linux/drivers/net/bonding/bond_alb.c
<<
>>
Prefs
   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#include <linux/skbuff.h>
  24#include <linux/netdevice.h>
  25#include <linux/etherdevice.h>
  26#include <linux/pkt_sched.h>
  27#include <linux/spinlock.h>
  28#include <linux/slab.h>
  29#include <linux/timer.h>
  30#include <linux/ip.h>
  31#include <linux/ipv6.h>
  32#include <linux/if_arp.h>
  33#include <linux/if_ether.h>
  34#include <linux/if_bonding.h>
  35#include <linux/if_vlan.h>
  36#include <linux/in.h>
  37#include <net/ipx.h>
  38#include <net/arp.h>
  39#include <net/ipv6.h>
  40#include <asm/byteorder.h>
  41#include "bonding.h"
  42#include "bond_alb.h"
  43
  44
  45#define ALB_TIMER_TICKS_PER_SEC     10  /* should be a divisor of HZ */
  46#define BOND_TLB_REBALANCE_INTERVAL 10  /* In seconds, periodic re-balancing.
  47                                         * Used for division - never set
  48                                         * to zero !!!
  49                                         */
  50#define BOND_ALB_LP_INTERVAL        1   /* In seconds, periodic send of
  51                                         * learning packets to the switch
  52                                         */
  53
  54#define BOND_TLB_REBALANCE_TICKS (BOND_TLB_REBALANCE_INTERVAL \
  55                                  * ALB_TIMER_TICKS_PER_SEC)
  56
  57#define BOND_ALB_LP_TICKS (BOND_ALB_LP_INTERVAL \
  58                           * ALB_TIMER_TICKS_PER_SEC)
  59
  60#define TLB_HASH_TABLE_SIZE 256 /* The size of the clients hash table.
  61                                 * Note that this value MUST NOT be smaller
  62                                 * because the key hash table is BYTE wide !
  63                                 */
  64
  65
  66#define TLB_NULL_INDEX          0xffffffff
  67#define MAX_LP_BURST            3
  68
  69/* rlb defs */
  70#define RLB_HASH_TABLE_SIZE     256
  71#define RLB_NULL_INDEX          0xffffffff
  72#define RLB_UPDATE_DELAY        2*ALB_TIMER_TICKS_PER_SEC /* 2 seconds */
  73#define RLB_ARP_BURST_SIZE      2
  74#define RLB_UPDATE_RETRY        3       /* 3-ticks - must be smaller than the rlb
  75                                         * rebalance interval (5 min).
  76                                         */
  77/* RLB_PROMISC_TIMEOUT = 10 sec equals the time that the current slave is
  78 * promiscuous after failover
  79 */
  80#define RLB_PROMISC_TIMEOUT     10*ALB_TIMER_TICKS_PER_SEC
  81
  82#ifndef __long_aligned
  83#define __long_aligned __attribute__((aligned((sizeof(long)))))
  84#endif
  85static const u8 mac_bcast[ETH_ALEN] __long_aligned = {
  86        0xff, 0xff, 0xff, 0xff, 0xff, 0xff
  87};
  88static const u8 mac_v6_allmcast[ETH_ALEN] __long_aligned = {
  89        0x33, 0x33, 0x00, 0x00, 0x00, 0x01
  90};
  91static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
  92
  93#pragma pack(1)
  94struct learning_pkt {
  95        u8 mac_dst[ETH_ALEN];
  96        u8 mac_src[ETH_ALEN];
  97        __be16 type;
  98        u8 padding[ETH_ZLEN - ETH_HLEN];
  99};
 100
 101struct arp_pkt {
 102        __be16  hw_addr_space;
 103        __be16  prot_addr_space;
 104        u8      hw_addr_len;
 105        u8      prot_addr_len;
 106        __be16  op_code;
 107        u8      mac_src[ETH_ALEN];      /* sender hardware address */
 108        __be32  ip_src;                 /* sender IP address */
 109        u8      mac_dst[ETH_ALEN];      /* target hardware address */
 110        __be32  ip_dst;                 /* target IP address */
 111};
 112#pragma pack()
 113
 114static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb)
 115{
 116        return (struct arp_pkt *)skb_network_header(skb);
 117}
 118
 119/* Forward declaration */
 120static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[]);
 121
 122static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
 123{
 124        int i;
 125        u8 hash = 0;
 126
 127        for (i = 0; i < hash_size; i++) {
 128                hash ^= hash_start[i];
 129        }
 130
 131        return hash;
 132}
 133
 134/*********************** tlb specific functions ***************************/
 135
 136static inline void _lock_tx_hashtbl(struct bonding *bond)
 137{
 138        spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 139}
 140
 141static inline void _unlock_tx_hashtbl(struct bonding *bond)
 142{
 143        spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 144}
 145
 146/* Caller must hold tx_hashtbl lock */
 147static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
 148{
 149        if (save_load) {
 150                entry->load_history = 1 + entry->tx_bytes /
 151                                      BOND_TLB_REBALANCE_INTERVAL;
 152                entry->tx_bytes = 0;
 153        }
 154
 155        entry->tx_slave = NULL;
 156        entry->next = TLB_NULL_INDEX;
 157        entry->prev = TLB_NULL_INDEX;
 158}
 159
 160static inline void tlb_init_slave(struct slave *slave)
 161{
 162        SLAVE_TLB_INFO(slave).load = 0;
 163        SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
 164}
 165
 166/* Caller must hold bond lock for read */
 167static void tlb_clear_slave(struct bonding *bond, struct slave *slave, int save_load)
 168{
 169        struct tlb_client_info *tx_hash_table;
 170        u32 index;
 171
 172        _lock_tx_hashtbl(bond);
 173
 174        /* clear slave from tx_hashtbl */
 175        tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
 176
 177        /* skip this if we've already freed the tx hash table */
 178        if (tx_hash_table) {
 179                index = SLAVE_TLB_INFO(slave).head;
 180                while (index != TLB_NULL_INDEX) {
 181                        u32 next_index = tx_hash_table[index].next;
 182                        tlb_init_table_entry(&tx_hash_table[index], save_load);
 183                        index = next_index;
 184                }
 185        }
 186
 187        tlb_init_slave(slave);
 188
 189        _unlock_tx_hashtbl(bond);
 190}
 191
 192/* Must be called before starting the monitor timer */
 193static int tlb_initialize(struct bonding *bond)
 194{
 195        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 196        int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 197        struct tlb_client_info *new_hashtbl;
 198        int i;
 199
 200        spin_lock_init(&(bond_info->tx_hashtbl_lock));
 201
 202        new_hashtbl = kzalloc(size, GFP_KERNEL);
 203        if (!new_hashtbl) {
 204                pr_err(DRV_NAME
 205                       ": %s: Error: Failed to allocate TLB hash table\n",
 206                       bond->dev->name);
 207                return -1;
 208        }
 209        _lock_tx_hashtbl(bond);
 210
 211        bond_info->tx_hashtbl = new_hashtbl;
 212
 213        for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
 214                tlb_init_table_entry(&bond_info->tx_hashtbl[i], 1);
 215        }
 216
 217        _unlock_tx_hashtbl(bond);
 218
 219        return 0;
 220}
 221
 222/* Must be called only after all slaves have been released */
 223static void tlb_deinitialize(struct bonding *bond)
 224{
 225        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 226
 227        _lock_tx_hashtbl(bond);
 228
 229        kfree(bond_info->tx_hashtbl);
 230        bond_info->tx_hashtbl = NULL;
 231
 232        _unlock_tx_hashtbl(bond);
 233}
 234
 235/* Caller must hold bond lock for read */
 236static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
 237{
 238        struct slave *slave, *least_loaded;
 239        s64 max_gap;
 240        int i, found = 0;
 241
 242        /* Find the first enabled slave */
 243        bond_for_each_slave(bond, slave, i) {
 244                if (SLAVE_IS_OK(slave)) {
 245                        found = 1;
 246                        break;
 247                }
 248        }
 249
 250        if (!found) {
 251                return NULL;
 252        }
 253
 254        least_loaded = slave;
 255        max_gap = (s64)(slave->speed << 20) - /* Convert to Megabit per sec */
 256                        (s64)(SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
 257
 258        /* Find the slave with the largest gap */
 259        bond_for_each_slave_from(bond, slave, i, least_loaded) {
 260                if (SLAVE_IS_OK(slave)) {
 261                        s64 gap = (s64)(slave->speed << 20) -
 262                                        (s64)(SLAVE_TLB_INFO(slave).load << 3);
 263                        if (max_gap < gap) {
 264                                least_loaded = slave;
 265                                max_gap = gap;
 266                        }
 267                }
 268        }
 269
 270        return least_loaded;
 271}
 272
 273/* Caller must hold bond lock for read */
 274static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index, u32 skb_len)
 275{
 276        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 277        struct tlb_client_info *hash_table;
 278        struct slave *assigned_slave;
 279
 280        _lock_tx_hashtbl(bond);
 281
 282        hash_table = bond_info->tx_hashtbl;
 283        assigned_slave = hash_table[hash_index].tx_slave;
 284        if (!assigned_slave) {
 285                assigned_slave = tlb_get_least_loaded_slave(bond);
 286
 287                if (assigned_slave) {
 288                        struct tlb_slave_info *slave_info =
 289                                &(SLAVE_TLB_INFO(assigned_slave));
 290                        u32 next_index = slave_info->head;
 291
 292                        hash_table[hash_index].tx_slave = assigned_slave;
 293                        hash_table[hash_index].next = next_index;
 294                        hash_table[hash_index].prev = TLB_NULL_INDEX;
 295
 296                        if (next_index != TLB_NULL_INDEX) {
 297                                hash_table[next_index].prev = hash_index;
 298                        }
 299
 300                        slave_info->head = hash_index;
 301                        slave_info->load +=
 302                                hash_table[hash_index].load_history;
 303                }
 304        }
 305
 306        if (assigned_slave) {
 307                hash_table[hash_index].tx_bytes += skb_len;
 308        }
 309
 310        _unlock_tx_hashtbl(bond);
 311
 312        return assigned_slave;
 313}
 314
 315/*********************** rlb specific functions ***************************/
 316static inline void _lock_rx_hashtbl(struct bonding *bond)
 317{
 318        spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 319}
 320
 321static inline void _unlock_rx_hashtbl(struct bonding *bond)
 322{
 323        spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 324}
 325
 326/* when an ARP REPLY is received from a client update its info
 327 * in the rx_hashtbl
 328 */
 329static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 330{
 331        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 332        struct rlb_client_info *client_info;
 333        u32 hash_index;
 334
 335        _lock_rx_hashtbl(bond);
 336
 337        hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
 338        client_info = &(bond_info->rx_hashtbl[hash_index]);
 339
 340        if ((client_info->assigned) &&
 341            (client_info->ip_src == arp->ip_dst) &&
 342            (client_info->ip_dst == arp->ip_src)) {
 343                /* update the clients MAC address */
 344                memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
 345                client_info->ntt = 1;
 346                bond_info->rx_ntt = 1;
 347        }
 348
 349        _unlock_rx_hashtbl(bond);
 350}
 351
 352static int rlb_arp_recv(struct sk_buff *skb, struct net_device *bond_dev, struct packet_type *ptype, struct net_device *orig_dev)
 353{
 354        struct bonding *bond;
 355        struct arp_pkt *arp = (struct arp_pkt *)skb->data;
 356        int res = NET_RX_DROP;
 357
 358        if (dev_net(bond_dev) != &init_net)
 359                goto out;
 360
 361        while (bond_dev->priv_flags & IFF_802_1Q_VLAN)
 362                bond_dev = vlan_dev_real_dev(bond_dev);
 363
 364        if (!(bond_dev->priv_flags & IFF_BONDING) ||
 365            !(bond_dev->flags & IFF_MASTER))
 366                goto out;
 367
 368        if (!arp) {
 369                pr_debug("Packet has no ARP data\n");
 370                goto out;
 371        }
 372
 373        if (skb->len < sizeof(struct arp_pkt)) {
 374                pr_debug("Packet is too small to be an ARP\n");
 375                goto out;
 376        }
 377
 378        if (arp->op_code == htons(ARPOP_REPLY)) {
 379                /* update rx hash table for this ARP */
 380                bond = netdev_priv(bond_dev);
 381                rlb_update_entry_from_arp(bond, arp);
 382                pr_debug("Server received an ARP Reply from client\n");
 383        }
 384
 385        res = NET_RX_SUCCESS;
 386
 387out:
 388        dev_kfree_skb(skb);
 389
 390        return res;
 391}
 392
 393/* Caller must hold bond lock for read */
 394static struct slave *rlb_next_rx_slave(struct bonding *bond)
 395{
 396        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 397        struct slave *rx_slave, *slave, *start_at;
 398        int i = 0;
 399
 400        if (bond_info->next_rx_slave) {
 401                start_at = bond_info->next_rx_slave;
 402        } else {
 403                start_at = bond->first_slave;
 404        }
 405
 406        rx_slave = NULL;
 407
 408        bond_for_each_slave_from(bond, slave, i, start_at) {
 409                if (SLAVE_IS_OK(slave)) {
 410                        if (!rx_slave) {
 411                                rx_slave = slave;
 412                        } else if (slave->speed > rx_slave->speed) {
 413                                rx_slave = slave;
 414                        }
 415                }
 416        }
 417
 418        if (rx_slave) {
 419                bond_info->next_rx_slave = rx_slave->next;
 420        }
 421
 422        return rx_slave;
 423}
 424
 425/* teach the switch the mac of a disabled slave
 426 * on the primary for fault tolerance
 427 *
 428 * Caller must hold bond->curr_slave_lock for write or bond lock for write
 429 */
 430static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
 431{
 432        if (!bond->curr_active_slave) {
 433                return;
 434        }
 435
 436        if (!bond->alb_info.primary_is_promisc) {
 437                if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1))
 438                        bond->alb_info.primary_is_promisc = 1;
 439                else
 440                        bond->alb_info.primary_is_promisc = 0;
 441        }
 442
 443        bond->alb_info.rlb_promisc_timeout_counter = 0;
 444
 445        alb_send_learning_packets(bond->curr_active_slave, addr);
 446}
 447
 448/* slave being removed should not be active at this point
 449 *
 450 * Caller must hold bond lock for read
 451 */
 452static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 453{
 454        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 455        struct rlb_client_info *rx_hash_table;
 456        u32 index, next_index;
 457
 458        /* clear slave from rx_hashtbl */
 459        _lock_rx_hashtbl(bond);
 460
 461        rx_hash_table = bond_info->rx_hashtbl;
 462        index = bond_info->rx_hashtbl_head;
 463        for (; index != RLB_NULL_INDEX; index = next_index) {
 464                next_index = rx_hash_table[index].next;
 465                if (rx_hash_table[index].slave == slave) {
 466                        struct slave *assigned_slave = rlb_next_rx_slave(bond);
 467
 468                        if (assigned_slave) {
 469                                rx_hash_table[index].slave = assigned_slave;
 470                                if (compare_ether_addr_64bits(rx_hash_table[index].mac_dst,
 471                                                              mac_bcast)) {
 472                                        bond_info->rx_hashtbl[index].ntt = 1;
 473                                        bond_info->rx_ntt = 1;
 474                                        /* A slave has been removed from the
 475                                         * table because it is either disabled
 476                                         * or being released. We must retry the
 477                                         * update to avoid clients from not
 478                                         * being updated & disconnecting when
 479                                         * there is stress
 480                                         */
 481                                        bond_info->rlb_update_retry_counter =
 482                                                RLB_UPDATE_RETRY;
 483                                }
 484                        } else {  /* there is no active slave */
 485                                rx_hash_table[index].slave = NULL;
 486                        }
 487                }
 488        }
 489
 490        _unlock_rx_hashtbl(bond);
 491
 492        write_lock_bh(&bond->curr_slave_lock);
 493
 494        if (slave != bond->curr_active_slave) {
 495                rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 496        }
 497
 498        write_unlock_bh(&bond->curr_slave_lock);
 499}
 500
 501static void rlb_update_client(struct rlb_client_info *client_info)
 502{
 503        int i;
 504
 505        if (!client_info->slave) {
 506                return;
 507        }
 508
 509        for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 510                struct sk_buff *skb;
 511
 512                skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 513                                 client_info->ip_dst,
 514                                 client_info->slave->dev,
 515                                 client_info->ip_src,
 516                                 client_info->mac_dst,
 517                                 client_info->slave->dev->dev_addr,
 518                                 client_info->mac_dst);
 519                if (!skb) {
 520                        pr_err(DRV_NAME
 521                               ": %s: Error: failed to create an ARP packet\n",
 522                               client_info->slave->dev->master->name);
 523                        continue;
 524                }
 525
 526                skb->dev = client_info->slave->dev;
 527
 528                if (client_info->tag) {
 529                        skb = vlan_put_tag(skb, client_info->vlan_id);
 530                        if (!skb) {
 531                                pr_err(DRV_NAME
 532                                       ": %s: Error: failed to insert VLAN tag\n",
 533                                       client_info->slave->dev->master->name);
 534                                continue;
 535                        }
 536                }
 537
 538                arp_xmit(skb);
 539        }
 540}
 541
 542/* sends ARP REPLIES that update the clients that need updating */
 543static void rlb_update_rx_clients(struct bonding *bond)
 544{
 545        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 546        struct rlb_client_info *client_info;
 547        u32 hash_index;
 548
 549        _lock_rx_hashtbl(bond);
 550
 551        hash_index = bond_info->rx_hashtbl_head;
 552        for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 553                client_info = &(bond_info->rx_hashtbl[hash_index]);
 554                if (client_info->ntt) {
 555                        rlb_update_client(client_info);
 556                        if (bond_info->rlb_update_retry_counter == 0) {
 557                                client_info->ntt = 0;
 558                        }
 559                }
 560        }
 561
 562        /* do not update the entries again untill this counter is zero so that
 563         * not to confuse the clients.
 564         */
 565        bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 566
 567        _unlock_rx_hashtbl(bond);
 568}
 569
 570/* The slave was assigned a new mac address - update the clients */
 571static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 572{
 573        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 574        struct rlb_client_info *client_info;
 575        int ntt = 0;
 576        u32 hash_index;
 577
 578        _lock_rx_hashtbl(bond);
 579
 580        hash_index = bond_info->rx_hashtbl_head;
 581        for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 582                client_info = &(bond_info->rx_hashtbl[hash_index]);
 583
 584                if ((client_info->slave == slave) &&
 585                    compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
 586                        client_info->ntt = 1;
 587                        ntt = 1;
 588                }
 589        }
 590
 591        // update the team's flag only after the whole iteration
 592        if (ntt) {
 593                bond_info->rx_ntt = 1;
 594                //fasten the change
 595                bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 596        }
 597
 598        _unlock_rx_hashtbl(bond);
 599}
 600
 601/* mark all clients using src_ip to be updated */
 602static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
 603{
 604        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 605        struct rlb_client_info *client_info;
 606        u32 hash_index;
 607
 608        _lock_rx_hashtbl(bond);
 609
 610        hash_index = bond_info->rx_hashtbl_head;
 611        for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 612                client_info = &(bond_info->rx_hashtbl[hash_index]);
 613
 614                if (!client_info->slave) {
 615                        pr_err(DRV_NAME
 616                               ": %s: Error: found a client with no channel in "
 617                               "the client's hash table\n",
 618                               bond->dev->name);
 619                        continue;
 620                }
 621                /*update all clients using this src_ip, that are not assigned
 622                 * to the team's address (curr_active_slave) and have a known
 623                 * unicast mac address.
 624                 */
 625                if ((client_info->ip_src == src_ip) &&
 626                    compare_ether_addr_64bits(client_info->slave->dev->dev_addr,
 627                           bond->dev->dev_addr) &&
 628                    compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
 629                        client_info->ntt = 1;
 630                        bond_info->rx_ntt = 1;
 631                }
 632        }
 633
 634        _unlock_rx_hashtbl(bond);
 635}
 636
 637/* Caller must hold both bond and ptr locks for read */
 638static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
 639{
 640        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 641        struct arp_pkt *arp = arp_pkt(skb);
 642        struct slave *assigned_slave;
 643        struct rlb_client_info *client_info;
 644        u32 hash_index = 0;
 645
 646        _lock_rx_hashtbl(bond);
 647
 648        hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_src));
 649        client_info = &(bond_info->rx_hashtbl[hash_index]);
 650
 651        if (client_info->assigned) {
 652                if ((client_info->ip_src == arp->ip_src) &&
 653                    (client_info->ip_dst == arp->ip_dst)) {
 654                        /* the entry is already assigned to this client */
 655                        if (compare_ether_addr_64bits(arp->mac_dst, mac_bcast)) {
 656                                /* update mac address from arp */
 657                                memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 658                        }
 659
 660                        assigned_slave = client_info->slave;
 661                        if (assigned_slave) {
 662                                _unlock_rx_hashtbl(bond);
 663                                return assigned_slave;
 664                        }
 665                } else {
 666                        /* the entry is already assigned to some other client,
 667                         * move the old client to primary (curr_active_slave) so
 668                         * that the new client can be assigned to this entry.
 669                         */
 670                        if (bond->curr_active_slave &&
 671                            client_info->slave != bond->curr_active_slave) {
 672                                client_info->slave = bond->curr_active_slave;
 673                                rlb_update_client(client_info);
 674                        }
 675                }
 676        }
 677        /* assign a new slave */
 678        assigned_slave = rlb_next_rx_slave(bond);
 679
 680        if (assigned_slave) {
 681                client_info->ip_src = arp->ip_src;
 682                client_info->ip_dst = arp->ip_dst;
 683                /* arp->mac_dst is broadcast for arp reqeusts.
 684                 * will be updated with clients actual unicast mac address
 685                 * upon receiving an arp reply.
 686                 */
 687                memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 688                client_info->slave = assigned_slave;
 689
 690                if (compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
 691                        client_info->ntt = 1;
 692                        bond->alb_info.rx_ntt = 1;
 693                } else {
 694                        client_info->ntt = 0;
 695                }
 696
 697                if (!list_empty(&bond->vlan_list)) {
 698                        if (!vlan_get_tag(skb, &client_info->vlan_id))
 699                                client_info->tag = 1;
 700                }
 701
 702                if (!client_info->assigned) {
 703                        u32 prev_tbl_head = bond_info->rx_hashtbl_head;
 704                        bond_info->rx_hashtbl_head = hash_index;
 705                        client_info->next = prev_tbl_head;
 706                        if (prev_tbl_head != RLB_NULL_INDEX) {
 707                                bond_info->rx_hashtbl[prev_tbl_head].prev =
 708                                        hash_index;
 709                        }
 710                        client_info->assigned = 1;
 711                }
 712        }
 713
 714        _unlock_rx_hashtbl(bond);
 715
 716        return assigned_slave;
 717}
 718
 719/* chooses (and returns) transmit channel for arp reply
 720 * does not choose channel for other arp types since they are
 721 * sent on the curr_active_slave
 722 */
 723static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 724{
 725        struct arp_pkt *arp = arp_pkt(skb);
 726        struct slave *tx_slave = NULL;
 727
 728        if (arp->op_code == htons(ARPOP_REPLY)) {
 729                /* the arp must be sent on the selected
 730                * rx channel
 731                */
 732                tx_slave = rlb_choose_channel(skb, bond);
 733                if (tx_slave) {
 734                        memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
 735                }
 736                pr_debug("Server sent ARP Reply packet\n");
 737        } else if (arp->op_code == htons(ARPOP_REQUEST)) {
 738                /* Create an entry in the rx_hashtbl for this client as a
 739                 * place holder.
 740                 * When the arp reply is received the entry will be updated
 741                 * with the correct unicast address of the client.
 742                 */
 743                rlb_choose_channel(skb, bond);
 744
 745                /* The ARP relpy packets must be delayed so that
 746                 * they can cancel out the influence of the ARP request.
 747                 */
 748                bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 749
 750                /* arp requests are broadcast and are sent on the primary
 751                 * the arp request will collapse all clients on the subnet to
 752                 * the primary slave. We must register these clients to be
 753                 * updated with their assigned mac.
 754                 */
 755                rlb_req_update_subnet_clients(bond, arp->ip_src);
 756                pr_debug("Server sent ARP Request packet\n");
 757        }
 758
 759        return tx_slave;
 760}
 761
 762/* Caller must hold bond lock for read */
 763static void rlb_rebalance(struct bonding *bond)
 764{
 765        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 766        struct slave *assigned_slave;
 767        struct rlb_client_info *client_info;
 768        int ntt;
 769        u32 hash_index;
 770
 771        _lock_rx_hashtbl(bond);
 772
 773        ntt = 0;
 774        hash_index = bond_info->rx_hashtbl_head;
 775        for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 776                client_info = &(bond_info->rx_hashtbl[hash_index]);
 777                assigned_slave = rlb_next_rx_slave(bond);
 778                if (assigned_slave && (client_info->slave != assigned_slave)) {
 779                        client_info->slave = assigned_slave;
 780                        client_info->ntt = 1;
 781                        ntt = 1;
 782                }
 783        }
 784
 785        /* update the team's flag only after the whole iteration */
 786        if (ntt) {
 787                bond_info->rx_ntt = 1;
 788        }
 789        _unlock_rx_hashtbl(bond);
 790}
 791
 792/* Caller must hold rx_hashtbl lock */
 793static void rlb_init_table_entry(struct rlb_client_info *entry)
 794{
 795        memset(entry, 0, sizeof(struct rlb_client_info));
 796        entry->next = RLB_NULL_INDEX;
 797        entry->prev = RLB_NULL_INDEX;
 798}
 799
 800static int rlb_initialize(struct bonding *bond)
 801{
 802        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 803        struct packet_type *pk_type = &(BOND_ALB_INFO(bond).rlb_pkt_type);
 804        struct rlb_client_info  *new_hashtbl;
 805        int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 806        int i;
 807
 808        spin_lock_init(&(bond_info->rx_hashtbl_lock));
 809
 810        new_hashtbl = kmalloc(size, GFP_KERNEL);
 811        if (!new_hashtbl) {
 812                pr_err(DRV_NAME
 813                       ": %s: Error: Failed to allocate RLB hash table\n",
 814                       bond->dev->name);
 815                return -1;
 816        }
 817        _lock_rx_hashtbl(bond);
 818
 819        bond_info->rx_hashtbl = new_hashtbl;
 820
 821        bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
 822
 823        for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
 824                rlb_init_table_entry(bond_info->rx_hashtbl + i);
 825        }
 826
 827        _unlock_rx_hashtbl(bond);
 828
 829        /*initialize packet type*/
 830        pk_type->type = cpu_to_be16(ETH_P_ARP);
 831        pk_type->dev = NULL;
 832        pk_type->func = rlb_arp_recv;
 833
 834        /* register to receive ARPs */
 835        dev_add_pack(pk_type);
 836
 837        return 0;
 838}
 839
 840static void rlb_deinitialize(struct bonding *bond)
 841{
 842        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 843
 844        dev_remove_pack(&(bond_info->rlb_pkt_type));
 845
 846        _lock_rx_hashtbl(bond);
 847
 848        kfree(bond_info->rx_hashtbl);
 849        bond_info->rx_hashtbl = NULL;
 850        bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
 851
 852        _unlock_rx_hashtbl(bond);
 853}
 854
 855static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 856{
 857        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 858        u32 curr_index;
 859
 860        _lock_rx_hashtbl(bond);
 861
 862        curr_index = bond_info->rx_hashtbl_head;
 863        while (curr_index != RLB_NULL_INDEX) {
 864                struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 865                u32 next_index = bond_info->rx_hashtbl[curr_index].next;
 866                u32 prev_index = bond_info->rx_hashtbl[curr_index].prev;
 867
 868                if (curr->tag && (curr->vlan_id == vlan_id)) {
 869                        if (curr_index == bond_info->rx_hashtbl_head) {
 870                                bond_info->rx_hashtbl_head = next_index;
 871                        }
 872                        if (prev_index != RLB_NULL_INDEX) {
 873                                bond_info->rx_hashtbl[prev_index].next = next_index;
 874                        }
 875                        if (next_index != RLB_NULL_INDEX) {
 876                                bond_info->rx_hashtbl[next_index].prev = prev_index;
 877                        }
 878
 879                        rlb_init_table_entry(curr);
 880                }
 881
 882                curr_index = next_index;
 883        }
 884
 885        _unlock_rx_hashtbl(bond);
 886}
 887
 888/*********************** tlb/rlb shared functions *********************/
 889
 890static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
 891{
 892        struct bonding *bond = bond_get_bond_by_slave(slave);
 893        struct learning_pkt pkt;
 894        int size = sizeof(struct learning_pkt);
 895        int i;
 896
 897        memset(&pkt, 0, size);
 898        memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
 899        memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
 900        pkt.type = cpu_to_be16(ETH_P_LOOP);
 901
 902        for (i = 0; i < MAX_LP_BURST; i++) {
 903                struct sk_buff *skb;
 904                char *data;
 905
 906                skb = dev_alloc_skb(size);
 907                if (!skb) {
 908                        return;
 909                }
 910
 911                data = skb_put(skb, size);
 912                memcpy(data, &pkt, size);
 913
 914                skb_reset_mac_header(skb);
 915                skb->network_header = skb->mac_header + ETH_HLEN;
 916                skb->protocol = pkt.type;
 917                skb->priority = TC_PRIO_CONTROL;
 918                skb->dev = slave->dev;
 919
 920                if (!list_empty(&bond->vlan_list)) {
 921                        struct vlan_entry *vlan;
 922
 923                        vlan = bond_next_vlan(bond,
 924                                              bond->alb_info.current_alb_vlan);
 925
 926                        bond->alb_info.current_alb_vlan = vlan;
 927                        if (!vlan) {
 928                                kfree_skb(skb);
 929                                continue;
 930                        }
 931
 932                        skb = vlan_put_tag(skb, vlan->vlan_id);
 933                        if (!skb) {
 934                                pr_err(DRV_NAME
 935                                       ": %s: Error: failed to insert VLAN tag\n",
 936                                       bond->dev->name);
 937                                continue;
 938                        }
 939                }
 940
 941                dev_queue_xmit(skb);
 942        }
 943}
 944
 945/* hw is a boolean parameter that determines whether we should try and
 946 * set the hw address of the device as well as the hw address of the
 947 * net_device
 948 */
 949static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[], int hw)
 950{
 951        struct net_device *dev = slave->dev;
 952        struct sockaddr s_addr;
 953
 954        if (!hw) {
 955                memcpy(dev->dev_addr, addr, dev->addr_len);
 956                return 0;
 957        }
 958
 959        /* for rlb each slave must have a unique hw mac addresses so that */
 960        /* each slave will receive packets destined to a different mac */
 961        memcpy(s_addr.sa_data, addr, dev->addr_len);
 962        s_addr.sa_family = dev->type;
 963        if (dev_set_mac_address(dev, &s_addr)) {
 964                pr_err(DRV_NAME
 965                       ": %s: Error: dev_set_mac_address of dev %s failed! ALB "
 966                       "mode requires that the base driver support setting "
 967                       "the hw address also when the network device's "
 968                       "interface is open\n",
 969                       dev->master->name, dev->name);
 970                return -EOPNOTSUPP;
 971        }
 972        return 0;
 973}
 974
 975/*
 976 * Swap MAC addresses between two slaves.
 977 *
 978 * Called with RTNL held, and no other locks.
 979 *
 980 */
 981
 982static void alb_swap_mac_addr(struct bonding *bond, struct slave *slave1, struct slave *slave2)
 983{
 984        u8 tmp_mac_addr[ETH_ALEN];
 985
 986        memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
 987        alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr, bond->alb_info.rlb_enabled);
 988        alb_set_slave_mac_addr(slave2, tmp_mac_addr, bond->alb_info.rlb_enabled);
 989
 990}
 991
 992/*
 993 * Send learning packets after MAC address swap.
 994 *
 995 * Called with RTNL and no other locks
 996 */
 997static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
 998                                struct slave *slave2)
 999{
1000        int slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
1001        struct slave *disabled_slave = NULL;
1002
1003        ASSERT_RTNL();
1004
1005        /* fasten the change in the switch */
1006        if (SLAVE_IS_OK(slave1)) {
1007                alb_send_learning_packets(slave1, slave1->dev->dev_addr);
1008                if (bond->alb_info.rlb_enabled) {
1009                        /* inform the clients that the mac address
1010                         * has changed
1011                         */
1012                        rlb_req_update_slave_clients(bond, slave1);
1013                }
1014        } else {
1015                disabled_slave = slave1;
1016        }
1017
1018        if (SLAVE_IS_OK(slave2)) {
1019                alb_send_learning_packets(slave2, slave2->dev->dev_addr);
1020                if (bond->alb_info.rlb_enabled) {
1021                        /* inform the clients that the mac address
1022                         * has changed
1023                         */
1024                        rlb_req_update_slave_clients(bond, slave2);
1025                }
1026        } else {
1027                disabled_slave = slave2;
1028        }
1029
1030        if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1031                /* A disabled slave was assigned an active mac addr */
1032                rlb_teach_disabled_mac_on_primary(bond,
1033                                                  disabled_slave->dev->dev_addr);
1034        }
1035}
1036
1037/**
1038 * alb_change_hw_addr_on_detach
1039 * @bond: bonding we're working on
1040 * @slave: the slave that was just detached
1041 *
1042 * We assume that @slave was already detached from the slave list.
1043 *
1044 * If @slave's permanent hw address is different both from its current
1045 * address and from @bond's address, then somewhere in the bond there's
1046 * a slave that has @slave's permanet address as its current address.
1047 * We'll make sure that that slave no longer uses @slave's permanent address.
1048 *
1049 * Caller must hold RTNL and no other locks
1050 */
1051static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1052{
1053        int perm_curr_diff;
1054        int perm_bond_diff;
1055
1056        perm_curr_diff = compare_ether_addr_64bits(slave->perm_hwaddr,
1057                                                   slave->dev->dev_addr);
1058        perm_bond_diff = compare_ether_addr_64bits(slave->perm_hwaddr,
1059                                                   bond->dev->dev_addr);
1060
1061        if (perm_curr_diff && perm_bond_diff) {
1062                struct slave *tmp_slave;
1063                int i, found = 0;
1064
1065                bond_for_each_slave(bond, tmp_slave, i) {
1066                        if (!compare_ether_addr_64bits(slave->perm_hwaddr,
1067                                                       tmp_slave->dev->dev_addr)) {
1068                                found = 1;
1069                                break;
1070                        }
1071                }
1072
1073                if (found) {
1074                        /* locking: needs RTNL and nothing else */
1075                        alb_swap_mac_addr(bond, slave, tmp_slave);
1076                        alb_fasten_mac_swap(bond, slave, tmp_slave);
1077                }
1078        }
1079}
1080
1081/**
1082 * alb_handle_addr_collision_on_attach
1083 * @bond: bonding we're working on
1084 * @slave: the slave that was just attached
1085 *
1086 * checks uniqueness of slave's mac address and handles the case the
1087 * new slave uses the bonds mac address.
1088 *
1089 * If the permanent hw address of @slave is @bond's hw address, we need to
1090 * find a different hw address to give @slave, that isn't in use by any other
1091 * slave in the bond. This address must be, of course, one of the premanent
1092 * addresses of the other slaves.
1093 *
1094 * We go over the slave list, and for each slave there we compare its
1095 * permanent hw address with the current address of all the other slaves.
1096 * If no match was found, then we've found a slave with a permanent address
1097 * that isn't used by any other slave in the bond, so we can assign it to
1098 * @slave.
1099 *
1100 * assumption: this function is called before @slave is attached to the
1101 *             bond slave list.
1102 *
1103 * caller must hold the bond lock for write since the mac addresses are compared
1104 * and may be swapped.
1105 */
1106static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1107{
1108        struct slave *tmp_slave1, *tmp_slave2, *free_mac_slave;
1109        struct slave *has_bond_addr = bond->curr_active_slave;
1110        int i, j, found = 0;
1111
1112        if (bond->slave_cnt == 0) {
1113                /* this is the first slave */
1114                return 0;
1115        }
1116
1117        /* if slave's mac address differs from bond's mac address
1118         * check uniqueness of slave's mac address against the other
1119         * slaves in the bond.
1120         */
1121        if (compare_ether_addr_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1122                bond_for_each_slave(bond, tmp_slave1, i) {
1123                        if (!compare_ether_addr_64bits(tmp_slave1->dev->dev_addr,
1124                                                       slave->dev->dev_addr)) {
1125                                found = 1;
1126                                break;
1127                        }
1128                }
1129
1130                if (!found)
1131                        return 0;
1132
1133                /* Try setting slave mac to bond address and fall-through
1134                   to code handling that situation below... */
1135                alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1136                                       bond->alb_info.rlb_enabled);
1137        }
1138
1139        /* The slave's address is equal to the address of the bond.
1140         * Search for a spare address in the bond for this slave.
1141         */
1142        free_mac_slave = NULL;
1143
1144        bond_for_each_slave(bond, tmp_slave1, i) {
1145                found = 0;
1146                bond_for_each_slave(bond, tmp_slave2, j) {
1147                        if (!compare_ether_addr_64bits(tmp_slave1->perm_hwaddr,
1148                                                       tmp_slave2->dev->dev_addr)) {
1149                                found = 1;
1150                                break;
1151                        }
1152                }
1153
1154                if (!found) {
1155                        /* no slave has tmp_slave1's perm addr
1156                         * as its curr addr
1157                         */
1158                        free_mac_slave = tmp_slave1;
1159                        break;
1160                }
1161
1162                if (!has_bond_addr) {
1163                        if (!compare_ether_addr_64bits(tmp_slave1->dev->dev_addr,
1164                                                       bond->dev->dev_addr)) {
1165
1166                                has_bond_addr = tmp_slave1;
1167                        }
1168                }
1169        }
1170
1171        if (free_mac_slave) {
1172                alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1173                                       bond->alb_info.rlb_enabled);
1174
1175                pr_warning(DRV_NAME
1176                           ": %s: Warning: the hw address of slave %s is "
1177                           "in use by the bond; giving it the hw address "
1178                           "of %s\n",
1179                           bond->dev->name, slave->dev->name,
1180                           free_mac_slave->dev->name);
1181
1182        } else if (has_bond_addr) {
1183                pr_err(DRV_NAME
1184                       ": %s: Error: the hw address of slave %s is in use by the "
1185                       "bond; couldn't find a slave with a free hw address to "
1186                       "give it (this should not have happened)\n",
1187                       bond->dev->name, slave->dev->name);
1188                return -EFAULT;
1189        }
1190
1191        return 0;
1192}
1193
1194/**
1195 * alb_set_mac_address
1196 * @bond:
1197 * @addr:
1198 *
1199 * In TLB mode all slaves are configured to the bond's hw address, but set
1200 * their dev_addr field to different addresses (based on their permanent hw
1201 * addresses).
1202 *
1203 * For each slave, this function sets the interface to the new address and then
1204 * changes its dev_addr field to its previous value.
1205 *
1206 * Unwinding assumes bond's mac address has not yet changed.
1207 */
1208static int alb_set_mac_address(struct bonding *bond, void *addr)
1209{
1210        struct sockaddr sa;
1211        struct slave *slave, *stop_at;
1212        char tmp_addr[ETH_ALEN];
1213        int res;
1214        int i;
1215
1216        if (bond->alb_info.rlb_enabled) {
1217                return 0;
1218        }
1219
1220        bond_for_each_slave(bond, slave, i) {
1221                /* save net_device's current hw address */
1222                memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1223
1224                res = dev_set_mac_address(slave->dev, addr);
1225
1226                /* restore net_device's hw address */
1227                memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1228
1229                if (res)
1230                        goto unwind;
1231        }
1232
1233        return 0;
1234
1235unwind:
1236        memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1237        sa.sa_family = bond->dev->type;
1238
1239        /* unwind from head to the slave that failed */
1240        stop_at = slave;
1241        bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
1242                memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1243                dev_set_mac_address(slave->dev, &sa);
1244                memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1245        }
1246
1247        return res;
1248}
1249
1250/************************ exported alb funcions ************************/
1251
1252int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1253{
1254        int res;
1255
1256        res = tlb_initialize(bond);
1257        if (res) {
1258                return res;
1259        }
1260
1261        if (rlb_enabled) {
1262                bond->alb_info.rlb_enabled = 1;
1263                /* initialize rlb */
1264                res = rlb_initialize(bond);
1265                if (res) {
1266                        tlb_deinitialize(bond);
1267                        return res;
1268                }
1269        } else {
1270                bond->alb_info.rlb_enabled = 0;
1271        }
1272
1273        return 0;
1274}
1275
1276void bond_alb_deinitialize(struct bonding *bond)
1277{
1278        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1279
1280        tlb_deinitialize(bond);
1281
1282        if (bond_info->rlb_enabled) {
1283                rlb_deinitialize(bond);
1284        }
1285}
1286
1287int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1288{
1289        struct bonding *bond = netdev_priv(bond_dev);
1290        struct ethhdr *eth_data;
1291        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1292        struct slave *tx_slave = NULL;
1293        static const __be32 ip_bcast = htonl(0xffffffff);
1294        int hash_size = 0;
1295        int do_tx_balance = 1;
1296        u32 hash_index = 0;
1297        const u8 *hash_start = NULL;
1298        int res = 1;
1299        struct ipv6hdr *ip6hdr;
1300
1301        skb_reset_mac_header(skb);
1302        eth_data = eth_hdr(skb);
1303
1304        /* make sure that the curr_active_slave and the slaves list do
1305         * not change during tx
1306         */
1307        read_lock(&bond->lock);
1308        read_lock(&bond->curr_slave_lock);
1309
1310        if (!BOND_IS_OK(bond)) {
1311                goto out;
1312        }
1313
1314        switch (ntohs(skb->protocol)) {
1315        case ETH_P_IP: {
1316                const struct iphdr *iph = ip_hdr(skb);
1317
1318                if (!compare_ether_addr_64bits(eth_data->h_dest, mac_bcast) ||
1319                    (iph->daddr == ip_bcast) ||
1320                    (iph->protocol == IPPROTO_IGMP)) {
1321                        do_tx_balance = 0;
1322                        break;
1323                }
1324                hash_start = (char *)&(iph->daddr);
1325                hash_size = sizeof(iph->daddr);
1326        }
1327                break;
1328        case ETH_P_IPV6:
1329                /* IPv6 doesn't really use broadcast mac address, but leave
1330                 * that here just in case.
1331                 */
1332                if (!compare_ether_addr_64bits(eth_data->h_dest, mac_bcast)) {
1333                        do_tx_balance = 0;
1334                        break;
1335                }
1336
1337                /* IPv6 uses all-nodes multicast as an equivalent to
1338                 * broadcasts in IPv4.
1339                 */
1340                if (!compare_ether_addr_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1341                        do_tx_balance = 0;
1342                        break;
1343                }
1344
1345                /* Additianally, DAD probes should not be tx-balanced as that
1346                 * will lead to false positives for duplicate addresses and
1347                 * prevent address configuration from working.
1348                 */
1349                ip6hdr = ipv6_hdr(skb);
1350                if (ipv6_addr_any(&ip6hdr->saddr)) {
1351                        do_tx_balance = 0;
1352                        break;
1353                }
1354
1355                hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1356                hash_size = sizeof(ipv6_hdr(skb)->daddr);
1357                break;
1358        case ETH_P_IPX:
1359                if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1360                        /* something is wrong with this packet */
1361                        do_tx_balance = 0;
1362                        break;
1363                }
1364
1365                if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1366                        /* The only protocol worth balancing in
1367                         * this family since it has an "ARP" like
1368                         * mechanism
1369                         */
1370                        do_tx_balance = 0;
1371                        break;
1372                }
1373
1374                hash_start = (char*)eth_data->h_dest;
1375                hash_size = ETH_ALEN;
1376                break;
1377        case ETH_P_ARP:
1378                do_tx_balance = 0;
1379                if (bond_info->rlb_enabled) {
1380                        tx_slave = rlb_arp_xmit(skb, bond);
1381                }
1382                break;
1383        default:
1384                do_tx_balance = 0;
1385                break;
1386        }
1387
1388        if (do_tx_balance) {
1389                hash_index = _simple_hash(hash_start, hash_size);
1390                tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1391        }
1392
1393        if (!tx_slave) {
1394                /* unbalanced or unassigned, send through primary */
1395                tx_slave = bond->curr_active_slave;
1396                bond_info->unbalanced_load += skb->len;
1397        }
1398
1399        if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1400                if (tx_slave != bond->curr_active_slave) {
1401                        memcpy(eth_data->h_source,
1402                               tx_slave->dev->dev_addr,
1403                               ETH_ALEN);
1404                }
1405
1406                res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1407        } else {
1408                if (tx_slave) {
1409                        tlb_clear_slave(bond, tx_slave, 0);
1410                }
1411        }
1412
1413out:
1414        if (res) {
1415                /* no suitable interface, frame not sent */
1416                dev_kfree_skb(skb);
1417        }
1418        read_unlock(&bond->curr_slave_lock);
1419        read_unlock(&bond->lock);
1420        return NETDEV_TX_OK;
1421}
1422
1423void bond_alb_monitor(struct work_struct *work)
1424{
1425        struct bonding *bond = container_of(work, struct bonding,
1426                                            alb_work.work);
1427        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1428        struct slave *slave;
1429        int i;
1430
1431        read_lock(&bond->lock);
1432
1433        if (bond->kill_timers) {
1434                goto out;
1435        }
1436
1437        if (bond->slave_cnt == 0) {
1438                bond_info->tx_rebalance_counter = 0;
1439                bond_info->lp_counter = 0;
1440                goto re_arm;
1441        }
1442
1443        bond_info->tx_rebalance_counter++;
1444        bond_info->lp_counter++;
1445
1446        /* send learning packets */
1447        if (bond_info->lp_counter >= BOND_ALB_LP_TICKS) {
1448                /* change of curr_active_slave involves swapping of mac addresses.
1449                 * in order to avoid this swapping from happening while
1450                 * sending the learning packets, the curr_slave_lock must be held for
1451                 * read.
1452                 */
1453                read_lock(&bond->curr_slave_lock);
1454
1455                bond_for_each_slave(bond, slave, i) {
1456                        alb_send_learning_packets(slave, slave->dev->dev_addr);
1457                }
1458
1459                read_unlock(&bond->curr_slave_lock);
1460
1461                bond_info->lp_counter = 0;
1462        }
1463
1464        /* rebalance tx traffic */
1465        if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1466
1467                read_lock(&bond->curr_slave_lock);
1468
1469                bond_for_each_slave(bond, slave, i) {
1470                        tlb_clear_slave(bond, slave, 1);
1471                        if (slave == bond->curr_active_slave) {
1472                                SLAVE_TLB_INFO(slave).load =
1473                                        bond_info->unbalanced_load /
1474                                                BOND_TLB_REBALANCE_INTERVAL;
1475                                bond_info->unbalanced_load = 0;
1476                        }
1477                }
1478
1479                read_unlock(&bond->curr_slave_lock);
1480
1481                bond_info->tx_rebalance_counter = 0;
1482        }
1483
1484        /* handle rlb stuff */
1485        if (bond_info->rlb_enabled) {
1486                if (bond_info->primary_is_promisc &&
1487                    (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1488
1489                        /*
1490                         * dev_set_promiscuity requires rtnl and
1491                         * nothing else.
1492                         */
1493                        read_unlock(&bond->lock);
1494                        rtnl_lock();
1495
1496                        bond_info->rlb_promisc_timeout_counter = 0;
1497
1498                        /* If the primary was set to promiscuous mode
1499                         * because a slave was disabled then
1500                         * it can now leave promiscuous mode.
1501                         */
1502                        dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1503                        bond_info->primary_is_promisc = 0;
1504
1505                        rtnl_unlock();
1506                        read_lock(&bond->lock);
1507                }
1508
1509                if (bond_info->rlb_rebalance) {
1510                        bond_info->rlb_rebalance = 0;
1511                        rlb_rebalance(bond);
1512                }
1513
1514                /* check if clients need updating */
1515                if (bond_info->rx_ntt) {
1516                        if (bond_info->rlb_update_delay_counter) {
1517                                --bond_info->rlb_update_delay_counter;
1518                        } else {
1519                                rlb_update_rx_clients(bond);
1520                                if (bond_info->rlb_update_retry_counter) {
1521                                        --bond_info->rlb_update_retry_counter;
1522                                } else {
1523                                        bond_info->rx_ntt = 0;
1524                                }
1525                        }
1526                }
1527        }
1528
1529re_arm:
1530        queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1531out:
1532        read_unlock(&bond->lock);
1533}
1534
1535/* assumption: called before the slave is attached to the bond
1536 * and not locked by the bond lock
1537 */
1538int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1539{
1540        int res;
1541
1542        res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1543                                     bond->alb_info.rlb_enabled);
1544        if (res) {
1545                return res;
1546        }
1547
1548        /* caller must hold the bond lock for write since the mac addresses
1549         * are compared and may be swapped.
1550         */
1551        read_lock(&bond->lock);
1552
1553        res = alb_handle_addr_collision_on_attach(bond, slave);
1554
1555        read_unlock(&bond->lock);
1556
1557        if (res) {
1558                return res;
1559        }
1560
1561        tlb_init_slave(slave);
1562
1563        /* order a rebalance ASAP */
1564        bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1565
1566        if (bond->alb_info.rlb_enabled) {
1567                bond->alb_info.rlb_rebalance = 1;
1568        }
1569
1570        return 0;
1571}
1572
1573/*
1574 * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1575 * if necessary.
1576 *
1577 * Caller must hold RTNL and no other locks
1578 */
1579void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1580{
1581        if (bond->slave_cnt > 1) {
1582                alb_change_hw_addr_on_detach(bond, slave);
1583        }
1584
1585        tlb_clear_slave(bond, slave, 0);
1586
1587        if (bond->alb_info.rlb_enabled) {
1588                bond->alb_info.next_rx_slave = NULL;
1589                rlb_clear_slave(bond, slave);
1590        }
1591}
1592
1593/* Caller must hold bond lock for read */
1594void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1595{
1596        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1597
1598        if (link == BOND_LINK_DOWN) {
1599                tlb_clear_slave(bond, slave, 0);
1600                if (bond->alb_info.rlb_enabled) {
1601                        rlb_clear_slave(bond, slave);
1602                }
1603        } else if (link == BOND_LINK_UP) {
1604                /* order a rebalance ASAP */
1605                bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1606                if (bond->alb_info.rlb_enabled) {
1607                        bond->alb_info.rlb_rebalance = 1;
1608                        /* If the updelay module parameter is smaller than the
1609                         * forwarding delay of the switch the rebalance will
1610                         * not work because the rebalance arp replies will
1611                         * not be forwarded to the clients..
1612                         */
1613                }
1614        }
1615}
1616
1617/**
1618 * bond_alb_handle_active_change - assign new curr_active_slave
1619 * @bond: our bonding struct
1620 * @new_slave: new slave to assign
1621 *
1622 * Set the bond->curr_active_slave to @new_slave and handle
1623 * mac address swapping and promiscuity changes as needed.
1624 *
1625 * If new_slave is NULL, caller must hold curr_slave_lock or
1626 * bond->lock for write.
1627 *
1628 * If new_slave is not NULL, caller must hold RTNL, bond->lock for
1629 * read and curr_slave_lock for write.  Processing here may sleep, so
1630 * no other locks may be held.
1631 */
1632void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1633        __releases(&bond->curr_slave_lock)
1634        __releases(&bond->lock)
1635        __acquires(&bond->lock)
1636        __acquires(&bond->curr_slave_lock)
1637{
1638        struct slave *swap_slave;
1639        int i;
1640
1641        if (bond->curr_active_slave == new_slave) {
1642                return;
1643        }
1644
1645        if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1646                dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1647                bond->alb_info.primary_is_promisc = 0;
1648                bond->alb_info.rlb_promisc_timeout_counter = 0;
1649        }
1650
1651        swap_slave = bond->curr_active_slave;
1652        bond->curr_active_slave = new_slave;
1653
1654        if (!new_slave || (bond->slave_cnt == 0)) {
1655                return;
1656        }
1657
1658        /* set the new curr_active_slave to the bonds mac address
1659         * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1660         */
1661        if (!swap_slave) {
1662                struct slave *tmp_slave;
1663                /* find slave that is holding the bond's mac address */
1664                bond_for_each_slave(bond, tmp_slave, i) {
1665                        if (!compare_ether_addr_64bits(tmp_slave->dev->dev_addr,
1666                                                       bond->dev->dev_addr)) {
1667                                swap_slave = tmp_slave;
1668                                break;
1669                        }
1670                }
1671        }
1672
1673        /*
1674         * Arrange for swap_slave and new_slave to temporarily be
1675         * ignored so we can mess with their MAC addresses without
1676         * fear of interference from transmit activity.
1677         */
1678        if (swap_slave) {
1679                tlb_clear_slave(bond, swap_slave, 1);
1680        }
1681        tlb_clear_slave(bond, new_slave, 1);
1682
1683        write_unlock_bh(&bond->curr_slave_lock);
1684        read_unlock(&bond->lock);
1685
1686        ASSERT_RTNL();
1687
1688        /* curr_active_slave must be set before calling alb_swap_mac_addr */
1689        if (swap_slave) {
1690                /* swap mac address */
1691                alb_swap_mac_addr(bond, swap_slave, new_slave);
1692        } else {
1693                /* set the new_slave to the bond mac address */
1694                alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1695                                       bond->alb_info.rlb_enabled);
1696        }
1697
1698        if (swap_slave) {
1699                alb_fasten_mac_swap(bond, swap_slave, new_slave);
1700                read_lock(&bond->lock);
1701        } else {
1702                read_lock(&bond->lock);
1703                alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1704        }
1705
1706        write_lock_bh(&bond->curr_slave_lock);
1707}
1708
1709/*
1710 * Called with RTNL
1711 */
1712int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1713        __acquires(&bond->lock)
1714        __releases(&bond->lock)
1715{
1716        struct bonding *bond = netdev_priv(bond_dev);
1717        struct sockaddr *sa = addr;
1718        struct slave *slave, *swap_slave;
1719        int res;
1720        int i;
1721
1722        if (!is_valid_ether_addr(sa->sa_data)) {
1723                return -EADDRNOTAVAIL;
1724        }
1725
1726        res = alb_set_mac_address(bond, addr);
1727        if (res) {
1728                return res;
1729        }
1730
1731        memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1732
1733        /* If there is no curr_active_slave there is nothing else to do.
1734         * Otherwise we'll need to pass the new address to it and handle
1735         * duplications.
1736         */
1737        if (!bond->curr_active_slave) {
1738                return 0;
1739        }
1740
1741        swap_slave = NULL;
1742
1743        bond_for_each_slave(bond, slave, i) {
1744                if (!compare_ether_addr_64bits(slave->dev->dev_addr,
1745                                               bond_dev->dev_addr)) {
1746                        swap_slave = slave;
1747                        break;
1748                }
1749        }
1750
1751        if (swap_slave) {
1752                alb_swap_mac_addr(bond, swap_slave, bond->curr_active_slave);
1753                alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave);
1754        } else {
1755                alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr,
1756                                       bond->alb_info.rlb_enabled);
1757
1758                read_lock(&bond->lock);
1759                alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1760                if (bond->alb_info.rlb_enabled) {
1761                        /* inform clients mac address has changed */
1762                        rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1763                }
1764                read_unlock(&bond->lock);
1765        }
1766
1767        return 0;
1768}
1769
1770void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1771{
1772        if (bond->alb_info.current_alb_vlan &&
1773            (bond->alb_info.current_alb_vlan->vlan_id == vlan_id)) {
1774                bond->alb_info.current_alb_vlan = NULL;
1775        }
1776
1777        if (bond->alb_info.rlb_enabled) {
1778                rlb_clear_vlan(bond, vlan_id);
1779        }
1780}
1781
1782