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 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp);
  88 static void rlb_src_unlink(struct bonding *bond, u32 index);
  89 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash,
  90                          u32 ip_dst_hash);
  91
  92 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
  93 {
  94         int i;
  95         u8 hash = 0;
  96
  97         for (i = 0; i < hash_size; i++) {
  98                 hash ^= hash_start[i];
  99         }
 100
 101         return hash;
 102 }
 103
 104 /*********************** tlb specific functions ***************************/
 105
 106 static inline void _lock_tx_hashtbl_bh(struct bonding *bond)
 107 {
 108         spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 109 }
 110
 111 static inline void _unlock_tx_hashtbl_bh(struct bonding *bond)
 112 {
 113         spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 114 }
 115
 116 static inline void _lock_tx_hashtbl(struct bonding *bond)
 117 {
 118         spin_lock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 119 }
 120
 121 static inline void _unlock_tx_hashtbl(struct bonding *bond)
 122 {
 123         spin_unlock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 124 }
 125
 126 /* Caller must hold tx_hashtbl lock */
 127 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
 128 {
 129         if (save_load) {
 130                 entry->load_history = 1 + entry->tx_bytes /
 131                                       BOND_TLB_REBALANCE_INTERVAL;
 132                 entry->tx_bytes = 0;
 133         }
 134
 135         entry->tx_slave = NULL;
 136         entry->next = TLB_NULL_INDEX;
 137         entry->prev = TLB_NULL_INDEX;
 138 }
 139
 140 static inline void tlb_init_slave(struct slave *slave)
 141 {
 142         SLAVE_TLB_INFO(slave).load = 0;
 143         SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
 144 }
 145
 146 /* Caller must hold bond lock for read, BH disabled */
 147 static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
 148                          int save_load)
 149 {
 150         struct tlb_client_info *tx_hash_table;
 151         u32 index;
 152
 153         /* clear slave from tx_hashtbl */
 154         tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
 155
 156         /* skip this if we've already freed the tx hash table */
 157         if (tx_hash_table) {
 158                 index = SLAVE_TLB_INFO(slave).head;
 159                 while (index != TLB_NULL_INDEX) {
 160                         u32 next_index = tx_hash_table[index].next;
 161                         tlb_init_table_entry(&tx_hash_table[index], save_load);
 162                         index = next_index;
 163                 }
 164         }
 165
 166         tlb_init_slave(slave);
 167 }
 168
 169 /* Caller must hold bond lock for read */
 170 static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
 171                          int save_load)
 172 {
 173         _lock_tx_hashtbl_bh(bond);
 174         __tlb_clear_slave(bond, slave, save_load);
 175         _unlock_tx_hashtbl_bh(bond);
 176 }
 177
 178 /* Must be called before starting the monitor timer */
 179 static int tlb_initialize(struct bonding *bond)
 180 {
 181         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 182         int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 183         struct tlb_client_info *new_hashtbl;
 184         int i;
 185
 186         new_hashtbl = kzalloc(size, GFP_KERNEL);
 187         if (!new_hashtbl)
 188                 return -1;
 189
 190         _lock_tx_hashtbl_bh(bond);
 191
 192         bond_info->tx_hashtbl = new_hashtbl;
 193
 194         for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
 195                 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
 196         }
 197
 198         _unlock_tx_hashtbl_bh(bond);
 199
 200         return 0;
 201 }
 202
 203 /* Must be called only after all slaves have been released */
 204 static void tlb_deinitialize(struct bonding *bond)
 205 {
 206         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 207
 208         _lock_tx_hashtbl_bh(bond);
 209
 210         kfree(bond_info->tx_hashtbl);
 211         bond_info->tx_hashtbl = NULL;
 212
 213         _unlock_tx_hashtbl_bh(bond);
 214 }
 215
 216 static long long compute_gap(struct slave *slave)
 217 {
 218         return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
 219                (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
 220 }
 221
 222 /* Caller must hold bond lock for read */
 223 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
 224 {
 225         struct slave *slave, *least_loaded;
 226         struct list_head *iter;
 227         long long max_gap;
 228
 229         least_loaded = NULL;
 230         max_gap = LLONG_MIN;
 231
 232         /* Find the slave with the largest gap */
 233         bond_for_each_slave_rcu(bond, slave, iter) {
 234                 if (SLAVE_IS_OK(slave)) {
 235                         long long gap = compute_gap(slave);
 236
 237                         if (max_gap < gap) {
 238                                 least_loaded = slave;
 239                                 max_gap = gap;
 240                         }
 241                 }
 242         }
 243
 244         return least_loaded;
 245 }
 246
 247 static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
 248                                                 u32 skb_len)
 249 {
 250         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 251         struct tlb_client_info *hash_table;
 252         struct slave *assigned_slave;
 253
 254         hash_table = bond_info->tx_hashtbl;
 255         assigned_slave = hash_table[hash_index].tx_slave;
 256         if (!assigned_slave) {
 257                 assigned_slave = tlb_get_least_loaded_slave(bond);
 258
 259                 if (assigned_slave) {
 260                         struct tlb_slave_info *slave_info =
 261                                 &(SLAVE_TLB_INFO(assigned_slave));
 262                         u32 next_index = slave_info->head;
 263
 264                         hash_table[hash_index].tx_slave = assigned_slave;
 265                         hash_table[hash_index].next = next_index;
 266                         hash_table[hash_index].prev = TLB_NULL_INDEX;
 267
 268                         if (next_index != TLB_NULL_INDEX) {
 269                                 hash_table[next_index].prev = hash_index;
 270                         }
 271
 272                         slave_info->head = hash_index;
 273                         slave_info->load +=
 274                                 hash_table[hash_index].load_history;
 275                 }
 276         }
 277
 278         if (assigned_slave) {
 279                 hash_table[hash_index].tx_bytes += skb_len;
 280         }
 281
 282         return assigned_slave;
 283 }
 284
 285 /* Caller must hold bond lock for read */
 286 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
 287                                         u32 skb_len)
 288 {
 289         struct slave *tx_slave;
 290         /*
 291          * We don't need to disable softirq here, becase
 292          * tlb_choose_channel() is only called by bond_alb_xmit()
 293          * which already has softirq disabled.
 294          */
 295         _lock_tx_hashtbl(bond);
 296         tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
 297         _unlock_tx_hashtbl(bond);
 298         return tx_slave;
 299 }
 300
 301 /*********************** rlb specific functions ***************************/
 302 static inline void _lock_rx_hashtbl_bh(struct bonding *bond)
 303 {
 304         spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 305 }
 306
 307 static inline void _unlock_rx_hashtbl_bh(struct bonding *bond)
 308 {
 309         spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 310 }
 311
 312 static inline void _lock_rx_hashtbl(struct bonding *bond)
 313 {
 314         spin_lock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 315 }
 316
 317 static inline void _unlock_rx_hashtbl(struct bonding *bond)
 318 {
 319         spin_unlock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 320 }
 321
 322 /* when an ARP REPLY is received from a client update its info
 323  * in the rx_hashtbl
 324  */
 325 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 326 {
 327         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 328         struct rlb_client_info *client_info;
 329         u32 hash_index;
 330
 331         _lock_rx_hashtbl_bh(bond);
 332
 333         hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
 334         client_info = &(bond_info->rx_hashtbl[hash_index]);
 335
 336         if ((client_info->assigned) &&
 337             (client_info->ip_src == arp->ip_dst) &&
 338             (client_info->ip_dst == arp->ip_src) &&
 339             (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) {
 340                 /* update the clients MAC address */
 341                 memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
 342                 client_info->ntt = 1;
 343                 bond_info->rx_ntt = 1;
 344         }
 345
 346         _unlock_rx_hashtbl_bh(bond);
 347 }
 348
 349 static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
 350                         struct slave *slave)
 351 {
 352         struct arp_pkt *arp, _arp;
 353
 354         if (skb->protocol != cpu_to_be16(ETH_P_ARP))
 355                 goto out;
 356
 357         arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp);
 358         if (!arp)
 359                 goto out;
 360
 361         /* We received an ARP from arp->ip_src.
 362          * We might have used this IP address previously (on the bonding host
 363          * itself or on a system that is bridged together with the bond).
 364          * However, if arp->mac_src is different than what is stored in
 365          * rx_hashtbl, some other host is now using the IP and we must prevent
 366          * sending out client updates with this IP address and the old MAC
 367          * address.
 368          * Clean up all hash table entries that have this address as ip_src but
 369          * have a different mac_src.
 370          */
 371         rlb_purge_src_ip(bond, arp);
 372
 373         if (arp->op_code == htons(ARPOP_REPLY)) {
 374                 /* update rx hash table for this ARP */
 375                 rlb_update_entry_from_arp(bond, arp);
 376                 pr_debug("Server received an ARP Reply from client\n");
 377         }
 378 out:
 379         return RX_HANDLER_ANOTHER;
 380 }
 381
 382 /* Caller must hold bond lock for read */
 383 static struct slave *rlb_next_rx_slave(struct bonding *bond)
 384 {
 385         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 386         struct slave *before = NULL, *rx_slave = NULL, *slave;
 387         struct list_head *iter;
 388         bool found = false;
 389
 390         bond_for_each_slave(bond, slave, iter) {
 391                 if (!SLAVE_IS_OK(slave))
 392                         continue;
 393                 if (!found) {
 394                         if (!before || before->speed < slave->speed)
 395                                 before = slave;
 396                 } else {
 397                         if (!rx_slave || rx_slave->speed < slave->speed)
 398                                 rx_slave = slave;
 399                 }
 400                 if (slave == bond_info->rx_slave)
 401                         found = true;
 402         }
 403         /* we didn't find anything after the current or we have something
 404          * better before and up to the current slave
 405          */
 406         if (!rx_slave || (before && rx_slave->speed < before->speed))
 407                 rx_slave = before;
 408
 409         if (rx_slave)
 410                 bond_info->rx_slave = rx_slave;
 411
 412         return rx_slave;
 413 }
 414
 415 /* Caller must hold rcu_read_lock() for read */
 416 static struct slave *__rlb_next_rx_slave(struct bonding *bond)
 417 {
 418         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 419         struct slave *before = NULL, *rx_slave = NULL, *slave;
 420         struct list_head *iter;
 421         bool found = false;
 422
 423         bond_for_each_slave_rcu(bond, slave, iter) {
 424                 if (!SLAVE_IS_OK(slave))
 425                         continue;
 426                 if (!found) {
 427                         if (!before || before->speed < slave->speed)
 428                                 before = slave;
 429                 } else {
 430                         if (!rx_slave || rx_slave->speed < slave->speed)
 431                                 rx_slave = slave;
 432                 }
 433                 if (slave == bond_info->rx_slave)
 434                         found = true;
 435         }
 436         /* we didn't find anything after the current or we have something
 437          * better before and up to the current slave
 438          */
 439         if (!rx_slave || (before && rx_slave->speed < before->speed))
 440                 rx_slave = before;
 441
 442         if (rx_slave)
 443                 bond_info->rx_slave = rx_slave;
 444
 445         return rx_slave;
 446 }
 447
 448 /* teach the switch the mac of a disabled slave
 449  * on the primary for fault tolerance
 450  *
 451  * Caller must hold bond->curr_slave_lock for write or bond lock for write
 452  */
 453 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
 454 {
 455         if (!bond->curr_active_slave) {
 456                 return;
 457         }
 458
 459         if (!bond->alb_info.primary_is_promisc) {
 460                 if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1))
 461                         bond->alb_info.primary_is_promisc = 1;
 462                 else
 463                         bond->alb_info.primary_is_promisc = 0;
 464         }
 465
 466         bond->alb_info.rlb_promisc_timeout_counter = 0;
 467
 468         alb_send_learning_packets(bond->curr_active_slave, addr);
 469 }
 470
 471 /* slave being removed should not be active at this point
 472  *
 473  * Caller must hold bond lock for read
 474  */
 475 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 476 {
 477         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 478         struct rlb_client_info *rx_hash_table;
 479         u32 index, next_index;
 480
 481         /* clear slave from rx_hashtbl */
 482         _lock_rx_hashtbl_bh(bond);
 483
 484         rx_hash_table = bond_info->rx_hashtbl;
 485         index = bond_info->rx_hashtbl_used_head;
 486         for (; index != RLB_NULL_INDEX; index = next_index) {
 487                 next_index = rx_hash_table[index].used_next;
 488                 if (rx_hash_table[index].slave == slave) {
 489                         struct slave *assigned_slave = rlb_next_rx_slave(bond);
 490
 491                         if (assigned_slave) {
 492                                 rx_hash_table[index].slave = assigned_slave;
 493                                 if (!ether_addr_equal_64bits(rx_hash_table[index].mac_dst,
 494                                                              mac_bcast)) {
 495                                         bond_info->rx_hashtbl[index].ntt = 1;
 496                                         bond_info->rx_ntt = 1;
 497                                         /* A slave has been removed from the
 498                                          * table because it is either disabled
 499                                          * or being released. We must retry the
 500                                          * update to avoid clients from not
 501                                          * being updated & disconnecting when
 502                                          * there is stress
 503                                          */
 504                                         bond_info->rlb_update_retry_counter =
 505                                                 RLB_UPDATE_RETRY;
 506                                 }
 507                         } else {  /* there is no active slave */
 508                                 rx_hash_table[index].slave = NULL;
 509                         }
 510                 }
 511         }
 512
 513         _unlock_rx_hashtbl_bh(bond);
 514
 515         write_lock_bh(&bond->curr_slave_lock);
 516
 517         if (slave != bond->curr_active_slave) {
 518                 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 519         }
 520
 521         write_unlock_bh(&bond->curr_slave_lock);
 522 }
 523
 524 static void rlb_update_client(struct rlb_client_info *client_info)
 525 {
 526         int i;
 527
 528         if (!client_info->slave) {
 529                 return;
 530         }
 531
 532         for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 533                 struct sk_buff *skb;
 534
 535                 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 536                                  client_info->ip_dst,
 537                                  client_info->slave->dev,
 538                                  client_info->ip_src,
 539                                  client_info->mac_dst,
 540                                  client_info->slave->dev->dev_addr,
 541                                  client_info->mac_dst);
 542                 if (!skb) {
 543                         pr_err("%s: Error: failed to create an ARP packet\n",
 544                                client_info->slave->bond->dev->name);
 545                         continue;
 546                 }
 547
 548                 skb->dev = client_info->slave->dev;
 549
 550                 if (client_info->vlan_id) {
 551                         skb = vlan_put_tag(skb, htons(ETH_P_8021Q), client_info->vlan_id);
 552                         if (!skb) {
 553                                 pr_err("%s: Error: failed to insert VLAN tag\n",
 554                                        client_info->slave->bond->dev->name);
 555                                 continue;
 556                         }
 557                 }
 558
 559                 arp_xmit(skb);
 560         }
 561 }
 562
 563 /* sends ARP REPLIES that update the clients that need updating */
 564 static void rlb_update_rx_clients(struct bonding *bond)
 565 {
 566         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 567         struct rlb_client_info *client_info;
 568         u32 hash_index;
 569
 570         _lock_rx_hashtbl_bh(bond);
 571
 572         hash_index = bond_info->rx_hashtbl_used_head;
 573         for (; hash_index != RLB_NULL_INDEX;
 574              hash_index = client_info->used_next) {
 575                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 576                 if (client_info->ntt) {
 577                         rlb_update_client(client_info);
 578                         if (bond_info->rlb_update_retry_counter == 0) {
 579                                 client_info->ntt = 0;
 580                         }
 581                 }
 582         }
 583
 584         /* do not update the entries again until this counter is zero so that
 585          * not to confuse the clients.
 586          */
 587         bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 588
 589         _unlock_rx_hashtbl_bh(bond);
 590 }
 591
 592 /* The slave was assigned a new mac address - update the clients */
 593 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 594 {
 595         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 596         struct rlb_client_info *client_info;
 597         int ntt = 0;
 598         u32 hash_index;
 599
 600         _lock_rx_hashtbl_bh(bond);
 601
 602         hash_index = bond_info->rx_hashtbl_used_head;
 603         for (; hash_index != RLB_NULL_INDEX;
 604              hash_index = client_info->used_next) {
 605                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 606
 607                 if ((client_info->slave == slave) &&
 608                     !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
 609                         client_info->ntt = 1;
 610                         ntt = 1;
 611                 }
 612         }
 613
 614         // update the team's flag only after the whole iteration
 615         if (ntt) {
 616                 bond_info->rx_ntt = 1;
 617                 //fasten the change
 618                 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 619         }
 620
 621         _unlock_rx_hashtbl_bh(bond);
 622 }
 623
 624 /* mark all clients using src_ip to be updated */
 625 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
 626 {
 627         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 628         struct rlb_client_info *client_info;
 629         u32 hash_index;
 630
 631         _lock_rx_hashtbl(bond);
 632
 633         hash_index = bond_info->rx_hashtbl_used_head;
 634         for (; hash_index != RLB_NULL_INDEX;
 635              hash_index = client_info->used_next) {
 636                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 637
 638                 if (!client_info->slave) {
 639                         pr_err("%s: Error: found a client with no channel in the client's hash table\n",
 640                                bond->dev->name);
 641                         continue;
 642                 }
 643                 /*update all clients using this src_ip, that are not assigned
 644                  * to the team's address (curr_active_slave) and have a known
 645                  * unicast mac address.
 646                  */
 647                 if ((client_info->ip_src == src_ip) &&
 648                     !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
 649                                              bond->dev->dev_addr) &&
 650                     !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
 651                         client_info->ntt = 1;
 652                         bond_info->rx_ntt = 1;
 653                 }
 654         }
 655
 656         _unlock_rx_hashtbl(bond);
 657 }
 658
 659 /* Caller must hold both bond and ptr locks for read */
 660 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
 661 {
 662         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 663         struct arp_pkt *arp = arp_pkt(skb);
 664         struct slave *assigned_slave, *curr_active_slave;
 665         struct rlb_client_info *client_info;
 666         u32 hash_index = 0;
 667
 668         _lock_rx_hashtbl(bond);
 669
 670         curr_active_slave = rcu_dereference(bond->curr_active_slave);
 671
 672         hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
 673         client_info = &(bond_info->rx_hashtbl[hash_index]);
 674
 675         if (client_info->assigned) {
 676                 if ((client_info->ip_src == arp->ip_src) &&
 677                     (client_info->ip_dst == arp->ip_dst)) {
 678                         /* the entry is already assigned to this client */
 679                         if (!ether_addr_equal_64bits(arp->mac_dst, mac_bcast)) {
 680                                 /* update mac address from arp */
 681                                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 682                         }
 683                         memcpy(client_info->mac_src, arp->mac_src, ETH_ALEN);
 684
 685                         assigned_slave = client_info->slave;
 686                         if (assigned_slave) {
 687                                 _unlock_rx_hashtbl(bond);
 688                                 return assigned_slave;
 689                         }
 690                 } else {
 691                         /* the entry is already assigned to some other client,
 692                          * move the old client to primary (curr_active_slave) so
 693                          * that the new client can be assigned to this entry.
 694                          */
 695                         if (bond->curr_active_slave &&
 696                             client_info->slave != curr_active_slave) {
 697                                 client_info->slave = curr_active_slave;
 698                                 rlb_update_client(client_info);
 699                         }
 700                 }
 701         }
 702         /* assign a new slave */
 703         assigned_slave = __rlb_next_rx_slave(bond);
 704
 705         if (assigned_slave) {
 706                 if (!(client_info->assigned &&
 707                       client_info->ip_src == arp->ip_src)) {
 708                         /* ip_src is going to be updated,
 709                          * fix the src hash list
 710                          */
 711                         u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
 712                                                     sizeof(arp->ip_src));
 713                         rlb_src_unlink(bond, hash_index);
 714                         rlb_src_link(bond, hash_src, hash_index);
 715                 }
 716
 717                 client_info->ip_src = arp->ip_src;
 718                 client_info->ip_dst = arp->ip_dst;
 719                 /* arp->mac_dst is broadcast for arp reqeusts.
 720                  * will be updated with clients actual unicast mac address
 721                  * upon receiving an arp reply.
 722                  */
 723                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 724                 memcpy(client_info->mac_src, arp->mac_src, ETH_ALEN);
 725                 client_info->slave = assigned_slave;
 726
 727                 if (!ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
 728                         client_info->ntt = 1;
 729                         bond->alb_info.rx_ntt = 1;
 730                 } else {
 731                         client_info->ntt = 0;
 732                 }
 733
 734                 if (!vlan_get_tag(skb, &client_info->vlan_id))
 735                         client_info->vlan_id = 0;
 736
 737                 if (!client_info->assigned) {
 738                         u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
 739                         bond_info->rx_hashtbl_used_head = hash_index;
 740                         client_info->used_next = prev_tbl_head;
 741                         if (prev_tbl_head != RLB_NULL_INDEX) {
 742                                 bond_info->rx_hashtbl[prev_tbl_head].used_prev =
 743                                         hash_index;
 744                         }
 745                         client_info->assigned = 1;
 746                 }
 747         }
 748
 749         _unlock_rx_hashtbl(bond);
 750
 751         return assigned_slave;
 752 }
 753
 754 /* chooses (and returns) transmit channel for arp reply
 755  * does not choose channel for other arp types since they are
 756  * sent on the curr_active_slave
 757  */
 758 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 759 {
 760         struct arp_pkt *arp = arp_pkt(skb);
 761         struct slave *tx_slave = NULL;
 762
 763         /* Don't modify or load balance ARPs that do not originate locally
 764          * (e.g.,arrive via a bridge).
 765          */
 766         if (!bond_slave_has_mac_rcu(bond, arp->mac_src))
 767                 return NULL;
 768
 769         if (arp->op_code == htons(ARPOP_REPLY)) {
 770                 /* the arp must be sent on the selected
 771                 * rx channel
 772                 */
 773                 tx_slave = rlb_choose_channel(skb, bond);
 774                 if (tx_slave) {
 775                         memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
 776                 }
 777                 pr_debug("Server sent ARP Reply packet\n");
 778         } else if (arp->op_code == htons(ARPOP_REQUEST)) {
 779                 /* Create an entry in the rx_hashtbl for this client as a
 780                  * place holder.
 781                  * When the arp reply is received the entry will be updated
 782                  * with the correct unicast address of the client.
 783                  */
 784                 rlb_choose_channel(skb, bond);
 785
 786                 /* The ARP reply packets must be delayed so that
 787                  * they can cancel out the influence of the ARP request.
 788                  */
 789                 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 790
 791                 /* arp requests are broadcast and are sent on the primary
 792                  * the arp request will collapse all clients on the subnet to
 793                  * the primary slave. We must register these clients to be
 794                  * updated with their assigned mac.
 795                  */
 796                 rlb_req_update_subnet_clients(bond, arp->ip_src);
 797                 pr_debug("Server sent ARP Request packet\n");
 798         }
 799
 800         return tx_slave;
 801 }
 802
 803 /* Caller must hold bond lock for read */
 804 static void rlb_rebalance(struct bonding *bond)
 805 {
 806         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 807         struct slave *assigned_slave;
 808         struct rlb_client_info *client_info;
 809         int ntt;
 810         u32 hash_index;
 811
 812         _lock_rx_hashtbl_bh(bond);
 813
 814         ntt = 0;
 815         hash_index = bond_info->rx_hashtbl_used_head;
 816         for (; hash_index != RLB_NULL_INDEX;
 817              hash_index = client_info->used_next) {
 818                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 819                 assigned_slave = rlb_next_rx_slave(bond);
 820                 if (assigned_slave && (client_info->slave != assigned_slave)) {
 821                         client_info->slave = assigned_slave;
 822                         client_info->ntt = 1;
 823                         ntt = 1;
 824                 }
 825         }
 826
 827         /* update the team's flag only after the whole iteration */
 828         if (ntt) {
 829                 bond_info->rx_ntt = 1;
 830         }
 831         _unlock_rx_hashtbl_bh(bond);
 832 }
 833
 834 /* Caller must hold rx_hashtbl lock */
 835 static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
 836 {
 837         entry->used_next = RLB_NULL_INDEX;
 838         entry->used_prev = RLB_NULL_INDEX;
 839         entry->assigned = 0;
 840         entry->slave = NULL;
 841         entry->vlan_id = 0;
 842 }
 843 static void rlb_init_table_entry_src(struct rlb_client_info *entry)
 844 {
 845         entry->src_first = RLB_NULL_INDEX;
 846         entry->src_prev = RLB_NULL_INDEX;
 847         entry->src_next = RLB_NULL_INDEX;
 848 }
 849
 850 static void rlb_init_table_entry(struct rlb_client_info *entry)
 851 {
 852         memset(entry, 0, sizeof(struct rlb_client_info));
 853         rlb_init_table_entry_dst(entry);
 854         rlb_init_table_entry_src(entry);
 855 }
 856
 857 static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
 858 {
 859         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 860         u32 next_index = bond_info->rx_hashtbl[index].used_next;
 861         u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
 862
 863         if (index == bond_info->rx_hashtbl_used_head)
 864                 bond_info->rx_hashtbl_used_head = next_index;
 865         if (prev_index != RLB_NULL_INDEX)
 866                 bond_info->rx_hashtbl[prev_index].used_next = next_index;
 867         if (next_index != RLB_NULL_INDEX)
 868                 bond_info->rx_hashtbl[next_index].used_prev = prev_index;
 869 }
 870
 871 /* unlink a rlb hash table entry from the src list */
 872 static void rlb_src_unlink(struct bonding *bond, u32 index)
 873 {
 874         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 875         u32 next_index = bond_info->rx_hashtbl[index].src_next;
 876         u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
 877
 878         bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
 879         bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
 880
 881         if (next_index != RLB_NULL_INDEX)
 882                 bond_info->rx_hashtbl[next_index].src_prev = prev_index;
 883
 884         if (prev_index == RLB_NULL_INDEX)
 885                 return;
 886
 887         /* is prev_index pointing to the head of this list? */
 888         if (bond_info->rx_hashtbl[prev_index].src_first == index)
 889                 bond_info->rx_hashtbl[prev_index].src_first = next_index;
 890         else
 891                 bond_info->rx_hashtbl[prev_index].src_next = next_index;
 892
 893 }
 894
 895 static void rlb_delete_table_entry(struct bonding *bond, u32 index)
 896 {
 897         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 898         struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 899
 900         rlb_delete_table_entry_dst(bond, index);
 901         rlb_init_table_entry_dst(entry);
 902
 903         rlb_src_unlink(bond, index);
 904 }
 905
 906 /* add the rx_hashtbl[ip_dst_hash] entry to the list
 907  * of entries with identical ip_src_hash
 908  */
 909 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
 910 {
 911         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 912         u32 next;
 913
 914         bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
 915         next = bond_info->rx_hashtbl[ip_src_hash].src_first;
 916         bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
 917         if (next != RLB_NULL_INDEX)
 918                 bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
 919         bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
 920 }
 921
 922 /* deletes all rx_hashtbl entries with  arp->ip_src if their mac_src does
 923  * not match arp->mac_src */
 924 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
 925 {
 926         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 927         u32 ip_src_hash = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
 928         u32 index;
 929
 930         _lock_rx_hashtbl_bh(bond);
 931
 932         index = bond_info->rx_hashtbl[ip_src_hash].src_first;
 933         while (index != RLB_NULL_INDEX) {
 934                 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 935                 u32 next_index = entry->src_next;
 936                 if (entry->ip_src == arp->ip_src &&
 937                     !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
 938                                 rlb_delete_table_entry(bond, index);
 939                 index = next_index;
 940         }
 941         _unlock_rx_hashtbl_bh(bond);
 942 }
 943
 944 static int rlb_initialize(struct bonding *bond)
 945 {
 946         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 947         struct rlb_client_info  *new_hashtbl;
 948         int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 949         int i;
 950
 951         new_hashtbl = kmalloc(size, GFP_KERNEL);
 952         if (!new_hashtbl)
 953                 return -1;
 954
 955         _lock_rx_hashtbl_bh(bond);
 956
 957         bond_info->rx_hashtbl = new_hashtbl;
 958
 959         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 960
 961         for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
 962                 rlb_init_table_entry(bond_info->rx_hashtbl + i);
 963         }
 964
 965         _unlock_rx_hashtbl_bh(bond);
 966
 967         /* register to receive ARPs */
 968         bond->recv_probe = rlb_arp_recv;
 969
 970         return 0;
 971 }
 972
 973 static void rlb_deinitialize(struct bonding *bond)
 974 {
 975         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 976
 977         _lock_rx_hashtbl_bh(bond);
 978
 979         kfree(bond_info->rx_hashtbl);
 980         bond_info->rx_hashtbl = NULL;
 981         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 982
 983         _unlock_rx_hashtbl_bh(bond);
 984 }
 985
 986 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 987 {
 988         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 989         u32 curr_index;
 990
 991         _lock_rx_hashtbl_bh(bond);
 992
 993         curr_index = bond_info->rx_hashtbl_used_head;
 994         while (curr_index != RLB_NULL_INDEX) {
 995                 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 996                 u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
 997
 998                 if (curr->vlan_id == vlan_id)
 999                         rlb_delete_table_entry(bond, curr_index);
1000
1001                 curr_index = next_index;
1002         }
1003
1004         _unlock_rx_hashtbl_bh(bond);
1005 }
1006
1007 /*********************** tlb/rlb shared functions *********************/
1008
1009 static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
1010                             u16 vid)
1011 {
1012         struct learning_pkt pkt;
1013         struct sk_buff *skb;
1014         int size = sizeof(struct learning_pkt);
1015         char *data;
1016
1017         memset(&pkt, 0, size);
1018         memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
1019         memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
1020         pkt.type = cpu_to_be16(ETH_P_LOOP);
1021
1022         skb = dev_alloc_skb(size);
1023         if (!skb)
1024                 return;
1025
1026         data = skb_put(skb, size);
1027         memcpy(data, &pkt, size);
1028
1029         skb_reset_mac_header(skb);
1030         skb->network_header = skb->mac_header + ETH_HLEN;
1031         skb->protocol = pkt.type;
1032         skb->priority = TC_PRIO_CONTROL;
1033         skb->dev = slave->dev;
1034
1035         if (vid) {
1036                 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vid);
1037                 if (!skb) {
1038                         pr_err("%s: Error: failed to insert VLAN tag\n",
1039                                slave->bond->dev->name);
1040                         return;
1041                 }
1042         }
1043
1044         dev_queue_xmit(skb);
1045 }
1046
1047
1048 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
1049 {
1050         struct bonding *bond = bond_get_bond_by_slave(slave);
1051         struct net_device *upper;
1052         struct list_head *iter;
1053
1054         /* send untagged */
1055         alb_send_lp_vid(slave, mac_addr, 0);
1056
1057         /* loop through vlans and send one packet for each */
1058         rcu_read_lock();
1059         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
1060                 if (upper->priv_flags & IFF_802_1Q_VLAN)
1061                         alb_send_lp_vid(slave, mac_addr,
1062                                         vlan_dev_vlan_id(upper));
1063         }
1064         rcu_read_unlock();
1065 }
1066
1067 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[])
1068 {
1069         struct net_device *dev = slave->dev;
1070         struct sockaddr s_addr;
1071
1072         if (slave->bond->params.mode == BOND_MODE_TLB) {
1073                 memcpy(dev->dev_addr, addr, dev->addr_len);
1074                 return 0;
1075         }
1076
1077         /* for rlb each slave must have a unique hw mac addresses so that */
1078         /* each slave will receive packets destined to a different mac */
1079         memcpy(s_addr.sa_data, addr, dev->addr_len);
1080         s_addr.sa_family = dev->type;
1081         if (dev_set_mac_address(dev, &s_addr)) {
1082                 pr_err("%s: Error: dev_set_mac_address of dev %s failed!\n"
1083                        "ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
1084                        slave->bond->dev->name, dev->name);
1085                 return -EOPNOTSUPP;
1086         }
1087         return 0;
1088 }
1089
1090 /*
1091  * Swap MAC addresses between two slaves.
1092  *
1093  * Called with RTNL held, and no other locks.
1094  *
1095  */
1096
1097 static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1098 {
1099         u8 tmp_mac_addr[ETH_ALEN];
1100
1101         memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
1102         alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr);
1103         alb_set_slave_mac_addr(slave2, tmp_mac_addr);
1104
1105 }
1106
1107 /*
1108  * Send learning packets after MAC address swap.
1109  *
1110  * Called with RTNL and no other locks
1111  */
1112 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1113                                 struct slave *slave2)
1114 {
1115         int slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
1116         struct slave *disabled_slave = NULL;
1117
1118         ASSERT_RTNL();
1119
1120         /* fasten the change in the switch */
1121         if (SLAVE_IS_OK(slave1)) {
1122                 alb_send_learning_packets(slave1, slave1->dev->dev_addr);
1123                 if (bond->alb_info.rlb_enabled) {
1124                         /* inform the clients that the mac address
1125                          * has changed
1126                          */
1127                         rlb_req_update_slave_clients(bond, slave1);
1128                 }
1129         } else {
1130                 disabled_slave = slave1;
1131         }
1132
1133         if (SLAVE_IS_OK(slave2)) {
1134                 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
1135                 if (bond->alb_info.rlb_enabled) {
1136                         /* inform the clients that the mac address
1137                          * has changed
1138                          */
1139                         rlb_req_update_slave_clients(bond, slave2);
1140                 }
1141         } else {
1142                 disabled_slave = slave2;
1143         }
1144
1145         if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1146                 /* A disabled slave was assigned an active mac addr */
1147                 rlb_teach_disabled_mac_on_primary(bond,
1148                                                   disabled_slave->dev->dev_addr);
1149         }
1150 }
1151
1152 /**
1153  * alb_change_hw_addr_on_detach
1154  * @bond: bonding we're working on
1155  * @slave: the slave that was just detached
1156  *
1157  * We assume that @slave was already detached from the slave list.
1158  *
1159  * If @slave's permanent hw address is different both from its current
1160  * address and from @bond's address, then somewhere in the bond there's
1161  * a slave that has @slave's permanet address as its current address.
1162  * We'll make sure that that slave no longer uses @slave's permanent address.
1163  *
1164  * Caller must hold RTNL and no other locks
1165  */
1166 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1167 {
1168         int perm_curr_diff;
1169         int perm_bond_diff;
1170         struct slave *found_slave;
1171
1172         perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1173                                                   slave->dev->dev_addr);
1174         perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1175                                                   bond->dev->dev_addr);
1176
1177         if (perm_curr_diff && perm_bond_diff) {
1178                 found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1179
1180                 if (found_slave) {
1181                         /* locking: needs RTNL and nothing else */
1182                         alb_swap_mac_addr(slave, found_slave);
1183                         alb_fasten_mac_swap(bond, slave, found_slave);
1184                 }
1185         }
1186 }
1187
1188 /**
1189  * alb_handle_addr_collision_on_attach
1190  * @bond: bonding we're working on
1191  * @slave: the slave that was just attached
1192  *
1193  * checks uniqueness of slave's mac address and handles the case the
1194  * new slave uses the bonds mac address.
1195  *
1196  * If the permanent hw address of @slave is @bond's hw address, we need to
1197  * find a different hw address to give @slave, that isn't in use by any other
1198  * slave in the bond. This address must be, of course, one of the permanent
1199  * addresses of the other slaves.
1200  *
1201  * We go over the slave list, and for each slave there we compare its
1202  * permanent hw address with the current address of all the other slaves.
1203  * If no match was found, then we've found a slave with a permanent address
1204  * that isn't used by any other slave in the bond, so we can assign it to
1205  * @slave.
1206  *
1207  * assumption: this function is called before @slave is attached to the
1208  *             bond slave list.
1209  */
1210 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1211 {
1212         struct slave *has_bond_addr = bond->curr_active_slave;
1213         struct slave *tmp_slave1, *free_mac_slave = NULL;
1214         struct list_head *iter;
1215
1216         if (!bond_has_slaves(bond)) {
1217                 /* this is the first slave */
1218                 return 0;
1219         }
1220
1221         /* if slave's mac address differs from bond's mac address
1222          * check uniqueness of slave's mac address against the other
1223          * slaves in the bond.
1224          */
1225         if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1226                 if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1227                         return 0;
1228
1229                 /* Try setting slave mac to bond address and fall-through
1230                    to code handling that situation below... */
1231                 alb_set_slave_mac_addr(slave, bond->dev->dev_addr);
1232         }
1233
1234         /* The slave's address is equal to the address of the bond.
1235          * Search for a spare address in the bond for this slave.
1236          */
1237         bond_for_each_slave(bond, tmp_slave1, iter) {
1238                 if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1239                         /* no slave has tmp_slave1's perm addr
1240                          * as its curr addr
1241                          */
1242                         free_mac_slave = tmp_slave1;
1243                         break;
1244                 }
1245
1246                 if (!has_bond_addr) {
1247                         if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1248                                                     bond->dev->dev_addr)) {
1249
1250                                 has_bond_addr = tmp_slave1;
1251                         }
1252                 }
1253         }
1254
1255         if (free_mac_slave) {
1256                 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr);
1257
1258                 pr_warning("%s: Warning: the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1259                            bond->dev->name, slave->dev->name,
1260                            free_mac_slave->dev->name);
1261
1262         } else if (has_bond_addr) {
1263                 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",
1264                        bond->dev->name, slave->dev->name);
1265                 return -EFAULT;
1266         }
1267
1268         return 0;
1269 }
1270
1271 /**
1272  * alb_set_mac_address
1273  * @bond:
1274  * @addr:
1275  *
1276  * In TLB mode all slaves are configured to the bond's hw address, but set
1277  * their dev_addr field to different addresses (based on their permanent hw
1278  * addresses).
1279  *
1280  * For each slave, this function sets the interface to the new address and then
1281  * changes its dev_addr field to its previous value.
1282  *
1283  * Unwinding assumes bond's mac address has not yet changed.
1284  */
1285 static int alb_set_mac_address(struct bonding *bond, void *addr)
1286 {
1287         struct slave *slave, *rollback_slave;
1288         struct list_head *iter;
1289         struct sockaddr sa;
1290         char tmp_addr[ETH_ALEN];
1291         int res;
1292
1293         if (bond->alb_info.rlb_enabled)
1294                 return 0;
1295
1296         bond_for_each_slave(bond, slave, iter) {
1297                 /* save net_device's current hw address */
1298                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1299
1300                 res = dev_set_mac_address(slave->dev, addr);
1301
1302                 /* restore net_device's hw address */
1303                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1304
1305                 if (res)
1306                         goto unwind;
1307         }
1308
1309         return 0;
1310
1311 unwind:
1312         memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1313         sa.sa_family = bond->dev->type;
1314
1315         /* unwind from head to the slave that failed */
1316         bond_for_each_slave(bond, rollback_slave, iter) {
1317                 if (rollback_slave == slave)
1318                         break;
1319                 memcpy(tmp_addr, rollback_slave->dev->dev_addr, ETH_ALEN);
1320                 dev_set_mac_address(rollback_slave->dev, &sa);
1321                 memcpy(rollback_slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1322         }
1323
1324         return res;
1325 }
1326
1327 /************************ exported alb funcions ************************/
1328
1329 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1330 {
1331         int res;
1332
1333         res = tlb_initialize(bond);
1334         if (res) {
1335                 return res;
1336         }
1337
1338         if (rlb_enabled) {
1339                 bond->alb_info.rlb_enabled = 1;
1340                 /* initialize rlb */
1341                 res = rlb_initialize(bond);
1342                 if (res) {
1343                         tlb_deinitialize(bond);
1344                         return res;
1345                 }
1346         } else {
1347                 bond->alb_info.rlb_enabled = 0;
1348         }
1349
1350         return 0;
1351 }
1352
1353 void bond_alb_deinitialize(struct bonding *bond)
1354 {
1355         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1356
1357         tlb_deinitialize(bond);
1358
1359         if (bond_info->rlb_enabled) {
1360                 rlb_deinitialize(bond);
1361         }
1362 }
1363
1364 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1365 {
1366         struct bonding *bond = netdev_priv(bond_dev);
1367         struct ethhdr *eth_data;
1368         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1369         struct slave *tx_slave = NULL;
1370         static const __be32 ip_bcast = htonl(0xffffffff);
1371         int hash_size = 0;
1372         int do_tx_balance = 1;
1373         u32 hash_index = 0;
1374         const u8 *hash_start = NULL;
1375         int res = 1;
1376         struct ipv6hdr *ip6hdr;
1377
1378         skb_reset_mac_header(skb);
1379         eth_data = eth_hdr(skb);
1380
1381         switch (ntohs(skb->protocol)) {
1382         case ETH_P_IP: {
1383                 const struct iphdr *iph = ip_hdr(skb);
1384
1385                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast) ||
1386                     (iph->daddr == ip_bcast) ||
1387                     (iph->protocol == IPPROTO_IGMP)) {
1388                         do_tx_balance = 0;
1389                         break;
1390                 }
1391                 hash_start = (char *)&(iph->daddr);
1392                 hash_size = sizeof(iph->daddr);
1393         }
1394                 break;
1395         case ETH_P_IPV6:
1396                 /* IPv6 doesn't really use broadcast mac address, but leave
1397                  * that here just in case.
1398                  */
1399                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast)) {
1400                         do_tx_balance = 0;
1401                         break;
1402                 }
1403
1404                 /* IPv6 uses all-nodes multicast as an equivalent to
1405                  * broadcasts in IPv4.
1406                  */
1407                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1408                         do_tx_balance = 0;
1409                         break;
1410                 }
1411
1412                 /* Additianally, DAD probes should not be tx-balanced as that
1413                  * will lead to false positives for duplicate addresses and
1414                  * prevent address configuration from working.
1415                  */
1416                 ip6hdr = ipv6_hdr(skb);
1417                 if (ipv6_addr_any(&ip6hdr->saddr)) {
1418                         do_tx_balance = 0;
1419                         break;
1420                 }
1421
1422                 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1423                 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1424                 break;
1425         case ETH_P_IPX:
1426                 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1427                         /* something is wrong with this packet */
1428                         do_tx_balance = 0;
1429                         break;
1430                 }
1431
1432                 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1433                         /* The only protocol worth balancing in
1434                          * this family since it has an "ARP" like
1435                          * mechanism
1436                          */
1437                         do_tx_balance = 0;
1438                         break;
1439                 }
1440
1441                 hash_start = (char*)eth_data->h_dest;
1442                 hash_size = ETH_ALEN;
1443                 break;
1444         case ETH_P_ARP:
1445                 do_tx_balance = 0;
1446                 if (bond_info->rlb_enabled) {
1447                         tx_slave = rlb_arp_xmit(skb, bond);
1448                 }
1449                 break;
1450         default:
1451                 do_tx_balance = 0;
1452                 break;
1453         }
1454
1455         if (do_tx_balance) {
1456                 hash_index = _simple_hash(hash_start, hash_size);
1457                 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1458         }
1459
1460         if (!tx_slave) {
1461                 /* unbalanced or unassigned, send through primary */
1462                 tx_slave = rcu_dereference(bond->curr_active_slave);
1463                 bond_info->unbalanced_load += skb->len;
1464         }
1465
1466         if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1467                 if (tx_slave != rcu_dereference(bond->curr_active_slave)) {
1468                         memcpy(eth_data->h_source,
1469                                tx_slave->dev->dev_addr,
1470                                ETH_ALEN);
1471                 }
1472
1473                 res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1474         } else {
1475                 if (tx_slave) {
1476                         _lock_tx_hashtbl(bond);
1477                         __tlb_clear_slave(bond, tx_slave, 0);
1478                         _unlock_tx_hashtbl(bond);
1479                 }
1480         }
1481
1482         if (res) {
1483                 /* no suitable interface, frame not sent */
1484                 kfree_skb(skb);
1485         }
1486
1487         return NETDEV_TX_OK;
1488 }
1489
1490 void bond_alb_monitor(struct work_struct *work)
1491 {
1492         struct bonding *bond = container_of(work, struct bonding,
1493                                             alb_work.work);
1494         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1495         struct list_head *iter;
1496         struct slave *slave;
1497
1498         read_lock(&bond->lock);
1499
1500         if (!bond_has_slaves(bond)) {
1501                 bond_info->tx_rebalance_counter = 0;
1502                 bond_info->lp_counter = 0;
1503                 goto re_arm;
1504         }
1505
1506         bond_info->tx_rebalance_counter++;
1507         bond_info->lp_counter++;
1508
1509         /* send learning packets */
1510         if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1511                 /* change of curr_active_slave involves swapping of mac addresses.
1512                  * in order to avoid this swapping from happening while
1513                  * sending the learning packets, the curr_slave_lock must be held for
1514                  * read.
1515                  */
1516                 read_lock(&bond->curr_slave_lock);
1517
1518                 bond_for_each_slave(bond, slave, iter)
1519                         alb_send_learning_packets(slave, slave->dev->dev_addr);
1520
1521                 read_unlock(&bond->curr_slave_lock);
1522
1523                 bond_info->lp_counter = 0;
1524         }
1525
1526         /* rebalance tx traffic */
1527         if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1528
1529                 read_lock(&bond->curr_slave_lock);
1530
1531                 bond_for_each_slave(bond, slave, iter) {
1532                         tlb_clear_slave(bond, slave, 1);
1533                         if (slave == bond->curr_active_slave) {
1534                                 SLAVE_TLB_INFO(slave).load =
1535                                         bond_info->unbalanced_load /
1536                                                 BOND_TLB_REBALANCE_INTERVAL;
1537                                 bond_info->unbalanced_load = 0;
1538                         }
1539                 }
1540
1541                 read_unlock(&bond->curr_slave_lock);
1542
1543                 bond_info->tx_rebalance_counter = 0;
1544         }
1545
1546         /* handle rlb stuff */
1547         if (bond_info->rlb_enabled) {
1548                 if (bond_info->primary_is_promisc &&
1549                     (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1550
1551                         /*
1552                          * dev_set_promiscuity requires rtnl and
1553                          * nothing else.  Avoid race with bond_close.
1554                          */
1555                         read_unlock(&bond->lock);
1556                         if (!rtnl_trylock()) {
1557                                 read_lock(&bond->lock);
1558                                 goto re_arm;
1559                         }
1560
1561                         bond_info->rlb_promisc_timeout_counter = 0;
1562
1563                         /* If the primary was set to promiscuous mode
1564                          * because a slave was disabled then
1565                          * it can now leave promiscuous mode.
1566                          */
1567                         dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1568                         bond_info->primary_is_promisc = 0;
1569
1570                         rtnl_unlock();
1571                         read_lock(&bond->lock);
1572                 }
1573
1574                 if (bond_info->rlb_rebalance) {
1575                         bond_info->rlb_rebalance = 0;
1576                         rlb_rebalance(bond);
1577                 }
1578
1579                 /* check if clients need updating */
1580                 if (bond_info->rx_ntt) {
1581                         if (bond_info->rlb_update_delay_counter) {
1582                                 --bond_info->rlb_update_delay_counter;
1583                         } else {
1584                                 rlb_update_rx_clients(bond);
1585                                 if (bond_info->rlb_update_retry_counter) {
1586                                         --bond_info->rlb_update_retry_counter;
1587                                 } else {
1588                                         bond_info->rx_ntt = 0;
1589                                 }
1590                         }
1591                 }
1592         }
1593
1594 re_arm:
1595         queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1596
1597         read_unlock(&bond->lock);
1598 }
1599
1600 /* assumption: called before the slave is attached to the bond
1601  * and not locked by the bond lock
1602  */
1603 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1604 {
1605         int res;
1606
1607         res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr);
1608         if (res) {
1609                 return res;
1610         }
1611
1612         res = alb_handle_addr_collision_on_attach(bond, slave);
1613         if (res) {
1614                 return res;
1615         }
1616
1617         tlb_init_slave(slave);
1618
1619         /* order a rebalance ASAP */
1620         bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1621
1622         if (bond->alb_info.rlb_enabled) {
1623                 bond->alb_info.rlb_rebalance = 1;
1624         }
1625
1626         return 0;
1627 }
1628
1629 /*
1630  * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1631  * if necessary.
1632  *
1633  * Caller must hold RTNL and no other locks
1634  */
1635 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1636 {
1637         if (bond_has_slaves(bond))
1638                 alb_change_hw_addr_on_detach(bond, slave);
1639
1640         tlb_clear_slave(bond, slave, 0);
1641
1642         if (bond->alb_info.rlb_enabled) {
1643                 bond->alb_info.rx_slave = NULL;
1644                 rlb_clear_slave(bond, slave);
1645         }
1646 }
1647
1648 /* Caller must hold bond lock for read */
1649 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1650 {
1651         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1652
1653         if (link == BOND_LINK_DOWN) {
1654                 tlb_clear_slave(bond, slave, 0);
1655                 if (bond->alb_info.rlb_enabled) {
1656                         rlb_clear_slave(bond, slave);
1657                 }
1658         } else if (link == BOND_LINK_UP) {
1659                 /* order a rebalance ASAP */
1660                 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1661                 if (bond->alb_info.rlb_enabled) {
1662                         bond->alb_info.rlb_rebalance = 1;
1663                         /* If the updelay module parameter is smaller than the
1664                          * forwarding delay of the switch the rebalance will
1665                          * not work because the rebalance arp replies will
1666                          * not be forwarded to the clients..
1667                          */
1668                 }
1669         }
1670 }
1671
1672 /**
1673  * bond_alb_handle_active_change - assign new curr_active_slave
1674  * @bond: our bonding struct
1675  * @new_slave: new slave to assign
1676  *
1677  * Set the bond->curr_active_slave to @new_slave and handle
1678  * mac address swapping and promiscuity changes as needed.
1679  *
1680  * If new_slave is NULL, caller must hold curr_slave_lock or
1681  * bond->lock for write.
1682  *
1683  * If new_slave is not NULL, caller must hold RTNL, bond->lock for
1684  * read and curr_slave_lock for write.  Processing here may sleep, so
1685  * no other locks may be held.
1686  */
1687 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1688         __releases(&bond->curr_slave_lock)
1689         __releases(&bond->lock)
1690         __acquires(&bond->lock)
1691         __acquires(&bond->curr_slave_lock)
1692 {
1693         struct slave *swap_slave;
1694
1695         if (bond->curr_active_slave == new_slave)
1696                 return;
1697
1698         if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1699                 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1700                 bond->alb_info.primary_is_promisc = 0;
1701                 bond->alb_info.rlb_promisc_timeout_counter = 0;
1702         }
1703
1704         swap_slave = bond->curr_active_slave;
1705         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1706
1707         if (!new_slave || !bond_has_slaves(bond))
1708                 return;
1709
1710         /* set the new curr_active_slave to the bonds mac address
1711          * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1712          */
1713         if (!swap_slave)
1714                 swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1715
1716         /*
1717          * Arrange for swap_slave and new_slave to temporarily be
1718          * ignored so we can mess with their MAC addresses without
1719          * fear of interference from transmit activity.
1720          */
1721         if (swap_slave)
1722                 tlb_clear_slave(bond, swap_slave, 1);
1723         tlb_clear_slave(bond, new_slave, 1);
1724
1725         write_unlock_bh(&bond->curr_slave_lock);
1726         read_unlock(&bond->lock);
1727
1728         ASSERT_RTNL();
1729
1730         /* in TLB mode, the slave might flip down/up with the old dev_addr,
1731          * and thus filter bond->dev_addr's packets, so force bond's mac
1732          */
1733         if (bond->params.mode == BOND_MODE_TLB) {
1734                 struct sockaddr sa;
1735                 u8 tmp_addr[ETH_ALEN];
1736
1737                 memcpy(tmp_addr, new_slave->dev->dev_addr, ETH_ALEN);
1738
1739                 memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1740                 sa.sa_family = bond->dev->type;
1741                 /* we don't care if it can't change its mac, best effort */
1742                 dev_set_mac_address(new_slave->dev, &sa);
1743
1744                 memcpy(new_slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1745         }
1746
1747         /* curr_active_slave must be set before calling alb_swap_mac_addr */
1748         if (swap_slave) {
1749                 /* swap mac address */
1750                 alb_swap_mac_addr(swap_slave, new_slave);
1751                 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1752                 read_lock(&bond->lock);
1753         } else {
1754                 /* set the new_slave to the bond mac address */
1755                 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr);
1756                 read_lock(&bond->lock);
1757                 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1758         }
1759
1760         write_lock_bh(&bond->curr_slave_lock);
1761 }
1762
1763 /*
1764  * Called with RTNL
1765  */
1766 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1767         __acquires(&bond->lock)
1768         __releases(&bond->lock)
1769 {
1770         struct bonding *bond = netdev_priv(bond_dev);
1771         struct sockaddr *sa = addr;
1772         struct slave *swap_slave;
1773         int res;
1774
1775         if (!is_valid_ether_addr(sa->sa_data)) {
1776                 return -EADDRNOTAVAIL;
1777         }
1778
1779         res = alb_set_mac_address(bond, addr);
1780         if (res) {
1781                 return res;
1782         }
1783
1784         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1785
1786         /* If there is no curr_active_slave there is nothing else to do.
1787          * Otherwise we'll need to pass the new address to it and handle
1788          * duplications.
1789          */
1790         if (!bond->curr_active_slave) {
1791                 return 0;
1792         }
1793
1794         swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1795
1796         if (swap_slave) {
1797                 alb_swap_mac_addr(swap_slave, bond->curr_active_slave);
1798                 alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave);
1799         } else {
1800                 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr);
1801
1802                 read_lock(&bond->lock);
1803                 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1804                 if (bond->alb_info.rlb_enabled) {
1805                         /* inform clients mac address has changed */
1806                         rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1807                 }
1808                 read_unlock(&bond->lock);
1809         }
1810
1811         return 0;
1812 }
1813
1814 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1815 {
1816         if (bond->alb_info.rlb_enabled) {
1817                 rlb_clear_vlan(bond, vlan_id);
1818         }
1819 }
1820