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_ARP_TARGETS 16
40
41#define BOND_DEFAULT_MIIMON 100
42
43/*
44 * Less bad way to call ioctl from within the kernel; this needs to be
45 * done some other way to get the call out of interrupt context.
46 * Needs "ioctl" variable to be supplied by calling context.
47 */
48#define IOCTL(dev, arg, cmd) ({ \
49 int res = 0; \
50 mm_segment_t fs = get_fs(); \
51 set_fs(get_ds()); \
52 res = ioctl(dev, arg, cmd); \
53 set_fs(fs); \
54 res; })
55
56#define BOND_MODE(bond) ((bond)->params.mode)
57
58/* slave list primitives */
59#define bond_slave_list(bond) (&(bond)->dev->adj_list.lower)
60
61#define bond_has_slaves(bond) !list_empty(bond_slave_list(bond))
62
63/* IMPORTANT: bond_first/last_slave can return NULL in case of an empty list */
64#define bond_first_slave(bond) \
65 (bond_has_slaves(bond) ? \
66 netdev_adjacent_get_private(bond_slave_list(bond)->next) : \
67 NULL)
68#define bond_last_slave(bond) \
69 (bond_has_slaves(bond) ? \
70 netdev_adjacent_get_private(bond_slave_list(bond)->prev) : \
71 NULL)
72
73/* Caller must have rcu_read_lock */
74#define bond_first_slave_rcu(bond) \
75 netdev_lower_get_first_private_rcu(bond->dev)
76
77#define bond_is_first_slave(bond, pos) (pos == bond_first_slave(bond))
78#define bond_is_last_slave(bond, pos) (pos == bond_last_slave(bond))
79
80/**
81 * bond_for_each_slave - iterate over all slaves
82 * @bond: the bond holding this list
83 * @pos: current slave
84 * @iter: list_head * iterator
85 *
86 * Caller must hold bond->lock
87 */
88#define bond_for_each_slave(bond, pos, iter) \
89 netdev_for_each_lower_private((bond)->dev, pos, iter)
90
91/* Caller must have rcu_read_lock */
92#define bond_for_each_slave_rcu(bond, pos, iter) \
93 netdev_for_each_lower_private_rcu((bond)->dev, pos, iter)
94
95#ifdef CONFIG_NET_POLL_CONTROLLER
96extern atomic_t netpoll_block_tx;
97
98static inline void block_netpoll_tx(void)
99{
100 atomic_inc(&netpoll_block_tx);
101}
102
103static inline void unblock_netpoll_tx(void)
104{
105 atomic_dec(&netpoll_block_tx);
106}
107
108static inline int is_netpoll_tx_blocked(struct net_device *dev)
109{
110 if (unlikely(netpoll_tx_running(dev)))
111 return atomic_read(&netpoll_block_tx);
112 return 0;
113}
114#else
115#define block_netpoll_tx()
116#define unblock_netpoll_tx()
117#define is_netpoll_tx_blocked(dev) (0)
118#endif
119
120struct bond_params {
121 int mode;
122 int xmit_policy;
123 int miimon;
124 u8 num_peer_notif;
125 int arp_interval;
126 int arp_validate;
127 int arp_all_targets;
128 int use_carrier;
129 int fail_over_mac;
130 int updelay;
131 int downdelay;
132 int lacp_fast;
133 unsigned int min_links;
134 int ad_select;
135 char primary[IFNAMSIZ];
136 int primary_reselect;
137 __be32 arp_targets[BOND_MAX_ARP_TARGETS];
138 int tx_queues;
139 int all_slaves_active;
140 int resend_igmp;
141 int lp_interval;
142 int packets_per_slave;
143 int tlb_dynamic_lb;
144 struct reciprocal_value reciprocal_packets_per_slave;
145};
146
147struct bond_parm_tbl {
148 char *modename;
149 int mode;
150};
151
152struct slave {
153 struct net_device *dev; /* first - useful for panic debug */
154 struct bonding *bond; /* our master */
155 int delay;
156 /* all three in jiffies */
157 unsigned long last_link_up;
158 unsigned long last_rx;
159 unsigned long target_last_arp_rx[BOND_MAX_ARP_TARGETS];
160 s8 link; /* one of BOND_LINK_XXXX */
161 s8 new_link;
162 u8 backup:1, /* indicates backup slave. Value corresponds with
163 BOND_STATE_ACTIVE and BOND_STATE_BACKUP */
164 inactive:1, /* indicates inactive slave */
165 should_notify:1; /* indicateds whether the state changed */
166 u8 duplex;
167 u32 original_mtu;
168 u32 link_failure_count;
169 u32 speed;
170 u16 queue_id;
171 u8 perm_hwaddr[ETH_ALEN];
172 struct ad_slave_info *ad_info;
173 struct tlb_slave_info tlb_info;
174#ifdef CONFIG_NET_POLL_CONTROLLER
175 struct netpoll *np;
176#endif
177 struct kobject kobj;
178};
179
180/*
181 * Link pseudo-state only used internally by monitors
182 */
183#define BOND_LINK_NOCHANGE -1
184
185/*
186 * Here are the locking policies for the two bonding locks:
187 *
188 * 1) Get bond->lock when reading/writing slave list.
189 * 2) Get bond->curr_slave_lock when reading/writing bond->curr_active_slave.
190 * (It is unnecessary when the write-lock is put with bond->lock.)
191 * 3) When we lock with bond->curr_slave_lock, we must lock with bond->lock
192 * beforehand.
193 */
194struct bonding {
195 struct net_device *dev; /* first - useful for panic debug */
196 struct slave __rcu *curr_active_slave;
197 struct slave __rcu *current_arp_slave;
198 struct slave *primary_slave;
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 rwlock_t lock;
204 rwlock_t curr_slave_lock;
205 u8 send_peer_notif;
206 u8 igmp_retrans;
207#ifdef CONFIG_PROC_FS
208 struct proc_dir_entry *proc_entry;
209 char proc_file_name[IFNAMSIZ];
210#endif /* CONFIG_PROC_FS */
211 struct list_head bond_list;
212 u32 rr_tx_counter;
213 struct ad_bond_info ad_info;
214 struct alb_bond_info alb_info;
215 struct bond_params params;
216 struct workqueue_struct *wq;
217 struct delayed_work mii_work;
218 struct delayed_work arp_work;
219 struct delayed_work alb_work;
220 struct delayed_work ad_work;
221 struct delayed_work mcast_work;
222#ifdef CONFIG_DEBUG_FS
223 /* debugging support via debugfs */
224 struct dentry *debug_dir;
225#endif /* CONFIG_DEBUG_FS */
226};
227
228#define bond_slave_get_rcu(dev) \
229 ((struct slave *) rcu_dereference(dev->rx_handler_data))
230
231#define bond_slave_get_rtnl(dev) \
232 ((struct slave *) rtnl_dereference(dev->rx_handler_data))
233
234#define bond_deref_active_protected(bond) \
235 rcu_dereference_protected(bond->curr_active_slave, \
236 lockdep_is_held(&bond->curr_slave_lock))
237
238struct bond_vlan_tag {
239 __be16 vlan_proto;
240 unsigned short vlan_id;
241};
242
243/**
244 * Returns NULL if the net_device does not belong to any of the bond's slaves
245 *
246 * Caller must hold bond lock for read
247 */
248static inline struct slave *bond_get_slave_by_dev(struct bonding *bond,
249 struct net_device *slave_dev)
250{
251 return netdev_lower_dev_get_private(bond->dev, slave_dev);
252}
253
254static inline struct bonding *bond_get_bond_by_slave(struct slave *slave)
255{
256 return slave->bond;
257}
258
259static inline bool bond_should_override_tx_queue(struct bonding *bond)
260{
261 return BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
262 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN;
263}
264
265static inline bool bond_is_lb(const struct bonding *bond)
266{
267 return BOND_MODE(bond) == BOND_MODE_TLB ||
268 BOND_MODE(bond) == BOND_MODE_ALB;
269}
270
271static inline bool bond_is_nondyn_tlb(const struct bonding *bond)
272{
273 return (BOND_MODE(bond) == BOND_MODE_TLB) &&
274 (bond->params.tlb_dynamic_lb == 0);
275}
276
277static inline bool bond_mode_uses_arp(int mode)
278{
279 return mode != BOND_MODE_8023AD && mode != BOND_MODE_TLB &&
280 mode != BOND_MODE_ALB;
281}
282
283static inline bool bond_mode_uses_primary(int mode)
284{
285 return mode == BOND_MODE_ACTIVEBACKUP || mode == BOND_MODE_TLB ||
286 mode == BOND_MODE_ALB;
287}
288
289static inline bool bond_uses_primary(struct bonding *bond)
290{
291 return bond_mode_uses_primary(BOND_MODE(bond));
292}
293
294static inline bool bond_slave_is_up(struct slave *slave)
295{
296 return netif_running(slave->dev) && netif_carrier_ok(slave->dev);
297}
298
299static inline void bond_set_active_slave(struct slave *slave)
300{
301 if (slave->backup) {
302 slave->backup = 0;
303 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
304 }
305}
306
307static inline void bond_set_backup_slave(struct slave *slave)
308{
309 if (!slave->backup) {
310 slave->backup = 1;
311 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
312 }
313}
314
315static inline void bond_set_slave_state(struct slave *slave,
316 int slave_state, bool notify)
317{
318 if (slave->backup == slave_state)
319 return;
320
321 slave->backup = slave_state;
322 if (notify) {
323 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
324 slave->should_notify = 0;
325 } else {
326 if (slave->should_notify)
327 slave->should_notify = 0;
328 else
329 slave->should_notify = 1;
330 }
331}
332
333static inline void bond_slave_state_change(struct bonding *bond)
334{
335 struct list_head *iter;
336 struct slave *tmp;
337
338 bond_for_each_slave(bond, tmp, iter) {
339 if (tmp->link == BOND_LINK_UP)
340 bond_set_active_slave(tmp);
341 else if (tmp->link == BOND_LINK_DOWN)
342 bond_set_backup_slave(tmp);
343 }
344}
345
346static inline void bond_slave_state_notify(struct bonding *bond)
347{
348 struct list_head *iter;
349 struct slave *tmp;
350
351 bond_for_each_slave(bond, tmp, iter) {
352 if (tmp->should_notify) {
353 rtmsg_ifinfo(RTM_NEWLINK, tmp->dev, 0, GFP_ATOMIC);
354 tmp->should_notify = 0;
355 }
356 }
357}
358
359static inline int bond_slave_state(struct slave *slave)
360{
361 return slave->backup;
362}
363
364static inline bool bond_is_active_slave(struct slave *slave)
365{
366 return !bond_slave_state(slave);
367}
368
369static inline bool bond_slave_can_tx(struct slave *slave)
370{
371 return bond_slave_is_up(slave) && slave->link == BOND_LINK_UP &&
372 bond_is_active_slave(slave);
373}
374
375#define BOND_PRI_RESELECT_ALWAYS 0
376#define BOND_PRI_RESELECT_BETTER 1
377#define BOND_PRI_RESELECT_FAILURE 2
378
379#define BOND_FOM_NONE 0
380#define BOND_FOM_ACTIVE 1
381#define BOND_FOM_FOLLOW 2
382
383#define BOND_ARP_TARGETS_ANY 0
384#define BOND_ARP_TARGETS_ALL 1
385
386#define BOND_ARP_VALIDATE_NONE 0
387#define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE)
388#define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP)
389#define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \
390 BOND_ARP_VALIDATE_BACKUP)
391#define BOND_ARP_FILTER (BOND_ARP_VALIDATE_ALL + 1)
392#define BOND_ARP_FILTER_ACTIVE (BOND_ARP_VALIDATE_ACTIVE | \
393 BOND_ARP_FILTER)
394#define BOND_ARP_FILTER_BACKUP (BOND_ARP_VALIDATE_BACKUP | \
395 BOND_ARP_FILTER)
396
397#define BOND_SLAVE_NOTIFY_NOW true
398#define BOND_SLAVE_NOTIFY_LATER false
399
400static inline int slave_do_arp_validate(struct bonding *bond,
401 struct slave *slave)
402{
403 return bond->params.arp_validate & (1 << bond_slave_state(slave));
404}
405
406static inline int slave_do_arp_validate_only(struct bonding *bond)
407{
408 return bond->params.arp_validate & BOND_ARP_FILTER;
409}
410
411static inline int bond_is_ip_target_ok(__be32 addr)
412{
413 return !ipv4_is_lbcast(addr) && !ipv4_is_zeronet(addr);
414}
415
416/* Get the oldest arp which we've received on this slave for bond's
417 * arp_targets.
418 */
419static inline unsigned long slave_oldest_target_arp_rx(struct bonding *bond,
420 struct slave *slave)
421{
422 int i = 1;
423 unsigned long ret = slave->target_last_arp_rx[0];
424
425 for (; (i < BOND_MAX_ARP_TARGETS) && bond->params.arp_targets[i]; i++)
426 if (time_before(slave->target_last_arp_rx[i], ret))
427 ret = slave->target_last_arp_rx[i];
428
429 return ret;
430}
431
432static inline unsigned long slave_last_rx(struct bonding *bond,
433 struct slave *slave)
434{
435 if (bond->params.arp_all_targets == BOND_ARP_TARGETS_ALL)
436 return slave_oldest_target_arp_rx(bond, slave);
437
438 return slave->last_rx;
439}
440
441#ifdef CONFIG_NET_POLL_CONTROLLER
442static inline void bond_netpoll_send_skb(const struct slave *slave,
443 struct sk_buff *skb)
444{
445 struct netpoll *np = slave->np;
446
447 if (np)
448 netpoll_send_skb(np, skb);
449}
450#else
451static inline void bond_netpoll_send_skb(const struct slave *slave,
452 struct sk_buff *skb)
453{
454}
455#endif
456
457static inline void bond_set_slave_inactive_flags(struct slave *slave,
458 bool notify)
459{
460 if (!bond_is_lb(slave->bond))
461 bond_set_slave_state(slave, BOND_STATE_BACKUP, notify);
462 if (!slave->bond->params.all_slaves_active)
463 slave->inactive = 1;
464}
465
466static inline void bond_set_slave_active_flags(struct slave *slave,
467 bool notify)
468{
469 bond_set_slave_state(slave, BOND_STATE_ACTIVE, notify);
470 slave->inactive = 0;
471}
472
473static inline bool bond_is_slave_inactive(struct slave *slave)
474{
475 return slave->inactive;
476}
477
478static inline __be32 bond_confirm_addr(struct net_device *dev, __be32 dst, __be32 local)
479{
480 struct in_device *in_dev;
481 __be32 addr = 0;
482
483 rcu_read_lock();
484 in_dev = __in_dev_get_rcu(dev);
485
486 if (in_dev)
487 addr = inet_confirm_addr(dev_net(dev), in_dev, dst, local,
488 RT_SCOPE_HOST);
489 rcu_read_unlock();
490 return addr;
491}
492
493struct bond_net {
494 struct net *net; /* Associated network namespace */
495 struct list_head dev_list;
496#ifdef CONFIG_PROC_FS
497 struct proc_dir_entry *proc_dir;
498#endif
499 struct class_attribute class_attr_bonding_masters;
500};
501
502int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, struct slave *slave);
503void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev);
504int bond_create(struct net *net, const char *name);
505int bond_create_sysfs(struct bond_net *net);
506void bond_destroy_sysfs(struct bond_net *net);
507void bond_prepare_sysfs_group(struct bonding *bond);
508int bond_sysfs_slave_add(struct slave *slave);
509void bond_sysfs_slave_del(struct slave *slave);
510int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev);
511int bond_release(struct net_device *bond_dev, struct net_device *slave_dev);
512u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb);
513void bond_select_active_slave(struct bonding *bond);
514void bond_change_active_slave(struct bonding *bond, struct slave *new_active);
515void bond_create_debugfs(void);
516void bond_destroy_debugfs(void);
517void bond_debug_register(struct bonding *bond);
518void bond_debug_unregister(struct bonding *bond);
519void bond_debug_reregister(struct bonding *bond);
520const char *bond_mode_name(int mode);
521void bond_setup(struct net_device *bond_dev);
522unsigned int bond_get_num_tx_queues(void);
523int bond_netlink_init(void);
524void bond_netlink_fini(void);
525struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond);
526const char *bond_slave_link_status(s8 link);
527struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
528 struct net_device *end_dev,
529 int level);
530
531#ifdef CONFIG_PROC_FS
532void bond_create_proc_entry(struct bonding *bond);
533void bond_remove_proc_entry(struct bonding *bond);
534void bond_create_proc_dir(struct bond_net *bn);
535void bond_destroy_proc_dir(struct bond_net *bn);
536#else
537static inline void bond_create_proc_entry(struct bonding *bond)
538{
539}
540
541static inline void bond_remove_proc_entry(struct bonding *bond)
542{
543}
544
545static inline void bond_create_proc_dir(struct bond_net *bn)
546{
547}
548
549static inline void bond_destroy_proc_dir(struct bond_net *bn)
550{
551}
552#endif
553
554static inline struct slave *bond_slave_has_mac(struct bonding *bond,
555 const u8 *mac)
556{
557 struct list_head *iter;
558 struct slave *tmp;
559
560 bond_for_each_slave(bond, tmp, iter)
561 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
562 return tmp;
563
564 return NULL;
565}
566
567/* Caller must hold rcu_read_lock() for read */
568static inline struct slave *bond_slave_has_mac_rcu(struct bonding *bond,
569 const u8 *mac)
570{
571 struct list_head *iter;
572 struct slave *tmp;
573
574 bond_for_each_slave_rcu(bond, tmp, iter)
575 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
576 return tmp;
577
578 return NULL;
579}
580
581/* Caller must hold rcu_read_lock() for read */
582static inline bool bond_slave_has_mac_rx(struct bonding *bond, const u8 *mac)
583{
584 struct list_head *iter;
585 struct slave *tmp;
586 struct netdev_hw_addr *ha;
587
588 bond_for_each_slave_rcu(bond, tmp, iter)
589 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
590 return true;
591
592 if (netdev_uc_empty(bond->dev))
593 return false;
594
595 netdev_for_each_uc_addr(ha, bond->dev)
596 if (ether_addr_equal_64bits(mac, ha->addr))
597 return true;
598
599 return false;
600}
601
602/* Check if the ip is present in arp ip list, or first free slot if ip == 0
603 * Returns -1 if not found, index if found
604 */
605static inline int bond_get_targets_ip(__be32 *targets, __be32 ip)
606{
607 int i;
608
609 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
610 if (targets[i] == ip)
611 return i;
612 else if (targets[i] == 0)
613 break;
614
615 return -1;
616}
617
618/* exported from bond_main.c */
619extern int bond_net_id;
620extern const struct bond_parm_tbl bond_lacp_tbl[];
621extern const struct bond_parm_tbl xmit_hashtype_tbl[];
622extern const struct bond_parm_tbl arp_validate_tbl[];
623extern const struct bond_parm_tbl arp_all_targets_tbl[];
624extern const struct bond_parm_tbl fail_over_mac_tbl[];
625extern const struct bond_parm_tbl pri_reselect_tbl[];
626extern struct bond_parm_tbl ad_select_tbl[];
627
628/* exported from bond_netlink.c */
629extern struct rtnl_link_ops bond_link_ops;
630
631#endif /* _LINUX_BONDING_H */