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