git.oblomov.eu Git - linux-2.6/blob - 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 static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb)
 108 {
 109         return (struct arp_pkt *)skb_network_header(skb);
 110 }
 111
 112 /* Forward declaration */
 113 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[]);
 114
 115 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
 116 {
 117         int i;
 118         u8 hash = 0;
 119
 120         for (i = 0; i < hash_size; i++) {
 121                 hash ^= hash_start[i];
 122         }
 123
 124         return hash;
 125 }
 126
 127 /*********************** tlb specific functions ***************************/
 128
 129 static inline void _lock_tx_hashtbl(struct bonding *bond)
 130 {
 131         spin_lock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 132 }
 133
 134 static inline void _unlock_tx_hashtbl(struct bonding *bond)
 135 {
 136         spin_unlock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 137 }
 138
 139 /* Caller must hold tx_hashtbl lock */
 140 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
 141 {
 142         if (save_load) {
 143                 entry->load_history = 1 + entry->tx_bytes /
 144                                       BOND_TLB_REBALANCE_INTERVAL;
 145                 entry->tx_bytes = 0;
 146         }
 147
 148         entry->tx_slave = NULL;
 149         entry->next = TLB_NULL_INDEX;
 150         entry->prev = TLB_NULL_INDEX;
 151 }
 152
 153 static inline void tlb_init_slave(struct slave *slave)
 154 {
 155         SLAVE_TLB_INFO(slave).load = 0;
 156         SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
 157 }
 158
 159 /* Caller must hold bond lock for read */
 160 static void tlb_clear_slave(struct bonding *bond, struct slave *slave, int save_load)
 161 {
 162         struct tlb_client_info *tx_hash_table;
 163         u32 index;
 164
 165         _lock_tx_hashtbl(bond);
 166
 167         /* clear slave from tx_hashtbl */
 168         tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
 169
 170         index = SLAVE_TLB_INFO(slave).head;
 171         while (index != TLB_NULL_INDEX) {
 172                 u32 next_index = tx_hash_table[index].next;
 173                 tlb_init_table_entry(&tx_hash_table[index], save_load);
 174                 index = next_index;
 175         }
 176
 177         tlb_init_slave(slave);
 178
 179         _unlock_tx_hashtbl(bond);
 180 }
 181
 182 /* Must be called before starting the monitor timer */
 183 static int tlb_initialize(struct bonding *bond)
 184 {
 185         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 186         int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 187         struct tlb_client_info *new_hashtbl;
 188         int i;
 189
 190         spin_lock_init(&(bond_info->tx_hashtbl_lock));
 191
 192         new_hashtbl = kzalloc(size, GFP_KERNEL);
 193         if (!new_hashtbl) {
 194                 printk(KERN_ERR DRV_NAME
 195                        ": %s: Error: Failed to allocate TLB hash table\n",
 196                        bond->dev->name);
 197                 return -1;
 198         }
 199         _lock_tx_hashtbl(bond);
 200
 201         bond_info->tx_hashtbl = new_hashtbl;
 202
 203         for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
 204                 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 1);
 205         }
 206
 207         _unlock_tx_hashtbl(bond);
 208
 209         return 0;
 210 }
 211
 212 /* Must be called only after all slaves have been released */
 213 static void tlb_deinitialize(struct bonding *bond)
 214 {
 215         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 216
 217         _lock_tx_hashtbl(bond);
 218
 219         kfree(bond_info->tx_hashtbl);
 220         bond_info->tx_hashtbl = NULL;
 221
 222         _unlock_tx_hashtbl(bond);
 223 }
 224
 225 /* Caller must hold bond lock for read */
 226 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
 227 {
 228         struct slave *slave, *least_loaded;
 229         s64 max_gap;
 230         int i, found = 0;
 231
 232         /* Find the first enabled slave */
 233         bond_for_each_slave(bond, slave, i) {
 234                 if (SLAVE_IS_OK(slave)) {
 235                         found = 1;
 236                         break;
 237                 }
 238         }
 239
 240         if (!found) {
 241                 return NULL;
 242         }
 243
 244         least_loaded = slave;
 245         max_gap = (s64)(slave->speed << 20) - /* Convert to Megabit per sec */
 246                         (s64)(SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
 247
 248         /* Find the slave with the largest gap */
 249         bond_for_each_slave_from(bond, slave, i, least_loaded) {
 250                 if (SLAVE_IS_OK(slave)) {
 251                         s64 gap = (s64)(slave->speed << 20) -
 252                                         (s64)(SLAVE_TLB_INFO(slave).load << 3);
 253                         if (max_gap < gap) {
 254                                 least_loaded = slave;
 255                                 max_gap = gap;
 256                         }
 257                 }
 258         }
 259
 260         return least_loaded;
 261 }
 262
 263 /* Caller must hold bond lock for read */
 264 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index, u32 skb_len)
 265 {
 266         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 267         struct tlb_client_info *hash_table;
 268         struct slave *assigned_slave;
 269
 270         _lock_tx_hashtbl(bond);
 271
 272         hash_table = bond_info->tx_hashtbl;
 273         assigned_slave = hash_table[hash_index].tx_slave;
 274         if (!assigned_slave) {
 275                 assigned_slave = tlb_get_least_loaded_slave(bond);
 276
 277                 if (assigned_slave) {
 278                         struct tlb_slave_info *slave_info =
 279                                 &(SLAVE_TLB_INFO(assigned_slave));
 280                         u32 next_index = slave_info->head;
 281
 282                         hash_table[hash_index].tx_slave = assigned_slave;
 283                         hash_table[hash_index].next = next_index;
 284                         hash_table[hash_index].prev = TLB_NULL_INDEX;
 285
 286                         if (next_index != TLB_NULL_INDEX) {
 287                                 hash_table[next_index].prev = hash_index;
 288                         }
 289
 290                         slave_info->head = hash_index;
 291                         slave_info->load +=
 292                                 hash_table[hash_index].load_history;
 293                 }
 294         }
 295
 296         if (assigned_slave) {
 297                 hash_table[hash_index].tx_bytes += skb_len;
 298         }
 299
 300         _unlock_tx_hashtbl(bond);
 301
 302         return assigned_slave;
 303 }
 304
 305 /*********************** rlb specific functions ***************************/
 306 static inline void _lock_rx_hashtbl(struct bonding *bond)
 307 {
 308         spin_lock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 309 }
 310
 311 static inline void _unlock_rx_hashtbl(struct bonding *bond)
 312 {
 313         spin_unlock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 314 }
 315
 316 /* when an ARP REPLY is received from a client update its info
 317  * in the rx_hashtbl
 318  */
 319 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 320 {
 321         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 322         struct rlb_client_info *client_info;
 323         u32 hash_index;
 324
 325         _lock_rx_hashtbl(bond);
 326
 327         hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
 328         client_info = &(bond_info->rx_hashtbl[hash_index]);
 329
 330         if ((client_info->assigned) &&
 331             (client_info->ip_src == arp->ip_dst) &&
 332             (client_info->ip_dst == arp->ip_src)) {
 333                 /* update the clients MAC address */
 334                 memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
 335                 client_info->ntt = 1;
 336                 bond_info->rx_ntt = 1;
 337         }
 338
 339         _unlock_rx_hashtbl(bond);
 340 }
 341
 342 static int rlb_arp_recv(struct sk_buff *skb, struct net_device *bond_dev, struct packet_type *ptype, struct net_device *orig_dev)
 343 {
 344         struct bonding *bond = bond_dev->priv;
 345         struct arp_pkt *arp = (struct arp_pkt *)skb->data;
 346         int res = NET_RX_DROP;
 347
 348         if (!(bond_dev->flags & IFF_MASTER))
 349                 goto out;
 350
 351         if (!arp) {
 352                 dprintk("Packet has no ARP data\n");
 353                 goto out;
 354         }
 355
 356         if (skb->len < sizeof(struct arp_pkt)) {
 357                 dprintk("Packet is too small to be an ARP\n");
 358                 goto out;
 359         }
 360
 361         if (arp->op_code == htons(ARPOP_REPLY)) {
 362                 /* update rx hash table for this ARP */
 363                 rlb_update_entry_from_arp(bond, arp);
 364                 dprintk("Server received an ARP Reply from client\n");
 365         }
 366
 367         res = NET_RX_SUCCESS;
 368
 369 out:
 370         dev_kfree_skb(skb);
 371
 372         return res;
 373 }
 374
 375 /* Caller must hold bond lock for read */
 376 static struct slave *rlb_next_rx_slave(struct bonding *bond)
 377 {
 378         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 379         struct slave *rx_slave, *slave, *start_at;
 380         int i = 0;
 381
 382         if (bond_info->next_rx_slave) {
 383                 start_at = bond_info->next_rx_slave;
 384         } else {
 385                 start_at = bond->first_slave;
 386         }
 387
 388         rx_slave = NULL;
 389
 390         bond_for_each_slave_from(bond, slave, i, start_at) {
 391                 if (SLAVE_IS_OK(slave)) {
 392                         if (!rx_slave) {
 393                                 rx_slave = slave;
 394                         } else if (slave->speed > rx_slave->speed) {
 395                                 rx_slave = slave;
 396                         }
 397                 }
 398         }
 399
 400         if (rx_slave) {
 401                 bond_info->next_rx_slave = rx_slave->next;
 402         }
 403
 404         return rx_slave;
 405 }
 406
 407 /* teach the switch the mac of a disabled slave
 408  * on the primary for fault tolerance
 409  *
 410  * Caller must hold bond->curr_slave_lock for write or bond lock for write
 411  */
 412 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
 413 {
 414         if (!bond->curr_active_slave) {
 415                 return;
 416         }
 417
 418         if (!bond->alb_info.primary_is_promisc) {
 419                 bond->alb_info.primary_is_promisc = 1;
 420                 dev_set_promiscuity(bond->curr_active_slave->dev, 1);
 421         }
 422
 423         bond->alb_info.rlb_promisc_timeout_counter = 0;
 424
 425         alb_send_learning_packets(bond->curr_active_slave, addr);
 426 }
 427
 428 /* slave being removed should not be active at this point
 429  *
 430  * Caller must hold bond lock for read
 431  */
 432 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 433 {
 434         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 435         struct rlb_client_info *rx_hash_table;
 436         u32 index, next_index;
 437
 438         /* clear slave from rx_hashtbl */
 439         _lock_rx_hashtbl(bond);
 440
 441         rx_hash_table = bond_info->rx_hashtbl;
 442         index = bond_info->rx_hashtbl_head;
 443         for (; index != RLB_NULL_INDEX; index = next_index) {
 444                 next_index = rx_hash_table[index].next;
 445                 if (rx_hash_table[index].slave == slave) {
 446                         struct slave *assigned_slave = rlb_next_rx_slave(bond);
 447
 448                         if (assigned_slave) {
 449                                 rx_hash_table[index].slave = assigned_slave;
 450                                 if (memcmp(rx_hash_table[index].mac_dst,
 451                                            mac_bcast, ETH_ALEN)) {
 452                                         bond_info->rx_hashtbl[index].ntt = 1;
 453                                         bond_info->rx_ntt = 1;
 454                                         /* A slave has been removed from the
 455                                          * table because it is either disabled
 456                                          * or being released. We must retry the
 457                                          * update to avoid clients from not
 458                                          * being updated & disconnecting when
 459                                          * there is stress
 460                                          */
 461                                         bond_info->rlb_update_retry_counter =
 462                                                 RLB_UPDATE_RETRY;
 463                                 }
 464                         } else {  /* there is no active slave */
 465                                 rx_hash_table[index].slave = NULL;
 466                         }
 467                 }
 468         }
 469
 470         _unlock_rx_hashtbl(bond);
 471
 472         write_lock(&bond->curr_slave_lock);
 473
 474         if (slave != bond->curr_active_slave) {
 475                 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 476         }
 477
 478         write_unlock(&bond->curr_slave_lock);
 479 }
 480
 481 static void rlb_update_client(struct rlb_client_info *client_info)
 482 {
 483         int i;
 484
 485         if (!client_info->slave) {
 486                 return;
 487         }
 488
 489         for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 490                 struct sk_buff *skb;
 491
 492                 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 493                                  client_info->ip_dst,
 494                                  client_info->slave->dev,
 495                                  client_info->ip_src,
 496                                  client_info->mac_dst,
 497                                  client_info->slave->dev->dev_addr,
 498                                  client_info->mac_dst);
 499                 if (!skb) {
 500                         printk(KERN_ERR DRV_NAME
 501                                ": %s: Error: failed to create an ARP packet\n",
 502                                client_info->slave->dev->master->name);
 503                         continue;
 504                 }
 505
 506                 skb->dev = client_info->slave->dev;
 507
 508                 if (client_info->tag) {
 509                         skb = vlan_put_tag(skb, client_info->vlan_id);
 510                         if (!skb) {
 511                                 printk(KERN_ERR DRV_NAME
 512                                        ": %s: Error: failed to insert VLAN tag\n",
 513                                        client_info->slave->dev->master->name);
 514                                 continue;
 515                         }
 516                 }
 517
 518                 arp_xmit(skb);
 519         }
 520 }
 521
 522 /* sends ARP REPLIES that update the clients that need updating */
 523 static void rlb_update_rx_clients(struct bonding *bond)
 524 {
 525         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 526         struct rlb_client_info *client_info;
 527         u32 hash_index;
 528
 529         _lock_rx_hashtbl(bond);
 530
 531         hash_index = bond_info->rx_hashtbl_head;
 532         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 533                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 534                 if (client_info->ntt) {
 535                         rlb_update_client(client_info);
 536                         if (bond_info->rlb_update_retry_counter == 0) {
 537                                 client_info->ntt = 0;
 538                         }
 539                 }
 540         }
 541
 542         /* do not update the entries again untill this counter is zero so that
 543          * not to confuse the clients.
 544          */
 545         bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 546
 547         _unlock_rx_hashtbl(bond);
 548 }
 549
 550 /* The slave was assigned a new mac address - update the clients */
 551 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 552 {
 553         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 554         struct rlb_client_info *client_info;
 555         int ntt = 0;
 556         u32 hash_index;
 557
 558         _lock_rx_hashtbl(bond);
 559
 560         hash_index = bond_info->rx_hashtbl_head;
 561         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 562                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 563
 564                 if ((client_info->slave == slave) &&
 565                     memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
 566                         client_info->ntt = 1;
 567                         ntt = 1;
 568                 }
 569         }
 570
 571         // update the team's flag only after the whole iteration
 572         if (ntt) {
 573                 bond_info->rx_ntt = 1;
 574                 //fasten the change
 575                 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 576         }
 577
 578         _unlock_rx_hashtbl(bond);
 579 }
 580
 581 /* mark all clients using src_ip to be updated */
 582 static void rlb_req_update_subnet_clients(struct bonding *bond, u32 src_ip)
 583 {
 584         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 585         struct rlb_client_info *client_info;
 586         u32 hash_index;
 587
 588         _lock_rx_hashtbl(bond);
 589
 590         hash_index = bond_info->rx_hashtbl_head;
 591         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 592                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 593
 594                 if (!client_info->slave) {
 595                         printk(KERN_ERR DRV_NAME
 596                                ": %s: Error: found a client with no channel in "
 597                                "the client's hash table\n",
 598                                bond->dev->name);
 599                         continue;
 600                 }
 601                 /*update all clients using this src_ip, that are not assigned
 602                  * to the team's address (curr_active_slave) and have a known
 603                  * unicast mac address.
 604                  */
 605                 if ((client_info->ip_src == src_ip) &&
 606                     memcmp(client_info->slave->dev->dev_addr,
 607                            bond->dev->dev_addr, ETH_ALEN) &&
 608                     memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
 609                         client_info->ntt = 1;
 610                         bond_info->rx_ntt = 1;
 611                 }
 612         }
 613
 614         _unlock_rx_hashtbl(bond);
 615 }
 616
 617 /* Caller must hold both bond and ptr locks for read */
 618 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
 619 {
 620         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 621         struct arp_pkt *arp = arp_pkt(skb);
 622         struct slave *assigned_slave;
 623         struct rlb_client_info *client_info;
 624         u32 hash_index = 0;
 625
 626         _lock_rx_hashtbl(bond);
 627
 628         hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_src));
 629         client_info = &(bond_info->rx_hashtbl[hash_index]);
 630
 631         if (client_info->assigned) {
 632                 if ((client_info->ip_src == arp->ip_src) &&
 633                     (client_info->ip_dst == arp->ip_dst)) {
 634                         /* the entry is already assigned to this client */
 635                         if (memcmp(arp->mac_dst, mac_bcast, ETH_ALEN)) {
 636                                 /* update mac address from arp */
 637                                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 638                         }
 639
 640                         assigned_slave = client_info->slave;
 641                         if (assigned_slave) {
 642                                 _unlock_rx_hashtbl(bond);
 643                                 return assigned_slave;
 644                         }
 645                 } else {
 646                         /* the entry is already assigned to some other client,
 647                          * move the old client to primary (curr_active_slave) so
 648                          * that the new client can be assigned to this entry.
 649                          */
 650                         if (bond->curr_active_slave &&
 651                             client_info->slave != bond->curr_active_slave) {
 652                                 client_info->slave = bond->curr_active_slave;
 653                                 rlb_update_client(client_info);
 654                         }
 655                 }
 656         }
 657         /* assign a new slave */
 658         assigned_slave = rlb_next_rx_slave(bond);
 659
 660         if (assigned_slave) {
 661                 client_info->ip_src = arp->ip_src;
 662                 client_info->ip_dst = arp->ip_dst;
 663                 /* arp->mac_dst is broadcast for arp reqeusts.
 664                  * will be updated with clients actual unicast mac address
 665                  * upon receiving an arp reply.
 666                  */
 667                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 668                 client_info->slave = assigned_slave;
 669
 670                 if (memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
 671                         client_info->ntt = 1;
 672                         bond->alb_info.rx_ntt = 1;
 673                 } else {
 674                         client_info->ntt = 0;
 675                 }
 676
 677                 if (!list_empty(&bond->vlan_list)) {
 678                         unsigned short vlan_id;
 679                         int res = vlan_get_tag(skb, &vlan_id);
 680                         if (!res) {
 681                                 client_info->tag = 1;
 682                                 client_info->vlan_id = vlan_id;
 683                         }
 684                 }
 685
 686                 if (!client_info->assigned) {
 687                         u32 prev_tbl_head = bond_info->rx_hashtbl_head;
 688                         bond_info->rx_hashtbl_head = hash_index;
 689                         client_info->next = prev_tbl_head;
 690                         if (prev_tbl_head != RLB_NULL_INDEX) {
 691                                 bond_info->rx_hashtbl[prev_tbl_head].prev =
 692                                         hash_index;
 693                         }
 694                         client_info->assigned = 1;
 695                 }
 696         }
 697
 698         _unlock_rx_hashtbl(bond);
 699
 700         return assigned_slave;
 701 }
 702
 703 /* chooses (and returns) transmit channel for arp reply
 704  * does not choose channel for other arp types since they are
 705  * sent on the curr_active_slave
 706  */
 707 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 708 {
 709         struct arp_pkt *arp = arp_pkt(skb);
 710         struct slave *tx_slave = NULL;
 711
 712         if (arp->op_code == __constant_htons(ARPOP_REPLY)) {
 713                 /* the arp must be sent on the selected
 714                 * rx channel
 715                 */
 716                 tx_slave = rlb_choose_channel(skb, bond);
 717                 if (tx_slave) {
 718                         memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
 719                 }
 720                 dprintk("Server sent ARP Reply packet\n");
 721         } else if (arp->op_code == __constant_htons(ARPOP_REQUEST)) {
 722                 /* Create an entry in the rx_hashtbl for this client as a
 723                  * place holder.
 724                  * When the arp reply is received the entry will be updated
 725                  * with the correct unicast address of the client.
 726                  */
 727                 rlb_choose_channel(skb, bond);
 728
 729                 /* The ARP relpy packets must be delayed so that
 730                  * they can cancel out the influence of the ARP request.
 731                  */
 732                 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 733
 734                 /* arp requests are broadcast and are sent on the primary
 735                  * the arp request will collapse all clients on the subnet to
 736                  * the primary slave. We must register these clients to be
 737                  * updated with their assigned mac.
 738                  */
 739                 rlb_req_update_subnet_clients(bond, arp->ip_src);
 740                 dprintk("Server sent ARP Request packet\n");
 741         }
 742
 743         return tx_slave;
 744 }
 745
 746 /* Caller must hold bond lock for read */
 747 static void rlb_rebalance(struct bonding *bond)
 748 {
 749         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 750         struct slave *assigned_slave;
 751         struct rlb_client_info *client_info;
 752         int ntt;
 753         u32 hash_index;
 754
 755         _lock_rx_hashtbl(bond);
 756
 757         ntt = 0;
 758         hash_index = bond_info->rx_hashtbl_head;
 759         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 760                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 761                 assigned_slave = rlb_next_rx_slave(bond);
 762                 if (assigned_slave && (client_info->slave != assigned_slave)) {
 763                         client_info->slave = assigned_slave;
 764                         client_info->ntt = 1;
 765                         ntt = 1;
 766                 }
 767         }
 768
 769         /* update the team's flag only after the whole iteration */
 770         if (ntt) {
 771                 bond_info->rx_ntt = 1;
 772         }
 773         _unlock_rx_hashtbl(bond);
 774 }
 775
 776 /* Caller must hold rx_hashtbl lock */
 777 static void rlb_init_table_entry(struct rlb_client_info *entry)
 778 {
 779         memset(entry, 0, sizeof(struct rlb_client_info));
 780         entry->next = RLB_NULL_INDEX;
 781         entry->prev = RLB_NULL_INDEX;
 782 }
 783
 784 static int rlb_initialize(struct bonding *bond)
 785 {
 786         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 787         struct packet_type *pk_type = &(BOND_ALB_INFO(bond).rlb_pkt_type);
 788         struct rlb_client_info  *new_hashtbl;
 789         int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 790         int i;
 791
 792         spin_lock_init(&(bond_info->rx_hashtbl_lock));
 793
 794         new_hashtbl = kmalloc(size, GFP_KERNEL);
 795         if (!new_hashtbl) {
 796                 printk(KERN_ERR DRV_NAME
 797                        ": %s: Error: Failed to allocate RLB hash table\n",
 798                        bond->dev->name);
 799                 return -1;
 800         }
 801         _lock_rx_hashtbl(bond);
 802
 803         bond_info->rx_hashtbl = new_hashtbl;
 804
 805         bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
 806
 807         for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
 808                 rlb_init_table_entry(bond_info->rx_hashtbl + i);
 809         }
 810
 811         _unlock_rx_hashtbl(bond);
 812
 813         /*initialize packet type*/
 814         pk_type->type = __constant_htons(ETH_P_ARP);
 815         pk_type->dev = bond->dev;
 816         pk_type->func = rlb_arp_recv;
 817
 818         /* register to receive ARPs */
 819         dev_add_pack(pk_type);
 820
 821         return 0;
 822 }
 823
 824 static void rlb_deinitialize(struct bonding *bond)
 825 {
 826         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 827
 828         dev_remove_pack(&(bond_info->rlb_pkt_type));
 829
 830         _lock_rx_hashtbl(bond);
 831
 832         kfree(bond_info->rx_hashtbl);
 833         bond_info->rx_hashtbl = NULL;
 834         bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
 835
 836         _unlock_rx_hashtbl(bond);
 837 }
 838
 839 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 840 {
 841         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 842         u32 curr_index;
 843
 844         _lock_rx_hashtbl(bond);
 845
 846         curr_index = bond_info->rx_hashtbl_head;
 847         while (curr_index != RLB_NULL_INDEX) {
 848                 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 849                 u32 next_index = bond_info->rx_hashtbl[curr_index].next;
 850                 u32 prev_index = bond_info->rx_hashtbl[curr_index].prev;
 851
 852                 if (curr->tag && (curr->vlan_id == vlan_id)) {
 853                         if (curr_index == bond_info->rx_hashtbl_head) {
 854                                 bond_info->rx_hashtbl_head = next_index;
 855                         }
 856                         if (prev_index != RLB_NULL_INDEX) {
 857                                 bond_info->rx_hashtbl[prev_index].next = next_index;
 858                         }
 859                         if (next_index != RLB_NULL_INDEX) {
 860                                 bond_info->rx_hashtbl[next_index].prev = prev_index;
 861                         }
 862
 863                         rlb_init_table_entry(curr);
 864                 }
 865
 866                 curr_index = next_index;
 867         }
 868
 869         _unlock_rx_hashtbl(bond);
 870 }
 871
 872 /*********************** tlb/rlb shared functions *********************/
 873
 874 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
 875 {
 876         struct bonding *bond = bond_get_bond_by_slave(slave);
 877         struct learning_pkt pkt;
 878         int size = sizeof(struct learning_pkt);
 879         int i;
 880
 881         memset(&pkt, 0, size);
 882         memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
 883         memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
 884         pkt.type = __constant_htons(ETH_P_LOOP);
 885
 886         for (i = 0; i < MAX_LP_BURST; i++) {
 887                 struct sk_buff *skb;
 888                 char *data;
 889
 890                 skb = dev_alloc_skb(size);
 891                 if (!skb) {
 892                         return;
 893                 }
 894
 895                 data = skb_put(skb, size);
 896                 memcpy(data, &pkt, size);
 897
 898                 skb_reset_mac_header(skb);
 899                 skb->network_header = skb->mac_header + ETH_HLEN;
 900                 skb->protocol = pkt.type;
 901                 skb->priority = TC_PRIO_CONTROL;
 902                 skb->dev = slave->dev;
 903
 904                 if (!list_empty(&bond->vlan_list)) {
 905                         struct vlan_entry *vlan;
 906
 907                         vlan = bond_next_vlan(bond,
 908                                               bond->alb_info.current_alb_vlan);
 909
 910                         bond->alb_info.current_alb_vlan = vlan;
 911                         if (!vlan) {
 912                                 kfree_skb(skb);
 913                                 continue;
 914                         }
 915
 916                         skb = vlan_put_tag(skb, vlan->vlan_id);
 917                         if (!skb) {
 918                                 printk(KERN_ERR DRV_NAME
 919                                        ": %s: Error: failed to insert VLAN tag\n",
 920                                        bond->dev->name);
 921                                 continue;
 922                         }
 923                 }
 924
 925                 dev_queue_xmit(skb);
 926         }
 927 }
 928
 929 /* hw is a boolean parameter that determines whether we should try and
 930  * set the hw address of the device as well as the hw address of the
 931  * net_device
 932  */
 933 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[], int hw)
 934 {
 935         struct net_device *dev = slave->dev;
 936         struct sockaddr s_addr;
 937
 938         if (!hw) {
 939                 memcpy(dev->dev_addr, addr, dev->addr_len);
 940                 return 0;
 941         }
 942
 943         /* for rlb each slave must have a unique hw mac addresses so that */
 944         /* each slave will receive packets destined to a different mac */
 945         memcpy(s_addr.sa_data, addr, dev->addr_len);
 946         s_addr.sa_family = dev->type;
 947         if (dev_set_mac_address(dev, &s_addr)) {
 948                 printk(KERN_ERR DRV_NAME
 949                        ": %s: Error: dev_set_mac_address of dev %s failed! ALB "
 950                        "mode requires that the base driver support setting "
 951                        "the hw address also when the network device's "
 952                        "interface is open\n",
 953                        dev->master->name, dev->name);
 954                 return -EOPNOTSUPP;
 955         }
 956         return 0;
 957 }
 958
 959 /* Caller must hold bond lock for write or curr_slave_lock for write*/
 960 static void alb_swap_mac_addr(struct bonding *bond, struct slave *slave1, struct slave *slave2)
 961 {
 962         struct slave *disabled_slave = NULL;
 963         u8 tmp_mac_addr[ETH_ALEN];
 964         int slaves_state_differ;
 965
 966         slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
 967
 968         memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
 969         alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr, bond->alb_info.rlb_enabled);
 970         alb_set_slave_mac_addr(slave2, tmp_mac_addr, bond->alb_info.rlb_enabled);
 971
 972         /* fasten the change in the switch */
 973         if (SLAVE_IS_OK(slave1)) {
 974                 alb_send_learning_packets(slave1, slave1->dev->dev_addr);
 975                 if (bond->alb_info.rlb_enabled) {
 976                         /* inform the clients that the mac address
 977                          * has changed
 978                          */
 979                         rlb_req_update_slave_clients(bond, slave1);
 980                 }
 981         } else {
 982                 disabled_slave = slave1;
 983         }
 984
 985         if (SLAVE_IS_OK(slave2)) {
 986                 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
 987                 if (bond->alb_info.rlb_enabled) {
 988                         /* inform the clients that the mac address
 989                          * has changed
 990                          */
 991                         rlb_req_update_slave_clients(bond, slave2);
 992                 }
 993         } else {
 994                 disabled_slave = slave2;
 995         }
 996
 997         if (bond->alb_info.rlb_enabled && slaves_state_differ) {
 998                 /* A disabled slave was assigned an active mac addr */
 999                 rlb_teach_disabled_mac_on_primary(bond,
1000                                                   disabled_slave->dev->dev_addr);
1001         }
1002 }
1003
1004 /**
1005  * alb_change_hw_addr_on_detach
1006  * @bond: bonding we're working on
1007  * @slave: the slave that was just detached
1008  *
1009  * We assume that @slave was already detached from the slave list.
1010  *
1011  * If @slave's permanent hw address is different both from its current
1012  * address and from @bond's address, then somewhere in the bond there's
1013  * a slave that has @slave's permanet address as its current address.
1014  * We'll make sure that that slave no longer uses @slave's permanent address.
1015  *
1016  * Caller must hold bond lock
1017  */
1018 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1019 {
1020         int perm_curr_diff;
1021         int perm_bond_diff;
1022
1023         perm_curr_diff = memcmp(slave->perm_hwaddr,
1024                                 slave->dev->dev_addr,
1025                                 ETH_ALEN);
1026         perm_bond_diff = memcmp(slave->perm_hwaddr,
1027                                 bond->dev->dev_addr,
1028                                 ETH_ALEN);
1029
1030         if (perm_curr_diff && perm_bond_diff) {
1031                 struct slave *tmp_slave;
1032                 int i, found = 0;
1033
1034                 bond_for_each_slave(bond, tmp_slave, i) {
1035                         if (!memcmp(slave->perm_hwaddr,
1036                                     tmp_slave->dev->dev_addr,
1037                                     ETH_ALEN)) {
1038                                 found = 1;
1039                                 break;
1040                         }
1041                 }
1042
1043                 if (found) {
1044                         alb_swap_mac_addr(bond, slave, tmp_slave);
1045                 }
1046         }
1047 }
1048
1049 /**
1050  * alb_handle_addr_collision_on_attach
1051  * @bond: bonding we're working on
1052  * @slave: the slave that was just attached
1053  *
1054  * checks uniqueness of slave's mac address and handles the case the
1055  * new slave uses the bonds mac address.
1056  *
1057  * If the permanent hw address of @slave is @bond's hw address, we need to
1058  * find a different hw address to give @slave, that isn't in use by any other
1059  * slave in the bond. This address must be, of course, one of the premanent
1060  * addresses of the other slaves.
1061  *
1062  * We go over the slave list, and for each slave there we compare its
1063  * permanent hw address with the current address of all the other slaves.
1064  * If no match was found, then we've found a slave with a permanent address
1065  * that isn't used by any other slave in the bond, so we can assign it to
1066  * @slave.
1067  *
1068  * assumption: this function is called before @slave is attached to the
1069  *             bond slave list.
1070  *
1071  * caller must hold the bond lock for write since the mac addresses are compared
1072  * and may be swapped.
1073  */
1074 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1075 {
1076         struct slave *tmp_slave1, *tmp_slave2, *free_mac_slave;
1077         struct slave *has_bond_addr = bond->curr_active_slave;
1078         int i, j, found = 0;
1079
1080         if (bond->slave_cnt == 0) {
1081                 /* this is the first slave */
1082                 return 0;
1083         }
1084
1085         /* if slave's mac address differs from bond's mac address
1086          * check uniqueness of slave's mac address against the other
1087          * slaves in the bond.
1088          */
1089         if (memcmp(slave->perm_hwaddr, bond->dev->dev_addr, ETH_ALEN)) {
1090                 bond_for_each_slave(bond, tmp_slave1, i) {
1091                         if (!memcmp(tmp_slave1->dev->dev_addr, slave->dev->dev_addr,
1092                                     ETH_ALEN)) {
1093                                 found = 1;
1094                                 break;
1095                         }
1096                 }
1097
1098                 if (!found)
1099                         return 0;
1100
1101                 /* Try setting slave mac to bond address and fall-through
1102                    to code handling that situation below... */
1103                 alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1104                                        bond->alb_info.rlb_enabled);
1105         }
1106
1107         /* The slave's address is equal to the address of the bond.
1108          * Search for a spare address in the bond for this slave.
1109          */
1110         free_mac_slave = NULL;
1111
1112         bond_for_each_slave(bond, tmp_slave1, i) {
1113                 found = 0;
1114                 bond_for_each_slave(bond, tmp_slave2, j) {
1115                         if (!memcmp(tmp_slave1->perm_hwaddr,
1116                                     tmp_slave2->dev->dev_addr,
1117                                     ETH_ALEN)) {
1118                                 found = 1;
1119                                 break;
1120                         }
1121                 }
1122
1123                 if (!found) {
1124                         /* no slave has tmp_slave1's perm addr
1125                          * as its curr addr
1126                          */
1127                         free_mac_slave = tmp_slave1;
1128                         break;
1129                 }
1130
1131                 if (!has_bond_addr) {
1132                         if (!memcmp(tmp_slave1->dev->dev_addr,
1133                                     bond->dev->dev_addr,
1134                                     ETH_ALEN)) {
1135
1136                                 has_bond_addr = tmp_slave1;
1137                         }
1138                 }
1139         }
1140
1141         if (free_mac_slave) {
1142                 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1143                                        bond->alb_info.rlb_enabled);
1144
1145                 printk(KERN_WARNING DRV_NAME
1146                        ": %s: Warning: the hw address of slave %s is in use by "
1147                        "the bond; giving it the hw address of %s\n",
1148                        bond->dev->name, slave->dev->name, free_mac_slave->dev->name);
1149
1150         } else if (has_bond_addr) {
1151                 printk(KERN_ERR DRV_NAME
1152                        ": %s: Error: the hw address of slave %s is in use by the "
1153                        "bond; couldn't find a slave with a free hw address to "
1154                        "give it (this should not have happened)\n",
1155                        bond->dev->name, slave->dev->name);
1156                 return -EFAULT;
1157         }
1158
1159         return 0;
1160 }
1161
1162 /**
1163  * alb_set_mac_address
1164  * @bond:
1165  * @addr:
1166  *
1167  * In TLB mode all slaves are configured to the bond's hw address, but set
1168  * their dev_addr field to different addresses (based on their permanent hw
1169  * addresses).
1170  *
1171  * For each slave, this function sets the interface to the new address and then
1172  * changes its dev_addr field to its previous value.
1173  *
1174  * Unwinding assumes bond's mac address has not yet changed.
1175  */
1176 static int alb_set_mac_address(struct bonding *bond, void *addr)
1177 {
1178         struct sockaddr sa;
1179         struct slave *slave, *stop_at;
1180         char tmp_addr[ETH_ALEN];
1181         int res;
1182         int i;
1183
1184         if (bond->alb_info.rlb_enabled) {
1185                 return 0;
1186         }
1187
1188         bond_for_each_slave(bond, slave, i) {
1189                 if (slave->dev->set_mac_address == NULL) {
1190                         res = -EOPNOTSUPP;
1191                         goto unwind;
1192                 }
1193
1194                 /* save net_device's current hw address */
1195                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1196
1197                 res = dev_set_mac_address(slave->dev, addr);
1198
1199                 /* restore net_device's hw address */
1200                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1201
1202                 if (res) {
1203                         goto unwind;
1204                 }
1205         }
1206
1207         return 0;
1208
1209 unwind:
1210         memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1211         sa.sa_family = bond->dev->type;
1212
1213         /* unwind from head to the slave that failed */
1214         stop_at = slave;
1215         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
1216                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1217                 dev_set_mac_address(slave->dev, &sa);
1218                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1219         }
1220
1221         return res;
1222 }
1223
1224 /************************ exported alb funcions ************************/
1225
1226 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1227 {
1228         int res;
1229
1230         res = tlb_initialize(bond);
1231         if (res) {
1232                 return res;
1233         }
1234
1235         if (rlb_enabled) {
1236                 bond->alb_info.rlb_enabled = 1;
1237                 /* initialize rlb */
1238                 res = rlb_initialize(bond);
1239                 if (res) {
1240                         tlb_deinitialize(bond);
1241                         return res;
1242                 }
1243         } else {
1244                 bond->alb_info.rlb_enabled = 0;
1245         }
1246
1247         return 0;
1248 }
1249
1250 void bond_alb_deinitialize(struct bonding *bond)
1251 {
1252         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1253
1254         tlb_deinitialize(bond);
1255
1256         if (bond_info->rlb_enabled) {
1257                 rlb_deinitialize(bond);
1258         }
1259 }
1260
1261 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1262 {
1263         struct bonding *bond = bond_dev->priv;
1264         struct ethhdr *eth_data;
1265         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1266         struct slave *tx_slave = NULL;
1267         static const u32 ip_bcast = 0xffffffff;
1268         int hash_size = 0;
1269         int do_tx_balance = 1;
1270         u32 hash_index = 0;
1271         const u8 *hash_start = NULL;
1272         int res = 1;
1273
1274         skb_reset_mac_header(skb);
1275         eth_data = eth_hdr(skb);
1276
1277         /* make sure that the curr_active_slave and the slaves list do
1278          * not change during tx
1279          */
1280         read_lock(&bond->lock);
1281         read_lock(&bond->curr_slave_lock);
1282
1283         if (!BOND_IS_OK(bond)) {
1284                 goto out;
1285         }
1286
1287         switch (ntohs(skb->protocol)) {
1288         case ETH_P_IP: {
1289                 const struct iphdr *iph = ip_hdr(skb);
1290
1291                 if ((memcmp(eth_data->h_dest, mac_bcast, ETH_ALEN) == 0) ||
1292                     (iph->daddr == ip_bcast) ||
1293                     (iph->protocol == IPPROTO_IGMP)) {
1294                         do_tx_balance = 0;
1295                         break;
1296                 }
1297                 hash_start = (char *)&(iph->daddr);
1298                 hash_size = sizeof(iph->daddr);
1299         }
1300                 break;
1301         case ETH_P_IPV6:
1302                 if (memcmp(eth_data->h_dest, mac_bcast, ETH_ALEN) == 0) {
1303                         do_tx_balance = 0;
1304                         break;
1305                 }
1306
1307                 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1308                 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1309                 break;
1310         case ETH_P_IPX:
1311                 if (ipx_hdr(skb)->ipx_checksum !=
1312                     __constant_htons(IPX_NO_CHECKSUM)) {
1313                         /* something is wrong with this packet */
1314                         do_tx_balance = 0;
1315                         break;
1316                 }
1317
1318                 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1319                         /* The only protocol worth balancing in
1320                          * this family since it has an "ARP" like
1321                          * mechanism
1322                          */
1323                         do_tx_balance = 0;
1324                         break;
1325                 }
1326
1327                 hash_start = (char*)eth_data->h_dest;
1328                 hash_size = ETH_ALEN;
1329                 break;
1330         case ETH_P_ARP:
1331                 do_tx_balance = 0;
1332                 if (bond_info->rlb_enabled) {
1333                         tx_slave = rlb_arp_xmit(skb, bond);
1334                 }
1335                 break;
1336         default:
1337                 do_tx_balance = 0;
1338                 break;
1339         }
1340
1341         if (do_tx_balance) {
1342                 hash_index = _simple_hash(hash_start, hash_size);
1343                 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1344         }
1345
1346         if (!tx_slave) {
1347                 /* unbalanced or unassigned, send through primary */
1348                 tx_slave = bond->curr_active_slave;
1349                 bond_info->unbalanced_load += skb->len;
1350         }
1351
1352         if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1353                 if (tx_slave != bond->curr_active_slave) {
1354                         memcpy(eth_data->h_source,
1355                                tx_slave->dev->dev_addr,
1356                                ETH_ALEN);
1357                 }
1358
1359                 res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1360         } else {
1361                 if (tx_slave) {
1362                         tlb_clear_slave(bond, tx_slave, 0);
1363                 }
1364         }
1365
1366 out:
1367         if (res) {
1368                 /* no suitable interface, frame not sent */
1369                 dev_kfree_skb(skb);
1370         }
1371         read_unlock(&bond->curr_slave_lock);
1372         read_unlock(&bond->lock);
1373         return 0;
1374 }
1375
1376 void bond_alb_monitor(struct bonding *bond)
1377 {
1378         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1379         struct slave *slave;
1380         int i;
1381
1382         read_lock(&bond->lock);
1383
1384         if (bond->kill_timers) {
1385                 goto out;
1386         }
1387
1388         if (bond->slave_cnt == 0) {
1389                 bond_info->tx_rebalance_counter = 0;
1390                 bond_info->lp_counter = 0;
1391                 goto re_arm;
1392         }
1393
1394         bond_info->tx_rebalance_counter++;
1395         bond_info->lp_counter++;
1396
1397         /* send learning packets */
1398         if (bond_info->lp_counter >= BOND_ALB_LP_TICKS) {
1399                 /* change of curr_active_slave involves swapping of mac addresses.
1400                  * in order to avoid this swapping from happening while
1401                  * sending the learning packets, the curr_slave_lock must be held for
1402                  * read.
1403                  */
1404                 read_lock(&bond->curr_slave_lock);
1405
1406                 bond_for_each_slave(bond, slave, i) {
1407                         alb_send_learning_packets(slave, slave->dev->dev_addr);
1408                 }
1409
1410                 read_unlock(&bond->curr_slave_lock);
1411
1412                 bond_info->lp_counter = 0;
1413         }
1414
1415         /* rebalance tx traffic */
1416         if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1417
1418                 read_lock(&bond->curr_slave_lock);
1419
1420                 bond_for_each_slave(bond, slave, i) {
1421                         tlb_clear_slave(bond, slave, 1);
1422                         if (slave == bond->curr_active_slave) {
1423                                 SLAVE_TLB_INFO(slave).load =
1424                                         bond_info->unbalanced_load /
1425                                                 BOND_TLB_REBALANCE_INTERVAL;
1426                                 bond_info->unbalanced_load = 0;
1427                         }
1428                 }
1429
1430                 read_unlock(&bond->curr_slave_lock);
1431
1432                 bond_info->tx_rebalance_counter = 0;
1433         }
1434
1435         /* handle rlb stuff */
1436         if (bond_info->rlb_enabled) {
1437                 /* the following code changes the promiscuity of the
1438                  * the curr_active_slave. It needs to be locked with a
1439                  * write lock to protect from other code that also
1440                  * sets the promiscuity.
1441                  */
1442                 write_lock_bh(&bond->curr_slave_lock);
1443
1444                 if (bond_info->primary_is_promisc &&
1445                     (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1446
1447                         bond_info->rlb_promisc_timeout_counter = 0;
1448
1449                         /* If the primary was set to promiscuous mode
1450                          * because a slave was disabled then
1451                          * it can now leave promiscuous mode.
1452                          */
1453                         dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1454                         bond_info->primary_is_promisc = 0;
1455                 }
1456
1457                 write_unlock_bh(&bond->curr_slave_lock);
1458
1459                 if (bond_info->rlb_rebalance) {
1460                         bond_info->rlb_rebalance = 0;
1461                         rlb_rebalance(bond);
1462                 }
1463
1464                 /* check if clients need updating */
1465                 if (bond_info->rx_ntt) {
1466                         if (bond_info->rlb_update_delay_counter) {
1467                                 --bond_info->rlb_update_delay_counter;
1468                         } else {
1469                                 rlb_update_rx_clients(bond);
1470                                 if (bond_info->rlb_update_retry_counter) {
1471                                         --bond_info->rlb_update_retry_counter;
1472                                 } else {
1473                                         bond_info->rx_ntt = 0;
1474                                 }
1475                         }
1476                 }
1477         }
1478
1479 re_arm:
1480         mod_timer(&(bond_info->alb_timer), jiffies + alb_delta_in_ticks);
1481 out:
1482         read_unlock(&bond->lock);
1483 }
1484
1485 /* assumption: called before the slave is attached to the bond
1486  * and not locked by the bond lock
1487  */
1488 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1489 {
1490         int res;
1491
1492         res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1493                                      bond->alb_info.rlb_enabled);
1494         if (res) {
1495                 return res;
1496         }
1497
1498         /* caller must hold the bond lock for write since the mac addresses
1499          * are compared and may be swapped.
1500          */
1501         write_lock_bh(&bond->lock);
1502
1503         res = alb_handle_addr_collision_on_attach(bond, slave);
1504
1505         write_unlock_bh(&bond->lock);
1506
1507         if (res) {
1508                 return res;
1509         }
1510
1511         tlb_init_slave(slave);
1512
1513         /* order a rebalance ASAP */
1514         bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1515
1516         if (bond->alb_info.rlb_enabled) {
1517                 bond->alb_info.rlb_rebalance = 1;
1518         }
1519
1520         return 0;
1521 }
1522
1523 /* Caller must hold bond lock for write */
1524 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1525 {
1526         if (bond->slave_cnt > 1) {
1527                 alb_change_hw_addr_on_detach(bond, slave);
1528         }
1529
1530         tlb_clear_slave(bond, slave, 0);
1531
1532         if (bond->alb_info.rlb_enabled) {
1533                 bond->alb_info.next_rx_slave = NULL;
1534                 rlb_clear_slave(bond, slave);
1535         }
1536 }
1537
1538 /* Caller must hold bond lock for read */
1539 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1540 {
1541         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1542
1543         if (link == BOND_LINK_DOWN) {
1544                 tlb_clear_slave(bond, slave, 0);
1545                 if (bond->alb_info.rlb_enabled) {
1546                         rlb_clear_slave(bond, slave);
1547                 }
1548         } else if (link == BOND_LINK_UP) {
1549                 /* order a rebalance ASAP */
1550                 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1551                 if (bond->alb_info.rlb_enabled) {
1552                         bond->alb_info.rlb_rebalance = 1;
1553                         /* If the updelay module parameter is smaller than the
1554                          * forwarding delay of the switch the rebalance will
1555                          * not work because the rebalance arp replies will
1556                          * not be forwarded to the clients..
1557                          */
1558                 }
1559         }
1560 }
1561
1562 /**
1563  * bond_alb_handle_active_change - assign new curr_active_slave
1564  * @bond: our bonding struct
1565  * @new_slave: new slave to assign
1566  *
1567  * Set the bond->curr_active_slave to @new_slave and handle
1568  * mac address swapping and promiscuity changes as needed.
1569  *
1570  * Caller must hold bond curr_slave_lock for write (or bond lock for write)
1571  */
1572 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1573 {
1574         struct slave *swap_slave;
1575         int i;
1576
1577         if (bond->curr_active_slave == new_slave) {
1578                 return;
1579         }
1580
1581         if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1582                 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1583                 bond->alb_info.primary_is_promisc = 0;
1584                 bond->alb_info.rlb_promisc_timeout_counter = 0;
1585         }
1586
1587         swap_slave = bond->curr_active_slave;
1588         bond->curr_active_slave = new_slave;
1589
1590         if (!new_slave || (bond->slave_cnt == 0)) {
1591                 return;
1592         }
1593
1594         /* set the new curr_active_slave to the bonds mac address
1595          * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1596          */
1597         if (!swap_slave) {
1598                 struct slave *tmp_slave;
1599                 /* find slave that is holding the bond's mac address */
1600                 bond_for_each_slave(bond, tmp_slave, i) {
1601                         if (!memcmp(tmp_slave->dev->dev_addr,
1602                                     bond->dev->dev_addr, ETH_ALEN)) {
1603                                 swap_slave = tmp_slave;
1604                                 break;
1605                         }
1606                 }
1607         }
1608
1609         /* curr_active_slave must be set before calling alb_swap_mac_addr */
1610         if (swap_slave) {
1611                 /* swap mac address */
1612                 alb_swap_mac_addr(bond, swap_slave, new_slave);
1613         } else {
1614                 /* set the new_slave to the bond mac address */
1615                 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1616                                        bond->alb_info.rlb_enabled);
1617                 /* fasten bond mac on new current slave */
1618                 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1619         }
1620 }
1621
1622 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1623 {
1624         struct bonding *bond = bond_dev->priv;
1625         struct sockaddr *sa = addr;
1626         struct slave *slave, *swap_slave;
1627         int res;
1628         int i;
1629
1630         if (!is_valid_ether_addr(sa->sa_data)) {
1631                 return -EADDRNOTAVAIL;
1632         }
1633
1634         res = alb_set_mac_address(bond, addr);
1635         if (res) {
1636                 return res;
1637         }
1638
1639         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1640
1641         /* If there is no curr_active_slave there is nothing else to do.
1642          * Otherwise we'll need to pass the new address to it and handle
1643          * duplications.
1644          */
1645         if (!bond->curr_active_slave) {
1646                 return 0;
1647         }
1648
1649         swap_slave = NULL;
1650
1651         bond_for_each_slave(bond, slave, i) {
1652                 if (!memcmp(slave->dev->dev_addr, bond_dev->dev_addr, ETH_ALEN)) {
1653                         swap_slave = slave;
1654                         break;
1655                 }
1656         }
1657
1658         if (swap_slave) {
1659                 alb_swap_mac_addr(bond, swap_slave, bond->curr_active_slave);
1660         } else {
1661                 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr,
1662                                        bond->alb_info.rlb_enabled);
1663
1664                 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1665                 if (bond->alb_info.rlb_enabled) {
1666                         /* inform clients mac address has changed */
1667                         rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1668                 }
1669         }
1670
1671         return 0;
1672 }
1673
1674 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1675 {
1676         if (bond->alb_info.current_alb_vlan &&
1677             (bond->alb_info.current_alb_vlan->vlan_id == vlan_id)) {
1678                 bond->alb_info.current_alb_vlan = NULL;
1679         }
1680
1681         if (bond->alb_info.rlb_enabled) {
1682                 rlb_clear_vlan(bond, vlan_id);
1683         }
1684 }
1685