drivers/net/bonding/bond_alb.c

   1 /*
   2  * Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms of the GNU General Public License as published by the
   6  * Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful, but
  10  * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  11  * or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License along
  15  * with this program; if not, write to the Free Software Foundation, Inc.,
  16  * 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
  17  *
  18  * The full GNU General Public License is included in this distribution in the
  19  * file called LICENSE.
  20  *
  21  */
  22
  23 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  24
  25 #include <linux/skbuff.h>
  26 #include <linux/netdevice.h>
  27 #include <linux/etherdevice.h>
  28 #include <linux/pkt_sched.h>
  29 #include <linux/spinlock.h>
  30 #include <linux/slab.h>
  31 #include <linux/timer.h>
  32 #include <linux/ip.h>
  33 #include <linux/ipv6.h>
  34 #include <linux/if_arp.h>
  35 #include <linux/if_ether.h>
  36 #include <linux/if_bonding.h>
  37 #include <linux/if_vlan.h>
  38 #include <linux/in.h>
  39 #include <net/ipx.h>
  40 #include <net/arp.h>
  41 #include <net/ipv6.h>
  42 #include <asm/byteorder.h>
  43 #include "bonding.h"
  44 #include "bond_alb.h"
  45
  46
  47
  48 #ifndef __long_aligned
  49 #define __long_aligned __attribute__((aligned((sizeof(long)))))
  50 #endif
  51 static const u8 mac_bcast[ETH_ALEN] __long_aligned = {
  52         0xff, 0xff, 0xff, 0xff, 0xff, 0xff
  53 };
  54 static const u8 mac_v6_allmcast[ETH_ALEN] __long_aligned = {
  55         0x33, 0x33, 0x00, 0x00, 0x00, 0x01
  56 };
  57 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
  58
  59 #pragma pack(1)
  60 struct learning_pkt {
  61         u8 mac_dst[ETH_ALEN];
  62         u8 mac_src[ETH_ALEN];
  63         __be16 type;
  64         u8 padding[ETH_ZLEN - ETH_HLEN];
  65 };
  66
  67 struct arp_pkt {
  68         __be16  hw_addr_space;
  69         __be16  prot_addr_space;
  70         u8      hw_addr_len;
  71         u8      prot_addr_len;
  72         __be16  op_code;
  73         u8      mac_src[ETH_ALEN];      /* sender hardware address */
  74         __be32  ip_src;                 /* sender IP address */
  75         u8      mac_dst[ETH_ALEN];      /* target hardware address */
  76         __be32  ip_dst;                 /* target IP address */
  77 };
  78 #pragma pack()
  79
  80 static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb)
  81 {
  82         return (struct arp_pkt *)skb_network_header(skb);
  83 }
  84
  85 /* Forward declaration */
  86 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[]);
  87
  88 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
  89 {
  90         int i;
  91         u8 hash = 0;
  92
  93         for (i = 0; i < hash_size; i++) {
  94                 hash ^= hash_start[i];
  95         }
  96
  97         return hash;
  98 }
  99
 100 /*********************** tlb specific functions ***************************/
 101
 102 static inline void _lock_tx_hashtbl(struct bonding *bond)
 103 {
 104         spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 105 }
 106
 107 static inline void _unlock_tx_hashtbl(struct bonding *bond)
 108 {
 109         spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 110 }
 111
 112 /* Caller must hold tx_hashtbl lock */
 113 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
 114 {
 115         if (save_load) {
 116                 entry->load_history = 1 + entry->tx_bytes /
 117                                       BOND_TLB_REBALANCE_INTERVAL;
 118                 entry->tx_bytes = 0;
 119         }
 120
 121         entry->tx_slave = NULL;
 122         entry->next = TLB_NULL_INDEX;
 123         entry->prev = TLB_NULL_INDEX;
 124 }
 125
 126 static inline void tlb_init_slave(struct slave *slave)
 127 {
 128         SLAVE_TLB_INFO(slave).load = 0;
 129         SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
 130 }
 131
 132 /* Caller must hold bond lock for read */
 133 static void tlb_clear_slave(struct bonding *bond, struct slave *slave, int save_load)
 134 {
 135         struct tlb_client_info *tx_hash_table;
 136         u32 index;
 137
 138         _lock_tx_hashtbl(bond);
 139
 140         /* clear slave from tx_hashtbl */
 141         tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
 142
 143         /* skip this if we've already freed the tx hash table */
 144         if (tx_hash_table) {
 145                 index = SLAVE_TLB_INFO(slave).head;
 146                 while (index != TLB_NULL_INDEX) {
 147                         u32 next_index = tx_hash_table[index].next;
 148                         tlb_init_table_entry(&tx_hash_table[index], save_load);
 149                         index = next_index;
 150                 }
 151         }
 152
 153         tlb_init_slave(slave);
 154
 155         _unlock_tx_hashtbl(bond);
 156 }
 157
 158 /* Must be called before starting the monitor timer */
 159 static int tlb_initialize(struct bonding *bond)
 160 {
 161         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 162         int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 163         struct tlb_client_info *new_hashtbl;
 164         int i;
 165
 166         spin_lock_init(&(bond_info->tx_hashtbl_lock));
 167
 168         new_hashtbl = kzalloc(size, GFP_KERNEL);
 169         if (!new_hashtbl) {
 170                 pr_err("%s: Error: Failed to allocate TLB hash table\n",
 171                        bond->dev->name);
 172                 return -1;
 173         }
 174         _lock_tx_hashtbl(bond);
 175
 176         bond_info->tx_hashtbl = new_hashtbl;
 177
 178         for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
 179                 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
 180         }
 181
 182         _unlock_tx_hashtbl(bond);
 183
 184         return 0;
 185 }
 186
 187 /* Must be called only after all slaves have been released */
 188 static void tlb_deinitialize(struct bonding *bond)
 189 {
 190         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 191
 192         _lock_tx_hashtbl(bond);
 193
 194         kfree(bond_info->tx_hashtbl);
 195         bond_info->tx_hashtbl = NULL;
 196
 197         _unlock_tx_hashtbl(bond);
 198 }
 199
 200 static long long compute_gap(struct slave *slave)
 201 {
 202         return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
 203                (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
 204 }
 205
 206 /* Caller must hold bond lock for read */
 207 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
 208 {
 209         struct slave *slave, *least_loaded;
 210         long long max_gap;
 211         int i;
 212
 213         least_loaded = NULL;
 214         max_gap = LLONG_MIN;
 215
 216         /* Find the slave with the largest gap */
 217         bond_for_each_slave(bond, slave, i) {
 218                 if (SLAVE_IS_OK(slave)) {
 219                         long long gap = compute_gap(slave);
 220
 221                         if (max_gap < gap) {
 222                                 least_loaded = slave;
 223                                 max_gap = gap;
 224                         }
 225                 }
 226         }
 227
 228         return least_loaded;
 229 }
 230
 231 /* Caller must hold bond lock for read */
 232 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index, u32 skb_len)
 233 {
 234         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 235         struct tlb_client_info *hash_table;
 236         struct slave *assigned_slave;
 237
 238         _lock_tx_hashtbl(bond);
 239
 240         hash_table = bond_info->tx_hashtbl;
 241         assigned_slave = hash_table[hash_index].tx_slave;
 242         if (!assigned_slave) {
 243                 assigned_slave = tlb_get_least_loaded_slave(bond);
 244
 245                 if (assigned_slave) {
 246                         struct tlb_slave_info *slave_info =
 247                                 &(SLAVE_TLB_INFO(assigned_slave));
 248                         u32 next_index = slave_info->head;
 249
 250                         hash_table[hash_index].tx_slave = assigned_slave;
 251                         hash_table[hash_index].next = next_index;
 252                         hash_table[hash_index].prev = TLB_NULL_INDEX;
 253
 254                         if (next_index != TLB_NULL_INDEX) {
 255                                 hash_table[next_index].prev = hash_index;
 256                         }
 257
 258                         slave_info->head = hash_index;
 259                         slave_info->load +=
 260                                 hash_table[hash_index].load_history;
 261                 }
 262         }
 263
 264         if (assigned_slave) {
 265                 hash_table[hash_index].tx_bytes += skb_len;
 266         }
 267
 268         _unlock_tx_hashtbl(bond);
 269
 270         return assigned_slave;
 271 }
 272
 273 /*********************** rlb specific functions ***************************/
 274 static inline void _lock_rx_hashtbl(struct bonding *bond)
 275 {
 276         spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 277 }
 278
 279 static inline void _unlock_rx_hashtbl(struct bonding *bond)
 280 {
 281         spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 282 }
 283
 284 /* when an ARP REPLY is received from a client update its info
 285  * in the rx_hashtbl
 286  */
 287 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 288 {
 289         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 290         struct rlb_client_info *client_info;
 291         u32 hash_index;
 292
 293         _lock_rx_hashtbl(bond);
 294
 295         hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
 296         client_info = &(bond_info->rx_hashtbl[hash_index]);
 297
 298         if ((client_info->assigned) &&
 299             (client_info->ip_src == arp->ip_dst) &&
 300             (client_info->ip_dst == arp->ip_src) &&
 301             (compare_ether_addr_64bits(client_info->mac_dst, arp->mac_src))) {
 302                 /* update the clients MAC address */
 303                 memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
 304                 client_info->ntt = 1;
 305                 bond_info->rx_ntt = 1;
 306         }
 307
 308         _unlock_rx_hashtbl(bond);
 309 }
 310
 311 static void rlb_arp_recv(struct sk_buff *skb, struct bonding *bond,
 312                          struct slave *slave)
 313 {
 314         struct arp_pkt *arp;
 315
 316         if (skb->protocol != cpu_to_be16(ETH_P_ARP))
 317                 return;
 318
 319         arp = (struct arp_pkt *) skb->data;
 320         if (!arp) {
 321                 pr_debug("Packet has no ARP data\n");
 322                 return;
 323         }
 324
 325         if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
 326                 return;
 327
 328         if (skb->len < sizeof(struct arp_pkt)) {
 329                 pr_debug("Packet is too small to be an ARP\n");
 330                 return;
 331         }
 332
 333         if (arp->op_code == htons(ARPOP_REPLY)) {
 334                 /* update rx hash table for this ARP */
 335                 rlb_update_entry_from_arp(bond, arp);
 336                 pr_debug("Server received an ARP Reply from client\n");
 337         }
 338 }
 339
 340 /* Caller must hold bond lock for read */
 341 static struct slave *rlb_next_rx_slave(struct bonding *bond)
 342 {
 343         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 344         struct slave *rx_slave, *slave, *start_at;
 345         int i = 0;
 346
 347         if (bond_info->next_rx_slave) {
 348                 start_at = bond_info->next_rx_slave;
 349         } else {
 350                 start_at = bond->first_slave;
 351         }
 352
 353         rx_slave = NULL;
 354
 355         bond_for_each_slave_from(bond, slave, i, start_at) {
 356                 if (SLAVE_IS_OK(slave)) {
 357                         if (!rx_slave) {
 358                                 rx_slave = slave;
 359                         } else if (slave->speed > rx_slave->speed) {
 360                                 rx_slave = slave;
 361                         }
 362                 }
 363         }
 364
 365         if (rx_slave) {
 366                 bond_info->next_rx_slave = rx_slave->next;
 367         }
 368
 369         return rx_slave;
 370 }
 371
 372 /* teach the switch the mac of a disabled slave
 373  * on the primary for fault tolerance
 374  *
 375  * Caller must hold bond->curr_slave_lock for write or bond lock for write
 376  */
 377 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
 378 {
 379         if (!bond->curr_active_slave) {
 380                 return;
 381         }
 382
 383         if (!bond->alb_info.primary_is_promisc) {
 384                 if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1))
 385                         bond->alb_info.primary_is_promisc = 1;
 386                 else
 387                         bond->alb_info.primary_is_promisc = 0;
 388         }
 389
 390         bond->alb_info.rlb_promisc_timeout_counter = 0;
 391
 392         alb_send_learning_packets(bond->curr_active_slave, addr);
 393 }
 394
 395 /* slave being removed should not be active at this point
 396  *
 397  * Caller must hold bond lock for read
 398  */
 399 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 400 {
 401         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 402         struct rlb_client_info *rx_hash_table;
 403         u32 index, next_index;
 404
 405         /* clear slave from rx_hashtbl */
 406         _lock_rx_hashtbl(bond);
 407
 408         rx_hash_table = bond_info->rx_hashtbl;
 409         index = bond_info->rx_hashtbl_head;
 410         for (; index != RLB_NULL_INDEX; index = next_index) {
 411                 next_index = rx_hash_table[index].next;
 412                 if (rx_hash_table[index].slave == slave) {
 413                         struct slave *assigned_slave = rlb_next_rx_slave(bond);
 414
 415                         if (assigned_slave) {
 416                                 rx_hash_table[index].slave = assigned_slave;
 417                                 if (compare_ether_addr_64bits(rx_hash_table[index].mac_dst,
 418                                                               mac_bcast)) {
 419                                         bond_info->rx_hashtbl[index].ntt = 1;
 420                                         bond_info->rx_ntt = 1;
 421                                         /* A slave has been removed from the
 422                                          * table because it is either disabled
 423                                          * or being released. We must retry the
 424                                          * update to avoid clients from not
 425                                          * being updated & disconnecting when
 426                                          * there is stress
 427                                          */
 428                                         bond_info->rlb_update_retry_counter =
 429                                                 RLB_UPDATE_RETRY;
 430                                 }
 431                         } else {  /* there is no active slave */
 432                                 rx_hash_table[index].slave = NULL;
 433                         }
 434                 }
 435         }
 436
 437         _unlock_rx_hashtbl(bond);
 438
 439         write_lock_bh(&bond->curr_slave_lock);
 440
 441         if (slave != bond->curr_active_slave) {
 442                 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 443         }
 444
 445         write_unlock_bh(&bond->curr_slave_lock);
 446 }
 447
 448 static void rlb_update_client(struct rlb_client_info *client_info)
 449 {
 450         int i;
 451
 452         if (!client_info->slave) {
 453                 return;
 454         }
 455
 456         for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 457                 struct sk_buff *skb;
 458
 459                 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 460                                  client_info->ip_dst,
 461                                  client_info->slave->dev,
 462                                  client_info->ip_src,
 463                                  client_info->mac_dst,
 464                                  client_info->slave->dev->dev_addr,
 465                                  client_info->mac_dst);
 466                 if (!skb) {
 467                         pr_err("%s: Error: failed to create an ARP packet\n",
 468                                client_info->slave->dev->master->name);
 469                         continue;
 470                 }
 471
 472                 skb->dev = client_info->slave->dev;
 473
 474                 if (client_info->tag) {
 475                         skb = vlan_put_tag(skb, client_info->vlan_id);
 476                         if (!skb) {
 477                                 pr_err("%s: Error: failed to insert VLAN tag\n",
 478                                        client_info->slave->dev->master->name);
 479                                 continue;
 480                         }
 481                 }
 482
 483                 arp_xmit(skb);
 484         }
 485 }
 486
 487 /* sends ARP REPLIES that update the clients that need updating */
 488 static void rlb_update_rx_clients(struct bonding *bond)
 489 {
 490         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 491         struct rlb_client_info *client_info;
 492         u32 hash_index;
 493
 494         _lock_rx_hashtbl(bond);
 495
 496         hash_index = bond_info->rx_hashtbl_head;
 497         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 498                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 499                 if (client_info->ntt) {
 500                         rlb_update_client(client_info);
 501                         if (bond_info->rlb_update_retry_counter == 0) {
 502                                 client_info->ntt = 0;
 503                         }
 504                 }
 505         }
 506
 507         /* do not update the entries again until this counter is zero so that
 508          * not to confuse the clients.
 509          */
 510         bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 511
 512         _unlock_rx_hashtbl(bond);
 513 }
 514
 515 /* The slave was assigned a new mac address - update the clients */
 516 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 517 {
 518         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 519         struct rlb_client_info *client_info;
 520         int ntt = 0;
 521         u32 hash_index;
 522
 523         _lock_rx_hashtbl(bond);
 524
 525         hash_index = bond_info->rx_hashtbl_head;
 526         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 527                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 528
 529                 if ((client_info->slave == slave) &&
 530                     compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
 531                         client_info->ntt = 1;
 532                         ntt = 1;
 533                 }
 534         }
 535
 536         // update the team's flag only after the whole iteration
 537         if (ntt) {
 538                 bond_info->rx_ntt = 1;
 539                 //fasten the change
 540                 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 541         }
 542
 543         _unlock_rx_hashtbl(bond);
 544 }
 545
 546 /* mark all clients using src_ip to be updated */
 547 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
 548 {
 549         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 550         struct rlb_client_info *client_info;
 551         u32 hash_index;
 552
 553         _lock_rx_hashtbl(bond);
 554
 555         hash_index = bond_info->rx_hashtbl_head;
 556         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 557                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 558
 559                 if (!client_info->slave) {
 560                         pr_err("%s: Error: found a client with no channel in the client's hash table\n",
 561                                bond->dev->name);
 562                         continue;
 563                 }
 564                 /*update all clients using this src_ip, that are not assigned
 565                  * to the team's address (curr_active_slave) and have a known
 566                  * unicast mac address.
 567                  */
 568                 if ((client_info->ip_src == src_ip) &&
 569                     compare_ether_addr_64bits(client_info->slave->dev->dev_addr,
 570                            bond->dev->dev_addr) &&
 571                     compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
 572                         client_info->ntt = 1;
 573                         bond_info->rx_ntt = 1;
 574                 }
 575         }
 576
 577         _unlock_rx_hashtbl(bond);
 578 }
 579
 580 /* Caller must hold both bond and ptr locks for read */
 581 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
 582 {
 583         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 584         struct arp_pkt *arp = arp_pkt(skb);
 585         struct slave *assigned_slave;
 586         struct rlb_client_info *client_info;
 587         u32 hash_index = 0;
 588
 589         _lock_rx_hashtbl(bond);
 590
 591         hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
 592         client_info = &(bond_info->rx_hashtbl[hash_index]);
 593
 594         if (client_info->assigned) {
 595                 if ((client_info->ip_src == arp->ip_src) &&
 596                     (client_info->ip_dst == arp->ip_dst)) {
 597                         /* the entry is already assigned to this client */
 598                         if (compare_ether_addr_64bits(arp->mac_dst, mac_bcast)) {
 599                                 /* update mac address from arp */
 600                                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 601                         }
 602
 603                         assigned_slave = client_info->slave;
 604                         if (assigned_slave) {
 605                                 _unlock_rx_hashtbl(bond);
 606                                 return assigned_slave;
 607                         }
 608                 } else {
 609                         /* the entry is already assigned to some other client,
 610                          * move the old client to primary (curr_active_slave) so
 611                          * that the new client can be assigned to this entry.
 612                          */
 613                         if (bond->curr_active_slave &&
 614                             client_info->slave != bond->curr_active_slave) {
 615                                 client_info->slave = bond->curr_active_slave;
 616                                 rlb_update_client(client_info);
 617                         }
 618                 }
 619         }
 620         /* assign a new slave */
 621         assigned_slave = rlb_next_rx_slave(bond);
 622
 623         if (assigned_slave) {
 624                 client_info->ip_src = arp->ip_src;
 625                 client_info->ip_dst = arp->ip_dst;
 626                 /* arp->mac_dst is broadcast for arp reqeusts.
 627                  * will be updated with clients actual unicast mac address
 628                  * upon receiving an arp reply.
 629                  */
 630                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 631                 client_info->slave = assigned_slave;
 632
 633                 if (compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
 634                         client_info->ntt = 1;
 635                         bond->alb_info.rx_ntt = 1;
 636                 } else {
 637                         client_info->ntt = 0;
 638                 }
 639
 640                 if (bond->vlgrp) {
 641                         if (!vlan_get_tag(skb, &client_info->vlan_id))
 642                                 client_info->tag = 1;
 643                 }
 644
 645                 if (!client_info->assigned) {
 646                         u32 prev_tbl_head = bond_info->rx_hashtbl_head;
 647                         bond_info->rx_hashtbl_head = hash_index;
 648                         client_info->next = prev_tbl_head;
 649                         if (prev_tbl_head != RLB_NULL_INDEX) {
 650                                 bond_info->rx_hashtbl[prev_tbl_head].prev =
 651                                         hash_index;
 652                         }
 653                         client_info->assigned = 1;
 654                 }
 655         }
 656
 657         _unlock_rx_hashtbl(bond);
 658
 659         return assigned_slave;
 660 }
 661
 662 /* chooses (and returns) transmit channel for arp reply
 663  * does not choose channel for other arp types since they are
 664  * sent on the curr_active_slave
 665  */
 666 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 667 {
 668         struct arp_pkt *arp = arp_pkt(skb);
 669         struct slave *tx_slave = NULL;
 670
 671         if (arp->op_code == htons(ARPOP_REPLY)) {
 672                 /* the arp must be sent on the selected
 673                 * rx channel
 674                 */
 675                 tx_slave = rlb_choose_channel(skb, bond);
 676                 if (tx_slave) {
 677                         memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
 678                 }
 679                 pr_debug("Server sent ARP Reply packet\n");
 680         } else if (arp->op_code == htons(ARPOP_REQUEST)) {
 681                 /* Create an entry in the rx_hashtbl for this client as a
 682                  * place holder.
 683                  * When the arp reply is received the entry will be updated
 684                  * with the correct unicast address of the client.
 685                  */
 686                 rlb_choose_channel(skb, bond);
 687
 688                 /* The ARP reply packets must be delayed so that
 689                  * they can cancel out the influence of the ARP request.
 690                  */
 691                 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 692
 693                 /* arp requests are broadcast and are sent on the primary
 694                  * the arp request will collapse all clients on the subnet to
 695                  * the primary slave. We must register these clients to be
 696                  * updated with their assigned mac.
 697                  */
 698                 rlb_req_update_subnet_clients(bond, arp->ip_src);
 699                 pr_debug("Server sent ARP Request packet\n");
 700         }
 701
 702         return tx_slave;
 703 }
 704
 705 /* Caller must hold bond lock for read */
 706 static void rlb_rebalance(struct bonding *bond)
 707 {
 708         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 709         struct slave *assigned_slave;
 710         struct rlb_client_info *client_info;
 711         int ntt;
 712         u32 hash_index;
 713
 714         _lock_rx_hashtbl(bond);
 715
 716         ntt = 0;
 717         hash_index = bond_info->rx_hashtbl_head;
 718         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 719                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 720                 assigned_slave = rlb_next_rx_slave(bond);
 721                 if (assigned_slave && (client_info->slave != assigned_slave)) {
 722                         client_info->slave = assigned_slave;
 723                         client_info->ntt = 1;
 724                         ntt = 1;
 725                 }
 726         }
 727
 728         /* update the team's flag only after the whole iteration */
 729         if (ntt) {
 730                 bond_info->rx_ntt = 1;
 731         }
 732         _unlock_rx_hashtbl(bond);
 733 }
 734
 735 /* Caller must hold rx_hashtbl lock */
 736 static void rlb_init_table_entry(struct rlb_client_info *entry)
 737 {
 738         memset(entry, 0, sizeof(struct rlb_client_info));
 739         entry->next = RLB_NULL_INDEX;
 740         entry->prev = RLB_NULL_INDEX;
 741 }
 742
 743 static int rlb_initialize(struct bonding *bond)
 744 {
 745         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 746         struct rlb_client_info  *new_hashtbl;
 747         int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 748         int i;
 749
 750         spin_lock_init(&(bond_info->rx_hashtbl_lock));
 751
 752         new_hashtbl = kmalloc(size, GFP_KERNEL);
 753         if (!new_hashtbl) {
 754                 pr_err("%s: Error: Failed to allocate RLB hash table\n",
 755                        bond->dev->name);
 756                 return -1;
 757         }
 758         _lock_rx_hashtbl(bond);
 759
 760         bond_info->rx_hashtbl = new_hashtbl;
 761
 762         bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
 763
 764         for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
 765                 rlb_init_table_entry(bond_info->rx_hashtbl + i);
 766         }
 767
 768         _unlock_rx_hashtbl(bond);
 769
 770         /* register to receive ARPs */
 771         bond->recv_probe = rlb_arp_recv;
 772
 773         return 0;
 774 }
 775
 776 static void rlb_deinitialize(struct bonding *bond)
 777 {
 778         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 779
 780         _lock_rx_hashtbl(bond);
 781
 782         kfree(bond_info->rx_hashtbl);
 783         bond_info->rx_hashtbl = NULL;
 784         bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
 785
 786         _unlock_rx_hashtbl(bond);
 787 }
 788
 789 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 790 {
 791         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 792         u32 curr_index;
 793
 794         _lock_rx_hashtbl(bond);
 795
 796         curr_index = bond_info->rx_hashtbl_head;
 797         while (curr_index != RLB_NULL_INDEX) {
 798                 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 799                 u32 next_index = bond_info->rx_hashtbl[curr_index].next;
 800                 u32 prev_index = bond_info->rx_hashtbl[curr_index].prev;
 801
 802                 if (curr->tag && (curr->vlan_id == vlan_id)) {
 803                         if (curr_index == bond_info->rx_hashtbl_head) {
 804                                 bond_info->rx_hashtbl_head = next_index;
 805                         }
 806                         if (prev_index != RLB_NULL_INDEX) {
 807                                 bond_info->rx_hashtbl[prev_index].next = next_index;
 808                         }
 809                         if (next_index != RLB_NULL_INDEX) {
 810                                 bond_info->rx_hashtbl[next_index].prev = prev_index;
 811                         }
 812
 813                         rlb_init_table_entry(curr);
 814                 }
 815
 816                 curr_index = next_index;
 817         }
 818
 819         _unlock_rx_hashtbl(bond);
 820 }
 821
 822 /*********************** tlb/rlb shared functions *********************/
 823
 824 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
 825 {
 826         struct bonding *bond = bond_get_bond_by_slave(slave);
 827         struct learning_pkt pkt;
 828         int size = sizeof(struct learning_pkt);
 829         int i;
 830
 831         memset(&pkt, 0, size);
 832         memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
 833         memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
 834         pkt.type = cpu_to_be16(ETH_P_LOOP);
 835
 836         for (i = 0; i < MAX_LP_BURST; i++) {
 837                 struct sk_buff *skb;
 838                 char *data;
 839
 840                 skb = dev_alloc_skb(size);
 841                 if (!skb) {
 842                         return;
 843                 }
 844
 845                 data = skb_put(skb, size);
 846                 memcpy(data, &pkt, size);
 847
 848                 skb_reset_mac_header(skb);
 849                 skb->network_header = skb->mac_header + ETH_HLEN;
 850                 skb->protocol = pkt.type;
 851                 skb->priority = TC_PRIO_CONTROL;
 852                 skb->dev = slave->dev;
 853
 854                 if (bond->vlgrp) {
 855                         struct vlan_entry *vlan;
 856
 857                         vlan = bond_next_vlan(bond,
 858                                               bond->alb_info.current_alb_vlan);
 859
 860                         bond->alb_info.current_alb_vlan = vlan;
 861                         if (!vlan) {
 862                                 kfree_skb(skb);
 863                                 continue;
 864                         }
 865
 866                         skb = vlan_put_tag(skb, vlan->vlan_id);
 867                         if (!skb) {
 868                                 pr_err("%s: Error: failed to insert VLAN tag\n",
 869                                        bond->dev->name);
 870                                 continue;
 871                         }
 872                 }
 873
 874                 dev_queue_xmit(skb);
 875         }
 876 }
 877
 878 /* hw is a boolean parameter that determines whether we should try and
 879  * set the hw address of the device as well as the hw address of the
 880  * net_device
 881  */
 882 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[], int hw)
 883 {
 884         struct net_device *dev = slave->dev;
 885         struct sockaddr s_addr;
 886
 887         if (!hw) {
 888                 memcpy(dev->dev_addr, addr, dev->addr_len);
 889                 return 0;
 890         }
 891
 892         /* for rlb each slave must have a unique hw mac addresses so that */
 893         /* each slave will receive packets destined to a different mac */
 894         memcpy(s_addr.sa_data, addr, dev->addr_len);
 895         s_addr.sa_family = dev->type;
 896         if (dev_set_mac_address(dev, &s_addr)) {
 897                 pr_err("%s: Error: dev_set_mac_address of dev %s failed!\n"
 898                        "ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
 899                        dev->master->name, dev->name);
 900                 return -EOPNOTSUPP;
 901         }
 902         return 0;
 903 }
 904
 905 /*
 906  * Swap MAC addresses between two slaves.
 907  *
 908  * Called with RTNL held, and no other locks.
 909  *
 910  */
 911
 912 static void alb_swap_mac_addr(struct bonding *bond, struct slave *slave1, struct slave *slave2)
 913 {
 914         u8 tmp_mac_addr[ETH_ALEN];
 915
 916         memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
 917         alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr, bond->alb_info.rlb_enabled);
 918         alb_set_slave_mac_addr(slave2, tmp_mac_addr, bond->alb_info.rlb_enabled);
 919
 920 }
 921
 922 /*
 923  * Send learning packets after MAC address swap.
 924  *
 925  * Called with RTNL and no other locks
 926  */
 927 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
 928                                 struct slave *slave2)
 929 {
 930         int slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
 931         struct slave *disabled_slave = NULL;
 932
 933         ASSERT_RTNL();
 934
 935         /* fasten the change in the switch */
 936         if (SLAVE_IS_OK(slave1)) {
 937                 alb_send_learning_packets(slave1, slave1->dev->dev_addr);
 938                 if (bond->alb_info.rlb_enabled) {
 939                         /* inform the clients that the mac address
 940                          * has changed
 941                          */
 942                         rlb_req_update_slave_clients(bond, slave1);
 943                 }
 944         } else {
 945                 disabled_slave = slave1;
 946         }
 947
 948         if (SLAVE_IS_OK(slave2)) {
 949                 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
 950                 if (bond->alb_info.rlb_enabled) {
 951                         /* inform the clients that the mac address
 952                          * has changed
 953                          */
 954                         rlb_req_update_slave_clients(bond, slave2);
 955                 }
 956         } else {
 957                 disabled_slave = slave2;
 958         }
 959
 960         if (bond->alb_info.rlb_enabled && slaves_state_differ) {
 961                 /* A disabled slave was assigned an active mac addr */
 962                 rlb_teach_disabled_mac_on_primary(bond,
 963                                                   disabled_slave->dev->dev_addr);
 964         }
 965 }
 966
 967 /**
 968  * alb_change_hw_addr_on_detach
 969  * @bond: bonding we're working on
 970  * @slave: the slave that was just detached
 971  *
 972  * We assume that @slave was already detached from the slave list.
 973  *
 974  * If @slave's permanent hw address is different both from its current
 975  * address and from @bond's address, then somewhere in the bond there's
 976  * a slave that has @slave's permanet address as its current address.
 977  * We'll make sure that that slave no longer uses @slave's permanent address.
 978  *
 979  * Caller must hold RTNL and no other locks
 980  */
 981 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
 982 {
 983         int perm_curr_diff;
 984         int perm_bond_diff;
 985
 986         perm_curr_diff = compare_ether_addr_64bits(slave->perm_hwaddr,
 987                                                    slave->dev->dev_addr);
 988         perm_bond_diff = compare_ether_addr_64bits(slave->perm_hwaddr,
 989                                                    bond->dev->dev_addr);
 990
 991         if (perm_curr_diff && perm_bond_diff) {
 992                 struct slave *tmp_slave;
 993                 int i, found = 0;
 994
 995                 bond_for_each_slave(bond, tmp_slave, i) {
 996                         if (!compare_ether_addr_64bits(slave->perm_hwaddr,
 997                                                        tmp_slave->dev->dev_addr)) {
 998                                 found = 1;
 999                                 break;
1000                         }
1001                 }
1002
1003                 if (found) {
1004                         /* locking: needs RTNL and nothing else */
1005                         alb_swap_mac_addr(bond, slave, tmp_slave);
1006                         alb_fasten_mac_swap(bond, slave, tmp_slave);
1007                 }
1008         }
1009 }
1010
1011 /**
1012  * alb_handle_addr_collision_on_attach
1013  * @bond: bonding we're working on
1014  * @slave: the slave that was just attached
1015  *
1016  * checks uniqueness of slave's mac address and handles the case the
1017  * new slave uses the bonds mac address.
1018  *
1019  * If the permanent hw address of @slave is @bond's hw address, we need to
1020  * find a different hw address to give @slave, that isn't in use by any other
1021  * slave in the bond. This address must be, of course, one of the permanent
1022  * addresses of the other slaves.
1023  *
1024  * We go over the slave list, and for each slave there we compare its
1025  * permanent hw address with the current address of all the other slaves.
1026  * If no match was found, then we've found a slave with a permanent address
1027  * that isn't used by any other slave in the bond, so we can assign it to
1028  * @slave.
1029  *
1030  * assumption: this function is called before @slave is attached to the
1031  *             bond slave list.
1032  *
1033  * caller must hold the bond lock for write since the mac addresses are compared
1034  * and may be swapped.
1035  */
1036 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1037 {
1038         struct slave *tmp_slave1, *tmp_slave2, *free_mac_slave;
1039         struct slave *has_bond_addr = bond->curr_active_slave;
1040         int i, j, found = 0;
1041
1042         if (bond->slave_cnt == 0) {
1043                 /* this is the first slave */
1044                 return 0;
1045         }
1046
1047         /* if slave's mac address differs from bond's mac address
1048          * check uniqueness of slave's mac address against the other
1049          * slaves in the bond.
1050          */
1051         if (compare_ether_addr_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1052                 bond_for_each_slave(bond, tmp_slave1, i) {
1053                         if (!compare_ether_addr_64bits(tmp_slave1->dev->dev_addr,
1054                                                        slave->dev->dev_addr)) {
1055                                 found = 1;
1056                                 break;
1057                         }
1058                 }
1059
1060                 if (!found)
1061                         return 0;
1062
1063                 /* Try setting slave mac to bond address and fall-through
1064                    to code handling that situation below... */
1065                 alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1066                                        bond->alb_info.rlb_enabled);
1067         }
1068
1069         /* The slave's address is equal to the address of the bond.
1070          * Search for a spare address in the bond for this slave.
1071          */
1072         free_mac_slave = NULL;
1073
1074         bond_for_each_slave(bond, tmp_slave1, i) {
1075                 found = 0;
1076                 bond_for_each_slave(bond, tmp_slave2, j) {
1077                         if (!compare_ether_addr_64bits(tmp_slave1->perm_hwaddr,
1078                                                        tmp_slave2->dev->dev_addr)) {
1079                                 found = 1;
1080                                 break;
1081                         }
1082                 }
1083
1084                 if (!found) {
1085                         /* no slave has tmp_slave1's perm addr
1086                          * as its curr addr
1087                          */
1088                         free_mac_slave = tmp_slave1;
1089                         break;
1090                 }
1091
1092                 if (!has_bond_addr) {
1093                         if (!compare_ether_addr_64bits(tmp_slave1->dev->dev_addr,
1094                                                        bond->dev->dev_addr)) {
1095
1096                                 has_bond_addr = tmp_slave1;
1097                         }
1098                 }
1099         }
1100
1101         if (free_mac_slave) {
1102                 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1103                                        bond->alb_info.rlb_enabled);
1104
1105                 pr_warning("%s: Warning: the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1106                            bond->dev->name, slave->dev->name,
1107                            free_mac_slave->dev->name);
1108
1109         } else if (has_bond_addr) {
1110                 pr_err("%s: Error: the hw address of slave %s is in use by the bond; couldn't find a slave with a free hw address to give it (this should not have happened)\n",
1111                        bond->dev->name, slave->dev->name);
1112                 return -EFAULT;
1113         }
1114
1115         return 0;
1116 }
1117
1118 /**
1119  * alb_set_mac_address
1120  * @bond:
1121  * @addr:
1122  *
1123  * In TLB mode all slaves are configured to the bond's hw address, but set
1124  * their dev_addr field to different addresses (based on their permanent hw
1125  * addresses).
1126  *
1127  * For each slave, this function sets the interface to the new address and then
1128  * changes its dev_addr field to its previous value.
1129  *
1130  * Unwinding assumes bond's mac address has not yet changed.
1131  */
1132 static int alb_set_mac_address(struct bonding *bond, void *addr)
1133 {
1134         struct sockaddr sa;
1135         struct slave *slave, *stop_at;
1136         char tmp_addr[ETH_ALEN];
1137         int res;
1138         int i;
1139
1140         if (bond->alb_info.rlb_enabled) {
1141                 return 0;
1142         }
1143
1144         bond_for_each_slave(bond, slave, i) {
1145                 /* save net_device's current hw address */
1146                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1147
1148                 res = dev_set_mac_address(slave->dev, addr);
1149
1150                 /* restore net_device's hw address */
1151                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1152
1153                 if (res)
1154                         goto unwind;
1155         }
1156
1157         return 0;
1158
1159 unwind:
1160         memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1161         sa.sa_family = bond->dev->type;
1162
1163         /* unwind from head to the slave that failed */
1164         stop_at = slave;
1165         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
1166                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1167                 dev_set_mac_address(slave->dev, &sa);
1168                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1169         }
1170
1171         return res;
1172 }
1173
1174 /************************ exported alb funcions ************************/
1175
1176 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1177 {
1178         int res;
1179
1180         res = tlb_initialize(bond);
1181         if (res) {
1182                 return res;
1183         }
1184
1185         if (rlb_enabled) {
1186                 bond->alb_info.rlb_enabled = 1;
1187                 /* initialize rlb */
1188                 res = rlb_initialize(bond);
1189                 if (res) {
1190                         tlb_deinitialize(bond);
1191                         return res;
1192                 }
1193         } else {
1194                 bond->alb_info.rlb_enabled = 0;
1195         }
1196
1197         return 0;
1198 }
1199
1200 void bond_alb_deinitialize(struct bonding *bond)
1201 {
1202         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1203
1204         tlb_deinitialize(bond);
1205
1206         if (bond_info->rlb_enabled) {
1207                 rlb_deinitialize(bond);
1208         }
1209 }
1210
1211 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1212 {
1213         struct bonding *bond = netdev_priv(bond_dev);
1214         struct ethhdr *eth_data;
1215         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1216         struct slave *tx_slave = NULL;
1217         static const __be32 ip_bcast = htonl(0xffffffff);
1218         int hash_size = 0;
1219         int do_tx_balance = 1;
1220         u32 hash_index = 0;
1221         const u8 *hash_start = NULL;
1222         int res = 1;
1223         struct ipv6hdr *ip6hdr;
1224
1225         skb_reset_mac_header(skb);
1226         eth_data = eth_hdr(skb);
1227
1228         /* make sure that the curr_active_slave do not change during tx
1229          */
1230         read_lock(&bond->curr_slave_lock);
1231
1232         switch (ntohs(skb->protocol)) {
1233         case ETH_P_IP: {
1234                 const struct iphdr *iph = ip_hdr(skb);
1235
1236                 if (!compare_ether_addr_64bits(eth_data->h_dest, mac_bcast) ||
1237                     (iph->daddr == ip_bcast) ||
1238                     (iph->protocol == IPPROTO_IGMP)) {
1239                         do_tx_balance = 0;
1240                         break;
1241                 }
1242                 hash_start = (char *)&(iph->daddr);
1243                 hash_size = sizeof(iph->daddr);
1244         }
1245                 break;
1246         case ETH_P_IPV6:
1247                 /* IPv6 doesn't really use broadcast mac address, but leave
1248                  * that here just in case.
1249                  */
1250                 if (!compare_ether_addr_64bits(eth_data->h_dest, mac_bcast)) {
1251                         do_tx_balance = 0;
1252                         break;
1253                 }
1254
1255                 /* IPv6 uses all-nodes multicast as an equivalent to
1256                  * broadcasts in IPv4.
1257                  */
1258                 if (!compare_ether_addr_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1259                         do_tx_balance = 0;
1260                         break;
1261                 }
1262
1263                 /* Additianally, DAD probes should not be tx-balanced as that
1264                  * will lead to false positives for duplicate addresses and
1265                  * prevent address configuration from working.
1266                  */
1267                 ip6hdr = ipv6_hdr(skb);
1268                 if (ipv6_addr_any(&ip6hdr->saddr)) {
1269                         do_tx_balance = 0;
1270                         break;
1271                 }
1272
1273                 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1274                 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1275                 break;
1276         case ETH_P_IPX:
1277                 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1278                         /* something is wrong with this packet */
1279                         do_tx_balance = 0;
1280                         break;
1281                 }
1282
1283                 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1284                         /* The only protocol worth balancing in
1285                          * this family since it has an "ARP" like
1286                          * mechanism
1287                          */
1288                         do_tx_balance = 0;
1289                         break;
1290                 }
1291
1292                 hash_start = (char*)eth_data->h_dest;
1293                 hash_size = ETH_ALEN;
1294                 break;
1295         case ETH_P_ARP:
1296                 do_tx_balance = 0;
1297                 if (bond_info->rlb_enabled) {
1298                         tx_slave = rlb_arp_xmit(skb, bond);
1299                 }
1300                 break;
1301         default:
1302                 do_tx_balance = 0;
1303                 break;
1304         }
1305
1306         if (do_tx_balance) {
1307                 hash_index = _simple_hash(hash_start, hash_size);
1308                 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1309         }
1310
1311         if (!tx_slave) {
1312                 /* unbalanced or unassigned, send through primary */
1313                 tx_slave = bond->curr_active_slave;
1314                 bond_info->unbalanced_load += skb->len;
1315         }
1316
1317         if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1318                 if (tx_slave != bond->curr_active_slave) {
1319                         memcpy(eth_data->h_source,
1320                                tx_slave->dev->dev_addr,
1321                                ETH_ALEN);
1322                 }
1323
1324                 res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1325         } else {
1326                 if (tx_slave) {
1327                         tlb_clear_slave(bond, tx_slave, 0);
1328                 }
1329         }
1330
1331         if (res) {
1332                 /* no suitable interface, frame not sent */
1333                 dev_kfree_skb(skb);
1334         }
1335         read_unlock(&bond->curr_slave_lock);
1336
1337         return NETDEV_TX_OK;
1338 }
1339
1340 void bond_alb_monitor(struct work_struct *work)
1341 {
1342         struct bonding *bond = container_of(work, struct bonding,
1343                                             alb_work.work);
1344         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1345         struct slave *slave;
1346         int i;
1347
1348         read_lock(&bond->lock);
1349
1350         if (bond->kill_timers) {
1351                 goto out;
1352         }
1353
1354         if (bond->slave_cnt == 0) {
1355                 bond_info->tx_rebalance_counter = 0;
1356                 bond_info->lp_counter = 0;
1357                 goto re_arm;
1358         }
1359
1360         bond_info->tx_rebalance_counter++;
1361         bond_info->lp_counter++;
1362
1363         /* send learning packets */
1364         if (bond_info->lp_counter >= BOND_ALB_LP_TICKS) {
1365                 /* change of curr_active_slave involves swapping of mac addresses.
1366                  * in order to avoid this swapping from happening while
1367                  * sending the learning packets, the curr_slave_lock must be held for
1368                  * read.
1369                  */
1370                 read_lock(&bond->curr_slave_lock);
1371
1372                 bond_for_each_slave(bond, slave, i) {
1373                         alb_send_learning_packets(slave, slave->dev->dev_addr);
1374                 }
1375
1376                 read_unlock(&bond->curr_slave_lock);
1377
1378                 bond_info->lp_counter = 0;
1379         }
1380
1381         /* rebalance tx traffic */
1382         if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1383
1384                 read_lock(&bond->curr_slave_lock);
1385
1386                 bond_for_each_slave(bond, slave, i) {
1387                         tlb_clear_slave(bond, slave, 1);
1388                         if (slave == bond->curr_active_slave) {
1389                                 SLAVE_TLB_INFO(slave).load =
1390                                         bond_info->unbalanced_load /
1391                                                 BOND_TLB_REBALANCE_INTERVAL;
1392                                 bond_info->unbalanced_load = 0;
1393                         }
1394                 }
1395
1396                 read_unlock(&bond->curr_slave_lock);
1397
1398                 bond_info->tx_rebalance_counter = 0;
1399         }
1400
1401         /* handle rlb stuff */
1402         if (bond_info->rlb_enabled) {
1403                 if (bond_info->primary_is_promisc &&
1404                     (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1405
1406                         /*
1407                          * dev_set_promiscuity requires rtnl and
1408                          * nothing else.
1409                          */
1410                         read_unlock(&bond->lock);
1411                         rtnl_lock();
1412
1413                         bond_info->rlb_promisc_timeout_counter = 0;
1414
1415                         /* If the primary was set to promiscuous mode
1416                          * because a slave was disabled then
1417                          * it can now leave promiscuous mode.
1418                          */
1419                         dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1420                         bond_info->primary_is_promisc = 0;
1421
1422                         rtnl_unlock();
1423                         read_lock(&bond->lock);
1424                 }
1425
1426                 if (bond_info->rlb_rebalance) {
1427                         bond_info->rlb_rebalance = 0;
1428                         rlb_rebalance(bond);
1429                 }
1430
1431                 /* check if clients need updating */
1432                 if (bond_info->rx_ntt) {
1433                         if (bond_info->rlb_update_delay_counter) {
1434                                 --bond_info->rlb_update_delay_counter;
1435                         } else {
1436                                 rlb_update_rx_clients(bond);
1437                                 if (bond_info->rlb_update_retry_counter) {
1438                                         --bond_info->rlb_update_retry_counter;
1439                                 } else {
1440                                         bond_info->rx_ntt = 0;
1441                                 }
1442                         }
1443                 }
1444         }
1445
1446 re_arm:
1447         queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1448 out:
1449         read_unlock(&bond->lock);
1450 }
1451
1452 /* assumption: called before the slave is attached to the bond
1453  * and not locked by the bond lock
1454  */
1455 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1456 {
1457         int res;
1458
1459         res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1460                                      bond->alb_info.rlb_enabled);
1461         if (res) {
1462                 return res;
1463         }
1464
1465         /* caller must hold the bond lock for write since the mac addresses
1466          * are compared and may be swapped.
1467          */
1468         read_lock(&bond->lock);
1469
1470         res = alb_handle_addr_collision_on_attach(bond, slave);
1471
1472         read_unlock(&bond->lock);
1473
1474         if (res) {
1475                 return res;
1476         }
1477
1478         tlb_init_slave(slave);
1479
1480         /* order a rebalance ASAP */
1481         bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1482
1483         if (bond->alb_info.rlb_enabled) {
1484                 bond->alb_info.rlb_rebalance = 1;
1485         }
1486
1487         return 0;
1488 }
1489
1490 /*
1491  * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1492  * if necessary.
1493  *
1494  * Caller must hold RTNL and no other locks
1495  */
1496 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1497 {
1498         if (bond->slave_cnt > 1) {
1499                 alb_change_hw_addr_on_detach(bond, slave);
1500         }
1501
1502         tlb_clear_slave(bond, slave, 0);
1503
1504         if (bond->alb_info.rlb_enabled) {
1505                 bond->alb_info.next_rx_slave = NULL;
1506                 rlb_clear_slave(bond, slave);
1507         }
1508 }
1509
1510 /* Caller must hold bond lock for read */
1511 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1512 {
1513         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1514
1515         if (link == BOND_LINK_DOWN) {
1516                 tlb_clear_slave(bond, slave, 0);
1517                 if (bond->alb_info.rlb_enabled) {
1518                         rlb_clear_slave(bond, slave);
1519                 }
1520         } else if (link == BOND_LINK_UP) {
1521                 /* order a rebalance ASAP */
1522                 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1523                 if (bond->alb_info.rlb_enabled) {
1524                         bond->alb_info.rlb_rebalance = 1;
1525                         /* If the updelay module parameter is smaller than the
1526                          * forwarding delay of the switch the rebalance will
1527                          * not work because the rebalance arp replies will
1528                          * not be forwarded to the clients..
1529                          */
1530                 }
1531         }
1532 }
1533
1534 /**
1535  * bond_alb_handle_active_change - assign new curr_active_slave
1536  * @bond: our bonding struct
1537  * @new_slave: new slave to assign
1538  *
1539  * Set the bond->curr_active_slave to @new_slave and handle
1540  * mac address swapping and promiscuity changes as needed.
1541  *
1542  * If new_slave is NULL, caller must hold curr_slave_lock or
1543  * bond->lock for write.
1544  *
1545  * If new_slave is not NULL, caller must hold RTNL, bond->lock for
1546  * read and curr_slave_lock for write.  Processing here may sleep, so
1547  * no other locks may be held.
1548  */
1549 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1550         __releases(&bond->curr_slave_lock)
1551         __releases(&bond->lock)
1552         __acquires(&bond->lock)
1553         __acquires(&bond->curr_slave_lock)
1554 {
1555         struct slave *swap_slave;
1556         int i;
1557
1558         if (bond->curr_active_slave == new_slave) {
1559                 return;
1560         }
1561
1562         if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1563                 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1564                 bond->alb_info.primary_is_promisc = 0;
1565                 bond->alb_info.rlb_promisc_timeout_counter = 0;
1566         }
1567
1568         swap_slave = bond->curr_active_slave;
1569         bond->curr_active_slave = new_slave;
1570
1571         if (!new_slave || (bond->slave_cnt == 0)) {
1572                 return;
1573         }
1574
1575         /* set the new curr_active_slave to the bonds mac address
1576          * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1577          */
1578         if (!swap_slave) {
1579                 struct slave *tmp_slave;
1580                 /* find slave that is holding the bond's mac address */
1581                 bond_for_each_slave(bond, tmp_slave, i) {
1582                         if (!compare_ether_addr_64bits(tmp_slave->dev->dev_addr,
1583                                                        bond->dev->dev_addr)) {
1584                                 swap_slave = tmp_slave;
1585                                 break;
1586                         }
1587                 }
1588         }
1589
1590         /*
1591          * Arrange for swap_slave and new_slave to temporarily be
1592          * ignored so we can mess with their MAC addresses without
1593          * fear of interference from transmit activity.
1594          */
1595         if (swap_slave) {
1596                 tlb_clear_slave(bond, swap_slave, 1);
1597         }
1598         tlb_clear_slave(bond, new_slave, 1);
1599
1600         write_unlock_bh(&bond->curr_slave_lock);
1601         read_unlock(&bond->lock);
1602
1603         ASSERT_RTNL();
1604
1605         /* curr_active_slave must be set before calling alb_swap_mac_addr */
1606         if (swap_slave) {
1607                 /* swap mac address */
1608                 alb_swap_mac_addr(bond, swap_slave, new_slave);
1609         } else {
1610                 /* set the new_slave to the bond mac address */
1611                 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1612                                        bond->alb_info.rlb_enabled);
1613         }
1614
1615         if (swap_slave) {
1616                 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1617                 read_lock(&bond->lock);
1618         } else {
1619                 read_lock(&bond->lock);
1620                 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1621         }
1622
1623         write_lock_bh(&bond->curr_slave_lock);
1624 }
1625
1626 /*
1627  * Called with RTNL
1628  */
1629 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1630         __acquires(&bond->lock)
1631         __releases(&bond->lock)
1632 {
1633         struct bonding *bond = netdev_priv(bond_dev);
1634         struct sockaddr *sa = addr;
1635         struct slave *slave, *swap_slave;
1636         int res;
1637         int i;
1638
1639         if (!is_valid_ether_addr(sa->sa_data)) {
1640                 return -EADDRNOTAVAIL;
1641         }
1642
1643         res = alb_set_mac_address(bond, addr);
1644         if (res) {
1645                 return res;
1646         }
1647
1648         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1649
1650         /* If there is no curr_active_slave there is nothing else to do.
1651          * Otherwise we'll need to pass the new address to it and handle
1652          * duplications.
1653          */
1654         if (!bond->curr_active_slave) {
1655                 return 0;
1656         }
1657
1658         swap_slave = NULL;
1659
1660         bond_for_each_slave(bond, slave, i) {
1661                 if (!compare_ether_addr_64bits(slave->dev->dev_addr,
1662                                                bond_dev->dev_addr)) {
1663                         swap_slave = slave;
1664                         break;
1665                 }
1666         }
1667
1668         if (swap_slave) {
1669                 alb_swap_mac_addr(bond, swap_slave, bond->curr_active_slave);
1670                 alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave);
1671         } else {
1672                 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr,
1673                                        bond->alb_info.rlb_enabled);
1674
1675                 read_lock(&bond->lock);
1676                 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1677                 if (bond->alb_info.rlb_enabled) {
1678                         /* inform clients mac address has changed */
1679                         rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1680                 }
1681                 read_unlock(&bond->lock);
1682         }
1683
1684         return 0;
1685 }
1686
1687 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1688 {
1689         if (bond->alb_info.current_alb_vlan &&
1690             (bond->alb_info.current_alb_vlan->vlan_id == vlan_id)) {
1691                 bond->alb_info.current_alb_vlan = NULL;
1692         }
1693
1694         if (bond->alb_info.rlb_enabled) {
1695                 rlb_clear_vlan(bond, vlan_id);
1696         }
1697 }
1698