git.oblomov.eu Git - linux-2.6/blob - drivers/net/bonding/bond_alb.c

   1 /*
   2  * Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms of the GNU General Public License as published by the
   6  * Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful, but
  10  * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  11  * or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License along
  15  * with this program; if not, write to the Free Software Foundation, Inc.,
  16  * 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
  17  *
  18  * The full GNU General Public License is included in this distribution in the
  19  * file called LICENSE.
  20  *
  21  */
  22
  23 //#define BONDING_DEBUG 1
  24
  25 #include <linux/skbuff.h>
  26 #include <linux/netdevice.h>
  27 #include <linux/etherdevice.h>
  28 #include <linux/pkt_sched.h>
  29 #include <linux/spinlock.h>
  30 #include <linux/slab.h>
  31 #include <linux/timer.h>
  32 #include <linux/ip.h>
  33 #include <linux/ipv6.h>
  34 #include <linux/if_arp.h>
  35 #include <linux/if_ether.h>
  36 #include <linux/if_bonding.h>
  37 #include <linux/if_vlan.h>
  38 #include <linux/in.h>
  39 #include <net/ipx.h>
  40 #include <net/arp.h>
  41 #include <net/ipv6.h>
  42 #include <asm/byteorder.h>
  43 #include "bonding.h"
  44 #include "bond_alb.h"
  45
  46
  47 #define ALB_TIMER_TICKS_PER_SEC     10  /* should be a divisor of HZ */
  48 #define BOND_TLB_REBALANCE_INTERVAL 10  /* In seconds, periodic re-balancing.
  49                                          * Used for division - never set
  50                                          * to zero !!!
  51                                          */
  52 #define BOND_ALB_LP_INTERVAL        1   /* In seconds, periodic send of
  53                                          * learning packets to the switch
  54                                          */
  55
  56 #define BOND_TLB_REBALANCE_TICKS (BOND_TLB_REBALANCE_INTERVAL \
  57                                   * ALB_TIMER_TICKS_PER_SEC)
  58
  59 #define BOND_ALB_LP_TICKS (BOND_ALB_LP_INTERVAL \
  60                            * ALB_TIMER_TICKS_PER_SEC)
  61
  62 #define TLB_HASH_TABLE_SIZE 256 /* The size of the clients hash table.
  63                                  * Note that this value MUST NOT be smaller
  64                                  * because the key hash table is BYTE wide !
  65                                  */
  66
  67
  68 #define TLB_NULL_INDEX          0xffffffff
  69 #define MAX_LP_BURST            3
  70
  71 /* rlb defs */
  72 #define RLB_HASH_TABLE_SIZE     256
  73 #define RLB_NULL_INDEX          0xffffffff
  74 #define RLB_UPDATE_DELAY        2*ALB_TIMER_TICKS_PER_SEC /* 2 seconds */
  75 #define RLB_ARP_BURST_SIZE      2
  76 #define RLB_UPDATE_RETRY        3       /* 3-ticks - must be smaller than the rlb
  77                                          * rebalance interval (5 min).
  78                                          */
  79 /* RLB_PROMISC_TIMEOUT = 10 sec equals the time that the current slave is
  80  * promiscuous after failover
  81  */
  82 #define RLB_PROMISC_TIMEOUT     10*ALB_TIMER_TICKS_PER_SEC
  83
  84 static const u8 mac_bcast[ETH_ALEN] = {0xff,0xff,0xff,0xff,0xff,0xff};
  85 static const u8 mac_v6_allmcast[ETH_ALEN] = {0x33,0x33,0x00,0x00,0x00,0x01};
  86 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
  87
  88 #pragma pack(1)
  89 struct learning_pkt {
  90         u8 mac_dst[ETH_ALEN];
  91         u8 mac_src[ETH_ALEN];
  92         __be16 type;
  93         u8 padding[ETH_ZLEN - ETH_HLEN];
  94 };
  95
  96 struct arp_pkt {
  97         __be16  hw_addr_space;
  98         __be16  prot_addr_space;
  99         u8      hw_addr_len;
 100         u8      prot_addr_len;
 101         __be16  op_code;
 102         u8      mac_src[ETH_ALEN];      /* sender hardware address */
 103         __be32  ip_src;                 /* sender IP address */
 104         u8      mac_dst[ETH_ALEN];      /* target hardware address */
 105         __be32  ip_dst;                 /* target IP address */
 106 };
 107 #pragma pack()
 108
 109 static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb)
 110 {
 111         return (struct arp_pkt *)skb_network_header(skb);
 112 }
 113
 114 /* Forward declaration */
 115 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[]);
 116
 117 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
 118 {
 119         int i;
 120         u8 hash = 0;
 121
 122         for (i = 0; i < hash_size; i++) {
 123                 hash ^= hash_start[i];
 124         }
 125
 126         return hash;
 127 }
 128
 129 /*********************** tlb specific functions ***************************/
 130
 131 static inline void _lock_tx_hashtbl(struct bonding *bond)
 132 {
 133         spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 134 }
 135
 136 static inline void _unlock_tx_hashtbl(struct bonding *bond)
 137 {
 138         spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 139 }
 140
 141 /* Caller must hold tx_hashtbl lock */
 142 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
 143 {
 144         if (save_load) {
 145                 entry->load_history = 1 + entry->tx_bytes /
 146                                       BOND_TLB_REBALANCE_INTERVAL;
 147                 entry->tx_bytes = 0;
 148         }
 149
 150         entry->tx_slave = NULL;
 151         entry->next = TLB_NULL_INDEX;
 152         entry->prev = TLB_NULL_INDEX;
 153 }
 154
 155 static inline void tlb_init_slave(struct slave *slave)
 156 {
 157         SLAVE_TLB_INFO(slave).load = 0;
 158         SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
 159 }
 160
 161 /* Caller must hold bond lock for read */
 162 static void tlb_clear_slave(struct bonding *bond, struct slave *slave, int save_load)
 163 {
 164         struct tlb_client_info *tx_hash_table;
 165         u32 index;
 166
 167         _lock_tx_hashtbl(bond);
 168
 169         /* clear slave from tx_hashtbl */
 170         tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
 171
 172         index = SLAVE_TLB_INFO(slave).head;
 173         while (index != TLB_NULL_INDEX) {
 174                 u32 next_index = tx_hash_table[index].next;
 175                 tlb_init_table_entry(&tx_hash_table[index], save_load);
 176                 index = next_index;
 177         }
 178
 179         tlb_init_slave(slave);
 180
 181         _unlock_tx_hashtbl(bond);
 182 }
 183
 184 /* Must be called before starting the monitor timer */
 185 static int tlb_initialize(struct bonding *bond)
 186 {
 187         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 188         int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 189         struct tlb_client_info *new_hashtbl;
 190         int i;
 191
 192         spin_lock_init(&(bond_info->tx_hashtbl_lock));
 193
 194         new_hashtbl = kzalloc(size, GFP_KERNEL);
 195         if (!new_hashtbl) {
 196                 printk(KERN_ERR DRV_NAME
 197                        ": %s: Error: Failed to allocate TLB hash table\n",
 198                        bond->dev->name);
 199                 return -1;
 200         }
 201         _lock_tx_hashtbl(bond);
 202
 203         bond_info->tx_hashtbl = new_hashtbl;
 204
 205         for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
 206                 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 1);
 207         }
 208
 209         _unlock_tx_hashtbl(bond);
 210
 211         return 0;
 212 }
 213
 214 /* Must be called only after all slaves have been released */
 215 static void tlb_deinitialize(struct bonding *bond)
 216 {
 217         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 218
 219         _lock_tx_hashtbl(bond);
 220
 221         kfree(bond_info->tx_hashtbl);
 222         bond_info->tx_hashtbl = NULL;
 223
 224         _unlock_tx_hashtbl(bond);
 225 }
 226
 227 /* Caller must hold bond lock for read */
 228 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
 229 {
 230         struct slave *slave, *least_loaded;
 231         s64 max_gap;
 232         int i, found = 0;
 233
 234         /* Find the first enabled slave */
 235         bond_for_each_slave(bond, slave, i) {
 236                 if (SLAVE_IS_OK(slave)) {
 237                         found = 1;
 238                         break;
 239                 }
 240         }
 241
 242         if (!found) {
 243                 return NULL;
 244         }
 245
 246         least_loaded = slave;
 247         max_gap = (s64)(slave->speed << 20) - /* Convert to Megabit per sec */
 248                         (s64)(SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
 249
 250         /* Find the slave with the largest gap */
 251         bond_for_each_slave_from(bond, slave, i, least_loaded) {
 252                 if (SLAVE_IS_OK(slave)) {
 253                         s64 gap = (s64)(slave->speed << 20) -
 254                                         (s64)(SLAVE_TLB_INFO(slave).load << 3);
 255                         if (max_gap < gap) {
 256                                 least_loaded = slave;
 257                                 max_gap = gap;
 258                         }
 259                 }
 260         }
 261
 262         return least_loaded;
 263 }
 264
 265 /* Caller must hold bond lock for read */
 266 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index, u32 skb_len)
 267 {
 268         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 269         struct tlb_client_info *hash_table;
 270         struct slave *assigned_slave;
 271
 272         _lock_tx_hashtbl(bond);
 273
 274         hash_table = bond_info->tx_hashtbl;
 275         assigned_slave = hash_table[hash_index].tx_slave;
 276         if (!assigned_slave) {
 277                 assigned_slave = tlb_get_least_loaded_slave(bond);
 278
 279                 if (assigned_slave) {
 280                         struct tlb_slave_info *slave_info =
 281                                 &(SLAVE_TLB_INFO(assigned_slave));
 282                         u32 next_index = slave_info->head;
 283
 284                         hash_table[hash_index].tx_slave = assigned_slave;
 285                         hash_table[hash_index].next = next_index;
 286                         hash_table[hash_index].prev = TLB_NULL_INDEX;
 287
 288                         if (next_index != TLB_NULL_INDEX) {
 289                                 hash_table[next_index].prev = hash_index;
 290                         }
 291
 292                         slave_info->head = hash_index;
 293                         slave_info->load +=
 294                                 hash_table[hash_index].load_history;
 295                 }
 296         }
 297
 298         if (assigned_slave) {
 299                 hash_table[hash_index].tx_bytes += skb_len;
 300         }
 301
 302         _unlock_tx_hashtbl(bond);
 303
 304         return assigned_slave;
 305 }
 306
 307 /*********************** rlb specific functions ***************************/
 308 static inline void _lock_rx_hashtbl(struct bonding *bond)
 309 {
 310         spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 311 }
 312
 313 static inline void _unlock_rx_hashtbl(struct bonding *bond)
 314 {
 315         spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 316 }
 317
 318 /* when an ARP REPLY is received from a client update its info
 319  * in the rx_hashtbl
 320  */
 321 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 322 {
 323         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 324         struct rlb_client_info *client_info;
 325         u32 hash_index;
 326
 327         _lock_rx_hashtbl(bond);
 328
 329         hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
 330         client_info = &(bond_info->rx_hashtbl[hash_index]);
 331
 332         if ((client_info->assigned) &&
 333             (client_info->ip_src == arp->ip_dst) &&
 334             (client_info->ip_dst == arp->ip_src)) {
 335                 /* update the clients MAC address */
 336                 memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
 337                 client_info->ntt = 1;
 338                 bond_info->rx_ntt = 1;
 339         }
 340
 341         _unlock_rx_hashtbl(bond);
 342 }
 343
 344 static int rlb_arp_recv(struct sk_buff *skb, struct net_device *bond_dev, struct packet_type *ptype, struct net_device *orig_dev)
 345 {
 346         struct bonding *bond = bond_dev->priv;
 347         struct arp_pkt *arp = (struct arp_pkt *)skb->data;
 348         int res = NET_RX_DROP;
 349
 350         if (dev_net(bond_dev) != &init_net)
 351                 goto out;
 352
 353         if (!(bond_dev->flags & IFF_MASTER))
 354                 goto out;
 355
 356         if (!arp) {
 357                 dprintk("Packet has no ARP data\n");
 358                 goto out;
 359         }
 360
 361         if (skb->len < sizeof(struct arp_pkt)) {
 362                 dprintk("Packet is too small to be an ARP\n");
 363                 goto out;
 364         }
 365
 366         if (arp->op_code == htons(ARPOP_REPLY)) {
 367                 /* update rx hash table for this ARP */
 368                 rlb_update_entry_from_arp(bond, arp);
 369                 dprintk("Server received an ARP Reply from client\n");
 370         }
 371
 372         res = NET_RX_SUCCESS;
 373
 374 out:
 375         dev_kfree_skb(skb);
 376
 377         return res;
 378 }
 379
 380 /* Caller must hold bond lock for read */
 381 static struct slave *rlb_next_rx_slave(struct bonding *bond)
 382 {
 383         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 384         struct slave *rx_slave, *slave, *start_at;
 385         int i = 0;
 386
 387         if (bond_info->next_rx_slave) {
 388                 start_at = bond_info->next_rx_slave;
 389         } else {
 390                 start_at = bond->first_slave;
 391         }
 392
 393         rx_slave = NULL;
 394
 395         bond_for_each_slave_from(bond, slave, i, start_at) {
 396                 if (SLAVE_IS_OK(slave)) {
 397                         if (!rx_slave) {
 398                                 rx_slave = slave;
 399                         } else if (slave->speed > rx_slave->speed) {
 400                                 rx_slave = slave;
 401                         }
 402                 }
 403         }
 404
 405         if (rx_slave) {
 406                 bond_info->next_rx_slave = rx_slave->next;
 407         }
 408
 409         return rx_slave;
 410 }
 411
 412 /* teach the switch the mac of a disabled slave
 413  * on the primary for fault tolerance
 414  *
 415  * Caller must hold bond->curr_slave_lock for write or bond lock for write
 416  */
 417 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
 418 {
 419         if (!bond->curr_active_slave) {
 420                 return;
 421         }
 422
 423         if (!bond->alb_info.primary_is_promisc) {
 424                 if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1))
 425                         bond->alb_info.primary_is_promisc = 1;
 426                 else
 427                         bond->alb_info.primary_is_promisc = 0;
 428         }
 429
 430         bond->alb_info.rlb_promisc_timeout_counter = 0;
 431
 432         alb_send_learning_packets(bond->curr_active_slave, addr);
 433 }
 434
 435 /* slave being removed should not be active at this point
 436  *
 437  * Caller must hold bond lock for read
 438  */
 439 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 440 {
 441         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 442         struct rlb_client_info *rx_hash_table;
 443         u32 index, next_index;
 444
 445         /* clear slave from rx_hashtbl */
 446         _lock_rx_hashtbl(bond);
 447
 448         rx_hash_table = bond_info->rx_hashtbl;
 449         index = bond_info->rx_hashtbl_head;
 450         for (; index != RLB_NULL_INDEX; index = next_index) {
 451                 next_index = rx_hash_table[index].next;
 452                 if (rx_hash_table[index].slave == slave) {
 453                         struct slave *assigned_slave = rlb_next_rx_slave(bond);
 454
 455                         if (assigned_slave) {
 456                                 rx_hash_table[index].slave = assigned_slave;
 457                                 if (memcmp(rx_hash_table[index].mac_dst,
 458                                            mac_bcast, ETH_ALEN)) {
 459                                         bond_info->rx_hashtbl[index].ntt = 1;
 460                                         bond_info->rx_ntt = 1;
 461                                         /* A slave has been removed from the
 462                                          * table because it is either disabled
 463                                          * or being released. We must retry the
 464                                          * update to avoid clients from not
 465                                          * being updated & disconnecting when
 466                                          * there is stress
 467                                          */
 468                                         bond_info->rlb_update_retry_counter =
 469                                                 RLB_UPDATE_RETRY;
 470                                 }
 471                         } else {  /* there is no active slave */
 472                                 rx_hash_table[index].slave = NULL;
 473                         }
 474                 }
 475         }
 476
 477         _unlock_rx_hashtbl(bond);
 478
 479         write_lock_bh(&bond->curr_slave_lock);
 480
 481         if (slave != bond->curr_active_slave) {
 482                 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 483         }
 484
 485         write_unlock_bh(&bond->curr_slave_lock);
 486 }
 487
 488 static void rlb_update_client(struct rlb_client_info *client_info)
 489 {
 490         int i;
 491
 492         if (!client_info->slave) {
 493                 return;
 494         }
 495
 496         for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 497                 struct sk_buff *skb;
 498
 499                 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 500                                  client_info->ip_dst,
 501                                  client_info->slave->dev,
 502                                  client_info->ip_src,
 503                                  client_info->mac_dst,
 504                                  client_info->slave->dev->dev_addr,
 505                                  client_info->mac_dst);
 506                 if (!skb) {
 507                         printk(KERN_ERR DRV_NAME
 508                                ": %s: Error: failed to create an ARP packet\n",
 509                                client_info->slave->dev->master->name);
 510                         continue;
 511                 }
 512
 513                 skb->dev = client_info->slave->dev;
 514
 515                 if (client_info->tag) {
 516                         skb = vlan_put_tag(skb, client_info->vlan_id);
 517                         if (!skb) {
 518                                 printk(KERN_ERR DRV_NAME
 519                                        ": %s: Error: failed to insert VLAN tag\n",
 520                                        client_info->slave->dev->master->name);
 521                                 continue;
 522                         }
 523                 }
 524
 525                 arp_xmit(skb);
 526         }
 527 }
 528
 529 /* sends ARP REPLIES that update the clients that need updating */
 530 static void rlb_update_rx_clients(struct bonding *bond)
 531 {
 532         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 533         struct rlb_client_info *client_info;
 534         u32 hash_index;
 535
 536         _lock_rx_hashtbl(bond);
 537
 538         hash_index = bond_info->rx_hashtbl_head;
 539         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 540                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 541                 if (client_info->ntt) {
 542                         rlb_update_client(client_info);
 543                         if (bond_info->rlb_update_retry_counter == 0) {
 544                                 client_info->ntt = 0;
 545                         }
 546                 }
 547         }
 548
 549         /* do not update the entries again untill this counter is zero so that
 550          * not to confuse the clients.
 551          */
 552         bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 553
 554         _unlock_rx_hashtbl(bond);
 555 }
 556
 557 /* The slave was assigned a new mac address - update the clients */
 558 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 559 {
 560         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 561         struct rlb_client_info *client_info;
 562         int ntt = 0;
 563         u32 hash_index;
 564
 565         _lock_rx_hashtbl(bond);
 566
 567         hash_index = bond_info->rx_hashtbl_head;
 568         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 569                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 570
 571                 if ((client_info->slave == slave) &&
 572                     memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
 573                         client_info->ntt = 1;
 574                         ntt = 1;
 575                 }
 576         }
 577
 578         // update the team's flag only after the whole iteration
 579         if (ntt) {
 580                 bond_info->rx_ntt = 1;
 581                 //fasten the change
 582                 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 583         }
 584
 585         _unlock_rx_hashtbl(bond);
 586 }
 587
 588 /* mark all clients using src_ip to be updated */
 589 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
 590 {
 591         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 592         struct rlb_client_info *client_info;
 593         u32 hash_index;
 594
 595         _lock_rx_hashtbl(bond);
 596
 597         hash_index = bond_info->rx_hashtbl_head;
 598         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 599                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 600
 601                 if (!client_info->slave) {
 602                         printk(KERN_ERR DRV_NAME
 603                                ": %s: Error: found a client with no channel in "
 604                                "the client's hash table\n",
 605                                bond->dev->name);
 606                         continue;
 607                 }
 608                 /*update all clients using this src_ip, that are not assigned
 609                  * to the team's address (curr_active_slave) and have a known
 610                  * unicast mac address.
 611                  */
 612                 if ((client_info->ip_src == src_ip) &&
 613                     memcmp(client_info->slave->dev->dev_addr,
 614                            bond->dev->dev_addr, ETH_ALEN) &&
 615                     memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
 616                         client_info->ntt = 1;
 617                         bond_info->rx_ntt = 1;
 618                 }
 619         }
 620
 621         _unlock_rx_hashtbl(bond);
 622 }
 623
 624 /* Caller must hold both bond and ptr locks for read */
 625 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
 626 {
 627         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 628         struct arp_pkt *arp = arp_pkt(skb);
 629         struct slave *assigned_slave;
 630         struct rlb_client_info *client_info;
 631         u32 hash_index = 0;
 632
 633         _lock_rx_hashtbl(bond);
 634
 635         hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_src));
 636         client_info = &(bond_info->rx_hashtbl[hash_index]);
 637
 638         if (client_info->assigned) {
 639                 if ((client_info->ip_src == arp->ip_src) &&
 640                     (client_info->ip_dst == arp->ip_dst)) {
 641                         /* the entry is already assigned to this client */
 642                         if (memcmp(arp->mac_dst, mac_bcast, ETH_ALEN)) {
 643                                 /* update mac address from arp */
 644                                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 645                         }
 646
 647                         assigned_slave = client_info->slave;
 648                         if (assigned_slave) {
 649                                 _unlock_rx_hashtbl(bond);
 650                                 return assigned_slave;
 651                         }
 652                 } else {
 653                         /* the entry is already assigned to some other client,
 654                          * move the old client to primary (curr_active_slave) so
 655                          * that the new client can be assigned to this entry.
 656                          */
 657                         if (bond->curr_active_slave &&
 658                             client_info->slave != bond->curr_active_slave) {
 659                                 client_info->slave = bond->curr_active_slave;
 660                                 rlb_update_client(client_info);
 661                         }
 662                 }
 663         }
 664         /* assign a new slave */
 665         assigned_slave = rlb_next_rx_slave(bond);
 666
 667         if (assigned_slave) {
 668                 client_info->ip_src = arp->ip_src;
 669                 client_info->ip_dst = arp->ip_dst;
 670                 /* arp->mac_dst is broadcast for arp reqeusts.
 671                  * will be updated with clients actual unicast mac address
 672                  * upon receiving an arp reply.
 673                  */
 674                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 675                 client_info->slave = assigned_slave;
 676
 677                 if (memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
 678                         client_info->ntt = 1;
 679                         bond->alb_info.rx_ntt = 1;
 680                 } else {
 681                         client_info->ntt = 0;
 682                 }
 683
 684                 if (!list_empty(&bond->vlan_list)) {
 685                         if (!vlan_get_tag(skb, &client_info->vlan_id))
 686                                 client_info->tag = 1;
 687                 }
 688
 689                 if (!client_info->assigned) {
 690                         u32 prev_tbl_head = bond_info->rx_hashtbl_head;
 691                         bond_info->rx_hashtbl_head = hash_index;
 692                         client_info->next = prev_tbl_head;
 693                         if (prev_tbl_head != RLB_NULL_INDEX) {
 694                                 bond_info->rx_hashtbl[prev_tbl_head].prev =
 695                                         hash_index;
 696                         }
 697                         client_info->assigned = 1;
 698                 }
 699         }
 700
 701         _unlock_rx_hashtbl(bond);
 702
 703         return assigned_slave;
 704 }
 705
 706 /* chooses (and returns) transmit channel for arp reply
 707  * does not choose channel for other arp types since they are
 708  * sent on the curr_active_slave
 709  */
 710 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 711 {
 712         struct arp_pkt *arp = arp_pkt(skb);
 713         struct slave *tx_slave = NULL;
 714
 715         if (arp->op_code == htons(ARPOP_REPLY)) {
 716                 /* the arp must be sent on the selected
 717                 * rx channel
 718                 */
 719                 tx_slave = rlb_choose_channel(skb, bond);
 720                 if (tx_slave) {
 721                         memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
 722                 }
 723                 dprintk("Server sent ARP Reply packet\n");
 724         } else if (arp->op_code == htons(ARPOP_REQUEST)) {
 725                 /* Create an entry in the rx_hashtbl for this client as a
 726                  * place holder.
 727                  * When the arp reply is received the entry will be updated
 728                  * with the correct unicast address of the client.
 729                  */
 730                 rlb_choose_channel(skb, bond);
 731
 732                 /* The ARP relpy packets must be delayed so that
 733                  * they can cancel out the influence of the ARP request.
 734                  */
 735                 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 736
 737                 /* arp requests are broadcast and are sent on the primary
 738                  * the arp request will collapse all clients on the subnet to
 739                  * the primary slave. We must register these clients to be
 740                  * updated with their assigned mac.
 741                  */
 742                 rlb_req_update_subnet_clients(bond, arp->ip_src);
 743                 dprintk("Server sent ARP Request packet\n");
 744         }
 745
 746         return tx_slave;
 747 }
 748
 749 /* Caller must hold bond lock for read */
 750 static void rlb_rebalance(struct bonding *bond)
 751 {
 752         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 753         struct slave *assigned_slave;
 754         struct rlb_client_info *client_info;
 755         int ntt;
 756         u32 hash_index;
 757
 758         _lock_rx_hashtbl(bond);
 759
 760         ntt = 0;
 761         hash_index = bond_info->rx_hashtbl_head;
 762         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 763                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 764                 assigned_slave = rlb_next_rx_slave(bond);
 765                 if (assigned_slave && (client_info->slave != assigned_slave)) {
 766                         client_info->slave = assigned_slave;
 767                         client_info->ntt = 1;
 768                         ntt = 1;
 769                 }
 770         }
 771
 772         /* update the team's flag only after the whole iteration */
 773         if (ntt) {
 774                 bond_info->rx_ntt = 1;
 775         }
 776         _unlock_rx_hashtbl(bond);
 777 }
 778
 779 /* Caller must hold rx_hashtbl lock */
 780 static void rlb_init_table_entry(struct rlb_client_info *entry)
 781 {
 782         memset(entry, 0, sizeof(struct rlb_client_info));
 783         entry->next = RLB_NULL_INDEX;
 784         entry->prev = RLB_NULL_INDEX;
 785 }
 786
 787 static int rlb_initialize(struct bonding *bond)
 788 {
 789         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 790         struct packet_type *pk_type = &(BOND_ALB_INFO(bond).rlb_pkt_type);
 791         struct rlb_client_info  *new_hashtbl;
 792         int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 793         int i;
 794
 795         spin_lock_init(&(bond_info->rx_hashtbl_lock));
 796
 797         new_hashtbl = kmalloc(size, GFP_KERNEL);
 798         if (!new_hashtbl) {
 799                 printk(KERN_ERR DRV_NAME
 800                        ": %s: Error: Failed to allocate RLB hash table\n",
 801                        bond->dev->name);
 802                 return -1;
 803         }
 804         _lock_rx_hashtbl(bond);
 805
 806         bond_info->rx_hashtbl = new_hashtbl;
 807
 808         bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
 809
 810         for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
 811                 rlb_init_table_entry(bond_info->rx_hashtbl + i);
 812         }
 813
 814         _unlock_rx_hashtbl(bond);
 815
 816         /*initialize packet type*/
 817         pk_type->type = __constant_htons(ETH_P_ARP);
 818         pk_type->dev = bond->dev;
 819         pk_type->func = rlb_arp_recv;
 820
 821         /* register to receive ARPs */
 822         dev_add_pack(pk_type);
 823
 824         return 0;
 825 }
 826
 827 static void rlb_deinitialize(struct bonding *bond)
 828 {
 829         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 830
 831         dev_remove_pack(&(bond_info->rlb_pkt_type));
 832
 833         _lock_rx_hashtbl(bond);
 834
 835         kfree(bond_info->rx_hashtbl);
 836         bond_info->rx_hashtbl = NULL;
 837         bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
 838
 839         _unlock_rx_hashtbl(bond);
 840 }
 841
 842 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 843 {
 844         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 845         u32 curr_index;
 846
 847         _lock_rx_hashtbl(bond);
 848
 849         curr_index = bond_info->rx_hashtbl_head;
 850         while (curr_index != RLB_NULL_INDEX) {
 851                 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 852                 u32 next_index = bond_info->rx_hashtbl[curr_index].next;
 853                 u32 prev_index = bond_info->rx_hashtbl[curr_index].prev;
 854
 855                 if (curr->tag && (curr->vlan_id == vlan_id)) {
 856                         if (curr_index == bond_info->rx_hashtbl_head) {
 857                                 bond_info->rx_hashtbl_head = next_index;
 858                         }
 859                         if (prev_index != RLB_NULL_INDEX) {
 860                                 bond_info->rx_hashtbl[prev_index].next = next_index;
 861                         }
 862                         if (next_index != RLB_NULL_INDEX) {
 863                                 bond_info->rx_hashtbl[next_index].prev = prev_index;
 864                         }
 865
 866                         rlb_init_table_entry(curr);
 867                 }
 868
 869                 curr_index = next_index;
 870         }
 871
 872         _unlock_rx_hashtbl(bond);
 873 }
 874
 875 /*********************** tlb/rlb shared functions *********************/
 876
 877 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
 878 {
 879         struct bonding *bond = bond_get_bond_by_slave(slave);
 880         struct learning_pkt pkt;
 881         int size = sizeof(struct learning_pkt);
 882         int i;
 883
 884         memset(&pkt, 0, size);
 885         memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
 886         memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
 887         pkt.type = __constant_htons(ETH_P_LOOP);
 888
 889         for (i = 0; i < MAX_LP_BURST; i++) {
 890                 struct sk_buff *skb;
 891                 char *data;
 892
 893                 skb = dev_alloc_skb(size);
 894                 if (!skb) {
 895                         return;
 896                 }
 897
 898                 data = skb_put(skb, size);
 899                 memcpy(data, &pkt, size);
 900
 901                 skb_reset_mac_header(skb);
 902                 skb->network_header = skb->mac_header + ETH_HLEN;
 903                 skb->protocol = pkt.type;
 904                 skb->priority = TC_PRIO_CONTROL;
 905                 skb->dev = slave->dev;
 906
 907                 if (!list_empty(&bond->vlan_list)) {
 908                         struct vlan_entry *vlan;
 909
 910                         vlan = bond_next_vlan(bond,
 911                                               bond->alb_info.current_alb_vlan);
 912
 913                         bond->alb_info.current_alb_vlan = vlan;
 914                         if (!vlan) {
 915                                 kfree_skb(skb);
 916                                 continue;
 917                         }
 918
 919                         skb = vlan_put_tag(skb, vlan->vlan_id);
 920                         if (!skb) {
 921                                 printk(KERN_ERR DRV_NAME
 922                                        ": %s: Error: failed to insert VLAN tag\n",
 923                                        bond->dev->name);
 924                                 continue;
 925                         }
 926                 }
 927
 928                 dev_queue_xmit(skb);
 929         }
 930 }
 931
 932 /* hw is a boolean parameter that determines whether we should try and
 933  * set the hw address of the device as well as the hw address of the
 934  * net_device
 935  */
 936 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[], int hw)
 937 {
 938         struct net_device *dev = slave->dev;
 939         struct sockaddr s_addr;
 940
 941         if (!hw) {
 942                 memcpy(dev->dev_addr, addr, dev->addr_len);
 943                 return 0;
 944         }
 945
 946         /* for rlb each slave must have a unique hw mac addresses so that */
 947         /* each slave will receive packets destined to a different mac */
 948         memcpy(s_addr.sa_data, addr, dev->addr_len);
 949         s_addr.sa_family = dev->type;
 950         if (dev_set_mac_address(dev, &s_addr)) {
 951                 printk(KERN_ERR DRV_NAME
 952                        ": %s: Error: dev_set_mac_address of dev %s failed! ALB "
 953                        "mode requires that the base driver support setting "
 954                        "the hw address also when the network device's "
 955                        "interface is open\n",
 956                        dev->master->name, dev->name);
 957                 return -EOPNOTSUPP;
 958         }
 959         return 0;
 960 }
 961
 962 /*
 963  * Swap MAC addresses between two slaves.
 964  *
 965  * Called with RTNL held, and no other locks.
 966  *
 967  */
 968
 969 static void alb_swap_mac_addr(struct bonding *bond, struct slave *slave1, struct slave *slave2)
 970 {
 971         u8 tmp_mac_addr[ETH_ALEN];
 972
 973         memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
 974         alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr, bond->alb_info.rlb_enabled);
 975         alb_set_slave_mac_addr(slave2, tmp_mac_addr, bond->alb_info.rlb_enabled);
 976
 977 }
 978
 979 /*
 980  * Send learning packets after MAC address swap.
 981  *
 982  * Called with RTNL and no other locks
 983  */
 984 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
 985                                 struct slave *slave2)
 986 {
 987         int slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
 988         struct slave *disabled_slave = NULL;
 989
 990         ASSERT_RTNL();
 991
 992         /* fasten the change in the switch */
 993         if (SLAVE_IS_OK(slave1)) {
 994                 alb_send_learning_packets(slave1, slave1->dev->dev_addr);
 995                 if (bond->alb_info.rlb_enabled) {
 996                         /* inform the clients that the mac address
 997                          * has changed
 998                          */
 999                         rlb_req_update_slave_clients(bond, slave1);
1000                 }
1001         } else {
1002                 disabled_slave = slave1;
1003         }
1004
1005         if (SLAVE_IS_OK(slave2)) {
1006                 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
1007                 if (bond->alb_info.rlb_enabled) {
1008                         /* inform the clients that the mac address
1009                          * has changed
1010                          */
1011                         rlb_req_update_slave_clients(bond, slave2);
1012                 }
1013         } else {
1014                 disabled_slave = slave2;
1015         }
1016
1017         if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1018                 /* A disabled slave was assigned an active mac addr */
1019                 rlb_teach_disabled_mac_on_primary(bond,
1020                                                   disabled_slave->dev->dev_addr);
1021         }
1022 }
1023
1024 /**
1025  * alb_change_hw_addr_on_detach
1026  * @bond: bonding we're working on
1027  * @slave: the slave that was just detached
1028  *
1029  * We assume that @slave was already detached from the slave list.
1030  *
1031  * If @slave's permanent hw address is different both from its current
1032  * address and from @bond's address, then somewhere in the bond there's
1033  * a slave that has @slave's permanet address as its current address.
1034  * We'll make sure that that slave no longer uses @slave's permanent address.
1035  *
1036  * Caller must hold RTNL and no other locks
1037  */
1038 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1039 {
1040         int perm_curr_diff;
1041         int perm_bond_diff;
1042
1043         perm_curr_diff = memcmp(slave->perm_hwaddr,
1044                                 slave->dev->dev_addr,
1045                                 ETH_ALEN);
1046         perm_bond_diff = memcmp(slave->perm_hwaddr,
1047                                 bond->dev->dev_addr,
1048                                 ETH_ALEN);
1049
1050         if (perm_curr_diff && perm_bond_diff) {
1051                 struct slave *tmp_slave;
1052                 int i, found = 0;
1053
1054                 bond_for_each_slave(bond, tmp_slave, i) {
1055                         if (!memcmp(slave->perm_hwaddr,
1056                                     tmp_slave->dev->dev_addr,
1057                                     ETH_ALEN)) {
1058                                 found = 1;
1059                                 break;
1060                         }
1061                 }
1062
1063                 if (found) {
1064                         /* locking: needs RTNL and nothing else */
1065                         alb_swap_mac_addr(bond, slave, tmp_slave);
1066                         alb_fasten_mac_swap(bond, slave, tmp_slave);
1067                 }
1068         }
1069 }
1070
1071 /**
1072  * alb_handle_addr_collision_on_attach
1073  * @bond: bonding we're working on
1074  * @slave: the slave that was just attached
1075  *
1076  * checks uniqueness of slave's mac address and handles the case the
1077  * new slave uses the bonds mac address.
1078  *
1079  * If the permanent hw address of @slave is @bond's hw address, we need to
1080  * find a different hw address to give @slave, that isn't in use by any other
1081  * slave in the bond. This address must be, of course, one of the premanent
1082  * addresses of the other slaves.
1083  *
1084  * We go over the slave list, and for each slave there we compare its
1085  * permanent hw address with the current address of all the other slaves.
1086  * If no match was found, then we've found a slave with a permanent address
1087  * that isn't used by any other slave in the bond, so we can assign it to
1088  * @slave.
1089  *
1090  * assumption: this function is called before @slave is attached to the
1091  *             bond slave list.
1092  *
1093  * caller must hold the bond lock for write since the mac addresses are compared
1094  * and may be swapped.
1095  */
1096 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1097 {
1098         struct slave *tmp_slave1, *tmp_slave2, *free_mac_slave;
1099         struct slave *has_bond_addr = bond->curr_active_slave;
1100         int i, j, found = 0;
1101
1102         if (bond->slave_cnt == 0) {
1103                 /* this is the first slave */
1104                 return 0;
1105         }
1106
1107         /* if slave's mac address differs from bond's mac address
1108          * check uniqueness of slave's mac address against the other
1109          * slaves in the bond.
1110          */
1111         if (memcmp(slave->perm_hwaddr, bond->dev->dev_addr, ETH_ALEN)) {
1112                 bond_for_each_slave(bond, tmp_slave1, i) {
1113                         if (!memcmp(tmp_slave1->dev->dev_addr, slave->dev->dev_addr,
1114                                     ETH_ALEN)) {
1115                                 found = 1;
1116                                 break;
1117                         }
1118                 }
1119
1120                 if (!found)
1121                         return 0;
1122
1123                 /* Try setting slave mac to bond address and fall-through
1124                    to code handling that situation below... */
1125                 alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1126                                        bond->alb_info.rlb_enabled);
1127         }
1128
1129         /* The slave's address is equal to the address of the bond.
1130          * Search for a spare address in the bond for this slave.
1131          */
1132         free_mac_slave = NULL;
1133
1134         bond_for_each_slave(bond, tmp_slave1, i) {
1135                 found = 0;
1136                 bond_for_each_slave(bond, tmp_slave2, j) {
1137                         if (!memcmp(tmp_slave1->perm_hwaddr,
1138                                     tmp_slave2->dev->dev_addr,
1139                                     ETH_ALEN)) {
1140                                 found = 1;
1141                                 break;
1142                         }
1143                 }
1144
1145                 if (!found) {
1146                         /* no slave has tmp_slave1's perm addr
1147                          * as its curr addr
1148                          */
1149                         free_mac_slave = tmp_slave1;
1150                         break;
1151                 }
1152
1153                 if (!has_bond_addr) {
1154                         if (!memcmp(tmp_slave1->dev->dev_addr,
1155                                     bond->dev->dev_addr,
1156                                     ETH_ALEN)) {
1157
1158                                 has_bond_addr = tmp_slave1;
1159                         }
1160                 }
1161         }
1162
1163         if (free_mac_slave) {
1164                 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1165                                        bond->alb_info.rlb_enabled);
1166
1167                 printk(KERN_WARNING DRV_NAME
1168                        ": %s: Warning: the hw address of slave %s is in use by "
1169                        "the bond; giving it the hw address of %s\n",
1170                        bond->dev->name, slave->dev->name, free_mac_slave->dev->name);
1171
1172         } else if (has_bond_addr) {
1173                 printk(KERN_ERR DRV_NAME
1174                        ": %s: Error: the hw address of slave %s is in use by the "
1175                        "bond; couldn't find a slave with a free hw address to "
1176                        "give it (this should not have happened)\n",
1177                        bond->dev->name, slave->dev->name);
1178                 return -EFAULT;
1179         }
1180
1181         return 0;
1182 }
1183
1184 /**
1185  * alb_set_mac_address
1186  * @bond:
1187  * @addr:
1188  *
1189  * In TLB mode all slaves are configured to the bond's hw address, but set
1190  * their dev_addr field to different addresses (based on their permanent hw
1191  * addresses).
1192  *
1193  * For each slave, this function sets the interface to the new address and then
1194  * changes its dev_addr field to its previous value.
1195  *
1196  * Unwinding assumes bond's mac address has not yet changed.
1197  */
1198 static int alb_set_mac_address(struct bonding *bond, void *addr)
1199 {
1200         struct sockaddr sa;
1201         struct slave *slave, *stop_at;
1202         char tmp_addr[ETH_ALEN];
1203         int res;
1204         int i;
1205
1206         if (bond->alb_info.rlb_enabled) {
1207                 return 0;
1208         }
1209
1210         bond_for_each_slave(bond, slave, i) {
1211                 if (slave->dev->set_mac_address == NULL) {
1212                         res = -EOPNOTSUPP;
1213                         goto unwind;
1214                 }
1215
1216                 /* save net_device's current hw address */
1217                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1218
1219                 res = dev_set_mac_address(slave->dev, addr);
1220
1221                 /* restore net_device's hw address */
1222                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1223
1224                 if (res) {
1225                         goto unwind;
1226                 }
1227         }
1228
1229         return 0;
1230
1231 unwind:
1232         memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1233         sa.sa_family = bond->dev->type;
1234
1235         /* unwind from head to the slave that failed */
1236         stop_at = slave;
1237         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
1238                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1239                 dev_set_mac_address(slave->dev, &sa);
1240                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1241         }
1242
1243         return res;
1244 }
1245
1246 /************************ exported alb funcions ************************/
1247
1248 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1249 {
1250         int res;
1251
1252         res = tlb_initialize(bond);
1253         if (res) {
1254                 return res;
1255         }
1256
1257         if (rlb_enabled) {
1258                 bond->alb_info.rlb_enabled = 1;
1259                 /* initialize rlb */
1260                 res = rlb_initialize(bond);
1261                 if (res) {
1262                         tlb_deinitialize(bond);
1263                         return res;
1264                 }
1265         } else {
1266                 bond->alb_info.rlb_enabled = 0;
1267         }
1268
1269         return 0;
1270 }
1271
1272 void bond_alb_deinitialize(struct bonding *bond)
1273 {
1274         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1275
1276         tlb_deinitialize(bond);
1277
1278         if (bond_info->rlb_enabled) {
1279                 rlb_deinitialize(bond);
1280         }
1281 }
1282
1283 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1284 {
1285         struct bonding *bond = bond_dev->priv;
1286         struct ethhdr *eth_data;
1287         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1288         struct slave *tx_slave = NULL;
1289         static const __be32 ip_bcast = htonl(0xffffffff);
1290         int hash_size = 0;
1291         int do_tx_balance = 1;
1292         u32 hash_index = 0;
1293         const u8 *hash_start = NULL;
1294         int res = 1;
1295         struct ipv6hdr *ip6hdr;
1296
1297         skb_reset_mac_header(skb);
1298         eth_data = eth_hdr(skb);
1299
1300         /* make sure that the curr_active_slave and the slaves list do
1301          * not change during tx
1302          */
1303         read_lock(&bond->lock);
1304         read_lock(&bond->curr_slave_lock);
1305
1306         if (!BOND_IS_OK(bond)) {
1307                 goto out;
1308         }
1309
1310         switch (ntohs(skb->protocol)) {
1311         case ETH_P_IP: {
1312                 const struct iphdr *iph = ip_hdr(skb);
1313
1314                 if ((memcmp(eth_data->h_dest, mac_bcast, ETH_ALEN) == 0) ||
1315                     (iph->daddr == ip_bcast) ||
1316                     (iph->protocol == IPPROTO_IGMP)) {
1317                         do_tx_balance = 0;
1318                         break;
1319                 }
1320                 hash_start = (char *)&(iph->daddr);
1321                 hash_size = sizeof(iph->daddr);
1322         }
1323                 break;
1324         case ETH_P_IPV6:
1325                 /* IPv6 doesn't really use broadcast mac address, but leave
1326                  * that here just in case.
1327                  */
1328                 if (memcmp(eth_data->h_dest, mac_bcast, ETH_ALEN) == 0) {
1329                         do_tx_balance = 0;
1330                         break;
1331                 }
1332
1333                 /* IPv6 uses all-nodes multicast as an equivalent to
1334                  * broadcasts in IPv4.
1335                  */
1336                 if (memcmp(eth_data->h_dest, mac_v6_allmcast, ETH_ALEN) == 0) {
1337                         do_tx_balance = 0;
1338                         break;
1339                 }
1340
1341                 /* Additianally, DAD probes should not be tx-balanced as that
1342                  * will lead to false positives for duplicate addresses and
1343                  * prevent address configuration from working.
1344                  */
1345                 ip6hdr = ipv6_hdr(skb);
1346                 if (ipv6_addr_any(&ip6hdr->saddr)) {
1347                         do_tx_balance = 0;
1348                         break;
1349                 }
1350
1351                 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1352                 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1353                 break;
1354         case ETH_P_IPX:
1355                 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1356                         /* something is wrong with this packet */
1357                         do_tx_balance = 0;
1358                         break;
1359                 }
1360
1361                 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1362                         /* The only protocol worth balancing in
1363                          * this family since it has an "ARP" like
1364                          * mechanism
1365                          */
1366                         do_tx_balance = 0;
1367                         break;
1368                 }
1369
1370                 hash_start = (char*)eth_data->h_dest;
1371                 hash_size = ETH_ALEN;
1372                 break;
1373         case ETH_P_ARP:
1374                 do_tx_balance = 0;
1375                 if (bond_info->rlb_enabled) {
1376                         tx_slave = rlb_arp_xmit(skb, bond);
1377                 }
1378                 break;
1379         default:
1380                 do_tx_balance = 0;
1381                 break;
1382         }
1383
1384         if (do_tx_balance) {
1385                 hash_index = _simple_hash(hash_start, hash_size);
1386                 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1387         }
1388
1389         if (!tx_slave) {
1390                 /* unbalanced or unassigned, send through primary */
1391                 tx_slave = bond->curr_active_slave;
1392                 bond_info->unbalanced_load += skb->len;
1393         }
1394
1395         if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1396                 if (tx_slave != bond->curr_active_slave) {
1397                         memcpy(eth_data->h_source,
1398                                tx_slave->dev->dev_addr,
1399                                ETH_ALEN);
1400                 }
1401
1402                 res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1403         } else {
1404                 if (tx_slave) {
1405                         tlb_clear_slave(bond, tx_slave, 0);
1406                 }
1407         }
1408
1409 out:
1410         if (res) {
1411                 /* no suitable interface, frame not sent */
1412                 dev_kfree_skb(skb);
1413         }
1414         read_unlock(&bond->curr_slave_lock);
1415         read_unlock(&bond->lock);
1416         return 0;
1417 }
1418
1419 void bond_alb_monitor(struct work_struct *work)
1420 {
1421         struct bonding *bond = container_of(work, struct bonding,
1422                                             alb_work.work);
1423         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1424         struct slave *slave;
1425         int i;
1426
1427         read_lock(&bond->lock);
1428
1429         if (bond->kill_timers) {
1430                 goto out;
1431         }
1432
1433         if (bond->slave_cnt == 0) {
1434                 bond_info->tx_rebalance_counter = 0;
1435                 bond_info->lp_counter = 0;
1436                 goto re_arm;
1437         }
1438
1439         bond_info->tx_rebalance_counter++;
1440         bond_info->lp_counter++;
1441
1442         /* send learning packets */
1443         if (bond_info->lp_counter >= BOND_ALB_LP_TICKS) {
1444                 /* change of curr_active_slave involves swapping of mac addresses.
1445                  * in order to avoid this swapping from happening while
1446                  * sending the learning packets, the curr_slave_lock must be held for
1447                  * read.
1448                  */
1449                 read_lock(&bond->curr_slave_lock);
1450
1451                 bond_for_each_slave(bond, slave, i) {
1452                         alb_send_learning_packets(slave, slave->dev->dev_addr);
1453                 }
1454
1455                 read_unlock(&bond->curr_slave_lock);
1456
1457                 bond_info->lp_counter = 0;
1458         }
1459
1460         /* rebalance tx traffic */
1461         if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1462
1463                 read_lock(&bond->curr_slave_lock);
1464
1465                 bond_for_each_slave(bond, slave, i) {
1466                         tlb_clear_slave(bond, slave, 1);
1467                         if (slave == bond->curr_active_slave) {
1468                                 SLAVE_TLB_INFO(slave).load =
1469                                         bond_info->unbalanced_load /
1470                                                 BOND_TLB_REBALANCE_INTERVAL;
1471                                 bond_info->unbalanced_load = 0;
1472                         }
1473                 }
1474
1475                 read_unlock(&bond->curr_slave_lock);
1476
1477                 bond_info->tx_rebalance_counter = 0;
1478         }
1479
1480         /* handle rlb stuff */
1481         if (bond_info->rlb_enabled) {
1482                 if (bond_info->primary_is_promisc &&
1483                     (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1484
1485                         /*
1486                          * dev_set_promiscuity requires rtnl and
1487                          * nothing else.
1488                          */
1489                         read_unlock(&bond->lock);
1490                         rtnl_lock();
1491
1492                         bond_info->rlb_promisc_timeout_counter = 0;
1493
1494                         /* If the primary was set to promiscuous mode
1495                          * because a slave was disabled then
1496                          * it can now leave promiscuous mode.
1497                          */
1498                         dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1499                         bond_info->primary_is_promisc = 0;
1500
1501                         rtnl_unlock();
1502                         read_lock(&bond->lock);
1503                 }
1504
1505                 if (bond_info->rlb_rebalance) {
1506                         bond_info->rlb_rebalance = 0;
1507                         rlb_rebalance(bond);
1508                 }
1509
1510                 /* check if clients need updating */
1511                 if (bond_info->rx_ntt) {
1512                         if (bond_info->rlb_update_delay_counter) {
1513                                 --bond_info->rlb_update_delay_counter;
1514                         } else {
1515                                 rlb_update_rx_clients(bond);
1516                                 if (bond_info->rlb_update_retry_counter) {
1517                                         --bond_info->rlb_update_retry_counter;
1518                                 } else {
1519                                         bond_info->rx_ntt = 0;
1520                                 }
1521                         }
1522                 }
1523         }
1524
1525 re_arm:
1526         queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1527 out:
1528         read_unlock(&bond->lock);
1529 }
1530
1531 /* assumption: called before the slave is attached to the bond
1532  * and not locked by the bond lock
1533  */
1534 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1535 {
1536         int res;
1537
1538         res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1539                                      bond->alb_info.rlb_enabled);
1540         if (res) {
1541                 return res;
1542         }
1543
1544         /* caller must hold the bond lock for write since the mac addresses
1545          * are compared and may be swapped.
1546          */
1547         read_lock(&bond->lock);
1548
1549         res = alb_handle_addr_collision_on_attach(bond, slave);
1550
1551         read_unlock(&bond->lock);
1552
1553         if (res) {
1554                 return res;
1555         }
1556
1557         tlb_init_slave(slave);
1558
1559         /* order a rebalance ASAP */
1560         bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1561
1562         if (bond->alb_info.rlb_enabled) {
1563                 bond->alb_info.rlb_rebalance = 1;
1564         }
1565
1566         return 0;
1567 }
1568
1569 /*
1570  * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1571  * if necessary.
1572  *
1573  * Caller must hold RTNL and no other locks
1574  */
1575 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1576 {
1577         if (bond->slave_cnt > 1) {
1578                 alb_change_hw_addr_on_detach(bond, slave);
1579         }
1580
1581         tlb_clear_slave(bond, slave, 0);
1582
1583         if (bond->alb_info.rlb_enabled) {
1584                 bond->alb_info.next_rx_slave = NULL;
1585                 rlb_clear_slave(bond, slave);
1586         }
1587 }
1588
1589 /* Caller must hold bond lock for read */
1590 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1591 {
1592         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1593
1594         if (link == BOND_LINK_DOWN) {
1595                 tlb_clear_slave(bond, slave, 0);
1596                 if (bond->alb_info.rlb_enabled) {
1597                         rlb_clear_slave(bond, slave);
1598                 }
1599         } else if (link == BOND_LINK_UP) {
1600                 /* order a rebalance ASAP */
1601                 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1602                 if (bond->alb_info.rlb_enabled) {
1603                         bond->alb_info.rlb_rebalance = 1;
1604                         /* If the updelay module parameter is smaller than the
1605                          * forwarding delay of the switch the rebalance will
1606                          * not work because the rebalance arp replies will
1607                          * not be forwarded to the clients..
1608                          */
1609                 }
1610         }
1611 }
1612
1613 /**
1614  * bond_alb_handle_active_change - assign new curr_active_slave
1615  * @bond: our bonding struct
1616  * @new_slave: new slave to assign
1617  *
1618  * Set the bond->curr_active_slave to @new_slave and handle
1619  * mac address swapping and promiscuity changes as needed.
1620  *
1621  * If new_slave is NULL, caller must hold curr_slave_lock or
1622  * bond->lock for write.
1623  *
1624  * If new_slave is not NULL, caller must hold RTNL, bond->lock for
1625  * read and curr_slave_lock for write.  Processing here may sleep, so
1626  * no other locks may be held.
1627  */
1628 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1629 {
1630         struct slave *swap_slave;
1631         int i;
1632
1633         if (bond->curr_active_slave == new_slave) {
1634                 return;
1635         }
1636
1637         if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1638                 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1639                 bond->alb_info.primary_is_promisc = 0;
1640                 bond->alb_info.rlb_promisc_timeout_counter = 0;
1641         }
1642
1643         swap_slave = bond->curr_active_slave;
1644         bond->curr_active_slave = new_slave;
1645
1646         if (!new_slave || (bond->slave_cnt == 0)) {
1647                 return;
1648         }
1649
1650         /* set the new curr_active_slave to the bonds mac address
1651          * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1652          */
1653         if (!swap_slave) {
1654                 struct slave *tmp_slave;
1655                 /* find slave that is holding the bond's mac address */
1656                 bond_for_each_slave(bond, tmp_slave, i) {
1657                         if (!memcmp(tmp_slave->dev->dev_addr,
1658                                     bond->dev->dev_addr, ETH_ALEN)) {
1659                                 swap_slave = tmp_slave;
1660                                 break;
1661                         }
1662                 }
1663         }
1664
1665         /*
1666          * Arrange for swap_slave and new_slave to temporarily be
1667          * ignored so we can mess with their MAC addresses without
1668          * fear of interference from transmit activity.
1669          */
1670         if (swap_slave) {
1671                 tlb_clear_slave(bond, swap_slave, 1);
1672         }
1673         tlb_clear_slave(bond, new_slave, 1);
1674
1675         write_unlock_bh(&bond->curr_slave_lock);
1676         read_unlock(&bond->lock);
1677
1678         ASSERT_RTNL();
1679
1680         /* curr_active_slave must be set before calling alb_swap_mac_addr */
1681         if (swap_slave) {
1682                 /* swap mac address */
1683                 alb_swap_mac_addr(bond, swap_slave, new_slave);
1684         } else {
1685                 /* set the new_slave to the bond mac address */
1686                 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1687                                        bond->alb_info.rlb_enabled);
1688         }
1689
1690         if (swap_slave) {
1691                 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1692                 read_lock(&bond->lock);
1693         } else {
1694                 read_lock(&bond->lock);
1695                 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1696         }
1697
1698         write_lock_bh(&bond->curr_slave_lock);
1699 }
1700
1701 /*
1702  * Called with RTNL
1703  */
1704 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1705 {
1706         struct bonding *bond = bond_dev->priv;
1707         struct sockaddr *sa = addr;
1708         struct slave *slave, *swap_slave;
1709         int res;
1710         int i;
1711
1712         if (!is_valid_ether_addr(sa->sa_data)) {
1713                 return -EADDRNOTAVAIL;
1714         }
1715
1716         res = alb_set_mac_address(bond, addr);
1717         if (res) {
1718                 return res;
1719         }
1720
1721         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1722
1723         /* If there is no curr_active_slave there is nothing else to do.
1724          * Otherwise we'll need to pass the new address to it and handle
1725          * duplications.
1726          */
1727         if (!bond->curr_active_slave) {
1728                 return 0;
1729         }
1730
1731         swap_slave = NULL;
1732
1733         bond_for_each_slave(bond, slave, i) {
1734                 if (!memcmp(slave->dev->dev_addr, bond_dev->dev_addr, ETH_ALEN)) {
1735                         swap_slave = slave;
1736                         break;
1737                 }
1738         }
1739
1740         write_unlock_bh(&bond->curr_slave_lock);
1741         read_unlock(&bond->lock);
1742
1743         if (swap_slave) {
1744                 alb_swap_mac_addr(bond, swap_slave, bond->curr_active_slave);
1745                 alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave);
1746         } else {
1747                 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr,
1748                                        bond->alb_info.rlb_enabled);
1749
1750                 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1751                 if (bond->alb_info.rlb_enabled) {
1752                         /* inform clients mac address has changed */
1753                         rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1754                 }
1755         }
1756
1757         read_lock(&bond->lock);
1758         write_lock_bh(&bond->curr_slave_lock);
1759
1760         return 0;
1761 }
1762
1763 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1764 {
1765         if (bond->alb_info.current_alb_vlan &&
1766             (bond->alb_info.current_alb_vlan->vlan_id == vlan_id)) {
1767                 bond->alb_info.current_alb_vlan = NULL;
1768         }
1769
1770         if (bond->alb_info.rlb_enabled) {
1771                 rlb_clear_vlan(bond, vlan_id);
1772         }
1773 }
1774