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 <asm/byteorder.h>
  42 #include "bonding.h"
  43 #include "bond_alb.h"
  44
  45
  46 #define ALB_TIMER_TICKS_PER_SEC     10  /* should be a divisor of HZ */
  47 #define BOND_TLB_REBALANCE_INTERVAL 10  /* In seconds, periodic re-balancing.
  48                                          * Used for division - never set
  49                                          * to zero !!!
  50                                          */
  51 #define BOND_ALB_LP_INTERVAL        1   /* In seconds, periodic send of
  52                                          * learning packets to the switch
  53                                          */
  54
  55 #define BOND_TLB_REBALANCE_TICKS (BOND_TLB_REBALANCE_INTERVAL \
  56                                   * ALB_TIMER_TICKS_PER_SEC)
  57
  58 #define BOND_ALB_LP_TICKS (BOND_ALB_LP_INTERVAL \
  59                            * ALB_TIMER_TICKS_PER_SEC)
  60
  61 #define TLB_HASH_TABLE_SIZE 256 /* The size of the clients hash table.
  62                                  * Note that this value MUST NOT be smaller
  63                                  * because the key hash table is BYTE wide !
  64                                  */
  65
  66
  67 #define TLB_NULL_INDEX          0xffffffff
  68 #define MAX_LP_BURST            3
  69
  70 /* rlb defs */
  71 #define RLB_HASH_TABLE_SIZE     256
  72 #define RLB_NULL_INDEX          0xffffffff
  73 #define RLB_UPDATE_DELAY        2*ALB_TIMER_TICKS_PER_SEC /* 2 seconds */
  74 #define RLB_ARP_BURST_SIZE      2
  75 #define RLB_UPDATE_RETRY        3       /* 3-ticks - must be smaller than the rlb
  76                                          * rebalance interval (5 min).
  77                                          */
  78 /* RLB_PROMISC_TIMEOUT = 10 sec equals the time that the current slave is
  79  * promiscuous after failover
  80  */
  81 #define RLB_PROMISC_TIMEOUT     10*ALB_TIMER_TICKS_PER_SEC
  82
  83 static const u8 mac_bcast[ETH_ALEN] = {0xff,0xff,0xff,0xff,0xff,0xff};
  84 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
  85
  86 #pragma pack(1)
  87 struct learning_pkt {
  88         u8 mac_dst[ETH_ALEN];
  89         u8 mac_src[ETH_ALEN];
  90         __be16 type;
  91         u8 padding[ETH_ZLEN - ETH_HLEN];
  92 };
  93
  94 struct arp_pkt {
  95         __be16  hw_addr_space;
  96         __be16  prot_addr_space;
  97         u8      hw_addr_len;
  98         u8      prot_addr_len;
  99         __be16  op_code;
 100         u8      mac_src[ETH_ALEN];      /* sender hardware address */
 101         __be32  ip_src;                 /* sender IP address */
 102         u8      mac_dst[ETH_ALEN];      /* target hardware address */
 103         __be32  ip_dst;                 /* target IP address */
 104 };
 105 #pragma pack()
 106
 107 static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb)
 108 {
 109         return (struct arp_pkt *)skb_network_header(skb);
 110 }
 111
 112 /* Forward declaration */
 113 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[]);
 114
 115 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
 116 {
 117         int i;
 118         u8 hash = 0;
 119
 120         for (i = 0; i < hash_size; i++) {
 121                 hash ^= hash_start[i];
 122         }
 123
 124         return hash;
 125 }
 126
 127 /*********************** tlb specific functions ***************************/
 128
 129 static inline void _lock_tx_hashtbl(struct bonding *bond)
 130 {
 131         spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 132 }
 133
 134 static inline void _unlock_tx_hashtbl(struct bonding *bond)
 135 {
 136         spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 137 }
 138
 139 /* Caller must hold tx_hashtbl lock */
 140 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
 141 {
 142         if (save_load) {
 143                 entry->load_history = 1 + entry->tx_bytes /
 144                                       BOND_TLB_REBALANCE_INTERVAL;
 145                 entry->tx_bytes = 0;
 146         }
 147
 148         entry->tx_slave = NULL;
 149         entry->next = TLB_NULL_INDEX;
 150         entry->prev = TLB_NULL_INDEX;
 151 }
 152
 153 static inline void tlb_init_slave(struct slave *slave)
 154 {
 155         SLAVE_TLB_INFO(slave).load = 0;
 156         SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
 157 }
 158
 159 /* Caller must hold bond lock for read */
 160 static void tlb_clear_slave(struct bonding *bond, struct slave *slave, int save_load)
 161 {
 162         struct tlb_client_info *tx_hash_table;
 163         u32 index;
 164
 165         _lock_tx_hashtbl(bond);
 166
 167         /* clear slave from tx_hashtbl */
 168         tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
 169
 170         index = SLAVE_TLB_INFO(slave).head;
 171         while (index != TLB_NULL_INDEX) {
 172                 u32 next_index = tx_hash_table[index].next;
 173                 tlb_init_table_entry(&tx_hash_table[index], save_load);
 174                 index = next_index;
 175         }
 176
 177         tlb_init_slave(slave);
 178
 179         _unlock_tx_hashtbl(bond);
 180 }
 181
 182 /* Must be called before starting the monitor timer */
 183 static int tlb_initialize(struct bonding *bond)
 184 {
 185         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 186         int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 187         struct tlb_client_info *new_hashtbl;
 188         int i;
 189
 190         spin_lock_init(&(bond_info->tx_hashtbl_lock));
 191
 192         new_hashtbl = kzalloc(size, GFP_KERNEL);
 193         if (!new_hashtbl) {
 194                 printk(KERN_ERR DRV_NAME
 195                        ": %s: Error: Failed to allocate TLB hash table\n",
 196                        bond->dev->name);
 197                 return -1;
 198         }
 199         _lock_tx_hashtbl(bond);
 200
 201         bond_info->tx_hashtbl = new_hashtbl;
 202
 203         for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
 204                 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 1);
 205         }
 206
 207         _unlock_tx_hashtbl(bond);
 208
 209         return 0;
 210 }
 211
 212 /* Must be called only after all slaves have been released */
 213 static void tlb_deinitialize(struct bonding *bond)
 214 {
 215         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 216
 217         _lock_tx_hashtbl(bond);
 218
 219         kfree(bond_info->tx_hashtbl);
 220         bond_info->tx_hashtbl = NULL;
 221
 222         _unlock_tx_hashtbl(bond);
 223 }
 224
 225 /* Caller must hold bond lock for read */
 226 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
 227 {
 228         struct slave *slave, *least_loaded;
 229         s64 max_gap;
 230         int i, found = 0;
 231
 232         /* Find the first enabled slave */
 233         bond_for_each_slave(bond, slave, i) {
 234                 if (SLAVE_IS_OK(slave)) {
 235                         found = 1;
 236                         break;
 237                 }
 238         }
 239
 240         if (!found) {
 241                 return NULL;
 242         }
 243
 244         least_loaded = slave;
 245         max_gap = (s64)(slave->speed << 20) - /* Convert to Megabit per sec */
 246                         (s64)(SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
 247
 248         /* Find the slave with the largest gap */
 249         bond_for_each_slave_from(bond, slave, i, least_loaded) {
 250                 if (SLAVE_IS_OK(slave)) {
 251                         s64 gap = (s64)(slave->speed << 20) -
 252                                         (s64)(SLAVE_TLB_INFO(slave).load << 3);
 253                         if (max_gap < gap) {
 254                                 least_loaded = slave;
 255                                 max_gap = gap;
 256                         }
 257                 }
 258         }
 259
 260         return least_loaded;
 261 }
 262
 263 /* Caller must hold bond lock for read */
 264 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index, u32 skb_len)
 265 {
 266         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 267         struct tlb_client_info *hash_table;
 268         struct slave *assigned_slave;
 269
 270         _lock_tx_hashtbl(bond);
 271
 272         hash_table = bond_info->tx_hashtbl;
 273         assigned_slave = hash_table[hash_index].tx_slave;
 274         if (!assigned_slave) {
 275                 assigned_slave = tlb_get_least_loaded_slave(bond);
 276
 277                 if (assigned_slave) {
 278                         struct tlb_slave_info *slave_info =
 279                                 &(SLAVE_TLB_INFO(assigned_slave));
 280                         u32 next_index = slave_info->head;
 281
 282                         hash_table[hash_index].tx_slave = assigned_slave;
 283                         hash_table[hash_index].next = next_index;
 284                         hash_table[hash_index].prev = TLB_NULL_INDEX;
 285
 286                         if (next_index != TLB_NULL_INDEX) {
 287                                 hash_table[next_index].prev = hash_index;
 288                         }
 289
 290                         slave_info->head = hash_index;
 291                         slave_info->load +=
 292                                 hash_table[hash_index].load_history;
 293                 }
 294         }
 295
 296         if (assigned_slave) {
 297                 hash_table[hash_index].tx_bytes += skb_len;
 298         }
 299
 300         _unlock_tx_hashtbl(bond);
 301
 302         return assigned_slave;
 303 }
 304
 305 /*********************** rlb specific functions ***************************/
 306 static inline void _lock_rx_hashtbl(struct bonding *bond)
 307 {
 308         spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 309 }
 310
 311 static inline void _unlock_rx_hashtbl(struct bonding *bond)
 312 {
 313         spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 314 }
 315
 316 /* when an ARP REPLY is received from a client update its info
 317  * in the rx_hashtbl
 318  */
 319 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 320 {
 321         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 322         struct rlb_client_info *client_info;
 323         u32 hash_index;
 324
 325         _lock_rx_hashtbl(bond);
 326
 327         hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
 328         client_info = &(bond_info->rx_hashtbl[hash_index]);
 329
 330         if ((client_info->assigned) &&
 331             (client_info->ip_src == arp->ip_dst) &&
 332             (client_info->ip_dst == arp->ip_src)) {
 333                 /* update the clients MAC address */
 334                 memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
 335                 client_info->ntt = 1;
 336                 bond_info->rx_ntt = 1;
 337         }
 338
 339         _unlock_rx_hashtbl(bond);
 340 }
 341
 342 static int rlb_arp_recv(struct sk_buff *skb, struct net_device *bond_dev, struct packet_type *ptype, struct net_device *orig_dev)
 343 {
 344         struct bonding *bond = bond_dev->priv;
 345         struct arp_pkt *arp = (struct arp_pkt *)skb->data;
 346         int res = NET_RX_DROP;
 347
 348         if (bond_dev->nd_net != &init_net)
 349                 goto out;
 350
 351         if (!(bond_dev->flags & IFF_MASTER))
 352                 goto out;
 353
 354         if (!arp) {
 355                 dprintk("Packet has no ARP data\n");
 356                 goto out;
 357         }
 358
 359         if (skb->len < sizeof(struct arp_pkt)) {
 360                 dprintk("Packet is too small to be an ARP\n");
 361                 goto out;
 362         }
 363
 364         if (arp->op_code == htons(ARPOP_REPLY)) {
 365                 /* update rx hash table for this ARP */
 366                 rlb_update_entry_from_arp(bond, arp);
 367                 dprintk("Server received an ARP Reply from client\n");
 368         }
 369
 370         res = NET_RX_SUCCESS;
 371
 372 out:
 373         dev_kfree_skb(skb);
 374
 375         return res;
 376 }
 377
 378 /* Caller must hold bond lock for read */
 379 static struct slave *rlb_next_rx_slave(struct bonding *bond)
 380 {
 381         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 382         struct slave *rx_slave, *slave, *start_at;
 383         int i = 0;
 384
 385         if (bond_info->next_rx_slave) {
 386                 start_at = bond_info->next_rx_slave;
 387         } else {
 388                 start_at = bond->first_slave;
 389         }
 390
 391         rx_slave = NULL;
 392
 393         bond_for_each_slave_from(bond, slave, i, start_at) {
 394                 if (SLAVE_IS_OK(slave)) {
 395                         if (!rx_slave) {
 396                                 rx_slave = slave;
 397                         } else if (slave->speed > rx_slave->speed) {
 398                                 rx_slave = slave;
 399                         }
 400                 }
 401         }
 402
 403         if (rx_slave) {
 404                 bond_info->next_rx_slave = rx_slave->next;
 405         }
 406
 407         return rx_slave;
 408 }
 409
 410 /* teach the switch the mac of a disabled slave
 411  * on the primary for fault tolerance
 412  *
 413  * Caller must hold bond->curr_slave_lock for write or bond lock for write
 414  */
 415 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
 416 {
 417         if (!bond->curr_active_slave) {
 418                 return;
 419         }
 420
 421         if (!bond->alb_info.primary_is_promisc) {
 422                 bond->alb_info.primary_is_promisc = 1;
 423                 dev_set_promiscuity(bond->curr_active_slave->dev, 1);
 424         }
 425
 426         bond->alb_info.rlb_promisc_timeout_counter = 0;
 427
 428         alb_send_learning_packets(bond->curr_active_slave, addr);
 429 }
 430
 431 /* slave being removed should not be active at this point
 432  *
 433  * Caller must hold bond lock for read
 434  */
 435 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 436 {
 437         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 438         struct rlb_client_info *rx_hash_table;
 439         u32 index, next_index;
 440
 441         /* clear slave from rx_hashtbl */
 442         _lock_rx_hashtbl(bond);
 443
 444         rx_hash_table = bond_info->rx_hashtbl;
 445         index = bond_info->rx_hashtbl_head;
 446         for (; index != RLB_NULL_INDEX; index = next_index) {
 447                 next_index = rx_hash_table[index].next;
 448                 if (rx_hash_table[index].slave == slave) {
 449                         struct slave *assigned_slave = rlb_next_rx_slave(bond);
 450
 451                         if (assigned_slave) {
 452                                 rx_hash_table[index].slave = assigned_slave;
 453                                 if (memcmp(rx_hash_table[index].mac_dst,
 454                                            mac_bcast, ETH_ALEN)) {
 455                                         bond_info->rx_hashtbl[index].ntt = 1;
 456                                         bond_info->rx_ntt = 1;
 457                                         /* A slave has been removed from the
 458                                          * table because it is either disabled
 459                                          * or being released. We must retry the
 460                                          * update to avoid clients from not
 461                                          * being updated & disconnecting when
 462                                          * there is stress
 463                                          */
 464                                         bond_info->rlb_update_retry_counter =
 465                                                 RLB_UPDATE_RETRY;
 466                                 }
 467                         } else {  /* there is no active slave */
 468                                 rx_hash_table[index].slave = NULL;
 469                         }
 470                 }
 471         }
 472
 473         _unlock_rx_hashtbl(bond);
 474
 475         write_lock_bh(&bond->curr_slave_lock);
 476
 477         if (slave != bond->curr_active_slave) {
 478                 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 479         }
 480
 481         write_unlock_bh(&bond->curr_slave_lock);
 482 }
 483
 484 static void rlb_update_client(struct rlb_client_info *client_info)
 485 {
 486         int i;
 487
 488         if (!client_info->slave) {
 489                 return;
 490         }
 491
 492         for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 493                 struct sk_buff *skb;
 494
 495                 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 496                                  client_info->ip_dst,
 497                                  client_info->slave->dev,
 498                                  client_info->ip_src,
 499                                  client_info->mac_dst,
 500                                  client_info->slave->dev->dev_addr,
 501                                  client_info->mac_dst);
 502                 if (!skb) {
 503                         printk(KERN_ERR DRV_NAME
 504                                ": %s: Error: failed to create an ARP packet\n",
 505                                client_info->slave->dev->master->name);
 506                         continue;
 507                 }
 508
 509                 skb->dev = client_info->slave->dev;
 510
 511                 if (client_info->tag) {
 512                         skb = vlan_put_tag(skb, client_info->vlan_id);
 513                         if (!skb) {
 514                                 printk(KERN_ERR DRV_NAME
 515                                        ": %s: Error: failed to insert VLAN tag\n",
 516                                        client_info->slave->dev->master->name);
 517                                 continue;
 518                         }
 519                 }
 520
 521                 arp_xmit(skb);
 522         }
 523 }
 524
 525 /* sends ARP REPLIES that update the clients that need updating */
 526 static void rlb_update_rx_clients(struct bonding *bond)
 527 {
 528         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 529         struct rlb_client_info *client_info;
 530         u32 hash_index;
 531
 532         _lock_rx_hashtbl(bond);
 533
 534         hash_index = bond_info->rx_hashtbl_head;
 535         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 536                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 537                 if (client_info->ntt) {
 538                         rlb_update_client(client_info);
 539                         if (bond_info->rlb_update_retry_counter == 0) {
 540                                 client_info->ntt = 0;
 541                         }
 542                 }
 543         }
 544
 545         /* do not update the entries again untill this counter is zero so that
 546          * not to confuse the clients.
 547          */
 548         bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 549
 550         _unlock_rx_hashtbl(bond);
 551 }
 552
 553 /* The slave was assigned a new mac address - update the clients */
 554 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 555 {
 556         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 557         struct rlb_client_info *client_info;
 558         int ntt = 0;
 559         u32 hash_index;
 560
 561         _lock_rx_hashtbl(bond);
 562
 563         hash_index = bond_info->rx_hashtbl_head;
 564         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 565                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 566
 567                 if ((client_info->slave == slave) &&
 568                     memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
 569                         client_info->ntt = 1;
 570                         ntt = 1;
 571                 }
 572         }
 573
 574         // update the team's flag only after the whole iteration
 575         if (ntt) {
 576                 bond_info->rx_ntt = 1;
 577                 //fasten the change
 578                 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 579         }
 580
 581         _unlock_rx_hashtbl(bond);
 582 }
 583
 584 /* mark all clients using src_ip to be updated */
 585 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
 586 {
 587         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 588         struct rlb_client_info *client_info;
 589         u32 hash_index;
 590
 591         _lock_rx_hashtbl(bond);
 592
 593         hash_index = bond_info->rx_hashtbl_head;
 594         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 595                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 596
 597                 if (!client_info->slave) {
 598                         printk(KERN_ERR DRV_NAME
 599                                ": %s: Error: found a client with no channel in "
 600                                "the client's hash table\n",
 601                                bond->dev->name);
 602                         continue;
 603                 }
 604                 /*update all clients using this src_ip, that are not assigned
 605                  * to the team's address (curr_active_slave) and have a known
 606                  * unicast mac address.
 607                  */
 608                 if ((client_info->ip_src == src_ip) &&
 609                     memcmp(client_info->slave->dev->dev_addr,
 610                            bond->dev->dev_addr, ETH_ALEN) &&
 611                     memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
 612                         client_info->ntt = 1;
 613                         bond_info->rx_ntt = 1;
 614                 }
 615         }
 616
 617         _unlock_rx_hashtbl(bond);
 618 }
 619
 620 /* Caller must hold both bond and ptr locks for read */
 621 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
 622 {
 623         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 624         struct arp_pkt *arp = arp_pkt(skb);
 625         struct slave *assigned_slave;
 626         struct rlb_client_info *client_info;
 627         u32 hash_index = 0;
 628
 629         _lock_rx_hashtbl(bond);
 630
 631         hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_src));
 632         client_info = &(bond_info->rx_hashtbl[hash_index]);
 633
 634         if (client_info->assigned) {
 635                 if ((client_info->ip_src == arp->ip_src) &&
 636                     (client_info->ip_dst == arp->ip_dst)) {
 637                         /* the entry is already assigned to this client */
 638                         if (memcmp(arp->mac_dst, mac_bcast, ETH_ALEN)) {
 639                                 /* update mac address from arp */
 640                                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 641                         }
 642
 643                         assigned_slave = client_info->slave;
 644                         if (assigned_slave) {
 645                                 _unlock_rx_hashtbl(bond);
 646                                 return assigned_slave;
 647                         }
 648                 } else {
 649                         /* the entry is already assigned to some other client,
 650                          * move the old client to primary (curr_active_slave) so
 651                          * that the new client can be assigned to this entry.
 652                          */
 653                         if (bond->curr_active_slave &&
 654                             client_info->slave != bond->curr_active_slave) {
 655                                 client_info->slave = bond->curr_active_slave;
 656                                 rlb_update_client(client_info);
 657                         }
 658                 }
 659         }
 660         /* assign a new slave */
 661         assigned_slave = rlb_next_rx_slave(bond);
 662
 663         if (assigned_slave) {
 664                 client_info->ip_src = arp->ip_src;
 665                 client_info->ip_dst = arp->ip_dst;
 666                 /* arp->mac_dst is broadcast for arp reqeusts.
 667                  * will be updated with clients actual unicast mac address
 668                  * upon receiving an arp reply.
 669                  */
 670                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 671                 client_info->slave = assigned_slave;
 672
 673                 if (memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
 674                         client_info->ntt = 1;
 675                         bond->alb_info.rx_ntt = 1;
 676                 } else {
 677                         client_info->ntt = 0;
 678                 }
 679
 680                 if (!list_empty(&bond->vlan_list)) {
 681                         unsigned short vlan_id;
 682                         int res = vlan_get_tag(skb, &vlan_id);
 683                         if (!res) {
 684                                 client_info->tag = 1;
 685                                 client_info->vlan_id = vlan_id;
 686                         }
 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 == __constant_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 == __constant_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
1296         skb_reset_mac_header(skb);
1297         eth_data = eth_hdr(skb);
1298
1299         /* make sure that the curr_active_slave and the slaves list do
1300          * not change during tx
1301          */
1302         read_lock(&bond->lock);
1303         read_lock(&bond->curr_slave_lock);
1304
1305         if (!BOND_IS_OK(bond)) {
1306                 goto out;
1307         }
1308
1309         switch (ntohs(skb->protocol)) {
1310         case ETH_P_IP: {
1311                 const struct iphdr *iph = ip_hdr(skb);
1312
1313                 if ((memcmp(eth_data->h_dest, mac_bcast, ETH_ALEN) == 0) ||
1314                     (iph->daddr == ip_bcast) ||
1315                     (iph->protocol == IPPROTO_IGMP)) {
1316                         do_tx_balance = 0;
1317                         break;
1318                 }
1319                 hash_start = (char *)&(iph->daddr);
1320                 hash_size = sizeof(iph->daddr);
1321         }
1322                 break;
1323         case ETH_P_IPV6:
1324                 if (memcmp(eth_data->h_dest, mac_bcast, ETH_ALEN) == 0) {
1325                         do_tx_balance = 0;
1326                         break;
1327                 }
1328
1329                 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1330                 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1331                 break;
1332         case ETH_P_IPX:
1333                 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1334                         /* something is wrong with this packet */
1335                         do_tx_balance = 0;
1336                         break;
1337                 }
1338
1339                 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1340                         /* The only protocol worth balancing in
1341                          * this family since it has an "ARP" like
1342                          * mechanism
1343                          */
1344                         do_tx_balance = 0;
1345                         break;
1346                 }
1347
1348                 hash_start = (char*)eth_data->h_dest;
1349                 hash_size = ETH_ALEN;
1350                 break;
1351         case ETH_P_ARP:
1352                 do_tx_balance = 0;
1353                 if (bond_info->rlb_enabled) {
1354                         tx_slave = rlb_arp_xmit(skb, bond);
1355                 }
1356                 break;
1357         default:
1358                 do_tx_balance = 0;
1359                 break;
1360         }
1361
1362         if (do_tx_balance) {
1363                 hash_index = _simple_hash(hash_start, hash_size);
1364                 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1365         }
1366
1367         if (!tx_slave) {
1368                 /* unbalanced or unassigned, send through primary */
1369                 tx_slave = bond->curr_active_slave;
1370                 bond_info->unbalanced_load += skb->len;
1371         }
1372
1373         if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1374                 if (tx_slave != bond->curr_active_slave) {
1375                         memcpy(eth_data->h_source,
1376                                tx_slave->dev->dev_addr,
1377                                ETH_ALEN);
1378                 }
1379
1380                 res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1381         } else {
1382                 if (tx_slave) {
1383                         tlb_clear_slave(bond, tx_slave, 0);
1384                 }
1385         }
1386
1387 out:
1388         if (res) {
1389                 /* no suitable interface, frame not sent */
1390                 dev_kfree_skb(skb);
1391         }
1392         read_unlock(&bond->curr_slave_lock);
1393         read_unlock(&bond->lock);
1394         return 0;
1395 }
1396
1397 void bond_alb_monitor(struct work_struct *work)
1398 {
1399         struct bonding *bond = container_of(work, struct bonding,
1400                                             alb_work.work);
1401         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1402         struct slave *slave;
1403         int i;
1404
1405         read_lock(&bond->lock);
1406
1407         if (bond->kill_timers) {
1408                 goto out;
1409         }
1410
1411         if (bond->slave_cnt == 0) {
1412                 bond_info->tx_rebalance_counter = 0;
1413                 bond_info->lp_counter = 0;
1414                 goto re_arm;
1415         }
1416
1417         bond_info->tx_rebalance_counter++;
1418         bond_info->lp_counter++;
1419
1420         /* send learning packets */
1421         if (bond_info->lp_counter >= BOND_ALB_LP_TICKS) {
1422                 /* change of curr_active_slave involves swapping of mac addresses.
1423                  * in order to avoid this swapping from happening while
1424                  * sending the learning packets, the curr_slave_lock must be held for
1425                  * read.
1426                  */
1427                 read_lock(&bond->curr_slave_lock);
1428
1429                 bond_for_each_slave(bond, slave, i) {
1430                         alb_send_learning_packets(slave, slave->dev->dev_addr);
1431                 }
1432
1433                 read_unlock(&bond->curr_slave_lock);
1434
1435                 bond_info->lp_counter = 0;
1436         }
1437
1438         /* rebalance tx traffic */
1439         if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1440
1441                 read_lock(&bond->curr_slave_lock);
1442
1443                 bond_for_each_slave(bond, slave, i) {
1444                         tlb_clear_slave(bond, slave, 1);
1445                         if (slave == bond->curr_active_slave) {
1446                                 SLAVE_TLB_INFO(slave).load =
1447                                         bond_info->unbalanced_load /
1448                                                 BOND_TLB_REBALANCE_INTERVAL;
1449                                 bond_info->unbalanced_load = 0;
1450                         }
1451                 }
1452
1453                 read_unlock(&bond->curr_slave_lock);
1454
1455                 bond_info->tx_rebalance_counter = 0;
1456         }
1457
1458         /* handle rlb stuff */
1459         if (bond_info->rlb_enabled) {
1460                 if (bond_info->primary_is_promisc &&
1461                     (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1462
1463                         /*
1464                          * dev_set_promiscuity requires rtnl and
1465                          * nothing else.
1466                          */
1467                         read_unlock(&bond->lock);
1468                         rtnl_lock();
1469
1470                         bond_info->rlb_promisc_timeout_counter = 0;
1471
1472                         /* If the primary was set to promiscuous mode
1473                          * because a slave was disabled then
1474                          * it can now leave promiscuous mode.
1475                          */
1476                         dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1477                         bond_info->primary_is_promisc = 0;
1478
1479                         rtnl_unlock();
1480                         read_lock(&bond->lock);
1481                 }
1482
1483                 if (bond_info->rlb_rebalance) {
1484                         bond_info->rlb_rebalance = 0;
1485                         rlb_rebalance(bond);
1486                 }
1487
1488                 /* check if clients need updating */
1489                 if (bond_info->rx_ntt) {
1490                         if (bond_info->rlb_update_delay_counter) {
1491                                 --bond_info->rlb_update_delay_counter;
1492                         } else {
1493                                 rlb_update_rx_clients(bond);
1494                                 if (bond_info->rlb_update_retry_counter) {
1495                                         --bond_info->rlb_update_retry_counter;
1496                                 } else {
1497                                         bond_info->rx_ntt = 0;
1498                                 }
1499                         }
1500                 }
1501         }
1502
1503 re_arm:
1504         queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1505 out:
1506         read_unlock(&bond->lock);
1507 }
1508
1509 /* assumption: called before the slave is attached to the bond
1510  * and not locked by the bond lock
1511  */
1512 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1513 {
1514         int res;
1515
1516         res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1517                                      bond->alb_info.rlb_enabled);
1518         if (res) {
1519                 return res;
1520         }
1521
1522         /* caller must hold the bond lock for write since the mac addresses
1523          * are compared and may be swapped.
1524          */
1525         read_lock(&bond->lock);
1526
1527         res = alb_handle_addr_collision_on_attach(bond, slave);
1528
1529         read_unlock(&bond->lock);
1530
1531         if (res) {
1532                 return res;
1533         }
1534
1535         tlb_init_slave(slave);
1536
1537         /* order a rebalance ASAP */
1538         bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1539
1540         if (bond->alb_info.rlb_enabled) {
1541                 bond->alb_info.rlb_rebalance = 1;
1542         }
1543
1544         return 0;
1545 }
1546
1547 /*
1548  * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1549  * if necessary.
1550  *
1551  * Caller must hold RTNL and no other locks
1552  */
1553 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1554 {
1555         if (bond->slave_cnt > 1) {
1556                 alb_change_hw_addr_on_detach(bond, slave);
1557         }
1558
1559         tlb_clear_slave(bond, slave, 0);
1560
1561         if (bond->alb_info.rlb_enabled) {
1562                 bond->alb_info.next_rx_slave = NULL;
1563                 rlb_clear_slave(bond, slave);
1564         }
1565 }
1566
1567 /* Caller must hold bond lock for read */
1568 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1569 {
1570         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1571
1572         if (link == BOND_LINK_DOWN) {
1573                 tlb_clear_slave(bond, slave, 0);
1574                 if (bond->alb_info.rlb_enabled) {
1575                         rlb_clear_slave(bond, slave);
1576                 }
1577         } else if (link == BOND_LINK_UP) {
1578                 /* order a rebalance ASAP */
1579                 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1580                 if (bond->alb_info.rlb_enabled) {
1581                         bond->alb_info.rlb_rebalance = 1;
1582                         /* If the updelay module parameter is smaller than the
1583                          * forwarding delay of the switch the rebalance will
1584                          * not work because the rebalance arp replies will
1585                          * not be forwarded to the clients..
1586                          */
1587                 }
1588         }
1589 }
1590
1591 /**
1592  * bond_alb_handle_active_change - assign new curr_active_slave
1593  * @bond: our bonding struct
1594  * @new_slave: new slave to assign
1595  *
1596  * Set the bond->curr_active_slave to @new_slave and handle
1597  * mac address swapping and promiscuity changes as needed.
1598  *
1599  * If new_slave is NULL, caller must hold curr_slave_lock or
1600  * bond->lock for write.
1601  *
1602  * If new_slave is not NULL, caller must hold RTNL, bond->lock for
1603  * read and curr_slave_lock for write.  Processing here may sleep, so
1604  * no other locks may be held.
1605  */
1606 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1607 {
1608         struct slave *swap_slave;
1609         int i;
1610
1611         if (bond->curr_active_slave == new_slave) {
1612                 return;
1613         }
1614
1615         if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1616                 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1617                 bond->alb_info.primary_is_promisc = 0;
1618                 bond->alb_info.rlb_promisc_timeout_counter = 0;
1619         }
1620
1621         swap_slave = bond->curr_active_slave;
1622         bond->curr_active_slave = new_slave;
1623
1624         if (!new_slave || (bond->slave_cnt == 0)) {
1625                 return;
1626         }
1627
1628         /* set the new curr_active_slave to the bonds mac address
1629          * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1630          */
1631         if (!swap_slave) {
1632                 struct slave *tmp_slave;
1633                 /* find slave that is holding the bond's mac address */
1634                 bond_for_each_slave(bond, tmp_slave, i) {
1635                         if (!memcmp(tmp_slave->dev->dev_addr,
1636                                     bond->dev->dev_addr, ETH_ALEN)) {
1637                                 swap_slave = tmp_slave;
1638                                 break;
1639                         }
1640                 }
1641         }
1642
1643         /*
1644          * Arrange for swap_slave and new_slave to temporarily be
1645          * ignored so we can mess with their MAC addresses without
1646          * fear of interference from transmit activity.
1647          */
1648         if (swap_slave) {
1649                 tlb_clear_slave(bond, swap_slave, 1);
1650         }
1651         tlb_clear_slave(bond, new_slave, 1);
1652
1653         write_unlock_bh(&bond->curr_slave_lock);
1654         read_unlock(&bond->lock);
1655
1656         ASSERT_RTNL();
1657
1658         /* curr_active_slave must be set before calling alb_swap_mac_addr */
1659         if (swap_slave) {
1660                 /* swap mac address */
1661                 alb_swap_mac_addr(bond, swap_slave, new_slave);
1662         } else {
1663                 /* set the new_slave to the bond mac address */
1664                 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1665                                        bond->alb_info.rlb_enabled);
1666         }
1667
1668         if (swap_slave) {
1669                 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1670                 read_lock(&bond->lock);
1671         } else {
1672                 read_lock(&bond->lock);
1673                 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1674         }
1675
1676         write_lock_bh(&bond->curr_slave_lock);
1677 }
1678
1679 /*
1680  * Called with RTNL
1681  */
1682 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1683 {
1684         struct bonding *bond = bond_dev->priv;
1685         struct sockaddr *sa = addr;
1686         struct slave *slave, *swap_slave;
1687         int res;
1688         int i;
1689
1690         if (!is_valid_ether_addr(sa->sa_data)) {
1691                 return -EADDRNOTAVAIL;
1692         }
1693
1694         res = alb_set_mac_address(bond, addr);
1695         if (res) {
1696                 return res;
1697         }
1698
1699         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1700
1701         /* If there is no curr_active_slave there is nothing else to do.
1702          * Otherwise we'll need to pass the new address to it and handle
1703          * duplications.
1704          */
1705         if (!bond->curr_active_slave) {
1706                 return 0;
1707         }
1708
1709         swap_slave = NULL;
1710
1711         bond_for_each_slave(bond, slave, i) {
1712                 if (!memcmp(slave->dev->dev_addr, bond_dev->dev_addr, ETH_ALEN)) {
1713                         swap_slave = slave;
1714                         break;
1715                 }
1716         }
1717
1718         write_unlock_bh(&bond->curr_slave_lock);
1719         read_unlock(&bond->lock);
1720
1721         if (swap_slave) {
1722                 alb_swap_mac_addr(bond, swap_slave, bond->curr_active_slave);
1723                 alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave);
1724         } else {
1725                 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr,
1726                                        bond->alb_info.rlb_enabled);
1727
1728                 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1729                 if (bond->alb_info.rlb_enabled) {
1730                         /* inform clients mac address has changed */
1731                         rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1732                 }
1733         }
1734
1735         read_lock(&bond->lock);
1736         write_lock_bh(&bond->curr_slave_lock);
1737
1738         return 0;
1739 }
1740
1741 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1742 {
1743         if (bond->alb_info.current_alb_vlan &&
1744             (bond->alb_info.current_alb_vlan->vlan_id == vlan_id)) {
1745                 bond->alb_info.current_alb_vlan = NULL;
1746         }
1747
1748         if (bond->alb_info.rlb_enabled) {
1749                 rlb_clear_vlan(bond, vlan_id);
1750         }
1751 }
1752