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         /* skip this if we've already freed the tx hash table */
 173         if (tx_hash_table) {
 174                 index = SLAVE_TLB_INFO(slave).head;
 175                 while (index != TLB_NULL_INDEX) {
 176                         u32 next_index = tx_hash_table[index].next;
 177                         tlb_init_table_entry(&tx_hash_table[index], save_load);
 178                         index = next_index;
 179                 }
 180         }
 181
 182         tlb_init_slave(slave);
 183
 184         _unlock_tx_hashtbl(bond);
 185 }
 186
 187 /* Must be called before starting the monitor timer */
 188 static int tlb_initialize(struct bonding *bond)
 189 {
 190         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 191         int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 192         struct tlb_client_info *new_hashtbl;
 193         int i;
 194
 195         spin_lock_init(&(bond_info->tx_hashtbl_lock));
 196
 197         new_hashtbl = kzalloc(size, GFP_KERNEL);
 198         if (!new_hashtbl) {
 199                 printk(KERN_ERR DRV_NAME
 200                        ": %s: Error: Failed to allocate TLB hash table\n",
 201                        bond->dev->name);
 202                 return -1;
 203         }
 204         _lock_tx_hashtbl(bond);
 205
 206         bond_info->tx_hashtbl = new_hashtbl;
 207
 208         for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
 209                 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 1);
 210         }
 211
 212         _unlock_tx_hashtbl(bond);
 213
 214         return 0;
 215 }
 216
 217 /* Must be called only after all slaves have been released */
 218 static void tlb_deinitialize(struct bonding *bond)
 219 {
 220         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 221
 222         _lock_tx_hashtbl(bond);
 223
 224         kfree(bond_info->tx_hashtbl);
 225         bond_info->tx_hashtbl = NULL;
 226
 227         _unlock_tx_hashtbl(bond);
 228 }
 229
 230 /* Caller must hold bond lock for read */
 231 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
 232 {
 233         struct slave *slave, *least_loaded;
 234         s64 max_gap;
 235         int i, found = 0;
 236
 237         /* Find the first enabled slave */
 238         bond_for_each_slave(bond, slave, i) {
 239                 if (SLAVE_IS_OK(slave)) {
 240                         found = 1;
 241                         break;
 242                 }
 243         }
 244
 245         if (!found) {
 246                 return NULL;
 247         }
 248
 249         least_loaded = slave;
 250         max_gap = (s64)(slave->speed << 20) - /* Convert to Megabit per sec */
 251                         (s64)(SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
 252
 253         /* Find the slave with the largest gap */
 254         bond_for_each_slave_from(bond, slave, i, least_loaded) {
 255                 if (SLAVE_IS_OK(slave)) {
 256                         s64 gap = (s64)(slave->speed << 20) -
 257                                         (s64)(SLAVE_TLB_INFO(slave).load << 3);
 258                         if (max_gap < gap) {
 259                                 least_loaded = slave;
 260                                 max_gap = gap;
 261                         }
 262                 }
 263         }
 264
 265         return least_loaded;
 266 }
 267
 268 /* Caller must hold bond lock for read */
 269 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index, u32 skb_len)
 270 {
 271         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 272         struct tlb_client_info *hash_table;
 273         struct slave *assigned_slave;
 274
 275         _lock_tx_hashtbl(bond);
 276
 277         hash_table = bond_info->tx_hashtbl;
 278         assigned_slave = hash_table[hash_index].tx_slave;
 279         if (!assigned_slave) {
 280                 assigned_slave = tlb_get_least_loaded_slave(bond);
 281
 282                 if (assigned_slave) {
 283                         struct tlb_slave_info *slave_info =
 284                                 &(SLAVE_TLB_INFO(assigned_slave));
 285                         u32 next_index = slave_info->head;
 286
 287                         hash_table[hash_index].tx_slave = assigned_slave;
 288                         hash_table[hash_index].next = next_index;
 289                         hash_table[hash_index].prev = TLB_NULL_INDEX;
 290
 291                         if (next_index != TLB_NULL_INDEX) {
 292                                 hash_table[next_index].prev = hash_index;
 293                         }
 294
 295                         slave_info->head = hash_index;
 296                         slave_info->load +=
 297                                 hash_table[hash_index].load_history;
 298                 }
 299         }
 300
 301         if (assigned_slave) {
 302                 hash_table[hash_index].tx_bytes += skb_len;
 303         }
 304
 305         _unlock_tx_hashtbl(bond);
 306
 307         return assigned_slave;
 308 }
 309
 310 /*********************** rlb specific functions ***************************/
 311 static inline void _lock_rx_hashtbl(struct bonding *bond)
 312 {
 313         spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 314 }
 315
 316 static inline void _unlock_rx_hashtbl(struct bonding *bond)
 317 {
 318         spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 319 }
 320
 321 /* when an ARP REPLY is received from a client update its info
 322  * in the rx_hashtbl
 323  */
 324 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 325 {
 326         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 327         struct rlb_client_info *client_info;
 328         u32 hash_index;
 329
 330         _lock_rx_hashtbl(bond);
 331
 332         hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
 333         client_info = &(bond_info->rx_hashtbl[hash_index]);
 334
 335         if ((client_info->assigned) &&
 336             (client_info->ip_src == arp->ip_dst) &&
 337             (client_info->ip_dst == arp->ip_src)) {
 338                 /* update the clients MAC address */
 339                 memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
 340                 client_info->ntt = 1;
 341                 bond_info->rx_ntt = 1;
 342         }
 343
 344         _unlock_rx_hashtbl(bond);
 345 }
 346
 347 static int rlb_arp_recv(struct sk_buff *skb, struct net_device *bond_dev, struct packet_type *ptype, struct net_device *orig_dev)
 348 {
 349         struct bonding *bond;
 350         struct arp_pkt *arp = (struct arp_pkt *)skb->data;
 351         int res = NET_RX_DROP;
 352
 353         if (dev_net(bond_dev) != &init_net)
 354                 goto out;
 355
 356         while (bond_dev->priv_flags & IFF_802_1Q_VLAN)
 357                 bond_dev = vlan_dev_real_dev(bond_dev);
 358
 359         if (!(bond_dev->priv_flags & IFF_BONDING) ||
 360             !(bond_dev->flags & IFF_MASTER))
 361                 goto out;
 362
 363         if (!arp) {
 364                 dprintk("Packet has no ARP data\n");
 365                 goto out;
 366         }
 367
 368         if (skb->len < sizeof(struct arp_pkt)) {
 369                 dprintk("Packet is too small to be an ARP\n");
 370                 goto out;
 371         }
 372
 373         if (arp->op_code == htons(ARPOP_REPLY)) {
 374                 /* update rx hash table for this ARP */
 375                 printk("rar: update orig %s bond_dev %s\n", orig_dev->name,
 376                        bond_dev->name);
 377                 bond = netdev_priv(bond_dev);
 378                 rlb_update_entry_from_arp(bond, arp);
 379                 dprintk("Server received an ARP Reply from client\n");
 380         }
 381
 382         res = NET_RX_SUCCESS;
 383
 384 out:
 385         dev_kfree_skb(skb);
 386
 387         return res;
 388 }
 389
 390 /* Caller must hold bond lock for read */
 391 static struct slave *rlb_next_rx_slave(struct bonding *bond)
 392 {
 393         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 394         struct slave *rx_slave, *slave, *start_at;
 395         int i = 0;
 396
 397         if (bond_info->next_rx_slave) {
 398                 start_at = bond_info->next_rx_slave;
 399         } else {
 400                 start_at = bond->first_slave;
 401         }
 402
 403         rx_slave = NULL;
 404
 405         bond_for_each_slave_from(bond, slave, i, start_at) {
 406                 if (SLAVE_IS_OK(slave)) {
 407                         if (!rx_slave) {
 408                                 rx_slave = slave;
 409                         } else if (slave->speed > rx_slave->speed) {
 410                                 rx_slave = slave;
 411                         }
 412                 }
 413         }
 414
 415         if (rx_slave) {
 416                 bond_info->next_rx_slave = rx_slave->next;
 417         }
 418
 419         return rx_slave;
 420 }
 421
 422 /* teach the switch the mac of a disabled slave
 423  * on the primary for fault tolerance
 424  *
 425  * Caller must hold bond->curr_slave_lock for write or bond lock for write
 426  */
 427 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
 428 {
 429         if (!bond->curr_active_slave) {
 430                 return;
 431         }
 432
 433         if (!bond->alb_info.primary_is_promisc) {
 434                 if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1))
 435                         bond->alb_info.primary_is_promisc = 1;
 436                 else
 437                         bond->alb_info.primary_is_promisc = 0;
 438         }
 439
 440         bond->alb_info.rlb_promisc_timeout_counter = 0;
 441
 442         alb_send_learning_packets(bond->curr_active_slave, addr);
 443 }
 444
 445 /* slave being removed should not be active at this point
 446  *
 447  * Caller must hold bond lock for read
 448  */
 449 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 450 {
 451         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 452         struct rlb_client_info *rx_hash_table;
 453         u32 index, next_index;
 454
 455         /* clear slave from rx_hashtbl */
 456         _lock_rx_hashtbl(bond);
 457
 458         rx_hash_table = bond_info->rx_hashtbl;
 459         index = bond_info->rx_hashtbl_head;
 460         for (; index != RLB_NULL_INDEX; index = next_index) {
 461                 next_index = rx_hash_table[index].next;
 462                 if (rx_hash_table[index].slave == slave) {
 463                         struct slave *assigned_slave = rlb_next_rx_slave(bond);
 464
 465                         if (assigned_slave) {
 466                                 rx_hash_table[index].slave = assigned_slave;
 467                                 if (memcmp(rx_hash_table[index].mac_dst,
 468                                            mac_bcast, ETH_ALEN)) {
 469                                         bond_info->rx_hashtbl[index].ntt = 1;
 470                                         bond_info->rx_ntt = 1;
 471                                         /* A slave has been removed from the
 472                                          * table because it is either disabled
 473                                          * or being released. We must retry the
 474                                          * update to avoid clients from not
 475                                          * being updated & disconnecting when
 476                                          * there is stress
 477                                          */
 478                                         bond_info->rlb_update_retry_counter =
 479                                                 RLB_UPDATE_RETRY;
 480                                 }
 481                         } else {  /* there is no active slave */
 482                                 rx_hash_table[index].slave = NULL;
 483                         }
 484                 }
 485         }
 486
 487         _unlock_rx_hashtbl(bond);
 488
 489         write_lock_bh(&bond->curr_slave_lock);
 490
 491         if (slave != bond->curr_active_slave) {
 492                 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 493         }
 494
 495         write_unlock_bh(&bond->curr_slave_lock);
 496 }
 497
 498 static void rlb_update_client(struct rlb_client_info *client_info)
 499 {
 500         int i;
 501
 502         if (!client_info->slave) {
 503                 return;
 504         }
 505
 506         for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 507                 struct sk_buff *skb;
 508
 509                 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 510                                  client_info->ip_dst,
 511                                  client_info->slave->dev,
 512                                  client_info->ip_src,
 513                                  client_info->mac_dst,
 514                                  client_info->slave->dev->dev_addr,
 515                                  client_info->mac_dst);
 516                 if (!skb) {
 517                         printk(KERN_ERR DRV_NAME
 518                                ": %s: Error: failed to create an ARP packet\n",
 519                                client_info->slave->dev->master->name);
 520                         continue;
 521                 }
 522
 523                 skb->dev = client_info->slave->dev;
 524
 525                 if (client_info->tag) {
 526                         skb = vlan_put_tag(skb, client_info->vlan_id);
 527                         if (!skb) {
 528                                 printk(KERN_ERR DRV_NAME
 529                                        ": %s: Error: failed to insert VLAN tag\n",
 530                                        client_info->slave->dev->master->name);
 531                                 continue;
 532                         }
 533                 }
 534
 535                 arp_xmit(skb);
 536         }
 537 }
 538
 539 /* sends ARP REPLIES that update the clients that need updating */
 540 static void rlb_update_rx_clients(struct bonding *bond)
 541 {
 542         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 543         struct rlb_client_info *client_info;
 544         u32 hash_index;
 545
 546         _lock_rx_hashtbl(bond);
 547
 548         hash_index = bond_info->rx_hashtbl_head;
 549         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 550                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 551                 if (client_info->ntt) {
 552                         rlb_update_client(client_info);
 553                         if (bond_info->rlb_update_retry_counter == 0) {
 554                                 client_info->ntt = 0;
 555                         }
 556                 }
 557         }
 558
 559         /* do not update the entries again untill this counter is zero so that
 560          * not to confuse the clients.
 561          */
 562         bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 563
 564         _unlock_rx_hashtbl(bond);
 565 }
 566
 567 /* The slave was assigned a new mac address - update the clients */
 568 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 569 {
 570         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 571         struct rlb_client_info *client_info;
 572         int ntt = 0;
 573         u32 hash_index;
 574
 575         _lock_rx_hashtbl(bond);
 576
 577         hash_index = bond_info->rx_hashtbl_head;
 578         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 579                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 580
 581                 if ((client_info->slave == slave) &&
 582                     memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
 583                         client_info->ntt = 1;
 584                         ntt = 1;
 585                 }
 586         }
 587
 588         // update the team's flag only after the whole iteration
 589         if (ntt) {
 590                 bond_info->rx_ntt = 1;
 591                 //fasten the change
 592                 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 593         }
 594
 595         _unlock_rx_hashtbl(bond);
 596 }
 597
 598 /* mark all clients using src_ip to be updated */
 599 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
 600 {
 601         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 602         struct rlb_client_info *client_info;
 603         u32 hash_index;
 604
 605         _lock_rx_hashtbl(bond);
 606
 607         hash_index = bond_info->rx_hashtbl_head;
 608         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 609                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 610
 611                 if (!client_info->slave) {
 612                         printk(KERN_ERR DRV_NAME
 613                                ": %s: Error: found a client with no channel in "
 614                                "the client's hash table\n",
 615                                bond->dev->name);
 616                         continue;
 617                 }
 618                 /*update all clients using this src_ip, that are not assigned
 619                  * to the team's address (curr_active_slave) and have a known
 620                  * unicast mac address.
 621                  */
 622                 if ((client_info->ip_src == src_ip) &&
 623                     memcmp(client_info->slave->dev->dev_addr,
 624                            bond->dev->dev_addr, ETH_ALEN) &&
 625                     memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
 626                         client_info->ntt = 1;
 627                         bond_info->rx_ntt = 1;
 628                 }
 629         }
 630
 631         _unlock_rx_hashtbl(bond);
 632 }
 633
 634 /* Caller must hold both bond and ptr locks for read */
 635 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
 636 {
 637         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 638         struct arp_pkt *arp = arp_pkt(skb);
 639         struct slave *assigned_slave;
 640         struct rlb_client_info *client_info;
 641         u32 hash_index = 0;
 642
 643         _lock_rx_hashtbl(bond);
 644
 645         hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_src));
 646         client_info = &(bond_info->rx_hashtbl[hash_index]);
 647
 648         if (client_info->assigned) {
 649                 if ((client_info->ip_src == arp->ip_src) &&
 650                     (client_info->ip_dst == arp->ip_dst)) {
 651                         /* the entry is already assigned to this client */
 652                         if (memcmp(arp->mac_dst, mac_bcast, ETH_ALEN)) {
 653                                 /* update mac address from arp */
 654                                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 655                         }
 656
 657                         assigned_slave = client_info->slave;
 658                         if (assigned_slave) {
 659                                 _unlock_rx_hashtbl(bond);
 660                                 return assigned_slave;
 661                         }
 662                 } else {
 663                         /* the entry is already assigned to some other client,
 664                          * move the old client to primary (curr_active_slave) so
 665                          * that the new client can be assigned to this entry.
 666                          */
 667                         if (bond->curr_active_slave &&
 668                             client_info->slave != bond->curr_active_slave) {
 669                                 client_info->slave = bond->curr_active_slave;
 670                                 rlb_update_client(client_info);
 671                         }
 672                 }
 673         }
 674         /* assign a new slave */
 675         assigned_slave = rlb_next_rx_slave(bond);
 676
 677         if (assigned_slave) {
 678                 client_info->ip_src = arp->ip_src;
 679                 client_info->ip_dst = arp->ip_dst;
 680                 /* arp->mac_dst is broadcast for arp reqeusts.
 681                  * will be updated with clients actual unicast mac address
 682                  * upon receiving an arp reply.
 683                  */
 684                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 685                 client_info->slave = assigned_slave;
 686
 687                 if (memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
 688                         client_info->ntt = 1;
 689                         bond->alb_info.rx_ntt = 1;
 690                 } else {
 691                         client_info->ntt = 0;
 692                 }
 693
 694                 if (!list_empty(&bond->vlan_list)) {
 695                         if (!vlan_get_tag(skb, &client_info->vlan_id))
 696                                 client_info->tag = 1;
 697                 }
 698
 699                 if (!client_info->assigned) {
 700                         u32 prev_tbl_head = bond_info->rx_hashtbl_head;
 701                         bond_info->rx_hashtbl_head = hash_index;
 702                         client_info->next = prev_tbl_head;
 703                         if (prev_tbl_head != RLB_NULL_INDEX) {
 704                                 bond_info->rx_hashtbl[prev_tbl_head].prev =
 705                                         hash_index;
 706                         }
 707                         client_info->assigned = 1;
 708                 }
 709         }
 710
 711         _unlock_rx_hashtbl(bond);
 712
 713         return assigned_slave;
 714 }
 715
 716 /* chooses (and returns) transmit channel for arp reply
 717  * does not choose channel for other arp types since they are
 718  * sent on the curr_active_slave
 719  */
 720 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 721 {
 722         struct arp_pkt *arp = arp_pkt(skb);
 723         struct slave *tx_slave = NULL;
 724
 725         if (arp->op_code == htons(ARPOP_REPLY)) {
 726                 /* the arp must be sent on the selected
 727                 * rx channel
 728                 */
 729                 tx_slave = rlb_choose_channel(skb, bond);
 730                 if (tx_slave) {
 731                         memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
 732                 }
 733                 dprintk("Server sent ARP Reply packet\n");
 734         } else if (arp->op_code == htons(ARPOP_REQUEST)) {
 735                 /* Create an entry in the rx_hashtbl for this client as a
 736                  * place holder.
 737                  * When the arp reply is received the entry will be updated
 738                  * with the correct unicast address of the client.
 739                  */
 740                 rlb_choose_channel(skb, bond);
 741
 742                 /* The ARP relpy packets must be delayed so that
 743                  * they can cancel out the influence of the ARP request.
 744                  */
 745                 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 746
 747                 /* arp requests are broadcast and are sent on the primary
 748                  * the arp request will collapse all clients on the subnet to
 749                  * the primary slave. We must register these clients to be
 750                  * updated with their assigned mac.
 751                  */
 752                 rlb_req_update_subnet_clients(bond, arp->ip_src);
 753                 dprintk("Server sent ARP Request packet\n");
 754         }
 755
 756         return tx_slave;
 757 }
 758
 759 /* Caller must hold bond lock for read */
 760 static void rlb_rebalance(struct bonding *bond)
 761 {
 762         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 763         struct slave *assigned_slave;
 764         struct rlb_client_info *client_info;
 765         int ntt;
 766         u32 hash_index;
 767
 768         _lock_rx_hashtbl(bond);
 769
 770         ntt = 0;
 771         hash_index = bond_info->rx_hashtbl_head;
 772         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 773                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 774                 assigned_slave = rlb_next_rx_slave(bond);
 775                 if (assigned_slave && (client_info->slave != assigned_slave)) {
 776                         client_info->slave = assigned_slave;
 777                         client_info->ntt = 1;
 778                         ntt = 1;
 779                 }
 780         }
 781
 782         /* update the team's flag only after the whole iteration */
 783         if (ntt) {
 784                 bond_info->rx_ntt = 1;
 785         }
 786         _unlock_rx_hashtbl(bond);
 787 }
 788
 789 /* Caller must hold rx_hashtbl lock */
 790 static void rlb_init_table_entry(struct rlb_client_info *entry)
 791 {
 792         memset(entry, 0, sizeof(struct rlb_client_info));
 793         entry->next = RLB_NULL_INDEX;
 794         entry->prev = RLB_NULL_INDEX;
 795 }
 796
 797 static int rlb_initialize(struct bonding *bond)
 798 {
 799         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 800         struct packet_type *pk_type = &(BOND_ALB_INFO(bond).rlb_pkt_type);
 801         struct rlb_client_info  *new_hashtbl;
 802         int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 803         int i;
 804
 805         spin_lock_init(&(bond_info->rx_hashtbl_lock));
 806
 807         new_hashtbl = kmalloc(size, GFP_KERNEL);
 808         if (!new_hashtbl) {
 809                 printk(KERN_ERR DRV_NAME
 810                        ": %s: Error: Failed to allocate RLB hash table\n",
 811                        bond->dev->name);
 812                 return -1;
 813         }
 814         _lock_rx_hashtbl(bond);
 815
 816         bond_info->rx_hashtbl = new_hashtbl;
 817
 818         bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
 819
 820         for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
 821                 rlb_init_table_entry(bond_info->rx_hashtbl + i);
 822         }
 823
 824         _unlock_rx_hashtbl(bond);
 825
 826         /*initialize packet type*/
 827         pk_type->type = __constant_htons(ETH_P_ARP);
 828         pk_type->dev = NULL;
 829         pk_type->func = rlb_arp_recv;
 830
 831         /* register to receive ARPs */
 832         dev_add_pack(pk_type);
 833
 834         return 0;
 835 }
 836
 837 static void rlb_deinitialize(struct bonding *bond)
 838 {
 839         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 840
 841         dev_remove_pack(&(bond_info->rlb_pkt_type));
 842
 843         _lock_rx_hashtbl(bond);
 844
 845         kfree(bond_info->rx_hashtbl);
 846         bond_info->rx_hashtbl = NULL;
 847         bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
 848
 849         _unlock_rx_hashtbl(bond);
 850 }
 851
 852 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 853 {
 854         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 855         u32 curr_index;
 856
 857         _lock_rx_hashtbl(bond);
 858
 859         curr_index = bond_info->rx_hashtbl_head;
 860         while (curr_index != RLB_NULL_INDEX) {
 861                 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 862                 u32 next_index = bond_info->rx_hashtbl[curr_index].next;
 863                 u32 prev_index = bond_info->rx_hashtbl[curr_index].prev;
 864
 865                 if (curr->tag && (curr->vlan_id == vlan_id)) {
 866                         if (curr_index == bond_info->rx_hashtbl_head) {
 867                                 bond_info->rx_hashtbl_head = next_index;
 868                         }
 869                         if (prev_index != RLB_NULL_INDEX) {
 870                                 bond_info->rx_hashtbl[prev_index].next = next_index;
 871                         }
 872                         if (next_index != RLB_NULL_INDEX) {
 873                                 bond_info->rx_hashtbl[next_index].prev = prev_index;
 874                         }
 875
 876                         rlb_init_table_entry(curr);
 877                 }
 878
 879                 curr_index = next_index;
 880         }
 881
 882         _unlock_rx_hashtbl(bond);
 883 }
 884
 885 /*********************** tlb/rlb shared functions *********************/
 886
 887 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
 888 {
 889         struct bonding *bond = bond_get_bond_by_slave(slave);
 890         struct learning_pkt pkt;
 891         int size = sizeof(struct learning_pkt);
 892         int i;
 893
 894         memset(&pkt, 0, size);
 895         memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
 896         memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
 897         pkt.type = __constant_htons(ETH_P_LOOP);
 898
 899         for (i = 0; i < MAX_LP_BURST; i++) {
 900                 struct sk_buff *skb;
 901                 char *data;
 902
 903                 skb = dev_alloc_skb(size);
 904                 if (!skb) {
 905                         return;
 906                 }
 907
 908                 data = skb_put(skb, size);
 909                 memcpy(data, &pkt, size);
 910
 911                 skb_reset_mac_header(skb);
 912                 skb->network_header = skb->mac_header + ETH_HLEN;
 913                 skb->protocol = pkt.type;
 914                 skb->priority = TC_PRIO_CONTROL;
 915                 skb->dev = slave->dev;
 916
 917                 if (!list_empty(&bond->vlan_list)) {
 918                         struct vlan_entry *vlan;
 919
 920                         vlan = bond_next_vlan(bond,
 921                                               bond->alb_info.current_alb_vlan);
 922
 923                         bond->alb_info.current_alb_vlan = vlan;
 924                         if (!vlan) {
 925                                 kfree_skb(skb);
 926                                 continue;
 927                         }
 928
 929                         skb = vlan_put_tag(skb, vlan->vlan_id);
 930                         if (!skb) {
 931                                 printk(KERN_ERR DRV_NAME
 932                                        ": %s: Error: failed to insert VLAN tag\n",
 933                                        bond->dev->name);
 934                                 continue;
 935                         }
 936                 }
 937
 938                 dev_queue_xmit(skb);
 939         }
 940 }
 941
 942 /* hw is a boolean parameter that determines whether we should try and
 943  * set the hw address of the device as well as the hw address of the
 944  * net_device
 945  */
 946 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[], int hw)
 947 {
 948         struct net_device *dev = slave->dev;
 949         struct sockaddr s_addr;
 950
 951         if (!hw) {
 952                 memcpy(dev->dev_addr, addr, dev->addr_len);
 953                 return 0;
 954         }
 955
 956         /* for rlb each slave must have a unique hw mac addresses so that */
 957         /* each slave will receive packets destined to a different mac */
 958         memcpy(s_addr.sa_data, addr, dev->addr_len);
 959         s_addr.sa_family = dev->type;
 960         if (dev_set_mac_address(dev, &s_addr)) {
 961                 printk(KERN_ERR DRV_NAME
 962                        ": %s: Error: dev_set_mac_address of dev %s failed! ALB "
 963                        "mode requires that the base driver support setting "
 964                        "the hw address also when the network device's "
 965                        "interface is open\n",
 966                        dev->master->name, dev->name);
 967                 return -EOPNOTSUPP;
 968         }
 969         return 0;
 970 }
 971
 972 /*
 973  * Swap MAC addresses between two slaves.
 974  *
 975  * Called with RTNL held, and no other locks.
 976  *
 977  */
 978
 979 static void alb_swap_mac_addr(struct bonding *bond, struct slave *slave1, struct slave *slave2)
 980 {
 981         u8 tmp_mac_addr[ETH_ALEN];
 982
 983         memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
 984         alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr, bond->alb_info.rlb_enabled);
 985         alb_set_slave_mac_addr(slave2, tmp_mac_addr, bond->alb_info.rlb_enabled);
 986
 987 }
 988
 989 /*
 990  * Send learning packets after MAC address swap.
 991  *
 992  * Called with RTNL and no other locks
 993  */
 994 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
 995                                 struct slave *slave2)
 996 {
 997         int slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
 998         struct slave *disabled_slave = NULL;
 999
1000         ASSERT_RTNL();
1001
1002         /* fasten the change in the switch */
1003         if (SLAVE_IS_OK(slave1)) {
1004                 alb_send_learning_packets(slave1, slave1->dev->dev_addr);
1005                 if (bond->alb_info.rlb_enabled) {
1006                         /* inform the clients that the mac address
1007                          * has changed
1008                          */
1009                         rlb_req_update_slave_clients(bond, slave1);
1010                 }
1011         } else {
1012                 disabled_slave = slave1;
1013         }
1014
1015         if (SLAVE_IS_OK(slave2)) {
1016                 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
1017                 if (bond->alb_info.rlb_enabled) {
1018                         /* inform the clients that the mac address
1019                          * has changed
1020                          */
1021                         rlb_req_update_slave_clients(bond, slave2);
1022                 }
1023         } else {
1024                 disabled_slave = slave2;
1025         }
1026
1027         if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1028                 /* A disabled slave was assigned an active mac addr */
1029                 rlb_teach_disabled_mac_on_primary(bond,
1030                                                   disabled_slave->dev->dev_addr);
1031         }
1032 }
1033
1034 /**
1035  * alb_change_hw_addr_on_detach
1036  * @bond: bonding we're working on
1037  * @slave: the slave that was just detached
1038  *
1039  * We assume that @slave was already detached from the slave list.
1040  *
1041  * If @slave's permanent hw address is different both from its current
1042  * address and from @bond's address, then somewhere in the bond there's
1043  * a slave that has @slave's permanet address as its current address.
1044  * We'll make sure that that slave no longer uses @slave's permanent address.
1045  *
1046  * Caller must hold RTNL and no other locks
1047  */
1048 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1049 {
1050         int perm_curr_diff;
1051         int perm_bond_diff;
1052
1053         perm_curr_diff = memcmp(slave->perm_hwaddr,
1054                                 slave->dev->dev_addr,
1055                                 ETH_ALEN);
1056         perm_bond_diff = memcmp(slave->perm_hwaddr,
1057                                 bond->dev->dev_addr,
1058                                 ETH_ALEN);
1059
1060         if (perm_curr_diff && perm_bond_diff) {
1061                 struct slave *tmp_slave;
1062                 int i, found = 0;
1063
1064                 bond_for_each_slave(bond, tmp_slave, i) {
1065                         if (!memcmp(slave->perm_hwaddr,
1066                                     tmp_slave->dev->dev_addr,
1067                                     ETH_ALEN)) {
1068                                 found = 1;
1069                                 break;
1070                         }
1071                 }
1072
1073                 if (found) {
1074                         /* locking: needs RTNL and nothing else */
1075                         alb_swap_mac_addr(bond, slave, tmp_slave);
1076                         alb_fasten_mac_swap(bond, slave, tmp_slave);
1077                 }
1078         }
1079 }
1080
1081 /**
1082  * alb_handle_addr_collision_on_attach
1083  * @bond: bonding we're working on
1084  * @slave: the slave that was just attached
1085  *
1086  * checks uniqueness of slave's mac address and handles the case the
1087  * new slave uses the bonds mac address.
1088  *
1089  * If the permanent hw address of @slave is @bond's hw address, we need to
1090  * find a different hw address to give @slave, that isn't in use by any other
1091  * slave in the bond. This address must be, of course, one of the premanent
1092  * addresses of the other slaves.
1093  *
1094  * We go over the slave list, and for each slave there we compare its
1095  * permanent hw address with the current address of all the other slaves.
1096  * If no match was found, then we've found a slave with a permanent address
1097  * that isn't used by any other slave in the bond, so we can assign it to
1098  * @slave.
1099  *
1100  * assumption: this function is called before @slave is attached to the
1101  *             bond slave list.
1102  *
1103  * caller must hold the bond lock for write since the mac addresses are compared
1104  * and may be swapped.
1105  */
1106 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1107 {
1108         struct slave *tmp_slave1, *tmp_slave2, *free_mac_slave;
1109         struct slave *has_bond_addr = bond->curr_active_slave;
1110         int i, j, found = 0;
1111
1112         if (bond->slave_cnt == 0) {
1113                 /* this is the first slave */
1114                 return 0;
1115         }
1116
1117         /* if slave's mac address differs from bond's mac address
1118          * check uniqueness of slave's mac address against the other
1119          * slaves in the bond.
1120          */
1121         if (memcmp(slave->perm_hwaddr, bond->dev->dev_addr, ETH_ALEN)) {
1122                 bond_for_each_slave(bond, tmp_slave1, i) {
1123                         if (!memcmp(tmp_slave1->dev->dev_addr, slave->dev->dev_addr,
1124                                     ETH_ALEN)) {
1125                                 found = 1;
1126                                 break;
1127                         }
1128                 }
1129
1130                 if (!found)
1131                         return 0;
1132
1133                 /* Try setting slave mac to bond address and fall-through
1134                    to code handling that situation below... */
1135                 alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1136                                        bond->alb_info.rlb_enabled);
1137         }
1138
1139         /* The slave's address is equal to the address of the bond.
1140          * Search for a spare address in the bond for this slave.
1141          */
1142         free_mac_slave = NULL;
1143
1144         bond_for_each_slave(bond, tmp_slave1, i) {
1145                 found = 0;
1146                 bond_for_each_slave(bond, tmp_slave2, j) {
1147                         if (!memcmp(tmp_slave1->perm_hwaddr,
1148                                     tmp_slave2->dev->dev_addr,
1149                                     ETH_ALEN)) {
1150                                 found = 1;
1151                                 break;
1152                         }
1153                 }
1154
1155                 if (!found) {
1156                         /* no slave has tmp_slave1's perm addr
1157                          * as its curr addr
1158                          */
1159                         free_mac_slave = tmp_slave1;
1160                         break;
1161                 }
1162
1163                 if (!has_bond_addr) {
1164                         if (!memcmp(tmp_slave1->dev->dev_addr,
1165                                     bond->dev->dev_addr,
1166                                     ETH_ALEN)) {
1167
1168                                 has_bond_addr = tmp_slave1;
1169                         }
1170                 }
1171         }
1172
1173         if (free_mac_slave) {
1174                 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1175                                        bond->alb_info.rlb_enabled);
1176
1177                 printk(KERN_WARNING DRV_NAME
1178                        ": %s: Warning: the hw address of slave %s is in use by "
1179                        "the bond; giving it the hw address of %s\n",
1180                        bond->dev->name, slave->dev->name, free_mac_slave->dev->name);
1181
1182         } else if (has_bond_addr) {
1183                 printk(KERN_ERR DRV_NAME
1184                        ": %s: Error: the hw address of slave %s is in use by the "
1185                        "bond; couldn't find a slave with a free hw address to "
1186                        "give it (this should not have happened)\n",
1187                        bond->dev->name, slave->dev->name);
1188                 return -EFAULT;
1189         }
1190
1191         return 0;
1192 }
1193
1194 /**
1195  * alb_set_mac_address
1196  * @bond:
1197  * @addr:
1198  *
1199  * In TLB mode all slaves are configured to the bond's hw address, but set
1200  * their dev_addr field to different addresses (based on their permanent hw
1201  * addresses).
1202  *
1203  * For each slave, this function sets the interface to the new address and then
1204  * changes its dev_addr field to its previous value.
1205  *
1206  * Unwinding assumes bond's mac address has not yet changed.
1207  */
1208 static int alb_set_mac_address(struct bonding *bond, void *addr)
1209 {
1210         struct sockaddr sa;
1211         struct slave *slave, *stop_at;
1212         char tmp_addr[ETH_ALEN];
1213         int res;
1214         int i;
1215
1216         if (bond->alb_info.rlb_enabled) {
1217                 return 0;
1218         }
1219
1220         bond_for_each_slave(bond, slave, i) {
1221                 if (slave->dev->set_mac_address == NULL) {
1222                         res = -EOPNOTSUPP;
1223                         goto unwind;
1224                 }
1225
1226                 /* save net_device's current hw address */
1227                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1228
1229                 res = dev_set_mac_address(slave->dev, addr);
1230
1231                 /* restore net_device's hw address */
1232                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1233
1234                 if (res) {
1235                         goto unwind;
1236                 }
1237         }
1238
1239         return 0;
1240
1241 unwind:
1242         memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1243         sa.sa_family = bond->dev->type;
1244
1245         /* unwind from head to the slave that failed */
1246         stop_at = slave;
1247         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
1248                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1249                 dev_set_mac_address(slave->dev, &sa);
1250                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1251         }
1252
1253         return res;
1254 }
1255
1256 /************************ exported alb funcions ************************/
1257
1258 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1259 {
1260         int res;
1261
1262         res = tlb_initialize(bond);
1263         if (res) {
1264                 return res;
1265         }
1266
1267         if (rlb_enabled) {
1268                 bond->alb_info.rlb_enabled = 1;
1269                 /* initialize rlb */
1270                 res = rlb_initialize(bond);
1271                 if (res) {
1272                         tlb_deinitialize(bond);
1273                         return res;
1274                 }
1275         } else {
1276                 bond->alb_info.rlb_enabled = 0;
1277         }
1278
1279         return 0;
1280 }
1281
1282 void bond_alb_deinitialize(struct bonding *bond)
1283 {
1284         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1285
1286         tlb_deinitialize(bond);
1287
1288         if (bond_info->rlb_enabled) {
1289                 rlb_deinitialize(bond);
1290         }
1291 }
1292
1293 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1294 {
1295         struct bonding *bond = netdev_priv(bond_dev);
1296         struct ethhdr *eth_data;
1297         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1298         struct slave *tx_slave = NULL;
1299         static const __be32 ip_bcast = htonl(0xffffffff);
1300         int hash_size = 0;
1301         int do_tx_balance = 1;
1302         u32 hash_index = 0;
1303         const u8 *hash_start = NULL;
1304         int res = 1;
1305         struct ipv6hdr *ip6hdr;
1306
1307         skb_reset_mac_header(skb);
1308         eth_data = eth_hdr(skb);
1309
1310         /* make sure that the curr_active_slave and the slaves list do
1311          * not change during tx
1312          */
1313         read_lock(&bond->lock);
1314         read_lock(&bond->curr_slave_lock);
1315
1316         if (!BOND_IS_OK(bond)) {
1317                 goto out;
1318         }
1319
1320         switch (ntohs(skb->protocol)) {
1321         case ETH_P_IP: {
1322                 const struct iphdr *iph = ip_hdr(skb);
1323
1324                 if ((memcmp(eth_data->h_dest, mac_bcast, ETH_ALEN) == 0) ||
1325                     (iph->daddr == ip_bcast) ||
1326                     (iph->protocol == IPPROTO_IGMP)) {
1327                         do_tx_balance = 0;
1328                         break;
1329                 }
1330                 hash_start = (char *)&(iph->daddr);
1331                 hash_size = sizeof(iph->daddr);
1332         }
1333                 break;
1334         case ETH_P_IPV6:
1335                 /* IPv6 doesn't really use broadcast mac address, but leave
1336                  * that here just in case.
1337                  */
1338                 if (memcmp(eth_data->h_dest, mac_bcast, ETH_ALEN) == 0) {
1339                         do_tx_balance = 0;
1340                         break;
1341                 }
1342
1343                 /* IPv6 uses all-nodes multicast as an equivalent to
1344                  * broadcasts in IPv4.
1345                  */
1346                 if (memcmp(eth_data->h_dest, mac_v6_allmcast, ETH_ALEN) == 0) {
1347                         do_tx_balance = 0;
1348                         break;
1349                 }
1350
1351                 /* Additianally, DAD probes should not be tx-balanced as that
1352                  * will lead to false positives for duplicate addresses and
1353                  * prevent address configuration from working.
1354                  */
1355                 ip6hdr = ipv6_hdr(skb);
1356                 if (ipv6_addr_any(&ip6hdr->saddr)) {
1357                         do_tx_balance = 0;
1358                         break;
1359                 }
1360
1361                 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1362                 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1363                 break;
1364         case ETH_P_IPX:
1365                 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1366                         /* something is wrong with this packet */
1367                         do_tx_balance = 0;
1368                         break;
1369                 }
1370
1371                 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1372                         /* The only protocol worth balancing in
1373                          * this family since it has an "ARP" like
1374                          * mechanism
1375                          */
1376                         do_tx_balance = 0;
1377                         break;
1378                 }
1379
1380                 hash_start = (char*)eth_data->h_dest;
1381                 hash_size = ETH_ALEN;
1382                 break;
1383         case ETH_P_ARP:
1384                 do_tx_balance = 0;
1385                 if (bond_info->rlb_enabled) {
1386                         tx_slave = rlb_arp_xmit(skb, bond);
1387                 }
1388                 break;
1389         default:
1390                 do_tx_balance = 0;
1391                 break;
1392         }
1393
1394         if (do_tx_balance) {
1395                 hash_index = _simple_hash(hash_start, hash_size);
1396                 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1397         }
1398
1399         if (!tx_slave) {
1400                 /* unbalanced or unassigned, send through primary */
1401                 tx_slave = bond->curr_active_slave;
1402                 bond_info->unbalanced_load += skb->len;
1403         }
1404
1405         if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1406                 if (tx_slave != bond->curr_active_slave) {
1407                         memcpy(eth_data->h_source,
1408                                tx_slave->dev->dev_addr,
1409                                ETH_ALEN);
1410                 }
1411
1412                 res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1413         } else {
1414                 if (tx_slave) {
1415                         tlb_clear_slave(bond, tx_slave, 0);
1416                 }
1417         }
1418
1419 out:
1420         if (res) {
1421                 /* no suitable interface, frame not sent */
1422                 dev_kfree_skb(skb);
1423         }
1424         read_unlock(&bond->curr_slave_lock);
1425         read_unlock(&bond->lock);
1426         return 0;
1427 }
1428
1429 void bond_alb_monitor(struct work_struct *work)
1430 {
1431         struct bonding *bond = container_of(work, struct bonding,
1432                                             alb_work.work);
1433         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1434         struct slave *slave;
1435         int i;
1436
1437         read_lock(&bond->lock);
1438
1439         if (bond->kill_timers) {
1440                 goto out;
1441         }
1442
1443         if (bond->slave_cnt == 0) {
1444                 bond_info->tx_rebalance_counter = 0;
1445                 bond_info->lp_counter = 0;
1446                 goto re_arm;
1447         }
1448
1449         bond_info->tx_rebalance_counter++;
1450         bond_info->lp_counter++;
1451
1452         /* send learning packets */
1453         if (bond_info->lp_counter >= BOND_ALB_LP_TICKS) {
1454                 /* change of curr_active_slave involves swapping of mac addresses.
1455                  * in order to avoid this swapping from happening while
1456                  * sending the learning packets, the curr_slave_lock must be held for
1457                  * read.
1458                  */
1459                 read_lock(&bond->curr_slave_lock);
1460
1461                 bond_for_each_slave(bond, slave, i) {
1462                         alb_send_learning_packets(slave, slave->dev->dev_addr);
1463                 }
1464
1465                 read_unlock(&bond->curr_slave_lock);
1466
1467                 bond_info->lp_counter = 0;
1468         }
1469
1470         /* rebalance tx traffic */
1471         if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1472
1473                 read_lock(&bond->curr_slave_lock);
1474
1475                 bond_for_each_slave(bond, slave, i) {
1476                         tlb_clear_slave(bond, slave, 1);
1477                         if (slave == bond->curr_active_slave) {
1478                                 SLAVE_TLB_INFO(slave).load =
1479                                         bond_info->unbalanced_load /
1480                                                 BOND_TLB_REBALANCE_INTERVAL;
1481                                 bond_info->unbalanced_load = 0;
1482                         }
1483                 }
1484
1485                 read_unlock(&bond->curr_slave_lock);
1486
1487                 bond_info->tx_rebalance_counter = 0;
1488         }
1489
1490         /* handle rlb stuff */
1491         if (bond_info->rlb_enabled) {
1492                 if (bond_info->primary_is_promisc &&
1493                     (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1494
1495                         /*
1496                          * dev_set_promiscuity requires rtnl and
1497                          * nothing else.
1498                          */
1499                         read_unlock(&bond->lock);
1500                         rtnl_lock();
1501
1502                         bond_info->rlb_promisc_timeout_counter = 0;
1503
1504                         /* If the primary was set to promiscuous mode
1505                          * because a slave was disabled then
1506                          * it can now leave promiscuous mode.
1507                          */
1508                         dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1509                         bond_info->primary_is_promisc = 0;
1510
1511                         rtnl_unlock();
1512                         read_lock(&bond->lock);
1513                 }
1514
1515                 if (bond_info->rlb_rebalance) {
1516                         bond_info->rlb_rebalance = 0;
1517                         rlb_rebalance(bond);
1518                 }
1519
1520                 /* check if clients need updating */
1521                 if (bond_info->rx_ntt) {
1522                         if (bond_info->rlb_update_delay_counter) {
1523                                 --bond_info->rlb_update_delay_counter;
1524                         } else {
1525                                 rlb_update_rx_clients(bond);
1526                                 if (bond_info->rlb_update_retry_counter) {
1527                                         --bond_info->rlb_update_retry_counter;
1528                                 } else {
1529                                         bond_info->rx_ntt = 0;
1530                                 }
1531                         }
1532                 }
1533         }
1534
1535 re_arm:
1536         queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1537 out:
1538         read_unlock(&bond->lock);
1539 }
1540
1541 /* assumption: called before the slave is attached to the bond
1542  * and not locked by the bond lock
1543  */
1544 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1545 {
1546         int res;
1547
1548         res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1549                                      bond->alb_info.rlb_enabled);
1550         if (res) {
1551                 return res;
1552         }
1553
1554         /* caller must hold the bond lock for write since the mac addresses
1555          * are compared and may be swapped.
1556          */
1557         read_lock(&bond->lock);
1558
1559         res = alb_handle_addr_collision_on_attach(bond, slave);
1560
1561         read_unlock(&bond->lock);
1562
1563         if (res) {
1564                 return res;
1565         }
1566
1567         tlb_init_slave(slave);
1568
1569         /* order a rebalance ASAP */
1570         bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1571
1572         if (bond->alb_info.rlb_enabled) {
1573                 bond->alb_info.rlb_rebalance = 1;
1574         }
1575
1576         return 0;
1577 }
1578
1579 /*
1580  * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1581  * if necessary.
1582  *
1583  * Caller must hold RTNL and no other locks
1584  */
1585 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1586 {
1587         if (bond->slave_cnt > 1) {
1588                 alb_change_hw_addr_on_detach(bond, slave);
1589         }
1590
1591         tlb_clear_slave(bond, slave, 0);
1592
1593         if (bond->alb_info.rlb_enabled) {
1594                 bond->alb_info.next_rx_slave = NULL;
1595                 rlb_clear_slave(bond, slave);
1596         }
1597 }
1598
1599 /* Caller must hold bond lock for read */
1600 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1601 {
1602         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1603
1604         if (link == BOND_LINK_DOWN) {
1605                 tlb_clear_slave(bond, slave, 0);
1606                 if (bond->alb_info.rlb_enabled) {
1607                         rlb_clear_slave(bond, slave);
1608                 }
1609         } else if (link == BOND_LINK_UP) {
1610                 /* order a rebalance ASAP */
1611                 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1612                 if (bond->alb_info.rlb_enabled) {
1613                         bond->alb_info.rlb_rebalance = 1;
1614                         /* If the updelay module parameter is smaller than the
1615                          * forwarding delay of the switch the rebalance will
1616                          * not work because the rebalance arp replies will
1617                          * not be forwarded to the clients..
1618                          */
1619                 }
1620         }
1621 }
1622
1623 /**
1624  * bond_alb_handle_active_change - assign new curr_active_slave
1625  * @bond: our bonding struct
1626  * @new_slave: new slave to assign
1627  *
1628  * Set the bond->curr_active_slave to @new_slave and handle
1629  * mac address swapping and promiscuity changes as needed.
1630  *
1631  * If new_slave is NULL, caller must hold curr_slave_lock or
1632  * bond->lock for write.
1633  *
1634  * If new_slave is not NULL, caller must hold RTNL, bond->lock for
1635  * read and curr_slave_lock for write.  Processing here may sleep, so
1636  * no other locks may be held.
1637  */
1638 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1639 {
1640         struct slave *swap_slave;
1641         int i;
1642
1643         if (bond->curr_active_slave == new_slave) {
1644                 return;
1645         }
1646
1647         if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1648                 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1649                 bond->alb_info.primary_is_promisc = 0;
1650                 bond->alb_info.rlb_promisc_timeout_counter = 0;
1651         }
1652
1653         swap_slave = bond->curr_active_slave;
1654         bond->curr_active_slave = new_slave;
1655
1656         if (!new_slave || (bond->slave_cnt == 0)) {
1657                 return;
1658         }
1659
1660         /* set the new curr_active_slave to the bonds mac address
1661          * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1662          */
1663         if (!swap_slave) {
1664                 struct slave *tmp_slave;
1665                 /* find slave that is holding the bond's mac address */
1666                 bond_for_each_slave(bond, tmp_slave, i) {
1667                         if (!memcmp(tmp_slave->dev->dev_addr,
1668                                     bond->dev->dev_addr, ETH_ALEN)) {
1669                                 swap_slave = tmp_slave;
1670                                 break;
1671                         }
1672                 }
1673         }
1674
1675         /*
1676          * Arrange for swap_slave and new_slave to temporarily be
1677          * ignored so we can mess with their MAC addresses without
1678          * fear of interference from transmit activity.
1679          */
1680         if (swap_slave) {
1681                 tlb_clear_slave(bond, swap_slave, 1);
1682         }
1683         tlb_clear_slave(bond, new_slave, 1);
1684
1685         write_unlock_bh(&bond->curr_slave_lock);
1686         read_unlock(&bond->lock);
1687
1688         ASSERT_RTNL();
1689
1690         /* curr_active_slave must be set before calling alb_swap_mac_addr */
1691         if (swap_slave) {
1692                 /* swap mac address */
1693                 alb_swap_mac_addr(bond, swap_slave, new_slave);
1694         } else {
1695                 /* set the new_slave to the bond mac address */
1696                 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1697                                        bond->alb_info.rlb_enabled);
1698         }
1699
1700         if (swap_slave) {
1701                 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1702                 read_lock(&bond->lock);
1703         } else {
1704                 read_lock(&bond->lock);
1705                 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1706         }
1707
1708         write_lock_bh(&bond->curr_slave_lock);
1709 }
1710
1711 /*
1712  * Called with RTNL
1713  */
1714 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1715 {
1716         struct bonding *bond = netdev_priv(bond_dev);
1717         struct sockaddr *sa = addr;
1718         struct slave *slave, *swap_slave;
1719         int res;
1720         int i;
1721
1722         if (!is_valid_ether_addr(sa->sa_data)) {
1723                 return -EADDRNOTAVAIL;
1724         }
1725
1726         res = alb_set_mac_address(bond, addr);
1727         if (res) {
1728                 return res;
1729         }
1730
1731         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1732
1733         /* If there is no curr_active_slave there is nothing else to do.
1734          * Otherwise we'll need to pass the new address to it and handle
1735          * duplications.
1736          */
1737         if (!bond->curr_active_slave) {
1738                 return 0;
1739         }
1740
1741         swap_slave = NULL;
1742
1743         bond_for_each_slave(bond, slave, i) {
1744                 if (!memcmp(slave->dev->dev_addr, bond_dev->dev_addr, ETH_ALEN)) {
1745                         swap_slave = slave;
1746                         break;
1747                 }
1748         }
1749
1750         write_unlock_bh(&bond->curr_slave_lock);
1751         read_unlock(&bond->lock);
1752
1753         if (swap_slave) {
1754                 alb_swap_mac_addr(bond, swap_slave, bond->curr_active_slave);
1755                 alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave);
1756         } else {
1757                 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr,
1758                                        bond->alb_info.rlb_enabled);
1759
1760                 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1761                 if (bond->alb_info.rlb_enabled) {
1762                         /* inform clients mac address has changed */
1763                         rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1764                 }
1765         }
1766
1767         read_lock(&bond->lock);
1768         write_lock_bh(&bond->curr_slave_lock);
1769
1770         return 0;
1771 }
1772
1773 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1774 {
1775         if (bond->alb_info.current_alb_vlan &&
1776             (bond->alb_info.current_alb_vlan->vlan_id == vlan_id)) {
1777                 bond->alb_info.current_alb_vlan = NULL;
1778         }
1779
1780         if (bond->alb_info.rlb_enabled) {
1781                 rlb_clear_vlan(bond, vlan_id);
1782         }
1783 }
1784