bonding.h 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511
  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. #ifndef _LINUX_BONDING_H
  15. #define _LINUX_BONDING_H
  16. #include <linux/timer.h>
  17. #include <linux/proc_fs.h>
  18. #include <linux/if_bonding.h>
  19. #include <linux/cpumask.h>
  20. #include <linux/in6.h>
  21. #include <linux/netpoll.h>
  22. #include <linux/inetdevice.h>
  23. #include <linux/etherdevice.h>
  24. #include "bond_3ad.h"
  25. #include "bond_alb.h"
  26. #define DRV_VERSION "3.7.1"
  27. #define DRV_RELDATE "April 27, 2011"
  28. #define DRV_NAME "bonding"
  29. #define DRV_DESCRIPTION "Ethernet Channel Bonding Driver"
  30. #define bond_version DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n"
  31. #define BOND_MAX_ARP_TARGETS 16
  32. #define IS_UP(dev) \
  33. ((((dev)->flags & IFF_UP) == IFF_UP) && \
  34. netif_running(dev) && \
  35. netif_carrier_ok(dev))
  36. /*
  37. * Checks whether slave is ready for transmit.
  38. */
  39. #define SLAVE_IS_OK(slave) \
  40. (((slave)->dev->flags & IFF_UP) && \
  41. netif_running((slave)->dev) && \
  42. ((slave)->link == BOND_LINK_UP) && \
  43. bond_is_active_slave(slave))
  44. #define USES_PRIMARY(mode) \
  45. (((mode) == BOND_MODE_ACTIVEBACKUP) || \
  46. ((mode) == BOND_MODE_TLB) || \
  47. ((mode) == BOND_MODE_ALB))
  48. #define TX_QUEUE_OVERRIDE(mode) \
  49. (((mode) == BOND_MODE_ACTIVEBACKUP) || \
  50. ((mode) == BOND_MODE_ROUNDROBIN))
  51. /*
  52. * Less bad way to call ioctl from within the kernel; this needs to be
  53. * done some other way to get the call out of interrupt context.
  54. * Needs "ioctl" variable to be supplied by calling context.
  55. */
  56. #define IOCTL(dev, arg, cmd) ({ \
  57. int res = 0; \
  58. mm_segment_t fs = get_fs(); \
  59. set_fs(get_ds()); \
  60. res = ioctl(dev, arg, cmd); \
  61. set_fs(fs); \
  62. res; })
  63. /* slave list primitives */
  64. #define bond_slave_list(bond) (&(bond)->dev->adj_list.lower)
  65. #define bond_has_slaves(bond) !list_empty(bond_slave_list(bond))
  66. #define bond_to_slave(ptr) list_entry(ptr, struct slave, list)
  67. /* IMPORTANT: bond_first/last_slave can return NULL in case of an empty list */
  68. #define bond_first_slave(bond) \
  69. (bond_has_slaves(bond) ? \
  70. netdev_adjacent_get_private(bond_slave_list(bond)->next) : \
  71. NULL)
  72. #define bond_last_slave(bond) \
  73. (bond_has_slaves(bond) ? \
  74. netdev_adjacent_get_private(bond_slave_list(bond)->prev) : \
  75. NULL)
  76. #define bond_is_first_slave(bond, pos) (pos == bond_first_slave(bond))
  77. #define bond_is_last_slave(bond, pos) (pos == bond_last_slave(bond))
  78. /* Since bond_first/last_slave can return NULL, these can return NULL too */
  79. #define bond_next_slave(bond, pos) \
  80. (bond_is_last_slave(bond, pos) ? bond_first_slave(bond) : \
  81. bond_to_slave((pos)->list.next))
  82. #define bond_prev_slave(bond, pos) \
  83. (bond_is_first_slave(bond, pos) ? bond_last_slave(bond) : \
  84. bond_to_slave((pos)->list.prev))
  85. /**
  86. * bond_for_each_slave - iterate over all slaves
  87. * @bond: the bond holding this list
  88. * @pos: current slave
  89. * @iter: list_head * iterator
  90. *
  91. * Caller must hold bond->lock
  92. */
  93. #define bond_for_each_slave(bond, pos, iter) \
  94. netdev_for_each_lower_private((bond)->dev, pos, iter)
  95. /* Caller must have rcu_read_lock */
  96. #define bond_for_each_slave_rcu(bond, pos, iter) \
  97. netdev_for_each_lower_private_rcu((bond)->dev, pos, iter)
  98. #ifdef CONFIG_NET_POLL_CONTROLLER
  99. extern atomic_t netpoll_block_tx;
  100. static inline void block_netpoll_tx(void)
  101. {
  102. atomic_inc(&netpoll_block_tx);
  103. }
  104. static inline void unblock_netpoll_tx(void)
  105. {
  106. atomic_dec(&netpoll_block_tx);
  107. }
  108. static 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. struct bond_params {
  120. int mode;
  121. int xmit_policy;
  122. int miimon;
  123. u8 num_peer_notif;
  124. int arp_interval;
  125. int arp_validate;
  126. int arp_all_targets;
  127. int use_carrier;
  128. int fail_over_mac;
  129. int updelay;
  130. int downdelay;
  131. int lacp_fast;
  132. unsigned int min_links;
  133. int ad_select;
  134. char primary[IFNAMSIZ];
  135. int primary_reselect;
  136. __be32 arp_targets[BOND_MAX_ARP_TARGETS];
  137. int tx_queues;
  138. int all_slaves_active;
  139. int resend_igmp;
  140. int lp_interval;
  141. };
  142. struct bond_parm_tbl {
  143. char *modename;
  144. int mode;
  145. };
  146. #define BOND_MAX_MODENAME_LEN 20
  147. struct slave {
  148. struct net_device *dev; /* first - useful for panic debug */
  149. struct list_head list;
  150. struct bonding *bond; /* our master */
  151. int delay;
  152. unsigned long jiffies;
  153. unsigned long last_arp_rx;
  154. unsigned long target_last_arp_rx[BOND_MAX_ARP_TARGETS];
  155. s8 link; /* one of BOND_LINK_XXXX */
  156. s8 new_link;
  157. u8 backup:1, /* indicates backup slave. Value corresponds with
  158. BOND_STATE_ACTIVE and BOND_STATE_BACKUP */
  159. inactive:1; /* indicates inactive slave */
  160. u8 duplex;
  161. u32 original_mtu;
  162. u32 link_failure_count;
  163. u32 speed;
  164. u16 queue_id;
  165. u8 perm_hwaddr[ETH_ALEN];
  166. struct ad_slave_info ad_info; /* HUGE - better to dynamically alloc */
  167. struct tlb_slave_info tlb_info;
  168. #ifdef CONFIG_NET_POLL_CONTROLLER
  169. struct netpoll *np;
  170. #endif
  171. };
  172. /*
  173. * Link pseudo-state only used internally by monitors
  174. */
  175. #define BOND_LINK_NOCHANGE -1
  176. /*
  177. * Here are the locking policies for the two bonding locks:
  178. *
  179. * 1) Get bond->lock when reading/writing slave list.
  180. * 2) Get bond->curr_slave_lock when reading/writing bond->curr_active_slave.
  181. * (It is unnecessary when the write-lock is put with bond->lock.)
  182. * 3) When we lock with bond->curr_slave_lock, we must lock with bond->lock
  183. * beforehand.
  184. */
  185. struct bonding {
  186. struct net_device *dev; /* first - useful for panic debug */
  187. struct list_head slave_list;
  188. struct slave *curr_active_slave;
  189. struct slave *current_arp_slave;
  190. struct slave *primary_slave;
  191. bool force_primary;
  192. s32 slave_cnt; /* never change this value outside the attach/detach wrappers */
  193. int (*recv_probe)(const struct sk_buff *, struct bonding *,
  194. struct slave *);
  195. rwlock_t lock;
  196. rwlock_t curr_slave_lock;
  197. u8 send_peer_notif;
  198. u8 igmp_retrans;
  199. #ifdef CONFIG_PROC_FS
  200. struct proc_dir_entry *proc_entry;
  201. char proc_file_name[IFNAMSIZ];
  202. #endif /* CONFIG_PROC_FS */
  203. struct list_head bond_list;
  204. int (*xmit_hash_policy)(struct sk_buff *, int);
  205. u16 rr_tx_counter;
  206. struct ad_bond_info ad_info;
  207. struct alb_bond_info alb_info;
  208. struct bond_params params;
  209. struct workqueue_struct *wq;
  210. struct delayed_work mii_work;
  211. struct delayed_work arp_work;
  212. struct delayed_work alb_work;
  213. struct delayed_work ad_work;
  214. struct delayed_work mcast_work;
  215. #ifdef CONFIG_DEBUG_FS
  216. /* debugging support via debugfs */
  217. struct dentry *debug_dir;
  218. #endif /* CONFIG_DEBUG_FS */
  219. };
  220. #define bond_slave_get_rcu(dev) \
  221. ((struct slave *) rcu_dereference(dev->rx_handler_data))
  222. #define bond_slave_get_rtnl(dev) \
  223. ((struct slave *) rtnl_dereference(dev->rx_handler_data))
  224. /**
  225. * Returns NULL if the net_device does not belong to any of the bond's slaves
  226. *
  227. * Caller must hold bond lock for read
  228. */
  229. static inline struct slave *bond_get_slave_by_dev(struct bonding *bond,
  230. struct net_device *slave_dev)
  231. {
  232. return netdev_lower_dev_get_private(bond->dev, slave_dev);
  233. }
  234. static inline struct bonding *bond_get_bond_by_slave(struct slave *slave)
  235. {
  236. if (!slave || !slave->bond)
  237. return NULL;
  238. return slave->bond;
  239. }
  240. static inline bool bond_is_lb(const struct bonding *bond)
  241. {
  242. return (bond->params.mode == BOND_MODE_TLB ||
  243. bond->params.mode == BOND_MODE_ALB);
  244. }
  245. static inline void bond_set_active_slave(struct slave *slave)
  246. {
  247. slave->backup = 0;
  248. }
  249. static inline void bond_set_backup_slave(struct slave *slave)
  250. {
  251. slave->backup = 1;
  252. }
  253. static inline int bond_slave_state(struct slave *slave)
  254. {
  255. return slave->backup;
  256. }
  257. static inline bool bond_is_active_slave(struct slave *slave)
  258. {
  259. return !bond_slave_state(slave);
  260. }
  261. #define BOND_PRI_RESELECT_ALWAYS 0
  262. #define BOND_PRI_RESELECT_BETTER 1
  263. #define BOND_PRI_RESELECT_FAILURE 2
  264. #define BOND_FOM_NONE 0
  265. #define BOND_FOM_ACTIVE 1
  266. #define BOND_FOM_FOLLOW 2
  267. #define BOND_ARP_TARGETS_ANY 0
  268. #define BOND_ARP_TARGETS_ALL 1
  269. #define BOND_ARP_VALIDATE_NONE 0
  270. #define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE)
  271. #define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP)
  272. #define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \
  273. BOND_ARP_VALIDATE_BACKUP)
  274. static inline int slave_do_arp_validate(struct bonding *bond,
  275. struct slave *slave)
  276. {
  277. return bond->params.arp_validate & (1 << bond_slave_state(slave));
  278. }
  279. /* Get the oldest arp which we've received on this slave for bond's
  280. * arp_targets.
  281. */
  282. static inline unsigned long slave_oldest_target_arp_rx(struct bonding *bond,
  283. struct slave *slave)
  284. {
  285. int i = 1;
  286. unsigned long ret = slave->target_last_arp_rx[0];
  287. for (; (i < BOND_MAX_ARP_TARGETS) && bond->params.arp_targets[i]; i++)
  288. if (time_before(slave->target_last_arp_rx[i], ret))
  289. ret = slave->target_last_arp_rx[i];
  290. return ret;
  291. }
  292. static inline unsigned long slave_last_rx(struct bonding *bond,
  293. struct slave *slave)
  294. {
  295. if (slave_do_arp_validate(bond, slave)) {
  296. if (bond->params.arp_all_targets == BOND_ARP_TARGETS_ALL)
  297. return slave_oldest_target_arp_rx(bond, slave);
  298. else
  299. return slave->last_arp_rx;
  300. }
  301. return slave->dev->last_rx;
  302. }
  303. #ifdef CONFIG_NET_POLL_CONTROLLER
  304. static inline void bond_netpoll_send_skb(const struct slave *slave,
  305. struct sk_buff *skb)
  306. {
  307. struct netpoll *np = slave->np;
  308. if (np)
  309. netpoll_send_skb(np, skb);
  310. }
  311. #else
  312. static inline void bond_netpoll_send_skb(const struct slave *slave,
  313. struct sk_buff *skb)
  314. {
  315. }
  316. #endif
  317. static inline void bond_set_slave_inactive_flags(struct slave *slave)
  318. {
  319. if (!bond_is_lb(slave->bond))
  320. bond_set_backup_slave(slave);
  321. if (!slave->bond->params.all_slaves_active)
  322. slave->inactive = 1;
  323. }
  324. static inline void bond_set_slave_active_flags(struct slave *slave)
  325. {
  326. bond_set_active_slave(slave);
  327. slave->inactive = 0;
  328. }
  329. static inline bool bond_is_slave_inactive(struct slave *slave)
  330. {
  331. return slave->inactive;
  332. }
  333. static inline __be32 bond_confirm_addr(struct net_device *dev, __be32 dst, __be32 local)
  334. {
  335. struct in_device *in_dev;
  336. __be32 addr = 0;
  337. rcu_read_lock();
  338. in_dev = __in_dev_get_rcu(dev);
  339. if (in_dev)
  340. addr = inet_confirm_addr(in_dev, dst, local, RT_SCOPE_HOST);
  341. rcu_read_unlock();
  342. return addr;
  343. }
  344. static inline bool slave_can_tx(struct slave *slave)
  345. {
  346. if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP &&
  347. bond_is_active_slave(slave))
  348. return true;
  349. else
  350. return false;
  351. }
  352. struct bond_net;
  353. int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, struct slave *slave);
  354. struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr);
  355. int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev);
  356. void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id);
  357. int bond_create(struct net *net, const char *name);
  358. int bond_create_sysfs(struct bond_net *net);
  359. void bond_destroy_sysfs(struct bond_net *net);
  360. void bond_prepare_sysfs_group(struct bonding *bond);
  361. int bond_create_slave_symlinks(struct net_device *master, struct net_device *slave);
  362. void bond_destroy_slave_symlinks(struct net_device *master, struct net_device *slave);
  363. int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev);
  364. int bond_release(struct net_device *bond_dev, struct net_device *slave_dev);
  365. void bond_mii_monitor(struct work_struct *);
  366. void bond_loadbalance_arp_mon(struct work_struct *);
  367. void bond_activebackup_arp_mon(struct work_struct *);
  368. void bond_set_mode_ops(struct bonding *bond, int mode);
  369. int bond_parse_parm(const char *mode_arg, const struct bond_parm_tbl *tbl);
  370. void bond_select_active_slave(struct bonding *bond);
  371. void bond_change_active_slave(struct bonding *bond, struct slave *new_active);
  372. void bond_create_debugfs(void);
  373. void bond_destroy_debugfs(void);
  374. void bond_debug_register(struct bonding *bond);
  375. void bond_debug_unregister(struct bonding *bond);
  376. void bond_debug_reregister(struct bonding *bond);
  377. const char *bond_mode_name(int mode);
  378. struct bond_net {
  379. struct net * net; /* Associated network namespace */
  380. struct list_head dev_list;
  381. #ifdef CONFIG_PROC_FS
  382. struct proc_dir_entry * proc_dir;
  383. #endif
  384. struct class_attribute class_attr_bonding_masters;
  385. };
  386. #ifdef CONFIG_PROC_FS
  387. void bond_create_proc_entry(struct bonding *bond);
  388. void bond_remove_proc_entry(struct bonding *bond);
  389. void bond_create_proc_dir(struct bond_net *bn);
  390. void bond_destroy_proc_dir(struct bond_net *bn);
  391. #else
  392. static inline void bond_create_proc_entry(struct bonding *bond)
  393. {
  394. }
  395. static inline void bond_remove_proc_entry(struct bonding *bond)
  396. {
  397. }
  398. static inline void bond_create_proc_dir(struct bond_net *bn)
  399. {
  400. }
  401. static inline void bond_destroy_proc_dir(struct bond_net *bn)
  402. {
  403. }
  404. #endif
  405. static inline struct slave *bond_slave_has_mac(struct bonding *bond,
  406. const u8 *mac)
  407. {
  408. struct list_head *iter;
  409. struct slave *tmp;
  410. bond_for_each_slave(bond, tmp, iter)
  411. if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
  412. return tmp;
  413. return NULL;
  414. }
  415. /* Check if the ip is present in arp ip list, or first free slot if ip == 0
  416. * Returns -1 if not found, index if found
  417. */
  418. static inline int bond_get_targets_ip(__be32 *targets, __be32 ip)
  419. {
  420. int i;
  421. for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
  422. if (targets[i] == ip)
  423. return i;
  424. else if (targets[i] == 0)
  425. break;
  426. return -1;
  427. }
  428. /* exported from bond_main.c */
  429. extern int bond_net_id;
  430. extern const struct bond_parm_tbl bond_lacp_tbl[];
  431. extern const struct bond_parm_tbl bond_mode_tbl[];
  432. extern const struct bond_parm_tbl xmit_hashtype_tbl[];
  433. extern const struct bond_parm_tbl arp_validate_tbl[];
  434. extern const struct bond_parm_tbl arp_all_targets_tbl[];
  435. extern const struct bond_parm_tbl fail_over_mac_tbl[];
  436. extern const struct bond_parm_tbl pri_reselect_tbl[];
  437. extern struct bond_parm_tbl ad_select_tbl[];
  438. #endif /* _LINUX_BONDING_H */