bond_alb.c 45 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716171717181719172017211722172317241725172617271728172917301731173217331734173517361737173817391740174117421743174417451746174717481749175017511752175317541755175617571758
  1. /*
  2. * Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved.
  3. *
  4. * This program is free software; you can redistribute it and/or modify it
  5. * under the terms of the GNU General Public License as published by the
  6. * Free Software Foundation; either version 2 of the License, or
  7. * (at your option) any later version.
  8. *
  9. * This program is distributed in the hope that it will be useful, but
  10. * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  11. * or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
  12. * for more details.
  13. *
  14. * You should have received a copy of the GNU General Public License along
  15. * with this program; if not, write to the Free Software Foundation, Inc.,
  16. * 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  17. *
  18. * The full GNU General Public License is included in this distribution in the
  19. * file called LICENSE.
  20. *
  21. */
  22. #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  23. #include <linux/skbuff.h>
  24. #include <linux/netdevice.h>
  25. #include <linux/etherdevice.h>
  26. #include <linux/pkt_sched.h>
  27. #include <linux/spinlock.h>
  28. #include <linux/slab.h>
  29. #include <linux/timer.h>
  30. #include <linux/ip.h>
  31. #include <linux/ipv6.h>
  32. #include <linux/if_arp.h>
  33. #include <linux/if_ether.h>
  34. #include <linux/if_bonding.h>
  35. #include <linux/if_vlan.h>
  36. #include <linux/in.h>
  37. #include <net/ipx.h>
  38. #include <net/arp.h>
  39. #include <net/ipv6.h>
  40. #include <asm/byteorder.h>
  41. #include "bonding.h"
  42. #include "bond_alb.h"
  43. #define ALB_TIMER_TICKS_PER_SEC 10 /* should be a divisor of HZ */
  44. #define BOND_TLB_REBALANCE_INTERVAL 10 /* In seconds, periodic re-balancing.
  45. * Used for division - never set
  46. * to zero !!!
  47. */
  48. #define BOND_ALB_LP_INTERVAL 1 /* In seconds, periodic send of
  49. * learning packets to the switch
  50. */
  51. #define BOND_TLB_REBALANCE_TICKS (BOND_TLB_REBALANCE_INTERVAL \
  52. * ALB_TIMER_TICKS_PER_SEC)
  53. #define BOND_ALB_LP_TICKS (BOND_ALB_LP_INTERVAL \
  54. * ALB_TIMER_TICKS_PER_SEC)
  55. #define TLB_HASH_TABLE_SIZE 256 /* The size of the clients hash table.
  56. * Note that this value MUST NOT be smaller
  57. * because the key hash table is BYTE wide !
  58. */
  59. #define TLB_NULL_INDEX 0xffffffff
  60. #define MAX_LP_BURST 3
  61. /* rlb defs */
  62. #define RLB_HASH_TABLE_SIZE 256
  63. #define RLB_NULL_INDEX 0xffffffff
  64. #define RLB_UPDATE_DELAY 2*ALB_TIMER_TICKS_PER_SEC /* 2 seconds */
  65. #define RLB_ARP_BURST_SIZE 2
  66. #define RLB_UPDATE_RETRY 3 /* 3-ticks - must be smaller than the rlb
  67. * rebalance interval (5 min).
  68. */
  69. /* RLB_PROMISC_TIMEOUT = 10 sec equals the time that the current slave is
  70. * promiscuous after failover
  71. */
  72. #define RLB_PROMISC_TIMEOUT 10*ALB_TIMER_TICKS_PER_SEC
  73. #ifndef __long_aligned
  74. #define __long_aligned __attribute__((aligned((sizeof(long)))))
  75. #endif
  76. static const u8 mac_bcast[ETH_ALEN] __long_aligned = {
  77. 0xff, 0xff, 0xff, 0xff, 0xff, 0xff
  78. };
  79. static const u8 mac_v6_allmcast[ETH_ALEN] __long_aligned = {
  80. 0x33, 0x33, 0x00, 0x00, 0x00, 0x01
  81. };
  82. static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
  83. #pragma pack(1)
  84. struct learning_pkt {
  85. u8 mac_dst[ETH_ALEN];
  86. u8 mac_src[ETH_ALEN];
  87. __be16 type;
  88. u8 padding[ETH_ZLEN - ETH_HLEN];
  89. };
  90. struct arp_pkt {
  91. __be16 hw_addr_space;
  92. __be16 prot_addr_space;
  93. u8 hw_addr_len;
  94. u8 prot_addr_len;
  95. __be16 op_code;
  96. u8 mac_src[ETH_ALEN]; /* sender hardware address */
  97. __be32 ip_src; /* sender IP address */
  98. u8 mac_dst[ETH_ALEN]; /* target hardware address */
  99. __be32 ip_dst; /* target IP address */
  100. };
  101. #pragma pack()
  102. static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb)
  103. {
  104. return (struct arp_pkt *)skb_network_header(skb);
  105. }
  106. /* Forward declaration */
  107. static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[]);
  108. static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
  109. {
  110. int i;
  111. u8 hash = 0;
  112. for (i = 0; i < hash_size; i++) {
  113. hash ^= hash_start[i];
  114. }
  115. return hash;
  116. }
  117. /*********************** tlb specific functions ***************************/
  118. static inline void _lock_tx_hashtbl(struct bonding *bond)
  119. {
  120. spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
  121. }
  122. static inline void _unlock_tx_hashtbl(struct bonding *bond)
  123. {
  124. spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
  125. }
  126. /* Caller must hold tx_hashtbl lock */
  127. static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
  128. {
  129. if (save_load) {
  130. entry->load_history = 1 + entry->tx_bytes /
  131. BOND_TLB_REBALANCE_INTERVAL;
  132. entry->tx_bytes = 0;
  133. }
  134. entry->tx_slave = NULL;
  135. entry->next = TLB_NULL_INDEX;
  136. entry->prev = TLB_NULL_INDEX;
  137. }
  138. static inline void tlb_init_slave(struct slave *slave)
  139. {
  140. SLAVE_TLB_INFO(slave).load = 0;
  141. SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
  142. }
  143. /* Caller must hold bond lock for read */
  144. static void tlb_clear_slave(struct bonding *bond, struct slave *slave, int save_load)
  145. {
  146. struct tlb_client_info *tx_hash_table;
  147. u32 index;
  148. _lock_tx_hashtbl(bond);
  149. /* clear slave from tx_hashtbl */
  150. tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
  151. /* skip this if we've already freed the tx hash table */
  152. if (tx_hash_table) {
  153. index = SLAVE_TLB_INFO(slave).head;
  154. while (index != TLB_NULL_INDEX) {
  155. u32 next_index = tx_hash_table[index].next;
  156. tlb_init_table_entry(&tx_hash_table[index], save_load);
  157. index = next_index;
  158. }
  159. }
  160. tlb_init_slave(slave);
  161. _unlock_tx_hashtbl(bond);
  162. }
  163. /* Must be called before starting the monitor timer */
  164. static int tlb_initialize(struct bonding *bond)
  165. {
  166. struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
  167. int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
  168. struct tlb_client_info *new_hashtbl;
  169. int i;
  170. spin_lock_init(&(bond_info->tx_hashtbl_lock));
  171. new_hashtbl = kzalloc(size, GFP_KERNEL);
  172. if (!new_hashtbl) {
  173. pr_err("%s: Error: Failed to allocate TLB hash table\n",
  174. bond->dev->name);
  175. return -1;
  176. }
  177. _lock_tx_hashtbl(bond);
  178. bond_info->tx_hashtbl = new_hashtbl;
  179. for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
  180. tlb_init_table_entry(&bond_info->tx_hashtbl[i], 1);
  181. }
  182. _unlock_tx_hashtbl(bond);
  183. return 0;
  184. }
  185. /* Must be called only after all slaves have been released */
  186. static void tlb_deinitialize(struct bonding *bond)
  187. {
  188. struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
  189. _lock_tx_hashtbl(bond);
  190. kfree(bond_info->tx_hashtbl);
  191. bond_info->tx_hashtbl = NULL;
  192. _unlock_tx_hashtbl(bond);
  193. }
  194. static long long compute_gap(struct slave *slave)
  195. {
  196. return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
  197. (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
  198. }
  199. /* Caller must hold bond lock for read */
  200. static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
  201. {
  202. struct slave *slave, *least_loaded;
  203. long long max_gap;
  204. int i;
  205. least_loaded = NULL;
  206. max_gap = LLONG_MIN;
  207. /* Find the slave with the largest gap */
  208. bond_for_each_slave(bond, slave, i) {
  209. if (SLAVE_IS_OK(slave)) {
  210. long long gap = compute_gap(slave);
  211. if (max_gap < gap) {
  212. least_loaded = slave;
  213. max_gap = gap;
  214. }
  215. }
  216. }
  217. return least_loaded;
  218. }
  219. /* Caller must hold bond lock for read */
  220. static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index, u32 skb_len)
  221. {
  222. struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
  223. struct tlb_client_info *hash_table;
  224. struct slave *assigned_slave;
  225. _lock_tx_hashtbl(bond);
  226. hash_table = bond_info->tx_hashtbl;
  227. assigned_slave = hash_table[hash_index].tx_slave;
  228. if (!assigned_slave) {
  229. assigned_slave = tlb_get_least_loaded_slave(bond);
  230. if (assigned_slave) {
  231. struct tlb_slave_info *slave_info =
  232. &(SLAVE_TLB_INFO(assigned_slave));
  233. u32 next_index = slave_info->head;
  234. hash_table[hash_index].tx_slave = assigned_slave;
  235. hash_table[hash_index].next = next_index;
  236. hash_table[hash_index].prev = TLB_NULL_INDEX;
  237. if (next_index != TLB_NULL_INDEX) {
  238. hash_table[next_index].prev = hash_index;
  239. }
  240. slave_info->head = hash_index;
  241. slave_info->load +=
  242. hash_table[hash_index].load_history;
  243. }
  244. }
  245. if (assigned_slave) {
  246. hash_table[hash_index].tx_bytes += skb_len;
  247. }
  248. _unlock_tx_hashtbl(bond);
  249. return assigned_slave;
  250. }
  251. /*********************** rlb specific functions ***************************/
  252. static inline void _lock_rx_hashtbl(struct bonding *bond)
  253. {
  254. spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
  255. }
  256. static inline void _unlock_rx_hashtbl(struct bonding *bond)
  257. {
  258. spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
  259. }
  260. /* when an ARP REPLY is received from a client update its info
  261. * in the rx_hashtbl
  262. */
  263. static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
  264. {
  265. struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
  266. struct rlb_client_info *client_info;
  267. u32 hash_index;
  268. _lock_rx_hashtbl(bond);
  269. hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
  270. client_info = &(bond_info->rx_hashtbl[hash_index]);
  271. if ((client_info->assigned) &&
  272. (client_info->ip_src == arp->ip_dst) &&
  273. (client_info->ip_dst == arp->ip_src) &&
  274. (compare_ether_addr_64bits(client_info->mac_dst, arp->mac_src))) {
  275. /* update the clients MAC address */
  276. memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
  277. client_info->ntt = 1;
  278. bond_info->rx_ntt = 1;
  279. }
  280. _unlock_rx_hashtbl(bond);
  281. }
  282. static int rlb_arp_recv(struct sk_buff *skb, struct net_device *bond_dev, struct packet_type *ptype, struct net_device *orig_dev)
  283. {
  284. struct bonding *bond;
  285. struct arp_pkt *arp = (struct arp_pkt *)skb->data;
  286. int res = NET_RX_DROP;
  287. while (bond_dev->priv_flags & IFF_802_1Q_VLAN)
  288. bond_dev = vlan_dev_real_dev(bond_dev);
  289. if (!(bond_dev->priv_flags & IFF_BONDING) ||
  290. !(bond_dev->flags & IFF_MASTER))
  291. goto out;
  292. if (!arp) {
  293. pr_debug("Packet has no ARP data\n");
  294. goto out;
  295. }
  296. if (skb->len < sizeof(struct arp_pkt)) {
  297. pr_debug("Packet is too small to be an ARP\n");
  298. goto out;
  299. }
  300. if (arp->op_code == htons(ARPOP_REPLY)) {
  301. /* update rx hash table for this ARP */
  302. bond = netdev_priv(bond_dev);
  303. rlb_update_entry_from_arp(bond, arp);
  304. pr_debug("Server received an ARP Reply from client\n");
  305. }
  306. res = NET_RX_SUCCESS;
  307. out:
  308. dev_kfree_skb(skb);
  309. return res;
  310. }
  311. /* Caller must hold bond lock for read */
  312. static struct slave *rlb_next_rx_slave(struct bonding *bond)
  313. {
  314. struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
  315. struct slave *rx_slave, *slave, *start_at;
  316. int i = 0;
  317. if (bond_info->next_rx_slave) {
  318. start_at = bond_info->next_rx_slave;
  319. } else {
  320. start_at = bond->first_slave;
  321. }
  322. rx_slave = NULL;
  323. bond_for_each_slave_from(bond, slave, i, start_at) {
  324. if (SLAVE_IS_OK(slave)) {
  325. if (!rx_slave) {
  326. rx_slave = slave;
  327. } else if (slave->speed > rx_slave->speed) {
  328. rx_slave = slave;
  329. }
  330. }
  331. }
  332. if (rx_slave) {
  333. bond_info->next_rx_slave = rx_slave->next;
  334. }
  335. return rx_slave;
  336. }
  337. /* teach the switch the mac of a disabled slave
  338. * on the primary for fault tolerance
  339. *
  340. * Caller must hold bond->curr_slave_lock for write or bond lock for write
  341. */
  342. static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
  343. {
  344. if (!bond->curr_active_slave) {
  345. return;
  346. }
  347. if (!bond->alb_info.primary_is_promisc) {
  348. if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1))
  349. bond->alb_info.primary_is_promisc = 1;
  350. else
  351. bond->alb_info.primary_is_promisc = 0;
  352. }
  353. bond->alb_info.rlb_promisc_timeout_counter = 0;
  354. alb_send_learning_packets(bond->curr_active_slave, addr);
  355. }
  356. /* slave being removed should not be active at this point
  357. *
  358. * Caller must hold bond lock for read
  359. */
  360. static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
  361. {
  362. struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
  363. struct rlb_client_info *rx_hash_table;
  364. u32 index, next_index;
  365. /* clear slave from rx_hashtbl */
  366. _lock_rx_hashtbl(bond);
  367. rx_hash_table = bond_info->rx_hashtbl;
  368. index = bond_info->rx_hashtbl_head;
  369. for (; index != RLB_NULL_INDEX; index = next_index) {
  370. next_index = rx_hash_table[index].next;
  371. if (rx_hash_table[index].slave == slave) {
  372. struct slave *assigned_slave = rlb_next_rx_slave(bond);
  373. if (assigned_slave) {
  374. rx_hash_table[index].slave = assigned_slave;
  375. if (compare_ether_addr_64bits(rx_hash_table[index].mac_dst,
  376. mac_bcast)) {
  377. bond_info->rx_hashtbl[index].ntt = 1;
  378. bond_info->rx_ntt = 1;
  379. /* A slave has been removed from the
  380. * table because it is either disabled
  381. * or being released. We must retry the
  382. * update to avoid clients from not
  383. * being updated & disconnecting when
  384. * there is stress
  385. */
  386. bond_info->rlb_update_retry_counter =
  387. RLB_UPDATE_RETRY;
  388. }
  389. } else { /* there is no active slave */
  390. rx_hash_table[index].slave = NULL;
  391. }
  392. }
  393. }
  394. _unlock_rx_hashtbl(bond);
  395. write_lock_bh(&bond->curr_slave_lock);
  396. if (slave != bond->curr_active_slave) {
  397. rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
  398. }
  399. write_unlock_bh(&bond->curr_slave_lock);
  400. }
  401. static void rlb_update_client(struct rlb_client_info *client_info)
  402. {
  403. int i;
  404. if (!client_info->slave) {
  405. return;
  406. }
  407. for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
  408. struct sk_buff *skb;
  409. skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
  410. client_info->ip_dst,
  411. client_info->slave->dev,
  412. client_info->ip_src,
  413. client_info->mac_dst,
  414. client_info->slave->dev->dev_addr,
  415. client_info->mac_dst);
  416. if (!skb) {
  417. pr_err("%s: Error: failed to create an ARP packet\n",
  418. client_info->slave->dev->master->name);
  419. continue;
  420. }
  421. skb->dev = client_info->slave->dev;
  422. if (client_info->tag) {
  423. skb = vlan_put_tag(skb, client_info->vlan_id);
  424. if (!skb) {
  425. pr_err("%s: Error: failed to insert VLAN tag\n",
  426. client_info->slave->dev->master->name);
  427. continue;
  428. }
  429. }
  430. arp_xmit(skb);
  431. }
  432. }
  433. /* sends ARP REPLIES that update the clients that need updating */
  434. static void rlb_update_rx_clients(struct bonding *bond)
  435. {
  436. struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
  437. struct rlb_client_info *client_info;
  438. u32 hash_index;
  439. _lock_rx_hashtbl(bond);
  440. hash_index = bond_info->rx_hashtbl_head;
  441. for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
  442. client_info = &(bond_info->rx_hashtbl[hash_index]);
  443. if (client_info->ntt) {
  444. rlb_update_client(client_info);
  445. if (bond_info->rlb_update_retry_counter == 0) {
  446. client_info->ntt = 0;
  447. }
  448. }
  449. }
  450. /* do not update the entries again until this counter is zero so that
  451. * not to confuse the clients.
  452. */
  453. bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
  454. _unlock_rx_hashtbl(bond);
  455. }
  456. /* The slave was assigned a new mac address - update the clients */
  457. static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
  458. {
  459. struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
  460. struct rlb_client_info *client_info;
  461. int ntt = 0;
  462. u32 hash_index;
  463. _lock_rx_hashtbl(bond);
  464. hash_index = bond_info->rx_hashtbl_head;
  465. for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
  466. client_info = &(bond_info->rx_hashtbl[hash_index]);
  467. if ((client_info->slave == slave) &&
  468. compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
  469. client_info->ntt = 1;
  470. ntt = 1;
  471. }
  472. }
  473. // update the team's flag only after the whole iteration
  474. if (ntt) {
  475. bond_info->rx_ntt = 1;
  476. //fasten the change
  477. bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
  478. }
  479. _unlock_rx_hashtbl(bond);
  480. }
  481. /* mark all clients using src_ip to be updated */
  482. static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
  483. {
  484. struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
  485. struct rlb_client_info *client_info;
  486. u32 hash_index;
  487. _lock_rx_hashtbl(bond);
  488. hash_index = bond_info->rx_hashtbl_head;
  489. for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
  490. client_info = &(bond_info->rx_hashtbl[hash_index]);
  491. if (!client_info->slave) {
  492. pr_err("%s: Error: found a client with no channel in the client's hash table\n",
  493. bond->dev->name);
  494. continue;
  495. }
  496. /*update all clients using this src_ip, that are not assigned
  497. * to the team's address (curr_active_slave) and have a known
  498. * unicast mac address.
  499. */
  500. if ((client_info->ip_src == src_ip) &&
  501. compare_ether_addr_64bits(client_info->slave->dev->dev_addr,
  502. bond->dev->dev_addr) &&
  503. compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
  504. client_info->ntt = 1;
  505. bond_info->rx_ntt = 1;
  506. }
  507. }
  508. _unlock_rx_hashtbl(bond);
  509. }
  510. /* Caller must hold both bond and ptr locks for read */
  511. static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
  512. {
  513. struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
  514. struct arp_pkt *arp = arp_pkt(skb);
  515. struct slave *assigned_slave;
  516. struct rlb_client_info *client_info;
  517. u32 hash_index = 0;
  518. _lock_rx_hashtbl(bond);
  519. hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_src));
  520. client_info = &(bond_info->rx_hashtbl[hash_index]);
  521. if (client_info->assigned) {
  522. if ((client_info->ip_src == arp->ip_src) &&
  523. (client_info->ip_dst == arp->ip_dst)) {
  524. /* the entry is already assigned to this client */
  525. if (compare_ether_addr_64bits(arp->mac_dst, mac_bcast)) {
  526. /* update mac address from arp */
  527. memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
  528. }
  529. assigned_slave = client_info->slave;
  530. if (assigned_slave) {
  531. _unlock_rx_hashtbl(bond);
  532. return assigned_slave;
  533. }
  534. } else {
  535. /* the entry is already assigned to some other client,
  536. * move the old client to primary (curr_active_slave) so
  537. * that the new client can be assigned to this entry.
  538. */
  539. if (bond->curr_active_slave &&
  540. client_info->slave != bond->curr_active_slave) {
  541. client_info->slave = bond->curr_active_slave;
  542. rlb_update_client(client_info);
  543. }
  544. }
  545. }
  546. /* assign a new slave */
  547. assigned_slave = rlb_next_rx_slave(bond);
  548. if (assigned_slave) {
  549. client_info->ip_src = arp->ip_src;
  550. client_info->ip_dst = arp->ip_dst;
  551. /* arp->mac_dst is broadcast for arp reqeusts.
  552. * will be updated with clients actual unicast mac address
  553. * upon receiving an arp reply.
  554. */
  555. memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
  556. client_info->slave = assigned_slave;
  557. if (compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
  558. client_info->ntt = 1;
  559. bond->alb_info.rx_ntt = 1;
  560. } else {
  561. client_info->ntt = 0;
  562. }
  563. if (!list_empty(&bond->vlan_list)) {
  564. if (!vlan_get_tag(skb, &client_info->vlan_id))
  565. client_info->tag = 1;
  566. }
  567. if (!client_info->assigned) {
  568. u32 prev_tbl_head = bond_info->rx_hashtbl_head;
  569. bond_info->rx_hashtbl_head = hash_index;
  570. client_info->next = prev_tbl_head;
  571. if (prev_tbl_head != RLB_NULL_INDEX) {
  572. bond_info->rx_hashtbl[prev_tbl_head].prev =
  573. hash_index;
  574. }
  575. client_info->assigned = 1;
  576. }
  577. }
  578. _unlock_rx_hashtbl(bond);
  579. return assigned_slave;
  580. }
  581. /* chooses (and returns) transmit channel for arp reply
  582. * does not choose channel for other arp types since they are
  583. * sent on the curr_active_slave
  584. */
  585. static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
  586. {
  587. struct arp_pkt *arp = arp_pkt(skb);
  588. struct slave *tx_slave = NULL;
  589. if (arp->op_code == htons(ARPOP_REPLY)) {
  590. /* the arp must be sent on the selected
  591. * rx channel
  592. */
  593. tx_slave = rlb_choose_channel(skb, bond);
  594. if (tx_slave) {
  595. memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
  596. }
  597. pr_debug("Server sent ARP Reply packet\n");
  598. } else if (arp->op_code == htons(ARPOP_REQUEST)) {
  599. /* Create an entry in the rx_hashtbl for this client as a
  600. * place holder.
  601. * When the arp reply is received the entry will be updated
  602. * with the correct unicast address of the client.
  603. */
  604. rlb_choose_channel(skb, bond);
  605. /* The ARP relpy packets must be delayed so that
  606. * they can cancel out the influence of the ARP request.
  607. */
  608. bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
  609. /* arp requests are broadcast and are sent on the primary
  610. * the arp request will collapse all clients on the subnet to
  611. * the primary slave. We must register these clients to be
  612. * updated with their assigned mac.
  613. */
  614. rlb_req_update_subnet_clients(bond, arp->ip_src);
  615. pr_debug("Server sent ARP Request packet\n");
  616. }
  617. return tx_slave;
  618. }
  619. /* Caller must hold bond lock for read */
  620. static void rlb_rebalance(struct bonding *bond)
  621. {
  622. struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
  623. struct slave *assigned_slave;
  624. struct rlb_client_info *client_info;
  625. int ntt;
  626. u32 hash_index;
  627. _lock_rx_hashtbl(bond);
  628. ntt = 0;
  629. hash_index = bond_info->rx_hashtbl_head;
  630. for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
  631. client_info = &(bond_info->rx_hashtbl[hash_index]);
  632. assigned_slave = rlb_next_rx_slave(bond);
  633. if (assigned_slave && (client_info->slave != assigned_slave)) {
  634. client_info->slave = assigned_slave;
  635. client_info->ntt = 1;
  636. ntt = 1;
  637. }
  638. }
  639. /* update the team's flag only after the whole iteration */
  640. if (ntt) {
  641. bond_info->rx_ntt = 1;
  642. }
  643. _unlock_rx_hashtbl(bond);
  644. }
  645. /* Caller must hold rx_hashtbl lock */
  646. static void rlb_init_table_entry(struct rlb_client_info *entry)
  647. {
  648. memset(entry, 0, sizeof(struct rlb_client_info));
  649. entry->next = RLB_NULL_INDEX;
  650. entry->prev = RLB_NULL_INDEX;
  651. }
  652. static int rlb_initialize(struct bonding *bond)
  653. {
  654. struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
  655. struct packet_type *pk_type = &(BOND_ALB_INFO(bond).rlb_pkt_type);
  656. struct rlb_client_info *new_hashtbl;
  657. int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
  658. int i;
  659. spin_lock_init(&(bond_info->rx_hashtbl_lock));
  660. new_hashtbl = kmalloc(size, GFP_KERNEL);
  661. if (!new_hashtbl) {
  662. pr_err("%s: Error: Failed to allocate RLB hash table\n",
  663. bond->dev->name);
  664. return -1;
  665. }
  666. _lock_rx_hashtbl(bond);
  667. bond_info->rx_hashtbl = new_hashtbl;
  668. bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
  669. for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
  670. rlb_init_table_entry(bond_info->rx_hashtbl + i);
  671. }
  672. _unlock_rx_hashtbl(bond);
  673. /*initialize packet type*/
  674. pk_type->type = cpu_to_be16(ETH_P_ARP);
  675. pk_type->dev = NULL;
  676. pk_type->func = rlb_arp_recv;
  677. /* register to receive ARPs */
  678. dev_add_pack(pk_type);
  679. return 0;
  680. }
  681. static void rlb_deinitialize(struct bonding *bond)
  682. {
  683. struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
  684. dev_remove_pack(&(bond_info->rlb_pkt_type));
  685. _lock_rx_hashtbl(bond);
  686. kfree(bond_info->rx_hashtbl);
  687. bond_info->rx_hashtbl = NULL;
  688. bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
  689. _unlock_rx_hashtbl(bond);
  690. }
  691. static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
  692. {
  693. struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
  694. u32 curr_index;
  695. _lock_rx_hashtbl(bond);
  696. curr_index = bond_info->rx_hashtbl_head;
  697. while (curr_index != RLB_NULL_INDEX) {
  698. struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
  699. u32 next_index = bond_info->rx_hashtbl[curr_index].next;
  700. u32 prev_index = bond_info->rx_hashtbl[curr_index].prev;
  701. if (curr->tag && (curr->vlan_id == vlan_id)) {
  702. if (curr_index == bond_info->rx_hashtbl_head) {
  703. bond_info->rx_hashtbl_head = next_index;
  704. }
  705. if (prev_index != RLB_NULL_INDEX) {
  706. bond_info->rx_hashtbl[prev_index].next = next_index;
  707. }
  708. if (next_index != RLB_NULL_INDEX) {
  709. bond_info->rx_hashtbl[next_index].prev = prev_index;
  710. }
  711. rlb_init_table_entry(curr);
  712. }
  713. curr_index = next_index;
  714. }
  715. _unlock_rx_hashtbl(bond);
  716. }
  717. /*********************** tlb/rlb shared functions *********************/
  718. static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
  719. {
  720. struct bonding *bond = bond_get_bond_by_slave(slave);
  721. struct learning_pkt pkt;
  722. int size = sizeof(struct learning_pkt);
  723. int i;
  724. memset(&pkt, 0, size);
  725. memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
  726. memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
  727. pkt.type = cpu_to_be16(ETH_P_LOOP);
  728. for (i = 0; i < MAX_LP_BURST; i++) {
  729. struct sk_buff *skb;
  730. char *data;
  731. skb = dev_alloc_skb(size);
  732. if (!skb) {
  733. return;
  734. }
  735. data = skb_put(skb, size);
  736. memcpy(data, &pkt, size);
  737. skb_reset_mac_header(skb);
  738. skb->network_header = skb->mac_header + ETH_HLEN;
  739. skb->protocol = pkt.type;
  740. skb->priority = TC_PRIO_CONTROL;
  741. skb->dev = slave->dev;
  742. if (!list_empty(&bond->vlan_list)) {
  743. struct vlan_entry *vlan;
  744. vlan = bond_next_vlan(bond,
  745. bond->alb_info.current_alb_vlan);
  746. bond->alb_info.current_alb_vlan = vlan;
  747. if (!vlan) {
  748. kfree_skb(skb);
  749. continue;
  750. }
  751. skb = vlan_put_tag(skb, vlan->vlan_id);
  752. if (!skb) {
  753. pr_err("%s: Error: failed to insert VLAN tag\n",
  754. bond->dev->name);
  755. continue;
  756. }
  757. }
  758. dev_queue_xmit(skb);
  759. }
  760. }
  761. /* hw is a boolean parameter that determines whether we should try and
  762. * set the hw address of the device as well as the hw address of the
  763. * net_device
  764. */
  765. static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[], int hw)
  766. {
  767. struct net_device *dev = slave->dev;
  768. struct sockaddr s_addr;
  769. if (!hw) {
  770. memcpy(dev->dev_addr, addr, dev->addr_len);
  771. return 0;
  772. }
  773. /* for rlb each slave must have a unique hw mac addresses so that */
  774. /* each slave will receive packets destined to a different mac */
  775. memcpy(s_addr.sa_data, addr, dev->addr_len);
  776. s_addr.sa_family = dev->type;
  777. if (dev_set_mac_address(dev, &s_addr)) {
  778. pr_err("%s: Error: dev_set_mac_address of dev %s failed!\n"
  779. "ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
  780. dev->master->name, dev->name);
  781. return -EOPNOTSUPP;
  782. }
  783. return 0;
  784. }
  785. /*
  786. * Swap MAC addresses between two slaves.
  787. *
  788. * Called with RTNL held, and no other locks.
  789. *
  790. */
  791. static void alb_swap_mac_addr(struct bonding *bond, struct slave *slave1, struct slave *slave2)
  792. {
  793. u8 tmp_mac_addr[ETH_ALEN];
  794. memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
  795. alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr, bond->alb_info.rlb_enabled);
  796. alb_set_slave_mac_addr(slave2, tmp_mac_addr, bond->alb_info.rlb_enabled);
  797. }
  798. /*
  799. * Send learning packets after MAC address swap.
  800. *
  801. * Called with RTNL and no other locks
  802. */
  803. static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
  804. struct slave *slave2)
  805. {
  806. int slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
  807. struct slave *disabled_slave = NULL;
  808. ASSERT_RTNL();
  809. /* fasten the change in the switch */
  810. if (SLAVE_IS_OK(slave1)) {
  811. alb_send_learning_packets(slave1, slave1->dev->dev_addr);
  812. if (bond->alb_info.rlb_enabled) {
  813. /* inform the clients that the mac address
  814. * has changed
  815. */
  816. rlb_req_update_slave_clients(bond, slave1);
  817. }
  818. } else {
  819. disabled_slave = slave1;
  820. }
  821. if (SLAVE_IS_OK(slave2)) {
  822. alb_send_learning_packets(slave2, slave2->dev->dev_addr);
  823. if (bond->alb_info.rlb_enabled) {
  824. /* inform the clients that the mac address
  825. * has changed
  826. */
  827. rlb_req_update_slave_clients(bond, slave2);
  828. }
  829. } else {
  830. disabled_slave = slave2;
  831. }
  832. if (bond->alb_info.rlb_enabled && slaves_state_differ) {
  833. /* A disabled slave was assigned an active mac addr */
  834. rlb_teach_disabled_mac_on_primary(bond,
  835. disabled_slave->dev->dev_addr);
  836. }
  837. }
  838. /**
  839. * alb_change_hw_addr_on_detach
  840. * @bond: bonding we're working on
  841. * @slave: the slave that was just detached
  842. *
  843. * We assume that @slave was already detached from the slave list.
  844. *
  845. * If @slave's permanent hw address is different both from its current
  846. * address and from @bond's address, then somewhere in the bond there's
  847. * a slave that has @slave's permanet address as its current address.
  848. * We'll make sure that that slave no longer uses @slave's permanent address.
  849. *
  850. * Caller must hold RTNL and no other locks
  851. */
  852. static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
  853. {
  854. int perm_curr_diff;
  855. int perm_bond_diff;
  856. perm_curr_diff = compare_ether_addr_64bits(slave->perm_hwaddr,
  857. slave->dev->dev_addr);
  858. perm_bond_diff = compare_ether_addr_64bits(slave->perm_hwaddr,
  859. bond->dev->dev_addr);
  860. if (perm_curr_diff && perm_bond_diff) {
  861. struct slave *tmp_slave;
  862. int i, found = 0;
  863. bond_for_each_slave(bond, tmp_slave, i) {
  864. if (!compare_ether_addr_64bits(slave->perm_hwaddr,
  865. tmp_slave->dev->dev_addr)) {
  866. found = 1;
  867. break;
  868. }
  869. }
  870. if (found) {
  871. /* locking: needs RTNL and nothing else */
  872. alb_swap_mac_addr(bond, slave, tmp_slave);
  873. alb_fasten_mac_swap(bond, slave, tmp_slave);
  874. }
  875. }
  876. }
  877. /**
  878. * alb_handle_addr_collision_on_attach
  879. * @bond: bonding we're working on
  880. * @slave: the slave that was just attached
  881. *
  882. * checks uniqueness of slave's mac address and handles the case the
  883. * new slave uses the bonds mac address.
  884. *
  885. * If the permanent hw address of @slave is @bond's hw address, we need to
  886. * find a different hw address to give @slave, that isn't in use by any other
  887. * slave in the bond. This address must be, of course, one of the premanent
  888. * addresses of the other slaves.
  889. *
  890. * We go over the slave list, and for each slave there we compare its
  891. * permanent hw address with the current address of all the other slaves.
  892. * If no match was found, then we've found a slave with a permanent address
  893. * that isn't used by any other slave in the bond, so we can assign it to
  894. * @slave.
  895. *
  896. * assumption: this function is called before @slave is attached to the
  897. * bond slave list.
  898. *
  899. * caller must hold the bond lock for write since the mac addresses are compared
  900. * and may be swapped.
  901. */
  902. static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
  903. {
  904. struct slave *tmp_slave1, *tmp_slave2, *free_mac_slave;
  905. struct slave *has_bond_addr = bond->curr_active_slave;
  906. int i, j, found = 0;
  907. if (bond->slave_cnt == 0) {
  908. /* this is the first slave */
  909. return 0;
  910. }
  911. /* if slave's mac address differs from bond's mac address
  912. * check uniqueness of slave's mac address against the other
  913. * slaves in the bond.
  914. */
  915. if (compare_ether_addr_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
  916. bond_for_each_slave(bond, tmp_slave1, i) {
  917. if (!compare_ether_addr_64bits(tmp_slave1->dev->dev_addr,
  918. slave->dev->dev_addr)) {
  919. found = 1;
  920. break;
  921. }
  922. }
  923. if (!found)
  924. return 0;
  925. /* Try setting slave mac to bond address and fall-through
  926. to code handling that situation below... */
  927. alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
  928. bond->alb_info.rlb_enabled);
  929. }
  930. /* The slave's address is equal to the address of the bond.
  931. * Search for a spare address in the bond for this slave.
  932. */
  933. free_mac_slave = NULL;
  934. bond_for_each_slave(bond, tmp_slave1, i) {
  935. found = 0;
  936. bond_for_each_slave(bond, tmp_slave2, j) {
  937. if (!compare_ether_addr_64bits(tmp_slave1->perm_hwaddr,
  938. tmp_slave2->dev->dev_addr)) {
  939. found = 1;
  940. break;
  941. }
  942. }
  943. if (!found) {
  944. /* no slave has tmp_slave1's perm addr
  945. * as its curr addr
  946. */
  947. free_mac_slave = tmp_slave1;
  948. break;
  949. }
  950. if (!has_bond_addr) {
  951. if (!compare_ether_addr_64bits(tmp_slave1->dev->dev_addr,
  952. bond->dev->dev_addr)) {
  953. has_bond_addr = tmp_slave1;
  954. }
  955. }
  956. }
  957. if (free_mac_slave) {
  958. alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
  959. bond->alb_info.rlb_enabled);
  960. pr_warning("%s: Warning: the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
  961. bond->dev->name, slave->dev->name,
  962. free_mac_slave->dev->name);
  963. } else if (has_bond_addr) {
  964. pr_err("%s: Error: the hw address of slave %s is in use by the bond; couldn't find a slave with a free hw address to give it (this should not have happened)\n",
  965. bond->dev->name, slave->dev->name);
  966. return -EFAULT;
  967. }
  968. return 0;
  969. }
  970. /**
  971. * alb_set_mac_address
  972. * @bond:
  973. * @addr:
  974. *
  975. * In TLB mode all slaves are configured to the bond's hw address, but set
  976. * their dev_addr field to different addresses (based on their permanent hw
  977. * addresses).
  978. *
  979. * For each slave, this function sets the interface to the new address and then
  980. * changes its dev_addr field to its previous value.
  981. *
  982. * Unwinding assumes bond's mac address has not yet changed.
  983. */
  984. static int alb_set_mac_address(struct bonding *bond, void *addr)
  985. {
  986. struct sockaddr sa;
  987. struct slave *slave, *stop_at;
  988. char tmp_addr[ETH_ALEN];
  989. int res;
  990. int i;
  991. if (bond->alb_info.rlb_enabled) {
  992. return 0;
  993. }
  994. bond_for_each_slave(bond, slave, i) {
  995. /* save net_device's current hw address */
  996. memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
  997. res = dev_set_mac_address(slave->dev, addr);
  998. /* restore net_device's hw address */
  999. memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
  1000. if (res)
  1001. goto unwind;
  1002. }
  1003. return 0;
  1004. unwind:
  1005. memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
  1006. sa.sa_family = bond->dev->type;
  1007. /* unwind from head to the slave that failed */
  1008. stop_at = slave;
  1009. bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
  1010. memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
  1011. dev_set_mac_address(slave->dev, &sa);
  1012. memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
  1013. }
  1014. return res;
  1015. }
  1016. /************************ exported alb funcions ************************/
  1017. int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
  1018. {
  1019. int res;
  1020. res = tlb_initialize(bond);
  1021. if (res) {
  1022. return res;
  1023. }
  1024. if (rlb_enabled) {
  1025. bond->alb_info.rlb_enabled = 1;
  1026. /* initialize rlb */
  1027. res = rlb_initialize(bond);
  1028. if (res) {
  1029. tlb_deinitialize(bond);
  1030. return res;
  1031. }
  1032. } else {
  1033. bond->alb_info.rlb_enabled = 0;
  1034. }
  1035. return 0;
  1036. }
  1037. void bond_alb_deinitialize(struct bonding *bond)
  1038. {
  1039. struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
  1040. tlb_deinitialize(bond);
  1041. if (bond_info->rlb_enabled) {
  1042. rlb_deinitialize(bond);
  1043. }
  1044. }
  1045. int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
  1046. {
  1047. struct bonding *bond = netdev_priv(bond_dev);
  1048. struct ethhdr *eth_data;
  1049. struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
  1050. struct slave *tx_slave = NULL;
  1051. static const __be32 ip_bcast = htonl(0xffffffff);
  1052. int hash_size = 0;
  1053. int do_tx_balance = 1;
  1054. u32 hash_index = 0;
  1055. const u8 *hash_start = NULL;
  1056. int res = 1;
  1057. struct ipv6hdr *ip6hdr;
  1058. skb_reset_mac_header(skb);
  1059. eth_data = eth_hdr(skb);
  1060. /* make sure that the curr_active_slave and the slaves list do
  1061. * not change during tx
  1062. */
  1063. read_lock(&bond->lock);
  1064. read_lock(&bond->curr_slave_lock);
  1065. if (!BOND_IS_OK(bond)) {
  1066. goto out;
  1067. }
  1068. switch (ntohs(skb->protocol)) {
  1069. case ETH_P_IP: {
  1070. const struct iphdr *iph = ip_hdr(skb);
  1071. if (!compare_ether_addr_64bits(eth_data->h_dest, mac_bcast) ||
  1072. (iph->daddr == ip_bcast) ||
  1073. (iph->protocol == IPPROTO_IGMP)) {
  1074. do_tx_balance = 0;
  1075. break;
  1076. }
  1077. hash_start = (char *)&(iph->daddr);
  1078. hash_size = sizeof(iph->daddr);
  1079. }
  1080. break;
  1081. case ETH_P_IPV6:
  1082. /* IPv6 doesn't really use broadcast mac address, but leave
  1083. * that here just in case.
  1084. */
  1085. if (!compare_ether_addr_64bits(eth_data->h_dest, mac_bcast)) {
  1086. do_tx_balance = 0;
  1087. break;
  1088. }
  1089. /* IPv6 uses all-nodes multicast as an equivalent to
  1090. * broadcasts in IPv4.
  1091. */
  1092. if (!compare_ether_addr_64bits(eth_data->h_dest, mac_v6_allmcast)) {
  1093. do_tx_balance = 0;
  1094. break;
  1095. }
  1096. /* Additianally, DAD probes should not be tx-balanced as that
  1097. * will lead to false positives for duplicate addresses and
  1098. * prevent address configuration from working.
  1099. */
  1100. ip6hdr = ipv6_hdr(skb);
  1101. if (ipv6_addr_any(&ip6hdr->saddr)) {
  1102. do_tx_balance = 0;
  1103. break;
  1104. }
  1105. hash_start = (char *)&(ipv6_hdr(skb)->daddr);
  1106. hash_size = sizeof(ipv6_hdr(skb)->daddr);
  1107. break;
  1108. case ETH_P_IPX:
  1109. if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
  1110. /* something is wrong with this packet */
  1111. do_tx_balance = 0;
  1112. break;
  1113. }
  1114. if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
  1115. /* The only protocol worth balancing in
  1116. * this family since it has an "ARP" like
  1117. * mechanism
  1118. */
  1119. do_tx_balance = 0;
  1120. break;
  1121. }
  1122. hash_start = (char*)eth_data->h_dest;
  1123. hash_size = ETH_ALEN;
  1124. break;
  1125. case ETH_P_ARP:
  1126. do_tx_balance = 0;
  1127. if (bond_info->rlb_enabled) {
  1128. tx_slave = rlb_arp_xmit(skb, bond);
  1129. }
  1130. break;
  1131. default:
  1132. do_tx_balance = 0;
  1133. break;
  1134. }
  1135. if (do_tx_balance) {
  1136. hash_index = _simple_hash(hash_start, hash_size);
  1137. tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
  1138. }
  1139. if (!tx_slave) {
  1140. /* unbalanced or unassigned, send through primary */
  1141. tx_slave = bond->curr_active_slave;
  1142. bond_info->unbalanced_load += skb->len;
  1143. }
  1144. if (tx_slave && SLAVE_IS_OK(tx_slave)) {
  1145. if (tx_slave != bond->curr_active_slave) {
  1146. memcpy(eth_data->h_source,
  1147. tx_slave->dev->dev_addr,
  1148. ETH_ALEN);
  1149. }
  1150. res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
  1151. } else {
  1152. if (tx_slave) {
  1153. tlb_clear_slave(bond, tx_slave, 0);
  1154. }
  1155. }
  1156. out:
  1157. if (res) {
  1158. /* no suitable interface, frame not sent */
  1159. dev_kfree_skb(skb);
  1160. }
  1161. read_unlock(&bond->curr_slave_lock);
  1162. read_unlock(&bond->lock);
  1163. return NETDEV_TX_OK;
  1164. }
  1165. void bond_alb_monitor(struct work_struct *work)
  1166. {
  1167. struct bonding *bond = container_of(work, struct bonding,
  1168. alb_work.work);
  1169. struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
  1170. struct slave *slave;
  1171. int i;
  1172. read_lock(&bond->lock);
  1173. if (bond->kill_timers) {
  1174. goto out;
  1175. }
  1176. if (bond->slave_cnt == 0) {
  1177. bond_info->tx_rebalance_counter = 0;
  1178. bond_info->lp_counter = 0;
  1179. goto re_arm;
  1180. }
  1181. bond_info->tx_rebalance_counter++;
  1182. bond_info->lp_counter++;
  1183. /* send learning packets */
  1184. if (bond_info->lp_counter >= BOND_ALB_LP_TICKS) {
  1185. /* change of curr_active_slave involves swapping of mac addresses.
  1186. * in order to avoid this swapping from happening while
  1187. * sending the learning packets, the curr_slave_lock must be held for
  1188. * read.
  1189. */
  1190. read_lock(&bond->curr_slave_lock);
  1191. bond_for_each_slave(bond, slave, i) {
  1192. alb_send_learning_packets(slave, slave->dev->dev_addr);
  1193. }
  1194. read_unlock(&bond->curr_slave_lock);
  1195. bond_info->lp_counter = 0;
  1196. }
  1197. /* rebalance tx traffic */
  1198. if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
  1199. read_lock(&bond->curr_slave_lock);
  1200. bond_for_each_slave(bond, slave, i) {
  1201. tlb_clear_slave(bond, slave, 1);
  1202. if (slave == bond->curr_active_slave) {
  1203. SLAVE_TLB_INFO(slave).load =
  1204. bond_info->unbalanced_load /
  1205. BOND_TLB_REBALANCE_INTERVAL;
  1206. bond_info->unbalanced_load = 0;
  1207. }
  1208. }
  1209. read_unlock(&bond->curr_slave_lock);
  1210. bond_info->tx_rebalance_counter = 0;
  1211. }
  1212. /* handle rlb stuff */
  1213. if (bond_info->rlb_enabled) {
  1214. if (bond_info->primary_is_promisc &&
  1215. (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
  1216. /*
  1217. * dev_set_promiscuity requires rtnl and
  1218. * nothing else.
  1219. */
  1220. read_unlock(&bond->lock);
  1221. rtnl_lock();
  1222. bond_info->rlb_promisc_timeout_counter = 0;
  1223. /* If the primary was set to promiscuous mode
  1224. * because a slave was disabled then
  1225. * it can now leave promiscuous mode.
  1226. */
  1227. dev_set_promiscuity(bond->curr_active_slave->dev, -1);
  1228. bond_info->primary_is_promisc = 0;
  1229. rtnl_unlock();
  1230. read_lock(&bond->lock);
  1231. }
  1232. if (bond_info->rlb_rebalance) {
  1233. bond_info->rlb_rebalance = 0;
  1234. rlb_rebalance(bond);
  1235. }
  1236. /* check if clients need updating */
  1237. if (bond_info->rx_ntt) {
  1238. if (bond_info->rlb_update_delay_counter) {
  1239. --bond_info->rlb_update_delay_counter;
  1240. } else {
  1241. rlb_update_rx_clients(bond);
  1242. if (bond_info->rlb_update_retry_counter) {
  1243. --bond_info->rlb_update_retry_counter;
  1244. } else {
  1245. bond_info->rx_ntt = 0;
  1246. }
  1247. }
  1248. }
  1249. }
  1250. re_arm:
  1251. queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
  1252. out:
  1253. read_unlock(&bond->lock);
  1254. }
  1255. /* assumption: called before the slave is attached to the bond
  1256. * and not locked by the bond lock
  1257. */
  1258. int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
  1259. {
  1260. int res;
  1261. res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
  1262. bond->alb_info.rlb_enabled);
  1263. if (res) {
  1264. return res;
  1265. }
  1266. /* caller must hold the bond lock for write since the mac addresses
  1267. * are compared and may be swapped.
  1268. */
  1269. read_lock(&bond->lock);
  1270. res = alb_handle_addr_collision_on_attach(bond, slave);
  1271. read_unlock(&bond->lock);
  1272. if (res) {
  1273. return res;
  1274. }
  1275. tlb_init_slave(slave);
  1276. /* order a rebalance ASAP */
  1277. bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
  1278. if (bond->alb_info.rlb_enabled) {
  1279. bond->alb_info.rlb_rebalance = 1;
  1280. }
  1281. return 0;
  1282. }
  1283. /*
  1284. * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
  1285. * if necessary.
  1286. *
  1287. * Caller must hold RTNL and no other locks
  1288. */
  1289. void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
  1290. {
  1291. if (bond->slave_cnt > 1) {
  1292. alb_change_hw_addr_on_detach(bond, slave);
  1293. }
  1294. tlb_clear_slave(bond, slave, 0);
  1295. if (bond->alb_info.rlb_enabled) {
  1296. bond->alb_info.next_rx_slave = NULL;
  1297. rlb_clear_slave(bond, slave);
  1298. }
  1299. }
  1300. /* Caller must hold bond lock for read */
  1301. void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
  1302. {
  1303. struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
  1304. if (link == BOND_LINK_DOWN) {
  1305. tlb_clear_slave(bond, slave, 0);
  1306. if (bond->alb_info.rlb_enabled) {
  1307. rlb_clear_slave(bond, slave);
  1308. }
  1309. } else if (link == BOND_LINK_UP) {
  1310. /* order a rebalance ASAP */
  1311. bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
  1312. if (bond->alb_info.rlb_enabled) {
  1313. bond->alb_info.rlb_rebalance = 1;
  1314. /* If the updelay module parameter is smaller than the
  1315. * forwarding delay of the switch the rebalance will
  1316. * not work because the rebalance arp replies will
  1317. * not be forwarded to the clients..
  1318. */
  1319. }
  1320. }
  1321. }
  1322. /**
  1323. * bond_alb_handle_active_change - assign new curr_active_slave
  1324. * @bond: our bonding struct
  1325. * @new_slave: new slave to assign
  1326. *
  1327. * Set the bond->curr_active_slave to @new_slave and handle
  1328. * mac address swapping and promiscuity changes as needed.
  1329. *
  1330. * If new_slave is NULL, caller must hold curr_slave_lock or
  1331. * bond->lock for write.
  1332. *
  1333. * If new_slave is not NULL, caller must hold RTNL, bond->lock for
  1334. * read and curr_slave_lock for write. Processing here may sleep, so
  1335. * no other locks may be held.
  1336. */
  1337. void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
  1338. __releases(&bond->curr_slave_lock)
  1339. __releases(&bond->lock)
  1340. __acquires(&bond->lock)
  1341. __acquires(&bond->curr_slave_lock)
  1342. {
  1343. struct slave *swap_slave;
  1344. int i;
  1345. if (bond->curr_active_slave == new_slave) {
  1346. return;
  1347. }
  1348. if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
  1349. dev_set_promiscuity(bond->curr_active_slave->dev, -1);
  1350. bond->alb_info.primary_is_promisc = 0;
  1351. bond->alb_info.rlb_promisc_timeout_counter = 0;
  1352. }
  1353. swap_slave = bond->curr_active_slave;
  1354. bond->curr_active_slave = new_slave;
  1355. if (!new_slave || (bond->slave_cnt == 0)) {
  1356. return;
  1357. }
  1358. /* set the new curr_active_slave to the bonds mac address
  1359. * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
  1360. */
  1361. if (!swap_slave) {
  1362. struct slave *tmp_slave;
  1363. /* find slave that is holding the bond's mac address */
  1364. bond_for_each_slave(bond, tmp_slave, i) {
  1365. if (!compare_ether_addr_64bits(tmp_slave->dev->dev_addr,
  1366. bond->dev->dev_addr)) {
  1367. swap_slave = tmp_slave;
  1368. break;
  1369. }
  1370. }
  1371. }
  1372. /*
  1373. * Arrange for swap_slave and new_slave to temporarily be
  1374. * ignored so we can mess with their MAC addresses without
  1375. * fear of interference from transmit activity.
  1376. */
  1377. if (swap_slave) {
  1378. tlb_clear_slave(bond, swap_slave, 1);
  1379. }
  1380. tlb_clear_slave(bond, new_slave, 1);
  1381. write_unlock_bh(&bond->curr_slave_lock);
  1382. read_unlock(&bond->lock);
  1383. ASSERT_RTNL();
  1384. /* curr_active_slave must be set before calling alb_swap_mac_addr */
  1385. if (swap_slave) {
  1386. /* swap mac address */
  1387. alb_swap_mac_addr(bond, swap_slave, new_slave);
  1388. } else {
  1389. /* set the new_slave to the bond mac address */
  1390. alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
  1391. bond->alb_info.rlb_enabled);
  1392. }
  1393. if (swap_slave) {
  1394. alb_fasten_mac_swap(bond, swap_slave, new_slave);
  1395. read_lock(&bond->lock);
  1396. } else {
  1397. read_lock(&bond->lock);
  1398. alb_send_learning_packets(new_slave, bond->dev->dev_addr);
  1399. }
  1400. write_lock_bh(&bond->curr_slave_lock);
  1401. }
  1402. /*
  1403. * Called with RTNL
  1404. */
  1405. int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
  1406. __acquires(&bond->lock)
  1407. __releases(&bond->lock)
  1408. {
  1409. struct bonding *bond = netdev_priv(bond_dev);
  1410. struct sockaddr *sa = addr;
  1411. struct slave *slave, *swap_slave;
  1412. int res;
  1413. int i;
  1414. if (!is_valid_ether_addr(sa->sa_data)) {
  1415. return -EADDRNOTAVAIL;
  1416. }
  1417. res = alb_set_mac_address(bond, addr);
  1418. if (res) {
  1419. return res;
  1420. }
  1421. memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
  1422. /* If there is no curr_active_slave there is nothing else to do.
  1423. * Otherwise we'll need to pass the new address to it and handle
  1424. * duplications.
  1425. */
  1426. if (!bond->curr_active_slave) {
  1427. return 0;
  1428. }
  1429. swap_slave = NULL;
  1430. bond_for_each_slave(bond, slave, i) {
  1431. if (!compare_ether_addr_64bits(slave->dev->dev_addr,
  1432. bond_dev->dev_addr)) {
  1433. swap_slave = slave;
  1434. break;
  1435. }
  1436. }
  1437. if (swap_slave) {
  1438. alb_swap_mac_addr(bond, swap_slave, bond->curr_active_slave);
  1439. alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave);
  1440. } else {
  1441. alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr,
  1442. bond->alb_info.rlb_enabled);
  1443. read_lock(&bond->lock);
  1444. alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
  1445. if (bond->alb_info.rlb_enabled) {
  1446. /* inform clients mac address has changed */
  1447. rlb_req_update_slave_clients(bond, bond->curr_active_slave);
  1448. }
  1449. read_unlock(&bond->lock);
  1450. }
  1451. return 0;
  1452. }
  1453. void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
  1454. {
  1455. if (bond->alb_info.current_alb_vlan &&
  1456. (bond->alb_info.current_alb_vlan->vlan_id == vlan_id)) {
  1457. bond->alb_info.current_alb_vlan = NULL;
  1458. }
  1459. if (bond->alb_info.rlb_enabled) {
  1460. rlb_clear_vlan(bond, vlan_id);
  1461. }
  1462. }