proc_dir(struct bond_net *bn);
600#else
601static inline void bond_create_proc_entry(struct bonding *bond)
602{
603}
604
605static inline void bond_remove_proc_entry(struct bonding *bond)
606{
607}
608
609static inline void bond_create_proc_dir(struct bond_net *bn)
610{
611}
612
613static inline void bond_destroy_proc_dir(struct bond_net *bn)
614{
615}
616#endif
617
618static inline struct slave *bond_slave_has_mac(struct bonding *bond,
619 const u8 *mac)
620{
621 struct list_head *iter;
622 struct slave *tmp;
623
624 bond_for_each_slave(bond, tmp, iter)
625 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
626 return tmp;
627
628 return NULL;
629}
630
631/* Caller must hold rcu_read_lock() for read */
632static inline struct slave *bond_slave_has_mac_rcu(struct bonding *bond,
633 const u8 *mac)
634{
635 struct list_head *iter;
636 struct slave *tmp;
637
638 bond_for_each_slave_rcu(bond, tmp, iter)
639 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
640 return tmp;
641
642 return NULL;
643}
644
645/* Caller must hold rcu_read_lock() for read */
646static inline bool bond_slave_has_mac_rx(struct bonding *bond, const u8 *mac)
647{
648 struct list_head *iter;
649 struct slave *tmp;
650 struct netdev_hw_addr *ha;
651
652 bond_for_each_slave_rcu(bond, tmp, iter)
653 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
654 return true;
655
656 if (netdev_uc_empty(bond->dev))
657 return false;
658
659 netdev_for_each_uc_addr(ha, bond->dev)
660 if (ether_addr_equal_64bits(mac, ha->addr))
661 return true;
662
663 return false;
664}
665
666/* Check if the ip is present in arp ip list, or first free slot if ip == 0
667 * Returns -1 if not found, index if found
668 */
669static inline int bond_get_targets_ip(__be32 *targets, __be32 ip)
670{
671 int i;
672
673 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
674 if (targets[i] == ip)
675 return i;
676 else if (targets[i] == 0)
677 break;
678
679 return -1;
680}
681
682/* exported from bond_main.c */
683extern int bond_net_id;
684extern const struct bond_parm_tbl bond_lacp_tbl[];
685extern const struct bond_parm_tbl xmit_hashtype_tbl[];
686extern const struct bond_parm_tbl arp_validate_tbl[];
687extern const struct bond_parm_tbl arp_all_targets_tbl[];
688extern const struct bond_parm_tbl fail_over_mac_tbl[];
689extern const struct bond_parm_tbl pri_reselect_tbl[];
690extern struct bond_parm_tbl ad_select_tbl[];
691
692/* exported from bond_netlink.c */
693extern struct rtnl_link_ops bond_link_ops;
694
695static inline void bond_tx_drop(struct net_device *dev, struct sk_buff *skb)
696{
697 atomic_long_inc(&dev->tx_dropped);
698 dev_kfree_skb_any(skb);
699}
700
701#endif /* _NET_BONDING_H */
This page took 0.030965 seconds and 5 git commands to generate.