drivers/net/bonding/bond_alb.c

   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
  23 //#define BONDING_DEBUG 1
  24
  25 #include <linux/skbuff.h>
  26 #include <linux/netdevice.h>
  27 #include <linux/etherdevice.h>
  28 #include <linux/pkt_sched.h>
  29 #include <linux/spinlock.h>
  30 #include <linux/slab.h>
  31 #include <linux/timer.h>
  32 #include <linux/ip.h>
  33 #include <linux/ipv6.h>
  34 #include <linux/if_arp.h>
  35 #include <linux/if_ether.h>
  36 #include <linux/if_bonding.h>
  37 #include <linux/if_vlan.h>
  38 #include <linux/in.h>
  39 #include <net/ipx.h>
  40 #include <net/arp.h>
  41 #include <asm/byteorder.h>
  42 #include "bonding.h"
  43 #include "bond_alb.h"
  44
  45
  46 #define ALB_TIMER_TICKS_PER_SEC     10  /* should be a divisor of HZ */
  47 #define BOND_TLB_REBALANCE_INTERVAL 10  /* In seconds, periodic re-balancing.
  48                                          * Used for division - never set
  49                                          * to zero !!!
  50                                          */
  51 #define BOND_ALB_LP_INTERVAL        1   /* In seconds, periodic send of
  52                                          * learning packets to the switch
  53                                          */
  54
  55 #define BOND_TLB_REBALANCE_TICKS (BOND_TLB_REBALANCE_INTERVAL \
  56                                   * ALB_TIMER_TICKS_PER_SEC)
  57
  58 #define BOND_ALB_LP_TICKS (BOND_ALB_LP_INTERVAL \
  59                            * ALB_TIMER_TICKS_PER_SEC)
  60
  61 #define TLB_HASH_TABLE_SIZE 256 /* The size of the clients hash table.
  62                                  * Note that this value MUST NOT be smaller
  63                                  * because the key hash table is BYTE wide !
  64                                  */
  65
  66
  67 #define TLB_NULL_INDEX          0xffffffff
  68 #define MAX_LP_BURST            3
  69
  70 /* rlb defs */
  71 #define RLB_HASH_TABLE_SIZE     256
  72 #define RLB_NULL_INDEX          0xffffffff
  73 #define RLB_UPDATE_DELAY        2*ALB_TIMER_TICKS_PER_SEC /* 2 seconds */
  74 #define RLB_ARP_BURST_SIZE      2
  75 #define RLB_UPDATE_RETRY        3       /* 3-ticks - must be smaller than the rlb
  76                                          * rebalance interval (5 min).
  77                                          */
  78 /* RLB_PROMISC_TIMEOUT = 10 sec equals the time that the current slave is
  79  * promiscuous after failover
  80  */
  81 #define RLB_PROMISC_TIMEOUT     10*ALB_TIMER_TICKS_PER_SEC
  82
  83 static const u8 mac_bcast[ETH_ALEN] = {0xff,0xff,0xff,0xff,0xff,0xff};
  84 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
  85
  86 #pragma pack(1)
  87 struct learning_pkt {
  88         u8 mac_dst[ETH_ALEN];
  89         u8 mac_src[ETH_ALEN];
  90         u16 type;
  91         u8 padding[ETH_ZLEN - ETH_HLEN];
  92 };
  93
  94 struct arp_pkt {
  95         u16     hw_addr_space;
  96         u16     prot_addr_space;
  97         u8      hw_addr_len;
  98         u8      prot_addr_len;
  99         u16     op_code;
 100         u8      mac_src[ETH_ALEN];      /* sender hardware address */
 101         u32     ip_src;                 /* sender IP address */
 102         u8      mac_dst[ETH_ALEN];      /* target hardware address */
 103         u32     ip_dst;                 /* target IP address */
 104 };
 105 #pragma pack()
 106
 107 /* Forward declaration */
 108 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[]);
 109
 110 static inline u8 _simple_hash(u8 *hash_start, int hash_size)
 111 {
 112         int i;
 113         u8 hash = 0;
 114
 115         for (i = 0; i < hash_size; i++) {
 116                 hash ^= hash_start[i];
 117         }
 118
 119         return hash;
 120 }
 121
 122 /*********************** tlb specific functions ***************************/
 123
 124 static inline void _lock_tx_hashtbl(struct bonding *bond)
 125 {
 126         spin_lock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 127 }
 128
 129 static inline void _unlock_tx_hashtbl(struct bonding *bond)
 130 {
 131         spin_unlock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 132 }
 133
 134 /* Caller must hold tx_hashtbl lock */
 135 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
 136 {
 137         if (save_load) {
 138                 entry->load_history = 1 + entry->tx_bytes /
 139                                       BOND_TLB_REBALANCE_INTERVAL;
 140                 entry->tx_bytes = 0;
 141         }
 142
 143         entry->tx_slave = NULL;
 144         entry->next = TLB_NULL_INDEX;
 145         entry->prev = TLB_NULL_INDEX;
 146 }
 147
 148 static inline void tlb_init_slave(struct slave *slave)
 149 {
 150         SLAVE_TLB_INFO(slave).load = 0;
 151         SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
 152 }
 153
 154 /* Caller must hold bond lock for read */
 155 static void tlb_clear_slave(struct bonding *bond, struct slave *slave, int save_load)
 156 {
 157         struct tlb_client_info *tx_hash_table;
 158         u32 index;
 159
 160         _lock_tx_hashtbl(bond);
 161
 162         /* clear slave from tx_hashtbl */
 163         tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
 164
 165         index = SLAVE_TLB_INFO(slave).head;
 166         while (index != TLB_NULL_INDEX) {
 167                 u32 next_index = tx_hash_table[index].next;
 168                 tlb_init_table_entry(&tx_hash_table[index], save_load);
 169                 index = next_index;
 170         }
 171
 172         tlb_init_slave(slave);
 173
 174         _unlock_tx_hashtbl(bond);
 175 }
 176
 177 /* Must be called before starting the monitor timer */
 178 static int tlb_initialize(struct bonding *bond)
 179 {
 180         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 181         int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 182         struct tlb_client_info *new_hashtbl;
 183         int i;
 184
 185         spin_lock_init(&(bond_info->tx_hashtbl_lock));
 186
 187         new_hashtbl = kzalloc(size, GFP_KERNEL);
 188         if (!new_hashtbl) {
 189                 printk(KERN_ERR DRV_NAME
 190                        ": %s: Error: Failed to allocate TLB hash table\n",
 191                        bond->dev->name);
 192                 return -1;
 193         }
 194         _lock_tx_hashtbl(bond);
 195
 196         bond_info->tx_hashtbl = new_hashtbl;
 197
 198         for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
 199                 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 1);
 200         }
 201
 202         _unlock_tx_hashtbl(bond);
 203
 204         return 0;
 205 }
 206
 207 /* Must be called only after all slaves have been released */
 208 static void tlb_deinitialize(struct bonding *bond)
 209 {
 210         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 211
 212         _lock_tx_hashtbl(bond);
 213
 214         kfree(bond_info->tx_hashtbl);
 215         bond_info->tx_hashtbl = NULL;
 216
 217         _unlock_tx_hashtbl(bond);
 218 }
 219
 220 /* Caller must hold bond lock for read */
 221 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
 222 {
 223         struct slave *slave, *least_loaded;
 224         s64 max_gap;
 225         int i, found = 0;
 226
 227         /* Find the first enabled slave */
 228         bond_for_each_slave(bond, slave, i) {
 229                 if (SLAVE_IS_OK(slave)) {
 230                         found = 1;
 231                         break;
 232                 }
 233         }
 234
 235         if (!found) {
 236                 return NULL;
 237         }
 238
 239         least_loaded = slave;
 240         max_gap = (s64)(slave->speed << 20) - /* Convert to Megabit per sec */
 241                         (s64)(SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
 242
 243         /* Find the slave with the largest gap */
 244         bond_for_each_slave_from(bond, slave, i, least_loaded) {
 245                 if (SLAVE_IS_OK(slave)) {
 246                         s64 gap = (s64)(slave->speed << 20) -
 247                                         (s64)(SLAVE_TLB_INFO(slave).load << 3);
 248                         if (max_gap < gap) {
 249                                 least_loaded = slave;
 250                                 max_gap = gap;
 251                         }
 252                 }
 253         }
 254
 255         return least_loaded;
 256 }
 257
 258 /* Caller must hold bond lock for read */
 259 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index, u32 skb_len)
 260 {
 261         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 262         struct tlb_client_info *hash_table;
 263         struct slave *assigned_slave;
 264
 265         _lock_tx_hashtbl(bond);
 266
 267         hash_table = bond_info->tx_hashtbl;
 268         assigned_slave = hash_table[hash_index].tx_slave;
 269         if (!assigned_slave) {
 270                 assigned_slave = tlb_get_least_loaded_slave(bond);
 271
 272                 if (assigned_slave) {
 273                         struct tlb_slave_info *slave_info =
 274                                 &(SLAVE_TLB_INFO(assigned_slave));
 275                         u32 next_index = slave_info->head;
 276
 277                         hash_table[hash_index].tx_slave = assigned_slave;
 278                         hash_table[hash_index].next = next_index;
 279                         hash_table[hash_index].prev = TLB_NULL_INDEX;
 280
 281                         if (next_index != TLB_NULL_INDEX) {
 282                                 hash_table[next_index].prev = hash_index;
 283                         }
 284
 285                         slave_info->head = hash_index;
 286                         slave_info->load +=
 287                                 hash_table[hash_index].load_history;
 288                 }
 289         }
 290
 291         if (assigned_slave) {
 292                 hash_table[hash_index].tx_bytes += skb_len;
 293         }
 294
 295         _unlock_tx_hashtbl(bond);
 296
 297         return assigned_slave;
 298 }
 299
 300 /*********************** rlb specific functions ***************************/
 301 static inline void _lock_rx_hashtbl(struct bonding *bond)
 302 {
 303         spin_lock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 304 }
 305
 306 static inline void _unlock_rx_hashtbl(struct bonding *bond)
 307 {
 308         spin_unlock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 309 }
 310
 311 /* when an ARP REPLY is received from a client update its info
 312  * in the rx_hashtbl
 313  */
 314 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 315 {
 316         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 317         struct rlb_client_info *client_info;
 318         u32 hash_index;
 319
 320         _lock_rx_hashtbl(bond);
 321
 322         hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
 323         client_info = &(bond_info->rx_hashtbl[hash_index]);
 324
 325         if ((client_info->assigned) &&
 326             (client_info->ip_src == arp->ip_dst) &&
 327             (client_info->ip_dst == arp->ip_src)) {
 328                 /* update the clients MAC address */
 329                 memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
 330                 client_info->ntt = 1;
 331                 bond_info->rx_ntt = 1;
 332         }
 333
 334         _unlock_rx_hashtbl(bond);
 335 }
 336
 337 static int rlb_arp_recv(struct sk_buff *skb, struct net_device *bond_dev, struct packet_type *ptype, struct net_device *orig_dev)
 338 {
 339         struct bonding *bond = bond_dev->priv;
 340         struct arp_pkt *arp = (struct arp_pkt *)skb->data;
 341         int res = NET_RX_DROP;
 342
 343         if (!(bond_dev->flags & IFF_MASTER))
 344                 goto out;
 345
 346         if (!arp) {
 347                 dprintk("Packet has no ARP data\n");
 348                 goto out;
 349         }
 350
 351         if (skb->len < sizeof(struct arp_pkt)) {
 352                 dprintk("Packet is too small to be an ARP\n");
 353                 goto out;
 354         }
 355
 356         if (arp->op_code == htons(ARPOP_REPLY)) {
 357                 /* update rx hash table for this ARP */
 358                 rlb_update_entry_from_arp(bond, arp);
 359                 dprintk("Server received an ARP Reply from client\n");
 360         }
 361
 362         res = NET_RX_SUCCESS;
 363
 364 out:
 365         dev_kfree_skb(skb);
 366
 367         return res;
 368 }
 369
 370 /* Caller must hold bond lock for read */
 371 static struct slave *rlb_next_rx_slave(struct bonding *bond)
 372 {
 373         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 374         struct slave *rx_slave, *slave, *start_at;
 375         int i = 0;
 376
 377         if (bond_info->next_rx_slave) {
 378                 start_at = bond_info->next_rx_slave;
 379         } else {
 380                 start_at = bond->first_slave;
 381         }
 382
 383         rx_slave = NULL;
 384
 385         bond_for_each_slave_from(bond, slave, i, start_at) {
 386                 if (SLAVE_IS_OK(slave)) {
 387                         if (!rx_slave) {
 388                                 rx_slave = slave;
 389                         } else if (slave->speed > rx_slave->speed) {
 390                                 rx_slave = slave;
 391                         }
 392                 }
 393         }
 394
 395         if (rx_slave) {
 396                 bond_info->next_rx_slave = rx_slave->next;
 397         }
 398
 399         return rx_slave;
 400 }
 401
 402 /* teach the switch the mac of a disabled slave
 403  * on the primary for fault tolerance
 404  *
 405  * Caller must hold bond->curr_slave_lock for write or bond lock for write
 406  */
 407 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
 408 {
 409         if (!bond->curr_active_slave) {
 410                 return;
 411         }
 412
 413         if (!bond->alb_info.primary_is_promisc) {
 414                 bond->alb_info.primary_is_promisc = 1;
 415                 dev_set_promiscuity(bond->curr_active_slave->dev, 1);
 416         }
 417
 418         bond->alb_info.rlb_promisc_timeout_counter = 0;
 419
 420         alb_send_learning_packets(bond->curr_active_slave, addr);
 421 }
 422
 423 /* slave being removed should not be active at this point
 424  *
 425  * Caller must hold bond lock for read
 426  */
 427 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 428 {
 429         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 430         struct rlb_client_info *rx_hash_table;
 431         u32 index, next_index;
 432
 433         /* clear slave from rx_hashtbl */
 434         _lock_rx_hashtbl(bond);
 435
 436         rx_hash_table = bond_info->rx_hashtbl;
 437         index = bond_info->rx_hashtbl_head;
 438         for (; index != RLB_NULL_INDEX; index = next_index) {
 439                 next_index = rx_hash_table[index].next;
 440                 if (rx_hash_table[index].slave == slave) {
 441                         struct slave *assigned_slave = rlb_next_rx_slave(bond);
 442
 443                         if (assigned_slave) {
 444                                 rx_hash_table[index].slave = assigned_slave;
 445                                 if (memcmp(rx_hash_table[index].mac_dst,
 446                                            mac_bcast, ETH_ALEN)) {
 447                                         bond_info->rx_hashtbl[index].ntt = 1;
 448                                         bond_info->rx_ntt = 1;
 449                                         /* A slave has been removed from the
 450                                          * table because it is either disabled
 451                                          * or being released. We must retry the
 452                                          * update to avoid clients from not
 453                                          * being updated & disconnecting when
 454                                          * there is stress
 455                                          */
 456                                         bond_info->rlb_update_retry_counter =
 457                                                 RLB_UPDATE_RETRY;
 458                                 }
 459                         } else {  /* there is no active slave */
 460                                 rx_hash_table[index].slave = NULL;
 461                         }
 462                 }
 463         }
 464
 465         _unlock_rx_hashtbl(bond);
 466
 467         write_lock(&bond->curr_slave_lock);
 468
 469         if (slave != bond->curr_active_slave) {
 470                 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 471         }
 472
 473         write_unlock(&bond->curr_slave_lock);
 474 }
 475
 476 static void rlb_update_client(struct rlb_client_info *client_info)
 477 {
 478         int i;
 479
 480         if (!client_info->slave) {
 481                 return;
 482         }
 483
 484         for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 485                 struct sk_buff *skb;
 486
 487                 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 488                                  client_info->ip_dst,
 489                                  client_info->slave->dev,
 490                                  client_info->ip_src,
 491                                  client_info->mac_dst,
 492                                  client_info->slave->dev->dev_addr,
 493                                  client_info->mac_dst);
 494                 if (!skb) {
 495                         printk(KERN_ERR DRV_NAME
 496                                ": %s: Error: failed to create an ARP packet\n",
 497                                client_info->slave->dev->master->name);
 498                         continue;
 499                 }
 500
 501                 skb->dev = client_info->slave->dev;
 502
 503                 if (client_info->tag) {
 504                         skb = vlan_put_tag(skb, client_info->vlan_id);
 505                         if (!skb) {
 506                                 printk(KERN_ERR DRV_NAME
 507                                        ": %s: Error: failed to insert VLAN tag\n",
 508                                        client_info->slave->dev->master->name);
 509                                 continue;
 510                         }
 511                 }
 512
 513                 arp_xmit(skb);
 514         }
 515 }
 516
 517 /* sends ARP REPLIES that update the clients that need updating */
 518 static void rlb_update_rx_clients(struct bonding *bond)
 519 {
 520         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 521         struct rlb_client_info *client_info;
 522         u32 hash_index;
 523
 524         _lock_rx_hashtbl(bond);
 525
 526         hash_index = bond_info->rx_hashtbl_head;
 527         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 528                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 529                 if (client_info->ntt) {
 530                         rlb_update_client(client_info);
 531                         if (bond_info->rlb_update_retry_counter == 0) {
 532                                 client_info->ntt = 0;
 533                         }
 534                 }
 535         }
 536
 537         /* do not update the entries again untill this counter is zero so that
 538          * not to confuse the clients.
 539          */
 540         bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 541
 542         _unlock_rx_hashtbl(bond);
 543 }
 544
 545 /* The slave was assigned a new mac address - update the clients */
 546 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 547 {
 548         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 549         struct rlb_client_info *client_info;
 550         int ntt = 0;
 551         u32 hash_index;
 552
 553         _lock_rx_hashtbl(bond);
 554
 555         hash_index = bond_info->rx_hashtbl_head;
 556         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 557                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 558
 559                 if ((client_info->slave == slave) &&
 560                     memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
 561                         client_info->ntt = 1;
 562                         ntt = 1;
 563                 }
 564         }
 565
 566         // update the team's flag only after the whole iteration
 567         if (ntt) {
 568                 bond_info->rx_ntt = 1;
 569                 //fasten the change
 570                 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 571         }
 572
 573         _unlock_rx_hashtbl(bond);
 574 }
 575
 576 /* mark all clients using src_ip to be updated */
 577 static void rlb_req_update_subnet_clients(struct bonding *bond, u32 src_ip)
 578 {
 579         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 580         struct rlb_client_info *client_info;
 581         u32 hash_index;
 582
 583         _lock_rx_hashtbl(bond);
 584
 585         hash_index = bond_info->rx_hashtbl_head;
 586         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 587                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 588
 589                 if (!client_info->slave) {
 590                         printk(KERN_ERR DRV_NAME
 591                                ": %s: Error: found a client with no channel in "
 592                                "the client's hash table\n",
 593                                bond->dev->name);
 594                         continue;
 595                 }
 596                 /*update all clients using this src_ip, that are not assigned
 597                  * to the team's address (curr_active_slave) and have a known
 598                  * unicast mac address.
 599                  */
 600                 if ((client_info->ip_src == src_ip) &&
 601                     memcmp(client_info->slave->dev->dev_addr,
 602                            bond->dev->dev_addr, ETH_ALEN) &&
 603                     memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
 604                         client_info->ntt = 1;
 605                         bond_info->rx_ntt = 1;
 606                 }
 607         }
 608
 609         _unlock_rx_hashtbl(bond);
 610 }
 611
 612 /* Caller must hold both bond and ptr locks for read */
 613 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
 614 {
 615         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 616         struct arp_pkt *arp = (struct arp_pkt *)skb->nh.raw;
 617         struct slave *assigned_slave;
 618         struct rlb_client_info *client_info;
 619         u32 hash_index = 0;
 620
 621         _lock_rx_hashtbl(bond);
 622
 623         hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_src));
 624         client_info = &(bond_info->rx_hashtbl[hash_index]);
 625
 626         if (client_info->assigned) {
 627                 if ((client_info->ip_src == arp->ip_src) &&
 628                     (client_info->ip_dst == arp->ip_dst)) {
 629                         /* the entry is already assigned to this client */
 630                         if (memcmp(arp->mac_dst, mac_bcast, ETH_ALEN)) {
 631                                 /* update mac address from arp */
 632                                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 633                         }
 634
 635                         assigned_slave = client_info->slave;
 636                         if (assigned_slave) {
 637                                 _unlock_rx_hashtbl(bond);
 638                                 return assigned_slave;
 639                         }
 640                 } else {
 641                         /* the entry is already assigned to some other client,
 642                          * move the old client to primary (curr_active_slave) so
 643                          * that the new client can be assigned to this entry.
 644                          */
 645                         if (bond->curr_active_slave &&
 646                             client_info->slave != bond->curr_active_slave) {
 647                                 client_info->slave = bond->curr_active_slave;
 648                                 rlb_update_client(client_info);
 649                         }
 650                 }
 651         }
 652         /* assign a new slave */
 653         assigned_slave = rlb_next_rx_slave(bond);
 654
 655         if (assigned_slave) {
 656                 client_info->ip_src = arp->ip_src;
 657                 client_info->ip_dst = arp->ip_dst;
 658                 /* arp->mac_dst is broadcast for arp reqeusts.
 659                  * will be updated with clients actual unicast mac address
 660                  * upon receiving an arp reply.
 661                  */
 662                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 663                 client_info->slave = assigned_slave;
 664
 665                 if (memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
 666                         client_info->ntt = 1;
 667                         bond->alb_info.rx_ntt = 1;
 668                 } else {
 669                         client_info->ntt = 0;
 670                 }
 671
 672                 if (!list_empty(&bond->vlan_list)) {
 673                         unsigned short vlan_id;
 674                         int res = vlan_get_tag(skb, &vlan_id);
 675                         if (!res) {
 676                                 client_info->tag = 1;
 677                                 client_info->vlan_id = vlan_id;
 678                         }
 679                 }
 680
 681                 if (!client_info->assigned) {
 682                         u32 prev_tbl_head = bond_info->rx_hashtbl_head;
 683                         bond_info->rx_hashtbl_head = hash_index;
 684                         client_info->next = prev_tbl_head;
 685                         if (prev_tbl_head != RLB_NULL_INDEX) {
 686                                 bond_info->rx_hashtbl[prev_tbl_head].prev =
 687                                         hash_index;
 688                         }
 689                         client_info->assigned = 1;
 690                 }
 691         }
 692
 693         _unlock_rx_hashtbl(bond);
 694
 695         return assigned_slave;
 696 }
 697
 698 /* chooses (and returns) transmit channel for arp reply
 699  * does not choose channel for other arp types since they are
 700  * sent on the curr_active_slave
 701  */
 702 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 703 {
 704         struct arp_pkt *arp = (struct arp_pkt *)skb->nh.raw;
 705         struct slave *tx_slave = NULL;
 706
 707         if (arp->op_code == __constant_htons(ARPOP_REPLY)) {
 708                 /* the arp must be sent on the selected
 709                 * rx channel
 710                 */
 711                 tx_slave = rlb_choose_channel(skb, bond);
 712                 if (tx_slave) {
 713                         memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
 714                 }
 715                 dprintk("Server sent ARP Reply packet\n");
 716         } else if (arp->op_code == __constant_htons(ARPOP_REQUEST)) {
 717                 /* Create an entry in the rx_hashtbl for this client as a
 718                  * place holder.
 719                  * When the arp reply is received the entry will be updated
 720                  * with the correct unicast address of the client.
 721                  */
 722                 rlb_choose_channel(skb, bond);
 723
 724                 /* The ARP relpy packets must be delayed so that
 725                  * they can cancel out the influence of the ARP request.
 726                  */
 727                 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 728
 729                 /* arp requests are broadcast and are sent on the primary
 730                  * the arp request will collapse all clients on the subnet to
 731                  * the primary slave. We must register these clients to be
 732                  * updated with their assigned mac.
 733                  */
 734                 rlb_req_update_subnet_clients(bond, arp->ip_src);
 735                 dprintk("Server sent ARP Request packet\n");
 736         }
 737
 738         return tx_slave;
 739 }
 740
 741 /* Caller must hold bond lock for read */
 742 static void rlb_rebalance(struct bonding *bond)
 743 {
 744         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 745         struct slave *assigned_slave;
 746         struct rlb_client_info *client_info;
 747         int ntt;
 748         u32 hash_index;
 749
 750         _lock_rx_hashtbl(bond);
 751
 752         ntt = 0;
 753         hash_index = bond_info->rx_hashtbl_head;
 754         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 755                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 756                 assigned_slave = rlb_next_rx_slave(bond);
 757                 if (assigned_slave && (client_info->slave != assigned_slave)) {
 758                         client_info->slave = assigned_slave;
 759                         client_info->ntt = 1;
 760                         ntt = 1;
 761                 }
 762         }
 763
 764         /* update the team's flag only after the whole iteration */
 765         if (ntt) {
 766                 bond_info->rx_ntt = 1;
 767         }
 768         _unlock_rx_hashtbl(bond);
 769 }
 770
 771 /* Caller must hold rx_hashtbl lock */
 772 static void rlb_init_table_entry(struct rlb_client_info *entry)
 773 {
 774         memset(entry, 0, sizeof(struct rlb_client_info));
 775         entry->next = RLB_NULL_INDEX;
 776         entry->prev = RLB_NULL_INDEX;
 777 }
 778
 779 static int rlb_initialize(struct bonding *bond)
 780 {
 781         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 782         struct packet_type *pk_type = &(BOND_ALB_INFO(bond).rlb_pkt_type);
 783         struct rlb_client_info  *new_hashtbl;
 784         int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 785         int i;
 786
 787         spin_lock_init(&(bond_info->rx_hashtbl_lock));
 788
 789         new_hashtbl = kmalloc(size, GFP_KERNEL);
 790         if (!new_hashtbl) {
 791                 printk(KERN_ERR DRV_NAME
 792                        ": %s: Error: Failed to allocate RLB hash table\n",
 793                        bond->dev->name);
 794                 return -1;
 795         }
 796         _lock_rx_hashtbl(bond);
 797
 798         bond_info->rx_hashtbl = new_hashtbl;
 799
 800         bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
 801
 802         for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
 803                 rlb_init_table_entry(bond_info->rx_hashtbl + i);
 804         }
 805
 806         _unlock_rx_hashtbl(bond);
 807
 808         /*initialize packet type*/
 809         pk_type->type = __constant_htons(ETH_P_ARP);
 810         pk_type->dev = bond->dev;
 811         pk_type->func = rlb_arp_recv;
 812
 813         /* register to receive ARPs */
 814         dev_add_pack(pk_type);
 815
 816         return 0;
 817 }
 818
 819 static void rlb_deinitialize(struct bonding *bond)
 820 {
 821         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 822
 823         dev_remove_pack(&(bond_info->rlb_pkt_type));
 824
 825         _lock_rx_hashtbl(bond);
 826
 827         kfree(bond_info->rx_hashtbl);
 828         bond_info->rx_hashtbl = NULL;
 829         bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
 830
 831         _unlock_rx_hashtbl(bond);
 832 }
 833
 834 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 835 {
 836         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 837         u32 curr_index;
 838
 839         _lock_rx_hashtbl(bond);
 840
 841         curr_index = bond_info->rx_hashtbl_head;
 842         while (curr_index != RLB_NULL_INDEX) {
 843                 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 844                 u32 next_index = bond_info->rx_hashtbl[curr_index].next;
 845                 u32 prev_index = bond_info->rx_hashtbl[curr_index].prev;
 846
 847                 if (curr->tag && (curr->vlan_id == vlan_id)) {
 848                         if (curr_index == bond_info->rx_hashtbl_head) {
 849                                 bond_info->rx_hashtbl_head = next_index;
 850                         }
 851                         if (prev_index != RLB_NULL_INDEX) {
 852                                 bond_info->rx_hashtbl[prev_index].next = next_index;
 853                         }
 854                         if (next_index != RLB_NULL_INDEX) {
 855                                 bond_info->rx_hashtbl[next_index].prev = prev_index;
 856                         }
 857
 858                         rlb_init_table_entry(curr);
 859                 }
 860
 861                 curr_index = next_index;
 862         }
 863
 864         _unlock_rx_hashtbl(bond);
 865 }
 866
 867 /*********************** tlb/rlb shared functions *********************/
 868
 869 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
 870 {
 871         struct bonding *bond = bond_get_bond_by_slave(slave);
 872         struct learning_pkt pkt;
 873         int size = sizeof(struct learning_pkt);
 874         int i;
 875
 876         memset(&pkt, 0, size);
 877         memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
 878         memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
 879         pkt.type = __constant_htons(ETH_P_LOOP);
 880
 881         for (i = 0; i < MAX_LP_BURST; i++) {
 882                 struct sk_buff *skb;
 883                 char *data;
 884
 885                 skb = dev_alloc_skb(size);
 886                 if (!skb) {
 887                         return;
 888                 }
 889
 890                 data = skb_put(skb, size);
 891                 memcpy(data, &pkt, size);
 892
 893                 skb->mac.raw = data;
 894                 skb->nh.raw = data + ETH_HLEN;
 895                 skb->protocol = pkt.type;
 896                 skb->priority = TC_PRIO_CONTROL;
 897                 skb->dev = slave->dev;
 898
 899                 if (!list_empty(&bond->vlan_list)) {
 900                         struct vlan_entry *vlan;
 901
 902                         vlan = bond_next_vlan(bond,
 903                                               bond->alb_info.current_alb_vlan);
 904
 905                         bond->alb_info.current_alb_vlan = vlan;
 906                         if (!vlan) {
 907                                 kfree_skb(skb);
 908                                 continue;
 909                         }
 910
 911                         skb = vlan_put_tag(skb, vlan->vlan_id);
 912                         if (!skb) {
 913                                 printk(KERN_ERR DRV_NAME
 914                                        ": %s: Error: failed to insert VLAN tag\n",
 915                                        bond->dev->name);
 916                                 continue;
 917                         }
 918                 }
 919
 920                 dev_queue_xmit(skb);
 921         }
 922 }
 923
 924 /* hw is a boolean parameter that determines whether we should try and
 925  * set the hw address of the device as well as the hw address of the
 926  * net_device
 927  */
 928 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[], int hw)
 929 {
 930         struct net_device *dev = slave->dev;
 931         struct sockaddr s_addr;
 932
 933         if (!hw) {
 934                 memcpy(dev->dev_addr, addr, dev->addr_len);
 935                 return 0;
 936         }
 937
 938         /* for rlb each slave must have a unique hw mac addresses so that */
 939         /* each slave will receive packets destined to a different mac */
 940         memcpy(s_addr.sa_data, addr, dev->addr_len);
 941         s_addr.sa_family = dev->type;
 942         if (dev_set_mac_address(dev, &s_addr)) {
 943                 printk(KERN_ERR DRV_NAME
 944                        ": %s: Error: dev_set_mac_address of dev %s failed! ALB "
 945                        "mode requires that the base driver support setting "
 946                        "the hw address also when the network device's "
 947                        "interface is open\n",
 948                        dev->master->name, dev->name);
 949                 return -EOPNOTSUPP;
 950         }
 951         return 0;
 952 }
 953
 954 /* Caller must hold bond lock for write or curr_slave_lock for write*/
 955 static void alb_swap_mac_addr(struct bonding *bond, struct slave *slave1, struct slave *slave2)
 956 {
 957         struct slave *disabled_slave = NULL;
 958         u8 tmp_mac_addr[ETH_ALEN];
 959         int slaves_state_differ;
 960
 961         slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
 962
 963         memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
 964         alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr, bond->alb_info.rlb_enabled);
 965         alb_set_slave_mac_addr(slave2, tmp_mac_addr, bond->alb_info.rlb_enabled);
 966
 967         /* fasten the change in the switch */
 968         if (SLAVE_IS_OK(slave1)) {
 969                 alb_send_learning_packets(slave1, slave1->dev->dev_addr);
 970                 if (bond->alb_info.rlb_enabled) {
 971                         /* inform the clients that the mac address
 972                          * has changed
 973                          */
 974                         rlb_req_update_slave_clients(bond, slave1);
 975                 }
 976         } else {
 977                 disabled_slave = slave1;
 978         }
 979
 980         if (SLAVE_IS_OK(slave2)) {
 981                 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
 982                 if (bond->alb_info.rlb_enabled) {
 983                         /* inform the clients that the mac address
 984                          * has changed
 985                          */
 986                         rlb_req_update_slave_clients(bond, slave2);
 987                 }
 988         } else {
 989                 disabled_slave = slave2;
 990         }
 991
 992         if (bond->alb_info.rlb_enabled && slaves_state_differ) {
 993                 /* A disabled slave was assigned an active mac addr */
 994                 rlb_teach_disabled_mac_on_primary(bond,
 995                                                   disabled_slave->dev->dev_addr);
 996         }
 997 }
 998
 999 /**
1000  * alb_change_hw_addr_on_detach
1001  * @bond: bonding we're working on
1002  * @slave: the slave that was just detached
1003  *
1004  * We assume that @slave was already detached from the slave list.
1005  *
1006  * If @slave's permanent hw address is different both from its current
1007  * address and from @bond's address, then somewhere in the bond there's
1008  * a slave that has @slave's permanet address as its current address.
1009  * We'll make sure that that slave no longer uses @slave's permanent address.
1010  *
1011  * Caller must hold bond lock
1012  */
1013 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1014 {
1015         int perm_curr_diff;
1016         int perm_bond_diff;
1017
1018         perm_curr_diff = memcmp(slave->perm_hwaddr,
1019                                 slave->dev->dev_addr,
1020                                 ETH_ALEN);
1021         perm_bond_diff = memcmp(slave->perm_hwaddr,
1022                                 bond->dev->dev_addr,
1023                                 ETH_ALEN);
1024
1025         if (perm_curr_diff && perm_bond_diff) {
1026                 struct slave *tmp_slave;
1027                 int i, found = 0;
1028
1029                 bond_for_each_slave(bond, tmp_slave, i) {
1030                         if (!memcmp(slave->perm_hwaddr,
1031                                     tmp_slave->dev->dev_addr,
1032                                     ETH_ALEN)) {
1033                                 found = 1;
1034                                 break;
1035                         }
1036                 }
1037
1038                 if (found) {
1039                         alb_swap_mac_addr(bond, slave, tmp_slave);
1040                 }
1041         }
1042 }
1043
1044 /**
1045  * alb_handle_addr_collision_on_attach
1046  * @bond: bonding we're working on
1047  * @slave: the slave that was just attached
1048  *
1049  * checks uniqueness of slave's mac address and handles the case the
1050  * new slave uses the bonds mac address.
1051  *
1052  * If the permanent hw address of @slave is @bond's hw address, we need to
1053  * find a different hw address to give @slave, that isn't in use by any other
1054  * slave in the bond. This address must be, of course, one of the premanent
1055  * addresses of the other slaves.
1056  *
1057  * We go over the slave list, and for each slave there we compare its
1058  * permanent hw address with the current address of all the other slaves.
1059  * If no match was found, then we've found a slave with a permanent address
1060  * that isn't used by any other slave in the bond, so we can assign it to
1061  * @slave.
1062  *
1063  * assumption: this function is called before @slave is attached to the
1064  *             bond slave list.
1065  *
1066  * caller must hold the bond lock for write since the mac addresses are compared
1067  * and may be swapped.
1068  */
1069 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1070 {
1071         struct slave *tmp_slave1, *tmp_slave2, *free_mac_slave;
1072         struct slave *has_bond_addr = bond->curr_active_slave;
1073         int i, j, found = 0;
1074
1075         if (bond->slave_cnt == 0) {
1076                 /* this is the first slave */
1077                 return 0;
1078         }
1079
1080         /* if slave's mac address differs from bond's mac address
1081          * check uniqueness of slave's mac address against the other
1082          * slaves in the bond.
1083          */
1084         if (memcmp(slave->perm_hwaddr, bond->dev->dev_addr, ETH_ALEN)) {
1085                 bond_for_each_slave(bond, tmp_slave1, i) {
1086                         if (!memcmp(tmp_slave1->dev->dev_addr, slave->dev->dev_addr,
1087                                     ETH_ALEN)) {
1088                                 found = 1;
1089                                 break;
1090                         }
1091                 }
1092
1093                 if (!found)
1094                         return 0;
1095
1096                 /* Try setting slave mac to bond address and fall-through
1097                    to code handling that situation below... */
1098                 alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1099                                        bond->alb_info.rlb_enabled);
1100         }
1101
1102         /* The slave's address is equal to the address of the bond.
1103          * Search for a spare address in the bond for this slave.
1104          */
1105         free_mac_slave = NULL;
1106
1107         bond_for_each_slave(bond, tmp_slave1, i) {
1108                 found = 0;
1109                 bond_for_each_slave(bond, tmp_slave2, j) {
1110                         if (!memcmp(tmp_slave1->perm_hwaddr,
1111                                     tmp_slave2->dev->dev_addr,
1112                                     ETH_ALEN)) {
1113                                 found = 1;
1114                                 break;
1115                         }
1116                 }
1117
1118                 if (!found) {
1119                         /* no slave has tmp_slave1's perm addr
1120                          * as its curr addr
1121                          */
1122                         free_mac_slave = tmp_slave1;
1123                         break;
1124                 }
1125
1126                 if (!has_bond_addr) {
1127                         if (!memcmp(tmp_slave1->dev->dev_addr,
1128                                     bond->dev->dev_addr,
1129                                     ETH_ALEN)) {
1130
1131                                 has_bond_addr = tmp_slave1;
1132                         }
1133                 }
1134         }
1135
1136         if (free_mac_slave) {
1137                 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1138                                        bond->alb_info.rlb_enabled);
1139
1140                 printk(KERN_WARNING DRV_NAME
1141                        ": %s: Warning: the hw address of slave %s is in use by "
1142                        "the bond; giving it the hw address of %s\n",
1143                        bond->dev->name, slave->dev->name, free_mac_slave->dev->name);
1144
1145         } else if (has_bond_addr) {
1146                 printk(KERN_ERR DRV_NAME
1147                        ": %s: Error: the hw address of slave %s is in use by the "
1148                        "bond; couldn't find a slave with a free hw address to "
1149                        "give it (this should not have happened)\n",
1150                        bond->dev->name, slave->dev->name);
1151                 return -EFAULT;
1152         }
1153
1154         return 0;
1155 }
1156
1157 /**
1158  * alb_set_mac_address
1159  * @bond:
1160  * @addr:
1161  *
1162  * In TLB mode all slaves are configured to the bond's hw address, but set
1163  * their dev_addr field to different addresses (based on their permanent hw
1164  * addresses).
1165  *
1166  * For each slave, this function sets the interface to the new address and then
1167  * changes its dev_addr field to its previous value.
1168  *
1169  * Unwinding assumes bond's mac address has not yet changed.
1170  */
1171 static int alb_set_mac_address(struct bonding *bond, void *addr)
1172 {
1173         struct sockaddr sa;
1174         struct slave *slave, *stop_at;
1175         char tmp_addr[ETH_ALEN];
1176         int res;
1177         int i;
1178
1179         if (bond->alb_info.rlb_enabled) {
1180                 return 0;
1181         }
1182
1183         bond_for_each_slave(bond, slave, i) {
1184                 if (slave->dev->set_mac_address == NULL) {
1185                         res = -EOPNOTSUPP;
1186                         goto unwind;
1187                 }
1188
1189                 /* save net_device's current hw address */
1190                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1191
1192                 res = dev_set_mac_address(slave->dev, addr);
1193
1194                 /* restore net_device's hw address */
1195                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1196
1197                 if (res) {
1198                         goto unwind;
1199                 }
1200         }
1201
1202         return 0;
1203
1204 unwind:
1205         memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1206         sa.sa_family = bond->dev->type;
1207
1208         /* unwind from head to the slave that failed */
1209         stop_at = slave;
1210         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
1211                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1212                 dev_set_mac_address(slave->dev, &sa);
1213                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1214         }
1215
1216         return res;
1217 }
1218
1219 /************************ exported alb funcions ************************/
1220
1221 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1222 {
1223         int res;
1224
1225         res = tlb_initialize(bond);
1226         if (res) {
1227                 return res;
1228         }
1229
1230         if (rlb_enabled) {
1231                 bond->alb_info.rlb_enabled = 1;
1232                 /* initialize rlb */
1233                 res = rlb_initialize(bond);
1234                 if (res) {
1235                         tlb_deinitialize(bond);
1236                         return res;
1237                 }
1238         } else {
1239                 bond->alb_info.rlb_enabled = 0;
1240         }
1241
1242         return 0;
1243 }
1244
1245 void bond_alb_deinitialize(struct bonding *bond)
1246 {
1247         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1248
1249         tlb_deinitialize(bond);
1250
1251         if (bond_info->rlb_enabled) {
1252                 rlb_deinitialize(bond);
1253         }
1254 }
1255
1256 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1257 {
1258         struct bonding *bond = bond_dev->priv;
1259         struct ethhdr *eth_data;
1260         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1261         struct slave *tx_slave = NULL;
1262         static const u32 ip_bcast = 0xffffffff;
1263         int hash_size = 0;
1264         int do_tx_balance = 1;
1265         u32 hash_index = 0;
1266         u8 *hash_start = NULL;
1267         int res = 1;
1268
1269         skb->mac.raw = (unsigned char *)skb->data;
1270         eth_data = eth_hdr(skb);
1271
1272         /* make sure that the curr_active_slave and the slaves list do
1273          * not change during tx
1274          */
1275         read_lock(&bond->lock);
1276         read_lock(&bond->curr_slave_lock);
1277
1278         if (!BOND_IS_OK(bond)) {
1279                 goto out;
1280         }
1281
1282         switch (ntohs(skb->protocol)) {
1283         case ETH_P_IP:
1284                 if ((memcmp(eth_data->h_dest, mac_bcast, ETH_ALEN) == 0) ||
1285                     (skb->nh.iph->daddr == ip_bcast) ||
1286                     (skb->nh.iph->protocol == IPPROTO_IGMP)) {
1287                         do_tx_balance = 0;
1288                         break;
1289                 }
1290                 hash_start = (char*)&(skb->nh.iph->daddr);
1291                 hash_size = sizeof(skb->nh.iph->daddr);
1292                 break;
1293         case ETH_P_IPV6:
1294                 if (memcmp(eth_data->h_dest, mac_bcast, ETH_ALEN) == 0) {
1295                         do_tx_balance = 0;
1296                         break;
1297                 }
1298
1299                 hash_start = (char*)&(skb->nh.ipv6h->daddr);
1300                 hash_size = sizeof(skb->nh.ipv6h->daddr);
1301                 break;
1302         case ETH_P_IPX:
1303                 if (ipx_hdr(skb)->ipx_checksum !=
1304                     __constant_htons(IPX_NO_CHECKSUM)) {
1305                         /* something is wrong with this packet */
1306                         do_tx_balance = 0;
1307                         break;
1308                 }
1309
1310                 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1311                         /* The only protocol worth balancing in
1312                          * this family since it has an "ARP" like
1313                          * mechanism
1314                          */
1315                         do_tx_balance = 0;
1316                         break;
1317                 }
1318
1319                 hash_start = (char*)eth_data->h_dest;
1320                 hash_size = ETH_ALEN;
1321                 break;
1322         case ETH_P_ARP:
1323                 do_tx_balance = 0;
1324                 if (bond_info->rlb_enabled) {
1325                         tx_slave = rlb_arp_xmit(skb, bond);
1326                 }
1327                 break;
1328         default:
1329                 do_tx_balance = 0;
1330                 break;
1331         }
1332
1333         if (do_tx_balance) {
1334                 hash_index = _simple_hash(hash_start, hash_size);
1335                 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1336         }
1337
1338         if (!tx_slave) {
1339                 /* unbalanced or unassigned, send through primary */
1340                 tx_slave = bond->curr_active_slave;
1341                 bond_info->unbalanced_load += skb->len;
1342         }
1343
1344         if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1345                 if (tx_slave != bond->curr_active_slave) {
1346                         memcpy(eth_data->h_source,
1347                                tx_slave->dev->dev_addr,
1348                                ETH_ALEN);
1349                 }
1350
1351                 res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1352         } else {
1353                 if (tx_slave) {
1354                         tlb_clear_slave(bond, tx_slave, 0);
1355                 }
1356         }
1357
1358 out:
1359         if (res) {
1360                 /* no suitable interface, frame not sent */
1361                 dev_kfree_skb(skb);
1362         }
1363         read_unlock(&bond->curr_slave_lock);
1364         read_unlock(&bond->lock);
1365         return 0;
1366 }
1367
1368 void bond_alb_monitor(struct bonding *bond)
1369 {
1370         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1371         struct slave *slave;
1372         int i;
1373
1374         read_lock(&bond->lock);
1375
1376         if (bond->kill_timers) {
1377                 goto out;
1378         }
1379
1380         if (bond->slave_cnt == 0) {
1381                 bond_info->tx_rebalance_counter = 0;
1382                 bond_info->lp_counter = 0;
1383                 goto re_arm;
1384         }
1385
1386         bond_info->tx_rebalance_counter++;
1387         bond_info->lp_counter++;
1388
1389         /* send learning packets */
1390         if (bond_info->lp_counter >= BOND_ALB_LP_TICKS) {
1391                 /* change of curr_active_slave involves swapping of mac addresses.
1392                  * in order to avoid this swapping from happening while
1393                  * sending the learning packets, the curr_slave_lock must be held for
1394                  * read.
1395                  */
1396                 read_lock(&bond->curr_slave_lock);
1397
1398                 bond_for_each_slave(bond, slave, i) {
1399                         alb_send_learning_packets(slave, slave->dev->dev_addr);
1400                 }
1401
1402                 read_unlock(&bond->curr_slave_lock);
1403
1404                 bond_info->lp_counter = 0;
1405         }
1406
1407         /* rebalance tx traffic */
1408         if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1409
1410                 read_lock(&bond->curr_slave_lock);
1411
1412                 bond_for_each_slave(bond, slave, i) {
1413                         tlb_clear_slave(bond, slave, 1);
1414                         if (slave == bond->curr_active_slave) {
1415                                 SLAVE_TLB_INFO(slave).load =
1416                                         bond_info->unbalanced_load /
1417                                                 BOND_TLB_REBALANCE_INTERVAL;
1418                                 bond_info->unbalanced_load = 0;
1419                         }
1420                 }
1421
1422                 read_unlock(&bond->curr_slave_lock);
1423
1424                 bond_info->tx_rebalance_counter = 0;
1425         }
1426
1427         /* handle rlb stuff */
1428         if (bond_info->rlb_enabled) {
1429                 /* the following code changes the promiscuity of the
1430                  * the curr_active_slave. It needs to be locked with a
1431                  * write lock to protect from other code that also
1432                  * sets the promiscuity.
1433                  */
1434                 write_lock_bh(&bond->curr_slave_lock);
1435
1436                 if (bond_info->primary_is_promisc &&
1437                     (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1438
1439                         bond_info->rlb_promisc_timeout_counter = 0;
1440
1441                         /* If the primary was set to promiscuous mode
1442                          * because a slave was disabled then
1443                          * it can now leave promiscuous mode.
1444                          */
1445                         dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1446                         bond_info->primary_is_promisc = 0;
1447                 }
1448
1449                 write_unlock_bh(&bond->curr_slave_lock);
1450
1451                 if (bond_info->rlb_rebalance) {
1452                         bond_info->rlb_rebalance = 0;
1453                         rlb_rebalance(bond);
1454                 }
1455
1456                 /* check if clients need updating */
1457                 if (bond_info->rx_ntt) {
1458                         if (bond_info->rlb_update_delay_counter) {
1459                                 --bond_info->rlb_update_delay_counter;
1460                         } else {
1461                                 rlb_update_rx_clients(bond);
1462                                 if (bond_info->rlb_update_retry_counter) {
1463                                         --bond_info->rlb_update_retry_counter;
1464                                 } else {
1465                                         bond_info->rx_ntt = 0;
1466                                 }
1467                         }
1468                 }
1469         }
1470
1471 re_arm:
1472         mod_timer(&(bond_info->alb_timer), jiffies + alb_delta_in_ticks);
1473 out:
1474         read_unlock(&bond->lock);
1475 }
1476
1477 /* assumption: called before the slave is attached to the bond
1478  * and not locked by the bond lock
1479  */
1480 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1481 {
1482         int res;
1483
1484         res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1485                                      bond->alb_info.rlb_enabled);
1486         if (res) {
1487                 return res;
1488         }
1489
1490         /* caller must hold the bond lock for write since the mac addresses
1491          * are compared and may be swapped.
1492          */
1493         write_lock_bh(&bond->lock);
1494
1495         res = alb_handle_addr_collision_on_attach(bond, slave);
1496
1497         write_unlock_bh(&bond->lock);
1498
1499         if (res) {
1500                 return res;
1501         }
1502
1503         tlb_init_slave(slave);
1504
1505         /* order a rebalance ASAP */
1506         bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1507
1508         if (bond->alb_info.rlb_enabled) {
1509                 bond->alb_info.rlb_rebalance = 1;
1510         }
1511
1512         return 0;
1513 }
1514
1515 /* Caller must hold bond lock for write */
1516 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1517 {
1518         if (bond->slave_cnt > 1) {
1519                 alb_change_hw_addr_on_detach(bond, slave);
1520         }
1521
1522         tlb_clear_slave(bond, slave, 0);
1523
1524         if (bond->alb_info.rlb_enabled) {
1525                 bond->alb_info.next_rx_slave = NULL;
1526                 rlb_clear_slave(bond, slave);
1527         }
1528 }
1529
1530 /* Caller must hold bond lock for read */
1531 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1532 {
1533         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1534
1535         if (link == BOND_LINK_DOWN) {
1536                 tlb_clear_slave(bond, slave, 0);
1537                 if (bond->alb_info.rlb_enabled) {
1538                         rlb_clear_slave(bond, slave);
1539                 }
1540         } else if (link == BOND_LINK_UP) {
1541                 /* order a rebalance ASAP */
1542                 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1543                 if (bond->alb_info.rlb_enabled) {
1544                         bond->alb_info.rlb_rebalance = 1;
1545                         /* If the updelay module parameter is smaller than the
1546                          * forwarding delay of the switch the rebalance will
1547                          * not work because the rebalance arp replies will
1548                          * not be forwarded to the clients..
1549                          */
1550                 }
1551         }
1552 }
1553
1554 /**
1555  * bond_alb_handle_active_change - assign new curr_active_slave
1556  * @bond: our bonding struct
1557  * @new_slave: new slave to assign
1558  *
1559  * Set the bond->curr_active_slave to @new_slave and handle
1560  * mac address swapping and promiscuity changes as needed.
1561  *
1562  * Caller must hold bond curr_slave_lock for write (or bond lock for write)
1563  */
1564 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1565 {
1566         struct slave *swap_slave;
1567         int i;
1568
1569         if (bond->curr_active_slave == new_slave) {
1570                 return;
1571         }
1572
1573         if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1574                 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1575                 bond->alb_info.primary_is_promisc = 0;
1576                 bond->alb_info.rlb_promisc_timeout_counter = 0;
1577         }
1578
1579         swap_slave = bond->curr_active_slave;
1580         bond->curr_active_slave = new_slave;
1581
1582         if (!new_slave || (bond->slave_cnt == 0)) {
1583                 return;
1584         }
1585
1586         /* set the new curr_active_slave to the bonds mac address
1587          * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1588          */
1589         if (!swap_slave) {
1590                 struct slave *tmp_slave;
1591                 /* find slave that is holding the bond's mac address */
1592                 bond_for_each_slave(bond, tmp_slave, i) {
1593                         if (!memcmp(tmp_slave->dev->dev_addr,
1594                                     bond->dev->dev_addr, ETH_ALEN)) {
1595                                 swap_slave = tmp_slave;
1596                                 break;
1597                         }
1598                 }
1599         }
1600
1601         /* curr_active_slave must be set before calling alb_swap_mac_addr */
1602         if (swap_slave) {
1603                 /* swap mac address */
1604                 alb_swap_mac_addr(bond, swap_slave, new_slave);
1605         } else {
1606                 /* set the new_slave to the bond mac address */
1607                 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1608                                        bond->alb_info.rlb_enabled);
1609                 /* fasten bond mac on new current slave */
1610                 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1611         }
1612 }
1613
1614 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1615 {
1616         struct bonding *bond = bond_dev->priv;
1617         struct sockaddr *sa = addr;
1618         struct slave *slave, *swap_slave;
1619         int res;
1620         int i;
1621
1622         if (!is_valid_ether_addr(sa->sa_data)) {
1623                 return -EADDRNOTAVAIL;
1624         }
1625
1626         res = alb_set_mac_address(bond, addr);
1627         if (res) {
1628                 return res;
1629         }
1630
1631         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1632
1633         /* If there is no curr_active_slave there is nothing else to do.
1634          * Otherwise we'll need to pass the new address to it and handle
1635          * duplications.
1636          */
1637         if (!bond->curr_active_slave) {
1638                 return 0;
1639         }
1640
1641         swap_slave = NULL;
1642
1643         bond_for_each_slave(bond, slave, i) {
1644                 if (!memcmp(slave->dev->dev_addr, bond_dev->dev_addr, ETH_ALEN)) {
1645                         swap_slave = slave;
1646                         break;
1647                 }
1648         }
1649
1650         if (swap_slave) {
1651                 alb_swap_mac_addr(bond, swap_slave, bond->curr_active_slave);
1652         } else {
1653                 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr,
1654                                        bond->alb_info.rlb_enabled);
1655
1656                 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1657                 if (bond->alb_info.rlb_enabled) {
1658                         /* inform clients mac address has changed */
1659                         rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1660                 }
1661         }
1662
1663         return 0;
1664 }
1665
1666 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1667 {
1668         if (bond->alb_info.current_alb_vlan &&
1669             (bond->alb_info.current_alb_vlan->vlan_id == vlan_id)) {
1670                 bond->alb_info.current_alb_vlan = NULL;
1671         }
1672
1673         if (bond->alb_info.rlb_enabled) {
1674                 rlb_clear_vlan(bond, vlan_id);
1675         }
1676 }
1677