drivers/net/bonding/bond_alb.c

   1 /*
   2  * Copyright(c) 1999 - 2003 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  * Changes:
  23  *
  24  * 2003/06/25 - Shmulik Hen <shmulik.hen at intel dot com>
  25  *      - Fixed signed/unsigned calculation errors that caused load sharing
  26  *        to collapse to one slave under very heavy UDP Tx stress.
  27  *
  28  * 2003/08/06 - Amir Noam <amir.noam at intel dot com>
  29  *      - Add support for setting bond's MAC address with special
  30  *        handling required for ALB/TLB.
  31  *
  32  * 2003/09/24 - Shmulik Hen <shmulik.hen at intel dot com>
  33  *      - Code cleanup and style changes
  34  */
  35
  36 //#define BONDING_DEBUG 1
  37
  38 #include <linux/skbuff.h>
  39 #include <linux/netdevice.h>
  40 #include <linux/etherdevice.h>
  41 #include <linux/pkt_sched.h>
  42 #include <linux/spinlock.h>
  43 #include <linux/slab.h>
  44 #include <linux/timer.h>
  45 #include <linux/ip.h>
  46 #include <linux/ipv6.h>
  47 #include <linux/if_arp.h>
  48 #include <linux/if_ether.h>
  49 #include <linux/if_bonding.h>
  50 #include <net/ipx.h>
  51 #include <net/arp.h>
  52 #include <asm/byteorder.h>
  53 #include "bonding.h"
  54 #include "bond_alb.h"
  55
  56
  57 #define ALB_TIMER_TICKS_PER_SEC     10  /* should be a divisor of HZ */
  58 #define BOND_TLB_REBALANCE_INTERVAL 10  /* In seconds, periodic re-balancing.
  59                                          * Used for division - never set
  60                                          * to zero !!!
  61                                          */
  62 #define BOND_ALB_LP_INTERVAL        1   /* In seconds, periodic send of
  63                                          * learning packets to the switch
  64                                          */
  65
  66 #define BOND_TLB_REBALANCE_TICKS (BOND_TLB_REBALANCE_INTERVAL \
  67                                   * ALB_TIMER_TICKS_PER_SEC)
  68
  69 #define BOND_ALB_LP_TICKS (BOND_ALB_LP_INTERVAL \
  70                            * ALB_TIMER_TICKS_PER_SEC)
  71
  72 #define TLB_HASH_TABLE_SIZE 256 /* The size of the clients hash table.
  73                                  * Note that this value MUST NOT be smaller
  74                                  * because the key hash table is BYTE wide !
  75                                  */
  76
  77
  78 #define TLB_NULL_INDEX          0xffffffff
  79 #define MAX_LP_RETRY            3
  80
  81 /* rlb defs */
  82 #define RLB_HASH_TABLE_SIZE     256
  83 #define RLB_NULL_INDEX          0xffffffff
  84 #define RLB_UPDATE_DELAY        2*ALB_TIMER_TICKS_PER_SEC /* 2 seconds */
  85 #define RLB_ARP_BURST_SIZE      2
  86 #define RLB_UPDATE_RETRY        3       /* 3-ticks - must be smaller than the rlb
  87                                          * rebalance interval (5 min).
  88                                          */
  89 /* RLB_PROMISC_TIMEOUT = 10 sec equals the time that the current slave is
  90  * promiscuous after failover
  91  */
  92 #define RLB_PROMISC_TIMEOUT     10*ALB_TIMER_TICKS_PER_SEC
  93
  94 static const u8 mac_bcast[ETH_ALEN] = {0xff,0xff,0xff,0xff,0xff,0xff};
  95 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
  96
  97 #pragma pack(1)
  98 struct learning_pkt {
  99         u8 mac_dst[ETH_ALEN];
 100         u8 mac_src[ETH_ALEN];
 101         u16 type;
 102         u8 padding[ETH_ZLEN - ETH_HLEN];
 103 };
 104
 105 struct arp_pkt {
 106         u16     hw_addr_space;
 107         u16     prot_addr_space;
 108         u8      hw_addr_len;
 109         u8      prot_addr_len;
 110         u16     op_code;
 111         u8      mac_src[ETH_ALEN];      /* sender hardware address */
 112         u32     ip_src;                 /* sender IP address */
 113         u8      mac_dst[ETH_ALEN];      /* target hardware address */
 114         u32     ip_dst;                 /* target IP address */
 115 };
 116 #pragma pack()
 117
 118 /* Forward declaration */
 119 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[]);
 120
 121 static inline u8 _simple_hash(u8 *hash_start, int hash_size)
 122 {
 123         int i;
 124         u8 hash = 0;
 125
 126         for (i = 0; i < hash_size; i++) {
 127                 hash ^= hash_start[i];
 128         }
 129
 130         return hash;
 131 }
 132
 133 /*********************** tlb specific functions ***************************/
 134
 135 static inline void _lock_tx_hashtbl(struct bonding *bond)
 136 {
 137         spin_lock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 138 }
 139
 140 static inline void _unlock_tx_hashtbl(struct bonding *bond)
 141 {
 142         spin_unlock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 143 }
 144
 145 /* Caller must hold tx_hashtbl lock */
 146 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
 147 {
 148         if (save_load) {
 149                 entry->load_history = 1 + entry->tx_bytes /
 150                                       BOND_TLB_REBALANCE_INTERVAL;
 151                 entry->tx_bytes = 0;
 152         }
 153
 154         entry->tx_slave = NULL;
 155         entry->next = TLB_NULL_INDEX;
 156         entry->prev = TLB_NULL_INDEX;
 157 }
 158
 159 static inline void tlb_init_slave(struct slave *slave)
 160 {
 161         SLAVE_TLB_INFO(slave).load = 0;
 162         SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
 163 }
 164
 165 /* Caller must hold bond lock for read */
 166 static void tlb_clear_slave(struct bonding *bond, struct slave *slave, int save_load)
 167 {
 168         struct tlb_client_info *tx_hash_table;
 169         u32 index;
 170
 171         _lock_tx_hashtbl(bond);
 172
 173         /* clear slave from tx_hashtbl */
 174         tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
 175
 176         index = SLAVE_TLB_INFO(slave).head;
 177         while (index != TLB_NULL_INDEX) {
 178                 u32 next_index = tx_hash_table[index].next;
 179                 tlb_init_table_entry(&tx_hash_table[index], save_load);
 180                 index = next_index;
 181         }
 182
 183         _unlock_tx_hashtbl(bond);
 184
 185         tlb_init_slave(slave);
 186 }
 187
 188 /* Must be called before starting the monitor timer */
 189 static int tlb_initialize(struct bonding *bond)
 190 {
 191         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 192         int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 193         int i;
 194
 195         spin_lock_init(&(bond_info->tx_hashtbl_lock));
 196
 197         _lock_tx_hashtbl(bond);
 198
 199         bond_info->tx_hashtbl = kmalloc(size, GFP_KERNEL);
 200         if (!bond_info->tx_hashtbl) {
 201                 printk(KERN_ERR DRV_NAME
 202                        ": Error: %s: Failed to allocate TLB hash table\n",
 203                        bond->dev->name);
 204                 _unlock_tx_hashtbl(bond);
 205                 return -1;
 206         }
 207
 208         memset(bond_info->tx_hashtbl, 0, size);
 209
 210         for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
 211                 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 1);
 212         }
 213
 214         _unlock_tx_hashtbl(bond);
 215
 216         return 0;
 217 }
 218
 219 /* Must be called only after all slaves have been released */
 220 static void tlb_deinitialize(struct bonding *bond)
 221 {
 222         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 223
 224         _lock_tx_hashtbl(bond);
 225
 226         kfree(bond_info->tx_hashtbl);
 227         bond_info->tx_hashtbl = NULL;
 228
 229         _unlock_tx_hashtbl(bond);
 230 }
 231
 232 /* Caller must hold bond lock for read */
 233 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
 234 {
 235         struct slave *slave, *least_loaded;
 236         s64 max_gap;
 237         int i, found = 0;
 238
 239         /* Find the first enabled slave */
 240         bond_for_each_slave(bond, slave, i) {
 241                 if (SLAVE_IS_OK(slave)) {
 242                         found = 1;
 243                         break;
 244                 }
 245         }
 246
 247         if (!found) {
 248                 return NULL;
 249         }
 250
 251         least_loaded = slave;
 252         max_gap = (s64)(slave->speed << 20) - /* Convert to Megabit per sec */
 253                         (s64)(SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
 254
 255         /* Find the slave with the largest gap */
 256         bond_for_each_slave_from(bond, slave, i, least_loaded) {
 257                 if (SLAVE_IS_OK(slave)) {
 258                         s64 gap = (s64)(slave->speed << 20) -
 259                                         (s64)(SLAVE_TLB_INFO(slave).load << 3);
 260                         if (max_gap < gap) {
 261                                 least_loaded = slave;
 262                                 max_gap = gap;
 263                         }
 264                 }
 265         }
 266
 267         return least_loaded;
 268 }
 269
 270 /* Caller must hold bond lock for read */
 271 struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index, u32 skb_len)
 272 {
 273         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 274         struct tlb_client_info *hash_table;
 275         struct slave *assigned_slave;
 276
 277         _lock_tx_hashtbl(bond);
 278
 279         hash_table = bond_info->tx_hashtbl;
 280         assigned_slave = hash_table[hash_index].tx_slave;
 281         if (!assigned_slave) {
 282                 assigned_slave = tlb_get_least_loaded_slave(bond);
 283
 284                 if (assigned_slave) {
 285                         struct tlb_slave_info *slave_info =
 286                                 &(SLAVE_TLB_INFO(assigned_slave));
 287                         u32 next_index = slave_info->head;
 288
 289                         hash_table[hash_index].tx_slave = assigned_slave;
 290                         hash_table[hash_index].next = next_index;
 291                         hash_table[hash_index].prev = TLB_NULL_INDEX;
 292
 293                         if (next_index != TLB_NULL_INDEX) {
 294                                 hash_table[next_index].prev = hash_index;
 295                         }
 296
 297                         slave_info->head = hash_index;
 298                         slave_info->load +=
 299                                 hash_table[hash_index].load_history;
 300                 }
 301         }
 302
 303         if (assigned_slave) {
 304                 hash_table[hash_index].tx_bytes += skb_len;
 305         }
 306
 307         _unlock_tx_hashtbl(bond);
 308
 309         return assigned_slave;
 310 }
 311
 312 /*********************** rlb specific functions ***************************/
 313 static inline void _lock_rx_hashtbl(struct bonding *bond)
 314 {
 315         spin_lock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 316 }
 317
 318 static inline void _unlock_rx_hashtbl(struct bonding *bond)
 319 {
 320         spin_unlock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 321 }
 322
 323 /* when an ARP REPLY is received from a client update its info
 324  * in the rx_hashtbl
 325  */
 326 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 327 {
 328         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 329         struct rlb_client_info *client_info;
 330         u32 hash_index;
 331
 332         _lock_rx_hashtbl(bond);
 333
 334         hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
 335         client_info = &(bond_info->rx_hashtbl[hash_index]);
 336
 337         if ((client_info->assigned) &&
 338             (client_info->ip_src == arp->ip_dst) &&
 339             (client_info->ip_dst == arp->ip_src)) {
 340                 /* update the clients MAC address */
 341                 memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
 342                 client_info->ntt = 1;
 343                 bond_info->rx_ntt = 1;
 344         }
 345
 346         _unlock_rx_hashtbl(bond);
 347 }
 348
 349 static int rlb_arp_recv(struct sk_buff *skb, struct net_device *bond_dev, struct packet_type *ptype)
 350 {
 351         struct bonding *bond = bond_dev->priv;
 352         struct arp_pkt *arp = (struct arp_pkt *)skb->data;
 353         int res = NET_RX_DROP;
 354
 355         if (!(bond_dev->flags & IFF_MASTER)) {
 356                 goto out;
 357         }
 358
 359         if (!arp) {
 360                 dprintk("Packet has no ARP data\n");
 361                 goto out;
 362         }
 363
 364         if (skb->len < sizeof(struct arp_pkt)) {
 365                 dprintk("Packet is too small to be an ARP\n");
 366                 goto out;
 367         }
 368
 369         if (arp->op_code == htons(ARPOP_REPLY)) {
 370                 /* update rx hash table for this ARP */
 371                 rlb_update_entry_from_arp(bond, arp);
 372                 dprintk("Server received an ARP Reply from client\n");
 373         }
 374
 375         res = NET_RX_SUCCESS;
 376
 377 out:
 378         dev_kfree_skb(skb);
 379
 380         return res;
 381 }
 382
 383 /* Caller must hold bond lock for read */
 384 static struct slave *rlb_next_rx_slave(struct bonding *bond)
 385 {
 386         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 387         struct slave *rx_slave, *slave, *start_at;
 388         int i = 0;
 389
 390         if (bond_info->next_rx_slave) {
 391                 start_at = bond_info->next_rx_slave;
 392         } else {
 393                 start_at = bond->first_slave;
 394         }
 395
 396         rx_slave = NULL;
 397
 398         bond_for_each_slave_from(bond, slave, i, start_at) {
 399                 if (SLAVE_IS_OK(slave)) {
 400                         if (!rx_slave) {
 401                                 rx_slave = slave;
 402                         } else if (slave->speed > rx_slave->speed) {
 403                                 rx_slave = slave;
 404                         }
 405                 }
 406         }
 407
 408         if (rx_slave) {
 409                 bond_info->next_rx_slave = rx_slave->next;
 410         }
 411
 412         return rx_slave;
 413 }
 414
 415 /* teach the switch the mac of a disabled slave
 416  * on the primary for fault tolerance
 417  *
 418  * Caller must hold bond->curr_slave_lock for write or bond lock for write
 419  */
 420 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
 421 {
 422         if (!bond->curr_active_slave) {
 423                 return;
 424         }
 425
 426         if (!bond->alb_info.primary_is_promisc) {
 427                 bond->alb_info.primary_is_promisc = 1;
 428                 dev_set_promiscuity(bond->curr_active_slave->dev, 1);
 429         }
 430
 431         bond->alb_info.rlb_promisc_timeout_counter = 0;
 432
 433         alb_send_learning_packets(bond->curr_active_slave, addr);
 434 }
 435
 436 /* slave being removed should not be active at this point
 437  *
 438  * Caller must hold bond lock for read
 439  */
 440 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 441 {
 442         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 443         struct rlb_client_info *rx_hash_table;
 444         u32 index, next_index;
 445
 446         /* clear slave from rx_hashtbl */
 447         _lock_rx_hashtbl(bond);
 448
 449         rx_hash_table = bond_info->rx_hashtbl;
 450         index = bond_info->rx_hashtbl_head;
 451         for (; index != RLB_NULL_INDEX; index = next_index) {
 452                 next_index = rx_hash_table[index].next;
 453                 if (rx_hash_table[index].slave == slave) {
 454                         struct slave *assigned_slave = rlb_next_rx_slave(bond);
 455
 456                         if (assigned_slave) {
 457                                 rx_hash_table[index].slave = assigned_slave;
 458                                 if (memcmp(rx_hash_table[index].mac_dst,
 459                                            mac_bcast, ETH_ALEN)) {
 460                                         bond_info->rx_hashtbl[index].ntt = 1;
 461                                         bond_info->rx_ntt = 1;
 462                                         /* A slave has been removed from the
 463                                          * table because it is either disabled
 464                                          * or being released. We must retry the
 465                                          * update to avoid clients from not
 466                                          * being updated & disconnecting when
 467                                          * there is stress
 468                                          */
 469                                         bond_info->rlb_update_retry_counter =
 470                                                 RLB_UPDATE_RETRY;
 471                                 }
 472                         } else {  /* there is no active slave */
 473                                 rx_hash_table[index].slave = NULL;
 474                         }
 475                 }
 476         }
 477
 478         _unlock_rx_hashtbl(bond);
 479
 480         write_lock(&bond->curr_slave_lock);
 481
 482         if (slave != bond->curr_active_slave) {
 483                 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 484         }
 485
 486         write_unlock(&bond->curr_slave_lock);
 487 }
 488
 489 static void rlb_update_client(struct rlb_client_info *client_info)
 490 {
 491         int i;
 492
 493         if (!client_info->slave) {
 494                 return;
 495         }
 496
 497         for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 498                 arp_send(ARPOP_REPLY, ETH_P_ARP,
 499                          client_info->ip_dst,
 500                          client_info->slave->dev,
 501                          client_info->ip_src,
 502                          client_info->mac_dst,
 503                          client_info->slave->dev->dev_addr,
 504                          client_info->mac_dst);
 505         }
 506 }
 507
 508 /* sends ARP REPLIES that update the clients that need updating */
 509 static void rlb_update_rx_clients(struct bonding *bond)
 510 {
 511         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 512         struct rlb_client_info *client_info;
 513         u32 hash_index;
 514
 515         _lock_rx_hashtbl(bond);
 516
 517         hash_index = bond_info->rx_hashtbl_head;
 518         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 519                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 520                 if (client_info->ntt) {
 521                         rlb_update_client(client_info);
 522                         if (bond_info->rlb_update_retry_counter == 0) {
 523                                 client_info->ntt = 0;
 524                         }
 525                 }
 526         }
 527
 528         /* do not update the entries again untill this counter is zero so that
 529          * not to confuse the clients.
 530          */
 531         bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 532
 533         _unlock_rx_hashtbl(bond);
 534 }
 535
 536 /* The slave was assigned a new mac address - update the clients */
 537 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 538 {
 539         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 540         struct rlb_client_info *client_info;
 541         int ntt = 0;
 542         u32 hash_index;
 543
 544         _lock_rx_hashtbl(bond);
 545
 546         hash_index = bond_info->rx_hashtbl_head;
 547         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 548                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 549
 550                 if ((client_info->slave == slave) &&
 551                     memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
 552                         client_info->ntt = 1;
 553                         ntt = 1;
 554                 }
 555         }
 556
 557         // update the team's flag only after the whole iteration
 558         if (ntt) {
 559                 bond_info->rx_ntt = 1;
 560                 //fasten the change
 561                 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 562         }
 563
 564         _unlock_rx_hashtbl(bond);
 565 }
 566
 567 /* mark all clients using src_ip to be updated */
 568 static void rlb_req_update_subnet_clients(struct bonding *bond, u32 src_ip)
 569 {
 570         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 571         struct rlb_client_info *client_info;
 572         u32 hash_index;
 573
 574         _lock_rx_hashtbl(bond);
 575
 576         hash_index = bond_info->rx_hashtbl_head;
 577         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 578                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 579
 580                 if (!client_info->slave) {
 581                         printk(KERN_ERR DRV_NAME
 582                                ": Error: found a client with no channel in "
 583                                "the client's hash table\n");
 584                         continue;
 585                 }
 586                 /*update all clients using this src_ip, that are not assigned
 587                  * to the team's address (curr_active_slave) and have a known
 588                  * unicast mac address.
 589                  */
 590                 if ((client_info->ip_src == src_ip) &&
 591                     memcmp(client_info->slave->dev->dev_addr,
 592                            bond->dev->dev_addr, ETH_ALEN) &&
 593                     memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
 594                         client_info->ntt = 1;
 595                         bond_info->rx_ntt = 1;
 596                 }
 597         }
 598
 599         _unlock_rx_hashtbl(bond);
 600 }
 601
 602 /* Caller must hold both bond and ptr locks for read */
 603 struct slave *rlb_choose_channel(struct bonding *bond, struct arp_pkt *arp)
 604 {
 605         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 606         struct slave *assigned_slave;
 607         struct rlb_client_info *client_info;
 608         u32 hash_index = 0;
 609
 610         _lock_rx_hashtbl(bond);
 611
 612         hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_src));
 613         client_info = &(bond_info->rx_hashtbl[hash_index]);
 614
 615         if (client_info->assigned) {
 616                 if ((client_info->ip_src == arp->ip_src) &&
 617                     (client_info->ip_dst == arp->ip_dst)) {
 618                         /* the entry is already assigned to this client */
 619                         if (memcmp(arp->mac_dst, mac_bcast, ETH_ALEN)) {
 620                                 /* update mac address from arp */
 621                                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 622                         }
 623
 624                         assigned_slave = client_info->slave;
 625                         if (assigned_slave) {
 626                                 _unlock_rx_hashtbl(bond);
 627                                 return assigned_slave;
 628                         }
 629                 } else {
 630                         /* the entry is already assigned to some other client,
 631                          * move the old client to primary (curr_active_slave) so
 632                          * that the new client can be assigned to this entry.
 633                          */
 634                         if (bond->curr_active_slave &&
 635                             client_info->slave != bond->curr_active_slave) {
 636                                 client_info->slave = bond->curr_active_slave;
 637                                 rlb_update_client(client_info);
 638                         }
 639                 }
 640         }
 641         /* assign a new slave */
 642         assigned_slave = rlb_next_rx_slave(bond);
 643
 644         if (assigned_slave) {
 645                 client_info->ip_src = arp->ip_src;
 646                 client_info->ip_dst = arp->ip_dst;
 647                 /* arp->mac_dst is broadcast for arp reqeusts.
 648                  * will be updated with clients actual unicast mac address
 649                  * upon receiving an arp reply.
 650                  */
 651                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 652                 client_info->slave = assigned_slave;
 653
 654                 if (memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
 655                         client_info->ntt = 1;
 656                         bond->alb_info.rx_ntt = 1;
 657                 } else {
 658                         client_info->ntt = 0;
 659                 }
 660
 661                 if (!client_info->assigned) {
 662                         u32 prev_tbl_head = bond_info->rx_hashtbl_head;
 663                         bond_info->rx_hashtbl_head = hash_index;
 664                         client_info->next = prev_tbl_head;
 665                         if (prev_tbl_head != RLB_NULL_INDEX) {
 666                                 bond_info->rx_hashtbl[prev_tbl_head].prev =
 667                                         hash_index;
 668                         }
 669                         client_info->assigned = 1;
 670                 }
 671         }
 672
 673         _unlock_rx_hashtbl(bond);
 674
 675         return assigned_slave;
 676 }
 677
 678 /* chooses (and returns) transmit channel for arp reply
 679  * does not choose channel for other arp types since they are
 680  * sent on the curr_active_slave
 681  */
 682 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 683 {
 684         struct arp_pkt *arp = (struct arp_pkt *)skb->nh.raw;
 685         struct slave *tx_slave = NULL;
 686
 687         if (arp->op_code == __constant_htons(ARPOP_REPLY)) {
 688                 /* the arp must be sent on the selected
 689                 * rx channel
 690                 */
 691                 tx_slave = rlb_choose_channel(bond, arp);
 692                 if (tx_slave) {
 693                         memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
 694                 }
 695                 dprintk("Server sent ARP Reply packet\n");
 696         } else if (arp->op_code == __constant_htons(ARPOP_REQUEST)) {
 697                 /* Create an entry in the rx_hashtbl for this client as a
 698                  * place holder.
 699                  * When the arp reply is received the entry will be updated
 700                  * with the correct unicast address of the client.
 701                  */
 702                 rlb_choose_channel(bond, arp);
 703
 704                 /* The ARP relpy packets must be delayed so that
 705                  * they can cancel out the influence of the ARP request.
 706                  */
 707                 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 708
 709                 /* arp requests are broadcast and are sent on the primary
 710                  * the arp request will collapse all clients on the subnet to
 711                  * the primary slave. We must register these clients to be
 712                  * updated with their assigned mac.
 713                  */
 714                 rlb_req_update_subnet_clients(bond, arp->ip_src);
 715                 dprintk("Server sent ARP Request packet\n");
 716         }
 717
 718         return tx_slave;
 719 }
 720
 721 /* Caller must hold bond lock for read */
 722 static void rlb_rebalance(struct bonding *bond)
 723 {
 724         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 725         struct slave *assigned_slave;
 726         struct rlb_client_info *client_info;
 727         int ntt;
 728         u32 hash_index;
 729
 730         _lock_rx_hashtbl(bond);
 731
 732         ntt = 0;
 733         hash_index = bond_info->rx_hashtbl_head;
 734         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 735                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 736                 assigned_slave = rlb_next_rx_slave(bond);
 737                 if (assigned_slave && (client_info->slave != assigned_slave)) {
 738                         client_info->slave = assigned_slave;
 739                         client_info->ntt = 1;
 740                         ntt = 1;
 741                 }
 742         }
 743
 744         /* update the team's flag only after the whole iteration */
 745         if (ntt) {
 746                 bond_info->rx_ntt = 1;
 747         }
 748         _unlock_rx_hashtbl(bond);
 749 }
 750
 751 /* Caller must hold rx_hashtbl lock */
 752 static void rlb_init_table_entry(struct rlb_client_info *entry)
 753 {
 754         memset(entry, 0, sizeof(struct rlb_client_info));
 755         entry->next = RLB_NULL_INDEX;
 756         entry->prev = RLB_NULL_INDEX;
 757 }
 758
 759 static int rlb_initialize(struct bonding *bond)
 760 {
 761         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 762         struct packet_type *pk_type = &(BOND_ALB_INFO(bond).rlb_pkt_type);
 763         int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 764         int i;
 765
 766         spin_lock_init(&(bond_info->rx_hashtbl_lock));
 767
 768         _lock_rx_hashtbl(bond);
 769
 770         bond_info->rx_hashtbl = kmalloc(size, GFP_KERNEL);
 771         if (!bond_info->rx_hashtbl) {
 772                 printk(KERN_ERR DRV_NAME
 773                        ": Error: %s: Failed to allocate RLB hash table\n",
 774                        bond->dev->name);
 775                 _unlock_rx_hashtbl(bond);
 776                 return -1;
 777         }
 778
 779         bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
 780
 781         for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
 782                 rlb_init_table_entry(bond_info->rx_hashtbl + i);
 783         }
 784
 785         _unlock_rx_hashtbl(bond);
 786
 787         /*initialize packet type*/
 788         pk_type->type = __constant_htons(ETH_P_ARP);
 789         pk_type->dev = bond->dev;
 790         pk_type->func = rlb_arp_recv;
 791
 792         /* register to receive ARPs */
 793         dev_add_pack(pk_type);
 794
 795         return 0;
 796 }
 797
 798 static void rlb_deinitialize(struct bonding *bond)
 799 {
 800         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 801
 802         dev_remove_pack(&(bond_info->rlb_pkt_type));
 803
 804         _lock_rx_hashtbl(bond);
 805
 806         kfree(bond_info->rx_hashtbl);
 807         bond_info->rx_hashtbl = NULL;
 808
 809         _unlock_rx_hashtbl(bond);
 810 }
 811
 812 /*********************** tlb/rlb shared functions *********************/
 813
 814 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
 815 {
 816         struct learning_pkt pkt;
 817         int size = sizeof(struct learning_pkt);
 818         int i;
 819
 820         memset(&pkt, 0, size);
 821         memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
 822         memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
 823         pkt.type = __constant_htons(ETH_P_LOOP);
 824
 825         for (i = 0; i < MAX_LP_RETRY; i++) {
 826                 struct sk_buff *skb;
 827                 char *data;
 828
 829                 skb = dev_alloc_skb(size);
 830                 if (!skb) {
 831                         return;
 832                 }
 833
 834                 data = skb_put(skb, size);
 835                 memcpy(data, &pkt, size);
 836
 837                 skb->mac.raw = data;
 838                 skb->nh.raw = data + ETH_HLEN;
 839                 skb->protocol = pkt.type;
 840                 skb->priority = TC_PRIO_CONTROL;
 841                 skb->dev = slave->dev;
 842
 843                 dev_queue_xmit(skb);
 844         }
 845 }
 846
 847 /* hw is a boolean parameter that determines whether we should try and
 848  * set the hw address of the device as well as the hw address of the
 849  * net_device
 850  */
 851 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[], int hw)
 852 {
 853         struct net_device *dev = slave->dev;
 854         struct sockaddr s_addr;
 855
 856         if (!hw) {
 857                 memcpy(dev->dev_addr, addr, dev->addr_len);
 858                 return 0;
 859         }
 860
 861         /* for rlb each slave must have a unique hw mac addresses so that */
 862         /* each slave will receive packets destined to a different mac */
 863         memcpy(s_addr.sa_data, addr, dev->addr_len);
 864         s_addr.sa_family = dev->type;
 865         if (dev->set_mac_address(dev, &s_addr)) {
 866                 printk(KERN_ERR DRV_NAME
 867                        ": Error: dev->set_mac_address of dev %s failed! ALB "
 868                        "mode requires that the base driver support setting "
 869                        "the hw address also when the network device's "
 870                        "interface is open\n",
 871                        dev->name);
 872                 return -EOPNOTSUPP;
 873         }
 874         return 0;
 875 }
 876
 877 /* Caller must hold bond lock for write or curr_slave_lock for write*/
 878 static void alb_swap_mac_addr(struct bonding *bond, struct slave *slave1, struct slave *slave2)
 879 {
 880         struct slave *disabled_slave = NULL;
 881         u8 tmp_mac_addr[ETH_ALEN];
 882         int slaves_state_differ;
 883
 884         slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
 885
 886         memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
 887         alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr, bond->alb_info.rlb_enabled);
 888         alb_set_slave_mac_addr(slave2, tmp_mac_addr, bond->alb_info.rlb_enabled);
 889
 890         /* fasten the change in the switch */
 891         if (SLAVE_IS_OK(slave1)) {
 892                 alb_send_learning_packets(slave1, slave1->dev->dev_addr);
 893                 if (bond->alb_info.rlb_enabled) {
 894                         /* inform the clients that the mac address
 895                          * has changed
 896                          */
 897                         rlb_req_update_slave_clients(bond, slave1);
 898                 }
 899         } else {
 900                 disabled_slave = slave1;
 901         }
 902
 903         if (SLAVE_IS_OK(slave2)) {
 904                 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
 905                 if (bond->alb_info.rlb_enabled) {
 906                         /* inform the clients that the mac address
 907                          * has changed
 908                          */
 909                         rlb_req_update_slave_clients(bond, slave2);
 910                 }
 911         } else {
 912                 disabled_slave = slave2;
 913         }
 914
 915         if (bond->alb_info.rlb_enabled && slaves_state_differ) {
 916                 /* A disabled slave was assigned an active mac addr */
 917                 rlb_teach_disabled_mac_on_primary(bond,
 918                                                   disabled_slave->dev->dev_addr);
 919         }
 920 }
 921
 922 /**
 923  * alb_change_hw_addr_on_detach
 924  * @bond: bonding we're working on
 925  * @slave: the slave that was just detached
 926  *
 927  * We assume that @slave was already detached from the slave list.
 928  *
 929  * If @slave's permanent hw address is different both from its current
 930  * address and from @bond's address, then somewhere in the bond there's
 931  * a slave that has @slave's permanet address as its current address.
 932  * We'll make sure that that slave no longer uses @slave's permanent address.
 933  *
 934  * Caller must hold bond lock
 935  */
 936 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
 937 {
 938         int perm_curr_diff;
 939         int perm_bond_diff;
 940
 941         perm_curr_diff = memcmp(slave->perm_hwaddr,
 942                                 slave->dev->dev_addr,
 943                                 ETH_ALEN);
 944         perm_bond_diff = memcmp(slave->perm_hwaddr,
 945                                 bond->dev->dev_addr,
 946                                 ETH_ALEN);
 947
 948         if (perm_curr_diff && perm_bond_diff) {
 949                 struct slave *tmp_slave;
 950                 int i, found = 0;
 951
 952                 bond_for_each_slave(bond, tmp_slave, i) {
 953                         if (!memcmp(slave->perm_hwaddr,
 954                                     tmp_slave->dev->dev_addr,
 955                                     ETH_ALEN)) {
 956                                 found = 1;
 957                                 break;
 958                         }
 959                 }
 960
 961                 if (found) {
 962                         alb_swap_mac_addr(bond, slave, tmp_slave);
 963                 }
 964         }
 965 }
 966
 967 /**
 968  * alb_handle_addr_collision_on_attach
 969  * @bond: bonding we're working on
 970  * @slave: the slave that was just attached
 971  *
 972  * checks uniqueness of slave's mac address and handles the case the
 973  * new slave uses the bonds mac address.
 974  *
 975  * If the permanent hw address of @slave is @bond's hw address, we need to
 976  * find a different hw address to give @slave, that isn't in use by any other
 977  * slave in the bond. This address must be, of course, one of the premanent
 978  * addresses of the other slaves.
 979  *
 980  * We go over the slave list, and for each slave there we compare its
 981  * permanent hw address with the current address of all the other slaves.
 982  * If no match was found, then we've found a slave with a permanent address
 983  * that isn't used by any other slave in the bond, so we can assign it to
 984  * @slave.
 985  *
 986  * assumption: this function is called before @slave is attached to the
 987  *             bond slave list.
 988  *
 989  * caller must hold the bond lock for write since the mac addresses are compared
 990  * and may be swapped.
 991  */
 992 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
 993 {
 994         struct slave *tmp_slave1, *tmp_slave2, *free_mac_slave;
 995         struct slave *has_bond_addr = bond->curr_active_slave;
 996         int i, j, found = 0;
 997
 998         if (bond->slave_cnt == 0) {
 999                 /* this is the first slave */
1000                 return 0;
1001         }
1002
1003         /* if slave's mac address differs from bond's mac address
1004          * check uniqueness of slave's mac address against the other
1005          * slaves in the bond.
1006          */
1007         if (memcmp(slave->perm_hwaddr, bond->dev->dev_addr, ETH_ALEN)) {
1008                 bond_for_each_slave(bond, tmp_slave1, i) {
1009                         if (!memcmp(tmp_slave1->dev->dev_addr, slave->dev->dev_addr,
1010                                     ETH_ALEN)) {
1011                                 found = 1;
1012                                 break;
1013                         }
1014                 }
1015
1016                 if (found) {
1017                         /* a slave was found that is using the mac address
1018                          * of the new slave
1019                          */
1020                         printk(KERN_ERR DRV_NAME
1021                                ": Error: the hw address of slave %s is not "
1022                                "unique - cannot enslave it!",
1023                                slave->dev->name);
1024                         return -EINVAL;
1025                 }
1026
1027                 return 0;
1028         }
1029
1030         /* The slave's address is equal to the address of the bond.
1031          * Search for a spare address in the bond for this slave.
1032          */
1033         free_mac_slave = NULL;
1034
1035         bond_for_each_slave(bond, tmp_slave1, i) {
1036                 found = 0;
1037                 bond_for_each_slave(bond, tmp_slave2, j) {
1038                         if (!memcmp(tmp_slave1->perm_hwaddr,
1039                                     tmp_slave2->dev->dev_addr,
1040                                     ETH_ALEN)) {
1041                                 found = 1;
1042                                 break;
1043                         }
1044                 }
1045
1046                 if (!found) {
1047                         /* no slave has tmp_slave1's perm addr
1048                          * as its curr addr
1049                          */
1050                         free_mac_slave = tmp_slave1;
1051                         break;
1052                 }
1053
1054                 if (!has_bond_addr) {
1055                         if (!memcmp(tmp_slave1->dev->dev_addr,
1056                                     bond->dev->dev_addr,
1057                                     ETH_ALEN)) {
1058
1059                                 has_bond_addr = tmp_slave1;
1060                         }
1061                 }
1062         }
1063
1064         if (free_mac_slave) {
1065                 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1066                                        bond->alb_info.rlb_enabled);
1067
1068                 printk(KERN_WARNING DRV_NAME
1069                        ": Warning: the hw address of slave %s is in use by "
1070                        "the bond; giving it the hw address of %s\n",
1071                        slave->dev->name, free_mac_slave->dev->name);
1072
1073         } else if (has_bond_addr) {
1074                 printk(KERN_ERR DRV_NAME
1075                        ": Error: the hw address of slave %s is in use by the "
1076                        "bond; couldn't find a slave with a free hw address to "
1077                        "give it (this should not have happened)\n",
1078                        slave->dev->name);
1079                 return -EFAULT;
1080         }
1081
1082         return 0;
1083 }
1084
1085 /**
1086  * alb_set_mac_address
1087  * @bond:
1088  * @addr:
1089  *
1090  * In TLB mode all slaves are configured to the bond's hw address, but set
1091  * their dev_addr field to different addresses (based on their permanent hw
1092  * addresses).
1093  *
1094  * For each slave, this function sets the interface to the new address and then
1095  * changes its dev_addr field to its previous value.
1096  *
1097  * Unwinding assumes bond's mac address has not yet changed.
1098  */
1099 static int alb_set_mac_address(struct bonding *bond, void *addr)
1100 {
1101         struct sockaddr sa;
1102         struct slave *slave, *stop_at;
1103         char tmp_addr[ETH_ALEN];
1104         int res;
1105         int i;
1106
1107         if (bond->alb_info.rlb_enabled) {
1108                 return 0;
1109         }
1110
1111         bond_for_each_slave(bond, slave, i) {
1112                 if (slave->dev->set_mac_address == NULL) {
1113                         res = -EOPNOTSUPP;
1114                         goto unwind;
1115                 }
1116
1117                 /* save net_device's current hw address */
1118                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1119
1120                 res = slave->dev->set_mac_address(slave->dev, addr);
1121
1122                 /* restore net_device's hw address */
1123                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1124
1125                 if (res) {
1126                         goto unwind;
1127                 }
1128         }
1129
1130         return 0;
1131
1132 unwind:
1133         memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1134         sa.sa_family = bond->dev->type;
1135
1136         /* unwind from head to the slave that failed */
1137         stop_at = slave;
1138         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
1139                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1140                 slave->dev->set_mac_address(slave->dev, &sa);
1141                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1142         }
1143
1144         return res;
1145 }
1146
1147 /************************ exported alb funcions ************************/
1148
1149 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1150 {
1151         int res;
1152
1153         res = tlb_initialize(bond);
1154         if (res) {
1155                 return res;
1156         }
1157
1158         if (rlb_enabled) {
1159                 bond->alb_info.rlb_enabled = 1;
1160                 /* initialize rlb */
1161                 res = rlb_initialize(bond);
1162                 if (res) {
1163                         tlb_deinitialize(bond);
1164                         return res;
1165                 }
1166         }
1167
1168         return 0;
1169 }
1170
1171 void bond_alb_deinitialize(struct bonding *bond)
1172 {
1173         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1174
1175         tlb_deinitialize(bond);
1176
1177         if (bond_info->rlb_enabled) {
1178                 rlb_deinitialize(bond);
1179         }
1180 }
1181
1182 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1183 {
1184         struct bonding *bond = bond_dev->priv;
1185         struct ethhdr *eth_data = (struct ethhdr *)skb->mac.raw = skb->data;
1186         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1187         struct slave *tx_slave = NULL;
1188         static u32 ip_bcast = 0xffffffff;
1189         int hash_size = 0;
1190         int do_tx_balance = 1;
1191         u32 hash_index = 0;
1192         u8 *hash_start = NULL;
1193
1194         /* make sure that the curr_active_slave and the slaves list do
1195          * not change during tx
1196          */
1197         read_lock(&bond->lock);
1198         read_lock(&bond->curr_slave_lock);
1199
1200         if (!BOND_IS_OK(bond)) {
1201                 goto free_out;
1202         }
1203
1204         switch (ntohs(skb->protocol)) {
1205         case ETH_P_IP:
1206                 if ((memcmp(eth_data->h_dest, mac_bcast, ETH_ALEN) == 0) ||
1207                     (skb->nh.iph->daddr == ip_bcast)) {
1208                         do_tx_balance = 0;
1209                         break;
1210                 }
1211                 hash_start = (char*)&(skb->nh.iph->daddr);
1212                 hash_size = sizeof(skb->nh.iph->daddr);
1213                 break;
1214         case ETH_P_IPV6:
1215                 if (memcmp(eth_data->h_dest, mac_bcast, ETH_ALEN) == 0) {
1216                         do_tx_balance = 0;
1217                         break;
1218                 }
1219
1220                 hash_start = (char*)&(skb->nh.ipv6h->daddr);
1221                 hash_size = sizeof(skb->nh.ipv6h->daddr);
1222                 break;
1223         case ETH_P_IPX:
1224                 if (ipx_hdr(skb)->ipx_checksum !=
1225                     __constant_htons(IPX_NO_CHECKSUM)) {
1226                         /* something is wrong with this packet */
1227                         do_tx_balance = 0;
1228                         break;
1229                 }
1230
1231                 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1232                         /* The only protocol worth balancing in
1233                          * this family since it has an "ARP" like
1234                          * mechanism
1235                          */
1236                         do_tx_balance = 0;
1237                         break;
1238                 }
1239
1240                 hash_start = (char*)eth_data->h_dest;
1241                 hash_size = ETH_ALEN;
1242                 break;
1243         case ETH_P_ARP:
1244                 do_tx_balance = 0;
1245                 if (bond_info->rlb_enabled) {
1246                         tx_slave = rlb_arp_xmit(skb, bond);
1247                 }
1248                 break;
1249         default:
1250                 do_tx_balance = 0;
1251                 break;
1252         }
1253
1254         if (do_tx_balance) {
1255                 hash_index = _simple_hash(hash_start, hash_size);
1256                 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1257         }
1258
1259         if (!tx_slave) {
1260                 /* unbalanced or unassigned, send through primary */
1261                 tx_slave = bond->curr_active_slave;
1262                 bond_info->unbalanced_load += skb->len;
1263         }
1264
1265         if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1266                 skb->dev = tx_slave->dev;
1267                 if (tx_slave != bond->curr_active_slave) {
1268                         memcpy(eth_data->h_source,
1269                                tx_slave->dev->dev_addr,
1270                                ETH_ALEN);
1271                 }
1272                 dev_queue_xmit(skb);
1273         } else {
1274                 /* no suitable interface, frame not sent */
1275                 if (tx_slave) {
1276                         tlb_clear_slave(bond, tx_slave, 0);
1277                 }
1278                 goto free_out;
1279         }
1280
1281 out:
1282         read_unlock(&bond->curr_slave_lock);
1283         read_unlock(&bond->lock);
1284         return 0;
1285
1286 free_out:
1287         dev_kfree_skb(skb);
1288         goto out;
1289 }
1290
1291 void bond_alb_monitor(struct bonding *bond)
1292 {
1293         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1294         struct slave *slave;
1295         int i;
1296
1297         read_lock(&bond->lock);
1298
1299         if (bond->kill_timers) {
1300                 goto out;
1301         }
1302
1303         if (bond->slave_cnt == 0) {
1304                 bond_info->tx_rebalance_counter = 0;
1305                 bond_info->lp_counter = 0;
1306                 goto re_arm;
1307         }
1308
1309         bond_info->tx_rebalance_counter++;
1310         bond_info->lp_counter++;
1311
1312         /* send learning packets */
1313         if (bond_info->lp_counter >= BOND_ALB_LP_TICKS) {
1314                 /* change of curr_active_slave involves swapping of mac addresses.
1315                  * in order to avoid this swapping from happening while
1316                  * sending the learning packets, the curr_slave_lock must be held for
1317                  * read.
1318                  */
1319                 read_lock(&bond->curr_slave_lock);
1320
1321                 bond_for_each_slave(bond, slave, i) {
1322                         alb_send_learning_packets(slave,slave->dev->dev_addr);
1323                 }
1324
1325                 read_unlock(&bond->curr_slave_lock);
1326
1327                 bond_info->lp_counter = 0;
1328         }
1329
1330         /* rebalance tx traffic */
1331         if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1332
1333                 read_lock(&bond->curr_slave_lock);
1334
1335                 bond_for_each_slave(bond, slave, i) {
1336                         tlb_clear_slave(bond, slave, 1);
1337                         if (slave == bond->curr_active_slave) {
1338                                 SLAVE_TLB_INFO(slave).load =
1339                                         bond_info->unbalanced_load /
1340                                                 BOND_TLB_REBALANCE_INTERVAL;
1341                                 bond_info->unbalanced_load = 0;
1342                         }
1343                 }
1344
1345                 read_unlock(&bond->curr_slave_lock);
1346
1347                 bond_info->tx_rebalance_counter = 0;
1348         }
1349
1350         /* handle rlb stuff */
1351         if (bond_info->rlb_enabled) {
1352                 /* the following code changes the promiscuity of the
1353                  * the curr_active_slave. It needs to be locked with a
1354                  * write lock to protect from other code that also
1355                  * sets the promiscuity.
1356                  */
1357                 write_lock(&bond->curr_slave_lock);
1358
1359                 if (bond_info->primary_is_promisc &&
1360                     (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1361
1362                         bond_info->rlb_promisc_timeout_counter = 0;
1363
1364                         /* If the primary was set to promiscuous mode
1365                          * because a slave was disabled then
1366                          * it can now leave promiscuous mode.
1367                          */
1368                         dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1369                         bond_info->primary_is_promisc = 0;
1370                 }
1371
1372                 write_unlock(&bond->curr_slave_lock);
1373
1374                 if (bond_info->rlb_rebalance) {
1375                         bond_info->rlb_rebalance = 0;
1376                         rlb_rebalance(bond);
1377                 }
1378
1379                 /* check if clients need updating */
1380                 if (bond_info->rx_ntt) {
1381                         if (bond_info->rlb_update_delay_counter) {
1382                                 --bond_info->rlb_update_delay_counter;
1383                         } else {
1384                                 rlb_update_rx_clients(bond);
1385                                 if (bond_info->rlb_update_retry_counter) {
1386                                         --bond_info->rlb_update_retry_counter;
1387                                 } else {
1388                                         bond_info->rx_ntt = 0;
1389                                 }
1390                         }
1391                 }
1392         }
1393
1394 re_arm:
1395         mod_timer(&(bond_info->alb_timer), jiffies + alb_delta_in_ticks);
1396 out:
1397         read_unlock(&bond->lock);
1398 }
1399
1400 /* assumption: called before the slave is attached to the bond
1401  * and not locked by the bond lock
1402  */
1403 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1404 {
1405         int res;
1406
1407         res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1408                                      bond->alb_info.rlb_enabled);
1409         if (res) {
1410                 return res;
1411         }
1412
1413         /* caller must hold the bond lock for write since the mac addresses
1414          * are compared and may be swapped.
1415          */
1416         write_lock_bh(&bond->lock);
1417
1418         res = alb_handle_addr_collision_on_attach(bond, slave);
1419
1420         write_unlock_bh(&bond->lock);
1421
1422         if (res) {
1423                 return res;
1424         }
1425
1426         tlb_init_slave(slave);
1427
1428         /* order a rebalance ASAP */
1429         bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1430
1431         if (bond->alb_info.rlb_enabled) {
1432                 bond->alb_info.rlb_rebalance = 1;
1433         }
1434
1435         return 0;
1436 }
1437
1438 /* Caller must hold bond lock for write */
1439 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1440 {
1441         if (bond->slave_cnt > 1) {
1442                 alb_change_hw_addr_on_detach(bond, slave);
1443         }
1444
1445         tlb_clear_slave(bond, slave, 0);
1446
1447         if (bond->alb_info.rlb_enabled) {
1448                 bond->alb_info.next_rx_slave = NULL;
1449                 rlb_clear_slave(bond, slave);
1450         }
1451 }
1452
1453 /* Caller must hold bond lock for read */
1454 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1455 {
1456         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1457
1458         if (link == BOND_LINK_DOWN) {
1459                 tlb_clear_slave(bond, slave, 0);
1460                 if (bond->alb_info.rlb_enabled) {
1461                         rlb_clear_slave(bond, slave);
1462                 }
1463         } else if (link == BOND_LINK_UP) {
1464                 /* order a rebalance ASAP */
1465                 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1466                 if (bond->alb_info.rlb_enabled) {
1467                         bond->alb_info.rlb_rebalance = 1;
1468                         /* If the updelay module parameter is smaller than the
1469                          * forwarding delay of the switch the rebalance will
1470                          * not work because the rebalance arp replies will
1471                          * not be forwarded to the clients..
1472                          */
1473                 }
1474         }
1475 }
1476
1477 /**
1478  * bond_alb_handle_active_change - assign new curr_active_slave
1479  * @bond: our bonding struct
1480  * @new_slave: new slave to assign
1481  *
1482  * Set the bond->curr_active_slave to @new_slave and handle
1483  * mac address swapping and promiscuity changes as needed.
1484  *
1485  * Caller must hold bond curr_slave_lock for write (or bond lock for write)
1486  */
1487 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1488 {
1489         struct slave *swap_slave;
1490         int i;
1491
1492         if (bond->curr_active_slave == new_slave) {
1493                 return;
1494         }
1495
1496         if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1497                 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1498                 bond->alb_info.primary_is_promisc = 0;
1499                 bond->alb_info.rlb_promisc_timeout_counter = 0;
1500         }
1501
1502         swap_slave = bond->curr_active_slave;
1503         bond->curr_active_slave = new_slave;
1504
1505         if (!new_slave || (bond->slave_cnt == 0)) {
1506                 return;
1507         }
1508
1509         /* set the new curr_active_slave to the bonds mac address
1510          * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1511          */
1512         if (!swap_slave) {
1513                 struct slave *tmp_slave;
1514                 /* find slave that is holding the bond's mac address */
1515                 bond_for_each_slave(bond, tmp_slave, i) {
1516                         if (!memcmp(tmp_slave->dev->dev_addr,
1517                                     bond->dev->dev_addr, ETH_ALEN)) {
1518                                 swap_slave = tmp_slave;
1519                                 break;
1520                         }
1521                 }
1522         }
1523
1524         /* curr_active_slave must be set before calling alb_swap_mac_addr */
1525         if (swap_slave) {
1526                 /* swap mac address */
1527                 alb_swap_mac_addr(bond, swap_slave, new_slave);
1528         } else {
1529                 /* set the new_slave to the bond mac address */
1530                 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1531                                        bond->alb_info.rlb_enabled);
1532                 /* fasten bond mac on new current slave */
1533                 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1534         }
1535 }
1536
1537 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1538 {
1539         struct bonding *bond = bond_dev->priv;
1540         struct sockaddr *sa = addr;
1541         struct slave *slave, *swap_slave;
1542         int res;
1543         int i;
1544
1545         if (!is_valid_ether_addr(sa->sa_data)) {
1546                 return -EADDRNOTAVAIL;
1547         }
1548
1549         res = alb_set_mac_address(bond, addr);
1550         if (res) {
1551                 return res;
1552         }
1553
1554         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1555
1556         /* If there is no curr_active_slave there is nothing else to do.
1557          * Otherwise we'll need to pass the new address to it and handle
1558          * duplications.
1559          */
1560         if (!bond->curr_active_slave) {
1561                 return 0;
1562         }
1563
1564         swap_slave = NULL;
1565
1566         bond_for_each_slave(bond, slave, i) {
1567                 if (!memcmp(slave->dev->dev_addr, bond_dev->dev_addr, ETH_ALEN)) {
1568                         swap_slave = slave;
1569                         break;
1570                 }
1571         }
1572
1573         if (swap_slave) {
1574                 alb_swap_mac_addr(bond, swap_slave, bond->curr_active_slave);
1575         } else {
1576                 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr,
1577                                        bond->alb_info.rlb_enabled);
1578
1579                 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1580                 if (bond->alb_info.rlb_enabled) {
1581                         /* inform clients mac address has changed */
1582                         rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1583                 }
1584         }
1585
1586         return 0;
1587 }
1588