tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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 _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>
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
28#include "bond_3ad.h"
29#include "bond_alb.h"
30#include "bond_options.h"
31
32#define DRV_VERSION "3.7.1"
33#define DRV_RELDATE "April 27, 2011"
34#define DRV_NAME "bonding"
35#define DRV_DESCRIPTION "Ethernet Channel Bonding Driver"
36
37#define bond_version DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n"
38
39#define BOND_MAX_VLAN_ENCAP 2
40#define BOND_MAX_ARP_TARGETS 16
41
42#define BOND_DEFAULT_MIIMON 100
43
44/*
45 * Less bad way to call ioctl from within the kernel; this needs to be
46 * done some other way to get the call out of interrupt context.
47 * Needs "ioctl" variable to be supplied by calling context.
48 */
49#define IOCTL(dev, arg, cmd) ({ \
50 int res = 0; \
51 mm_segment_t fs = get_fs(); \
52 set_fs(get_ds()); \
53 res = ioctl(dev, arg, cmd); \
54 set_fs(fs); \
55 res; })
56
57#define BOND_MODE(bond) ((bond)->params.mode)
58
59/* slave list primitives */
60#define bond_slave_list(bond) (&(bond)->dev->adj_list.lower)
61
62#define bond_has_slaves(bond) !list_empty(bond_slave_list(bond))
63
64/* IMPORTANT: bond_first/last_slave can return NULL in case of an empty list */
65#define bond_first_slave(bond) \
66 (bond_has_slaves(bond) ? \
67 netdev_adjacent_get_private(bond_slave_list(bond)->next) : \
68 NULL)
69#define bond_last_slave(bond) \
70 (bond_has_slaves(bond) ? \
71 netdev_adjacent_get_private(bond_slave_list(bond)->prev) : \
72 NULL)
73
74/* Caller must have rcu_read_lock */
75#define bond_first_slave_rcu(bond) \
76 netdev_lower_get_first_private_rcu(bond->dev)
77
78#define bond_is_first_slave(bond, pos) (pos == bond_first_slave(bond))
79#define bond_is_last_slave(bond, pos) (pos == bond_last_slave(bond))
80
81/**
82 * bond_for_each_slave - iterate over all slaves
83 * @bond: the bond holding this list
84 * @pos: current slave
85 * @iter: list_head * iterator
86 *
87 * Caller must hold bond->lock
88 */
89#define bond_for_each_slave(bond, pos, iter) \
90 netdev_for_each_lower_private((bond)->dev, pos, iter)
91
92/* Caller must have rcu_read_lock */
93#define bond_for_each_slave_rcu(bond, pos, iter) \
94 netdev_for_each_lower_private_rcu((bond)->dev, pos, iter)
95
96#ifdef CONFIG_NET_POLL_CONTROLLER
97extern atomic_t netpoll_block_tx;
98
99static inline void block_netpoll_tx(void)
100{
101 atomic_inc(&netpoll_block_tx);
102}
103
104static inline void unblock_netpoll_tx(void)
105{
106 atomic_dec(&netpoll_block_tx);
107}
108
109static inline int is_netpoll_tx_blocked(struct net_device *dev)
110{
111 if (unlikely(netpoll_tx_running(dev)))
112 return atomic_read(&netpoll_block_tx);
113 return 0;
114}
115#else
116#define block_netpoll_tx()
117#define unblock_netpoll_tx()
118#define is_netpoll_tx_blocked(dev) (0)
119#endif
120
121struct bond_params {
122 int mode;
123 int xmit_policy;
124 int miimon;
125 u8 num_peer_notif;
126 int arp_interval;
127 int arp_validate;
128 int arp_all_targets;
129 int use_carrier;
130 int fail_over_mac;
131 int updelay;
132 int downdelay;
133 int lacp_fast;
134 unsigned int min_links;
135 int ad_select;
136 char primary[IFNAMSIZ];
137 int primary_reselect;
138 __be32 arp_targets[BOND_MAX_ARP_TARGETS];
139 int tx_queues;
140 int all_slaves_active;
141 int resend_igmp;
142 int lp_interval;
143 int packets_per_slave;
144 int tlb_dynamic_lb;
145 struct reciprocal_value reciprocal_packets_per_slave;
146};
147
148struct bond_parm_tbl {
149 char *modename;
150 int mode;
151};
152
153struct slave {
154 struct net_device *dev; /* first - useful for panic debug */
155 struct bonding *bond; /* our master */
156 int delay;
157 /* all three in jiffies */
158 unsigned long last_link_up;
159 unsigned long last_rx;
160 unsigned long target_last_arp_rx[BOND_MAX_ARP_TARGETS];
161 s8 link; /* one of BOND_LINK_XXXX */
162 s8 new_link;
163 u8 backup:1, /* indicates backup slave. Value corresponds with
164 BOND_STATE_ACTIVE and BOND_STATE_BACKUP */
165 inactive:1, /* indicates inactive slave */
166 should_notify:1; /* indicateds whether the state changed */
167 u8 duplex;
168 u32 original_mtu;
169 u32 link_failure_count;
170 u32 speed;
171 u16 queue_id;
172 u8 perm_hwaddr[ETH_ALEN];
173 struct ad_slave_info *ad_info;
174 struct tlb_slave_info tlb_info;
175#ifdef CONFIG_NET_POLL_CONTROLLER
176 struct netpoll *np;
177#endif
178 struct kobject kobj;
179};
180
181/*
182 * Link pseudo-state only used internally by monitors
183 */
184#define BOND_LINK_NOCHANGE -1
185
186/*
187 * Here are the locking policies for the two bonding locks:
188 *
189 * 1) Get bond->lock when reading/writing slave list.
190 * 2) Get bond->curr_slave_lock when reading/writing bond->curr_active_slave.
191 * (It is unnecessary when the write-lock is put with bond->lock.)
192 * 3) When we lock with bond->curr_slave_lock, we must lock with bond->lock
193 * beforehand.
194 */
195struct bonding {
196 struct net_device *dev; /* first - useful for panic debug */
197 struct slave *curr_active_slave;
198 struct slave *current_arp_slave;
199 struct slave *primary_slave;
200 bool force_primary;
201 s32 slave_cnt; /* never change this value outside the attach/detach wrappers */
202 int (*recv_probe)(const struct sk_buff *, struct bonding *,
203 struct slave *);
204 rwlock_t lock;
205 rwlock_t curr_slave_lock;
206 u8 send_peer_notif;
207 u8 igmp_retrans;
208#ifdef CONFIG_PROC_FS
209 struct proc_dir_entry *proc_entry;
210 char proc_file_name[IFNAMSIZ];
211#endif /* CONFIG_PROC_FS */
212 struct list_head bond_list;
213 u32 rr_tx_counter;
214 struct ad_bond_info ad_info;
215 struct alb_bond_info alb_info;
216 struct bond_params params;
217 struct workqueue_struct *wq;
218 struct delayed_work mii_work;
219 struct delayed_work arp_work;
220 struct delayed_work alb_work;
221 struct delayed_work ad_work;
222 struct delayed_work mcast_work;
223#ifdef CONFIG_DEBUG_FS
224 /* debugging support via debugfs */
225 struct dentry *debug_dir;
226#endif /* CONFIG_DEBUG_FS */
227};
228
229#define bond_slave_get_rcu(dev) \
230 ((struct slave *) rcu_dereference(dev->rx_handler_data))
231
232#define bond_slave_get_rtnl(dev) \
233 ((struct slave *) rtnl_dereference(dev->rx_handler_data))
234
235struct bond_vlan_tag {
236 __be16 vlan_proto;
237 unsigned short vlan_id;
238};
239
240/**
241 * Returns NULL if the net_device does not belong to any of the bond's slaves
242 *
243 * Caller must hold bond lock for read
244 */
245static inline struct slave *bond_get_slave_by_dev(struct bonding *bond,
246 struct net_device *slave_dev)
247{
248 return netdev_lower_dev_get_private(bond->dev, slave_dev);
249}
250
251static inline struct bonding *bond_get_bond_by_slave(struct slave *slave)
252{
253 return slave->bond;
254}
255
256static inline bool bond_should_override_tx_queue(struct bonding *bond)
257{
258 return BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
259 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN;
260}
261
262static inline bool bond_is_lb(const struct bonding *bond)
263{
264 return BOND_MODE(bond) == BOND_MODE_TLB ||
265 BOND_MODE(bond) == BOND_MODE_ALB;
266}
267
268static inline bool bond_mode_uses_arp(int mode)
269{
270 return mode != BOND_MODE_8023AD && mode != BOND_MODE_TLB &&
271 mode != BOND_MODE_ALB;
272}
273
274static inline bool bond_mode_uses_primary(int mode)
275{
276 return mode == BOND_MODE_ACTIVEBACKUP || mode == BOND_MODE_TLB ||
277 mode == BOND_MODE_ALB;
278}
279
280static inline bool bond_uses_primary(struct bonding *bond)
281{
282 return bond_mode_uses_primary(BOND_MODE(bond));
283}
284
285static inline bool bond_slave_is_up(struct slave *slave)
286{
287 return netif_running(slave->dev) && netif_carrier_ok(slave->dev);
288}
289
290static inline void bond_set_active_slave(struct slave *slave)
291{
292 if (slave->backup) {
293 slave->backup = 0;
294 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
295 }
296}
297
298static inline void bond_set_backup_slave(struct slave *slave)
299{
300 if (!slave->backup) {
301 slave->backup = 1;
302 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
303 }
304}
305
306static inline void bond_set_slave_state(struct slave *slave,
307 int slave_state, bool notify)
308{
309 if (slave->backup == slave_state)
310 return;
311
312 slave->backup = slave_state;
313 if (notify) {
314 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
315 slave->should_notify = 0;
316 } else {
317 if (slave->should_notify)
318 slave->should_notify = 0;
319 else
320 slave->should_notify = 1;
321 }
322}
323
324static inline void bond_slave_state_change(struct bonding *bond)
325{
326 struct list_head *iter;
327 struct slave *tmp;
328
329 bond_for_each_slave(bond, tmp, iter) {
330 if (tmp->link == BOND_LINK_UP)
331 bond_set_active_slave(tmp);
332 else if (tmp->link == BOND_LINK_DOWN)
333 bond_set_backup_slave(tmp);
334 }
335}
336
337static inline void bond_slave_state_notify(struct bonding *bond)
338{
339 struct list_head *iter;
340 struct slave *tmp;
341
342 bond_for_each_slave(bond, tmp, iter) {
343 if (tmp->should_notify) {
344 rtmsg_ifinfo(RTM_NEWLINK, tmp->dev, 0, GFP_ATOMIC);
345 tmp->should_notify = 0;
346 }
347 }
348}
349
350static inline int bond_slave_state(struct slave *slave)
351{
352 return slave->backup;
353}
354
355static inline bool bond_is_active_slave(struct slave *slave)
356{
357 return !bond_slave_state(slave);
358}
359
360static inline bool bond_slave_can_tx(struct slave *slave)
361{
362 return bond_slave_is_up(slave) && slave->link == BOND_LINK_UP &&
363 bond_is_active_slave(slave);
364}
365
366#define BOND_PRI_RESELECT_ALWAYS 0
367#define BOND_PRI_RESELECT_BETTER 1
368#define BOND_PRI_RESELECT_FAILURE 2
369
370#define BOND_FOM_NONE 0
371#define BOND_FOM_ACTIVE 1
372#define BOND_FOM_FOLLOW 2
373
374#define BOND_ARP_TARGETS_ANY 0
375#define BOND_ARP_TARGETS_ALL 1
376
377#define BOND_ARP_VALIDATE_NONE 0
378#define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE)
379#define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP)
380#define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \
381 BOND_ARP_VALIDATE_BACKUP)
382#define BOND_ARP_FILTER (BOND_ARP_VALIDATE_ALL + 1)
383#define BOND_ARP_FILTER_ACTIVE (BOND_ARP_VALIDATE_ACTIVE | \
384 BOND_ARP_FILTER)
385#define BOND_ARP_FILTER_BACKUP (BOND_ARP_VALIDATE_BACKUP | \
386 BOND_ARP_FILTER)
387
388#define BOND_SLAVE_NOTIFY_NOW true
389#define BOND_SLAVE_NOTIFY_LATER false
390
391static inline int slave_do_arp_validate(struct bonding *bond,
392 struct slave *slave)
393{
394 return bond->params.arp_validate & (1 << bond_slave_state(slave));
395}
396
397static inline int slave_do_arp_validate_only(struct bonding *bond)
398{
399 return bond->params.arp_validate & BOND_ARP_FILTER;
400}
401
402static inline int bond_is_ip_target_ok(__be32 addr)
403{
404 return !ipv4_is_lbcast(addr) && !ipv4_is_zeronet(addr);
405}
406
407/* Get the oldest arp which we've received on this slave for bond's
408 * arp_targets.
409 */
410static inline unsigned long slave_oldest_target_arp_rx(struct bonding *bond,
411 struct slave *slave)
412{
413 int i = 1;
414 unsigned long ret = slave->target_last_arp_rx[0];
415
416 for (; (i < BOND_MAX_ARP_TARGETS) && bond->params.arp_targets[i]; i++)
417 if (time_before(slave->target_last_arp_rx[i], ret))
418 ret = slave->target_last_arp_rx[i];
419
420 return ret;
421}
422
423static inline unsigned long slave_last_rx(struct bonding *bond,
424 struct slave *slave)
425{
426 if (bond->params.arp_all_targets == BOND_ARP_TARGETS_ALL)
427 return slave_oldest_target_arp_rx(bond, slave);
428
429 return slave->last_rx;
430}
431
432#ifdef CONFIG_NET_POLL_CONTROLLER
433static inline void bond_netpoll_send_skb(const struct slave *slave,
434 struct sk_buff *skb)
435{
436 struct netpoll *np = slave->np;
437
438 if (np)
439 netpoll_send_skb(np, skb);
440}
441#else
442static inline void bond_netpoll_send_skb(const struct slave *slave,
443 struct sk_buff *skb)
444{
445}
446#endif
447
448static inline void bond_set_slave_inactive_flags(struct slave *slave,
449 bool notify)
450{
451 if (!bond_is_lb(slave->bond))
452 bond_set_slave_state(slave, BOND_STATE_BACKUP, notify);
453 if (!slave->bond->params.all_slaves_active)
454 slave->inactive = 1;
455}
456
457static inline void bond_set_slave_active_flags(struct slave *slave,
458 bool notify)
459{
460 bond_set_slave_state(slave, BOND_STATE_ACTIVE, notify);
461 slave->inactive = 0;
462}
463
464static inline bool bond_is_slave_inactive(struct slave *slave)
465{
466 return slave->inactive;
467}
468
469static inline __be32 bond_confirm_addr(struct net_device *dev, __be32 dst, __be32 local)
470{
471 struct in_device *in_dev;
472 __be32 addr = 0;
473
474 rcu_read_lock();
475 in_dev = __in_dev_get_rcu(dev);
476
477 if (in_dev)
478 addr = inet_confirm_addr(dev_net(dev), in_dev, dst, local,
479 RT_SCOPE_HOST);
480 rcu_read_unlock();
481 return addr;
482}
483
484struct bond_net {
485 struct net *net; /* Associated network namespace */
486 struct list_head dev_list;
487#ifdef CONFIG_PROC_FS
488 struct proc_dir_entry *proc_dir;
489#endif
490 struct class_attribute class_attr_bonding_masters;
491};
492
493int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, struct slave *slave);
494void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev);
495int bond_create(struct net *net, const char *name);
496int bond_create_sysfs(struct bond_net *net);
497void bond_destroy_sysfs(struct bond_net *net);
498void bond_prepare_sysfs_group(struct bonding *bond);
499int bond_sysfs_slave_add(struct slave *slave);
500void bond_sysfs_slave_del(struct slave *slave);
501int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev);
502int bond_release(struct net_device *bond_dev, struct net_device *slave_dev);
503u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb);
504void bond_select_active_slave(struct bonding *bond);
505void bond_change_active_slave(struct bonding *bond, struct slave *new_active);
506void bond_create_debugfs(void);
507void bond_destroy_debugfs(void);
508void bond_debug_register(struct bonding *bond);
509void bond_debug_unregister(struct bonding *bond);
510void bond_debug_reregister(struct bonding *bond);
511const char *bond_mode_name(int mode);
512void bond_setup(struct net_device *bond_dev);
513unsigned int bond_get_num_tx_queues(void);
514int bond_netlink_init(void);
515void bond_netlink_fini(void);
516struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond);
517struct net_device *bond_option_active_slave_get(struct bonding *bond);
518const char *bond_slave_link_status(s8 link);
519bool bond_verify_device_path(struct net_device *start_dev,
520 struct net_device *end_dev,
521 struct bond_vlan_tag *tags);
522
523#ifdef CONFIG_PROC_FS
524void bond_create_proc_entry(struct bonding *bond);
525void bond_remove_proc_entry(struct bonding *bond);
526void bond_create_proc_dir(struct bond_net *bn);
527void bond_destroy_proc_dir(struct bond_net *bn);
528#else
529static inline void bond_create_proc_entry(struct bonding *bond)
530{
531}
532
533static inline void bond_remove_proc_entry(struct bonding *bond)
534{
535}
536
537static inline void bond_create_proc_dir(struct bond_net *bn)
538{
539}
540
541static inline void bond_destroy_proc_dir(struct bond_net *bn)
542{
543}
544#endif
545
546static inline struct slave *bond_slave_has_mac(struct bonding *bond,
547 const u8 *mac)
548{
549 struct list_head *iter;
550 struct slave *tmp;
551
552 bond_for_each_slave(bond, tmp, iter)
553 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
554 return tmp;
555
556 return NULL;
557}
558
559/* Caller must hold rcu_read_lock() for read */
560static inline struct slave *bond_slave_has_mac_rcu(struct bonding *bond,
561 const u8 *mac)
562{
563 struct list_head *iter;
564 struct slave *tmp;
565
566 bond_for_each_slave_rcu(bond, tmp, iter)
567 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
568 return tmp;
569
570 return NULL;
571}
572
573/* Caller must hold rcu_read_lock() for read */
574static inline bool bond_slave_has_mac_rx(struct bonding *bond, const u8 *mac)
575{
576 struct list_head *iter;
577 struct slave *tmp;
578 struct netdev_hw_addr *ha;
579
580 bond_for_each_slave_rcu(bond, tmp, iter)
581 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
582 return true;
583
584 if (netdev_uc_empty(bond->dev))
585 return false;
586
587 netdev_for_each_uc_addr(ha, bond->dev)
588 if (ether_addr_equal_64bits(mac, ha->addr))
589 return true;
590
591 return false;
592}
593
594/* Check if the ip is present in arp ip list, or first free slot if ip == 0
595 * Returns -1 if not found, index if found
596 */
597static inline int bond_get_targets_ip(__be32 *targets, __be32 ip)
598{
599 int i;
600
601 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
602 if (targets[i] == ip)
603 return i;
604 else if (targets[i] == 0)
605 break;
606
607 return -1;
608}
609
610/* exported from bond_main.c */
611extern int bond_net_id;
612extern const struct bond_parm_tbl bond_lacp_tbl[];
613extern const struct bond_parm_tbl xmit_hashtype_tbl[];
614extern const struct bond_parm_tbl arp_validate_tbl[];
615extern const struct bond_parm_tbl arp_all_targets_tbl[];
616extern const struct bond_parm_tbl fail_over_mac_tbl[];
617extern const struct bond_parm_tbl pri_reselect_tbl[];
618extern struct bond_parm_tbl ad_select_tbl[];
619
620/* exported from bond_netlink.c */
621extern struct rtnl_link_ops bond_link_ops;
622
623#endif /* _LINUX_BONDING_H */