blob: 4feeb2d650a4f478bc6ab4f3c70db3f3563c622d [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 */
14
15#ifndef _LINUX_BONDING_H
16#define _LINUX_BONDING_H
17
18#include <linux/timer.h>
19#include <linux/proc_fs.h>
20#include <linux/if_bonding.h>
Mitch Williamsb76cdba2005-11-09 10:36:41 -080021#include <linux/kobject.h>
Neil Hormane843fa52010-10-13 16:01:50 +000022#include <linux/cpumask.h>
Brian Haley305d5522008-11-04 17:51:14 -080023#include <linux/in6.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "bond_3ad.h"
25#include "bond_alb.h"
26
Andy Gospodarekbb1d9122010-06-02 08:40:18 +000027#define DRV_VERSION "3.7.0"
28#define DRV_RELDATE "June 2, 2010"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#define DRV_NAME "bonding"
30#define DRV_DESCRIPTION "Ethernet Channel Bonding Driver"
31
32#define BOND_MAX_ARP_TARGETS 16
33
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#define IS_UP(dev) \
35 ((((dev)->flags & IFF_UP) == IFF_UP) && \
36 netif_running(dev) && \
37 netif_carrier_ok(dev))
38
39/*
40 * Checks whether bond is ready for transmit.
41 *
42 * Caller must hold bond->lock
43 */
44#define BOND_IS_OK(bond) \
45 (((bond)->dev->flags & IFF_UP) && \
46 netif_running((bond)->dev) && \
47 ((bond)->slave_cnt > 0))
48
49/*
50 * Checks whether slave is ready for transmit.
51 */
52#define SLAVE_IS_OK(slave) \
53 (((slave)->dev->flags & IFF_UP) && \
54 netif_running((slave)->dev) && \
55 ((slave)->link == BOND_LINK_UP) && \
56 ((slave)->state == BOND_STATE_ACTIVE))
57
58
59#define USES_PRIMARY(mode) \
60 (((mode) == BOND_MODE_ACTIVEBACKUP) || \
61 ((mode) == BOND_MODE_TLB) || \
62 ((mode) == BOND_MODE_ALB))
63
Andy Gospodarekbb1d9122010-06-02 08:40:18 +000064#define TX_QUEUE_OVERRIDE(mode) \
65 (((mode) == BOND_MODE_ACTIVEBACKUP) || \
66 ((mode) == BOND_MODE_ROUNDROBIN))
Linus Torvalds1da177e2005-04-16 15:20:36 -070067/*
68 * Less bad way to call ioctl from within the kernel; this needs to be
69 * done some other way to get the call out of interrupt context.
70 * Needs "ioctl" variable to be supplied by calling context.
71 */
72#define IOCTL(dev, arg, cmd) ({ \
73 int res = 0; \
74 mm_segment_t fs = get_fs(); \
75 set_fs(get_ds()); \
76 res = ioctl(dev, arg, cmd); \
77 set_fs(fs); \
78 res; })
79
80/**
81 * bond_for_each_slave_from - iterate the slaves list from a starting point
82 * @bond: the bond holding this list.
83 * @pos: current slave.
84 * @cnt: counter for max number of moves
85 * @start: starting point.
86 *
87 * Caller must hold bond->lock
88 */
89#define bond_for_each_slave_from(bond, pos, cnt, start) \
90 for (cnt = 0, pos = start; \
91 cnt < (bond)->slave_cnt; \
92 cnt++, pos = (pos)->next)
93
94/**
95 * bond_for_each_slave_from_to - iterate the slaves list from start point to stop point
96 * @bond: the bond holding this list.
97 * @pos: current slave.
98 * @cnt: counter for number max of moves
99 * @start: start point.
100 * @stop: stop point.
101 *
102 * Caller must hold bond->lock
103 */
104#define bond_for_each_slave_from_to(bond, pos, cnt, start, stop) \
105 for (cnt = 0, pos = start; \
106 ((cnt < (bond)->slave_cnt) && (pos != (stop)->next)); \
107 cnt++, pos = (pos)->next)
108
109/**
110 * bond_for_each_slave - iterate the slaves list from head
111 * @bond: the bond holding this list.
112 * @pos: current slave.
113 * @cnt: counter for max number of moves
114 *
115 * Caller must hold bond->lock
116 */
117#define bond_for_each_slave(bond, pos, cnt) \
118 bond_for_each_slave_from(bond, pos, cnt, (bond)->first_slave)
119
120
Neil Hormane843fa52010-10-13 16:01:50 +0000121#ifdef CONFIG_NET_POLL_CONTROLLER
Neil Hormanfb4fa762010-12-06 09:05:50 +0000122extern atomic_t netpoll_block_tx;
Neil Hormane843fa52010-10-13 16:01:50 +0000123
124static inline void block_netpoll_tx(void)
125{
Neil Hormanfb4fa762010-12-06 09:05:50 +0000126 atomic_inc(&netpoll_block_tx);
Neil Hormane843fa52010-10-13 16:01:50 +0000127}
128
129static inline void unblock_netpoll_tx(void)
130{
Neil Hormanfb4fa762010-12-06 09:05:50 +0000131 atomic_dec(&netpoll_block_tx);
Neil Hormane843fa52010-10-13 16:01:50 +0000132}
133
134static inline int is_netpoll_tx_blocked(struct net_device *dev)
135{
136 if (unlikely(dev->priv_flags & IFF_IN_NETPOLL))
Neil Hormanfb4fa762010-12-06 09:05:50 +0000137 return atomic_read(&netpoll_block_tx);
Neil Hormane843fa52010-10-13 16:01:50 +0000138 return 0;
139}
140#else
141#define block_netpoll_tx()
142#define unblock_netpoll_tx()
143#define is_netpoll_tx_blocked(dev) (0)
144#endif
145
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146struct bond_params {
147 int mode;
Jay Vosburgh169a3e62005-06-26 17:54:11 -0400148 int xmit_policy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 int miimon;
Moni Shoua7893b242008-05-17 21:10:12 -0700150 int num_grat_arp;
Brian Haley305d5522008-11-04 17:51:14 -0800151 int num_unsol_na;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 int arp_interval;
Jay Vosburghf5b2b962006-09-22 21:54:53 -0700153 int arp_validate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 int use_carrier;
Jay Vosburghdd957c52007-10-09 19:57:24 -0700155 int fail_over_mac;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 int updelay;
157 int downdelay;
158 int lacp_fast;
Jay Vosburghfd989c82008-11-04 17:51:16 -0800159 int ad_select;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 char primary[IFNAMSIZ];
Jiri Pirkoa5499522009-09-25 03:28:09 +0000161 int primary_reselect;
Al Virod3bb52b2007-08-22 20:06:58 -0400162 __be32 arp_targets[BOND_MAX_ARP_TARGETS];
Andy Gospodarekbb1d9122010-06-02 08:40:18 +0000163 int tx_queues;
Andy Gospodarekebd8e492010-06-02 08:39:21 +0000164 int all_slaves_active;
Flavio Leitnerc2952c32010-10-05 14:23:59 +0000165 int resend_igmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166};
167
Mitch Williams12479f92005-11-09 10:35:44 -0800168struct bond_parm_tbl {
169 char *modename;
170 int mode;
171};
172
Jay Vosburghece95f72008-01-17 16:25:01 -0800173#define BOND_MAX_MODENAME_LEN 20
174
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175struct vlan_entry {
176 struct list_head vlan_list;
Al Virod3bb52b2007-08-22 20:06:58 -0400177 __be32 vlan_ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 unsigned short vlan_id;
Brian Haley305d5522008-11-04 17:51:14 -0800179#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
180 struct in6_addr vlan_ipv6;
181#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182};
183
184struct slave {
Mitch Williamse944ef72005-11-09 10:36:50 -0800185 struct net_device *dev; /* first - useful for panic debug */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 struct slave *next;
187 struct slave *prev;
Jay Vosburgh8bb5f962006-09-22 21:52:51 -0700188 int delay;
Andy Gospodarekf8a8ccd2007-01-29 12:08:38 -0800189 unsigned long jiffies;
190 unsigned long last_arp_rx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 s8 link; /* one of BOND_LINK_XXXX */
Jay Vosburghb2220ca2008-05-17 21:10:13 -0700192 s8 new_link;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 s8 state; /* one of BOND_STATE_XXXX */
Moni Shoua3158bf72007-10-09 19:43:41 -0700194 u32 original_mtu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 u32 link_failure_count;
Eric Dumazet885a1362009-09-01 06:31:18 +0000196 u8 perm_hwaddr[ETH_ALEN];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 u16 speed;
198 u8 duplex;
Andy Gospodarekbb1d9122010-06-02 08:40:18 +0000199 u16 queue_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 struct ad_slave_info ad_info; /* HUGE - better to dynamically alloc */
201 struct tlb_slave_info tlb_info;
202};
203
204/*
Jay Vosburghb2220ca2008-05-17 21:10:13 -0700205 * Link pseudo-state only used internally by monitors
206 */
207#define BOND_LINK_NOCHANGE -1
208
209/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 * Here are the locking policies for the two bonding locks:
211 *
212 * 1) Get bond->lock when reading/writing slave list.
213 * 2) Get bond->curr_slave_lock when reading/writing bond->curr_active_slave.
214 * (It is unnecessary when the write-lock is put with bond->lock.)
215 * 3) When we lock with bond->curr_slave_lock, we must lock with bond->lock
216 * beforehand.
217 */
218struct bonding {
Mitch Williamse944ef72005-11-09 10:36:50 -0800219 struct net_device *dev; /* first - useful for panic debug */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 struct slave *first_slave;
221 struct slave *curr_active_slave;
222 struct slave *current_arp_slave;
223 struct slave *primary_slave;
Jiri Pirkoa5499522009-09-25 03:28:09 +0000224 bool force_primary;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 s32 slave_cnt; /* never change this value outside the attach/detach wrappers */
226 rwlock_t lock;
227 rwlock_t curr_slave_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 s8 kill_timers;
Moni Shoua1053f622007-10-09 19:43:42 -0700229 s8 send_grat_arp;
Brian Haley305d5522008-11-04 17:51:14 -0800230 s8 send_unsol_na;
Moni Shouad90a1622007-10-09 19:43:43 -0700231 s8 setup_by_slave;
Flavio Leitnerc2952c32010-10-05 14:23:59 +0000232 s8 igmp_retrans;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233#ifdef CONFIG_PROC_FS
234 struct proc_dir_entry *proc_entry;
235 char proc_file_name[IFNAMSIZ];
236#endif /* CONFIG_PROC_FS */
237 struct list_head bond_list;
Jiri Pirko22bedad32010-04-01 21:22:57 +0000238 struct netdev_hw_addr_list mc_list;
Jasper Spaansa361c832009-10-23 04:09:24 +0000239 int (*xmit_hash_policy)(struct sk_buff *, int);
Al Virod3bb52b2007-08-22 20:06:58 -0400240 __be32 master_ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 u16 flags;
Jay Vosburghcf5f9042007-10-17 17:37:47 -0700242 u16 rr_tx_counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 struct ad_bond_info ad_info;
244 struct alb_bond_info alb_info;
245 struct bond_params params;
246 struct list_head vlan_list;
247 struct vlan_group *vlgrp;
Jay Vosburghf5b2b962006-09-22 21:54:53 -0700248 struct packet_type arp_mon_pt;
Jay Vosburgh1b76b312007-10-17 17:37:45 -0700249 struct workqueue_struct *wq;
250 struct delayed_work mii_work;
251 struct delayed_work arp_work;
252 struct delayed_work alb_work;
253 struct delayed_work ad_work;
Flavio Leitner5a37e8c2010-10-05 14:23:57 +0000254 struct delayed_work mcast_work;
Brian Haley305d5522008-11-04 17:51:14 -0800255#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
256 struct in6_addr master_ipv6;
257#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258};
259
260/**
261 * Returns NULL if the net_device does not belong to any of the bond's slaves
262 *
263 * Caller must hold bond lock for read
264 */
Adrian Bunkcceb9042006-01-05 22:45:42 -0800265static inline struct slave *bond_get_slave_by_dev(struct bonding *bond, struct net_device *slave_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266{
267 struct slave *slave = NULL;
268 int i;
269
270 bond_for_each_slave(bond, slave, i) {
271 if (slave->dev == slave_dev) {
Hillf Dantonaf3e5bd2010-12-10 18:54:11 +0000272 return slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 }
274 }
275
Hillf Dantonaf3e5bd2010-12-10 18:54:11 +0000276 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277}
278
Adrian Bunkcceb9042006-01-05 22:45:42 -0800279static inline struct bonding *bond_get_bond_by_slave(struct slave *slave)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280{
281 if (!slave || !slave->dev->master) {
282 return NULL;
283 }
284
Wang Chen454d7c92008-11-12 23:37:49 -0800285 return (struct bonding *)netdev_priv(slave->dev->master);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286}
287
Holger Eitzenberger58402052008-12-09 23:07:13 -0800288static inline bool bond_is_lb(const struct bonding *bond)
289{
Joe Perches8e95a202009-12-03 07:58:21 +0000290 return (bond->params.mode == BOND_MODE_TLB ||
291 bond->params.mode == BOND_MODE_ALB);
Holger Eitzenberger58402052008-12-09 23:07:13 -0800292}
293
Jiri Pirkoa5499522009-09-25 03:28:09 +0000294#define BOND_PRI_RESELECT_ALWAYS 0
295#define BOND_PRI_RESELECT_BETTER 1
296#define BOND_PRI_RESELECT_FAILURE 2
297
Jay Vosburgh3915c1e82008-05-17 21:10:14 -0700298#define BOND_FOM_NONE 0
299#define BOND_FOM_ACTIVE 1
300#define BOND_FOM_FOLLOW 2
301
Jay Vosburghf5b2b962006-09-22 21:54:53 -0700302#define BOND_ARP_VALIDATE_NONE 0
303#define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE)
304#define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP)
305#define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \
306 BOND_ARP_VALIDATE_BACKUP)
307
Adrian Bunk079ca7d2006-12-12 17:24:39 +0100308static inline int slave_do_arp_validate(struct bonding *bond,
309 struct slave *slave)
Jay Vosburghf5b2b962006-09-22 21:54:53 -0700310{
311 return bond->params.arp_validate & (1 << slave->state);
312}
313
Adrian Bunk079ca7d2006-12-12 17:24:39 +0100314static inline unsigned long slave_last_rx(struct bonding *bond,
Andy Gospodarekf8a8ccd2007-01-29 12:08:38 -0800315 struct slave *slave)
Jay Vosburghf5b2b962006-09-22 21:54:53 -0700316{
317 if (slave_do_arp_validate(bond, slave))
318 return slave->last_arp_rx;
319
320 return slave->dev->last_rx;
321}
322
Adrian Bunkcceb9042006-01-05 22:45:42 -0800323static inline void bond_set_slave_inactive_flags(struct slave *slave)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324{
Wang Chen454d7c92008-11-12 23:37:49 -0800325 struct bonding *bond = netdev_priv(slave->dev->master);
Jiri Pirkoae63e802009-05-27 05:42:36 +0000326 if (!bond_is_lb(bond))
Jay Vosburgh8f903c72006-02-21 16:36:44 -0800327 slave->state = BOND_STATE_BACKUP;
Andy Gospodarekebd8e492010-06-02 08:39:21 +0000328 if (!bond->params.all_slaves_active)
329 slave->dev->priv_flags |= IFF_SLAVE_INACTIVE;
Jay Vosburghf5b2b962006-09-22 21:54:53 -0700330 if (slave_do_arp_validate(bond, slave))
331 slave->dev->priv_flags |= IFF_SLAVE_NEEDARP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332}
333
Adrian Bunkcceb9042006-01-05 22:45:42 -0800334static inline void bond_set_slave_active_flags(struct slave *slave)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335{
336 slave->state = BOND_STATE_ACTIVE;
Jay Vosburghf5b2b962006-09-22 21:54:53 -0700337 slave->dev->priv_flags &= ~(IFF_SLAVE_INACTIVE | IFF_SLAVE_NEEDARP);
Jay Vosburgh8f903c72006-02-21 16:36:44 -0800338}
339
340static inline void bond_set_master_3ad_flags(struct bonding *bond)
341{
342 bond->dev->priv_flags |= IFF_MASTER_8023AD;
343}
344
345static inline void bond_unset_master_3ad_flags(struct bonding *bond)
346{
347 bond->dev->priv_flags &= ~IFF_MASTER_8023AD;
348}
349
350static inline void bond_set_master_alb_flags(struct bonding *bond)
351{
352 bond->dev->priv_flags |= IFF_MASTER_ALB;
353}
354
355static inline void bond_unset_master_alb_flags(struct bonding *bond)
356{
357 bond->dev->priv_flags &= ~IFF_MASTER_ALB;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358}
359
360struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr);
361int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev);
Eric W. Biedermanec87fd32009-10-29 14:18:26 +0000362int bond_create(struct net *net, const char *name);
Mitch Williamsb76cdba2005-11-09 10:36:41 -0800363int bond_create_sysfs(void);
364void bond_destroy_sysfs(void);
Eric W. Biederman6151b3d2009-10-29 14:18:22 +0000365void bond_prepare_sysfs_group(struct bonding *bond);
Mitch Williamsb76cdba2005-11-09 10:36:41 -0800366int bond_create_slave_symlinks(struct net_device *master, struct net_device *slave);
367void bond_destroy_slave_symlinks(struct net_device *master, struct net_device *slave);
Mitch Williamsa77b5322005-11-09 10:35:51 -0800368int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev);
369int bond_release(struct net_device *bond_dev, struct net_device *slave_dev);
Jay Vosburgh1b76b312007-10-17 17:37:45 -0700370void bond_mii_monitor(struct work_struct *);
371void bond_loadbalance_arp_mon(struct work_struct *);
372void bond_activebackup_arp_mon(struct work_struct *);
Mitch Williamsa77b5322005-11-09 10:35:51 -0800373void bond_set_mode_ops(struct bonding *bond, int mode);
Holger Eitzenberger325dcf72008-12-09 23:10:17 -0800374int bond_parse_parm(const char *mode_arg, const struct bond_parm_tbl *tbl);
Mitch Williamsa77b5322005-11-09 10:35:51 -0800375void bond_select_active_slave(struct bonding *bond);
376void bond_change_active_slave(struct bonding *bond, struct slave *new_active);
Jay Vosburghf5b2b962006-09-22 21:54:53 -0700377void bond_register_arp(struct bonding *);
378void bond_unregister_arp(struct bonding *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379
Eric W. Biedermanec87fd32009-10-29 14:18:26 +0000380struct bond_net {
381 struct net * net; /* Associated network namespace */
382 struct list_head dev_list;
383#ifdef CONFIG_PROC_FS
384 struct proc_dir_entry * proc_dir;
385#endif
386};
387
Ben Dooksb2259672008-09-14 16:56:12 +0100388/* exported from bond_main.c */
Eric W. Biedermanec87fd32009-10-29 14:18:26 +0000389extern int bond_net_id;
Holger Eitzenbergere97fd7c2008-12-09 23:10:38 -0800390extern const struct bond_parm_tbl bond_lacp_tbl[];
391extern const struct bond_parm_tbl bond_mode_tbl[];
392extern const struct bond_parm_tbl xmit_hashtype_tbl[];
393extern const struct bond_parm_tbl arp_validate_tbl[];
394extern const struct bond_parm_tbl fail_over_mac_tbl[];
Jiri Pirkoa5499522009-09-25 03:28:09 +0000395extern const struct bond_parm_tbl pri_reselect_tbl[];
Hannes Ederb06715b2008-12-25 23:58:57 -0800396extern struct bond_parm_tbl ad_select_tbl[];
397
Brian Haley305d5522008-11-04 17:51:14 -0800398#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
399void bond_send_unsolicited_na(struct bonding *bond);
400void bond_register_ipv6_notifier(void);
401void bond_unregister_ipv6_notifier(void);
402#else
403static inline void bond_send_unsolicited_na(struct bonding *bond)
404{
405 return;
406}
407static inline void bond_register_ipv6_notifier(void)
408{
409 return;
410}
411static inline void bond_unregister_ipv6_notifier(void)
412{
413 return;
414}
415#endif
416
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417#endif /* _LINUX_BONDING_H */