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  * Changes:
  23  *
  24  * 2003/06/25 - Shmulik Hen <shmulik.hen at intel dot com>
  25  *      - Fixed signed/unsigned calculation errors that caused load sharing
  26  *        to collapse to one slave under very heavy UDP Tx stress.
  27  *
  28  * 2003/08/06 - Amir Noam <amir.noam at intel dot com>
  29  *      - Add support for setting bond's MAC address with special
  30  *        handling required for ALB/TLB.
  31  *
  32  * 2003/12/01 - Shmulik Hen <shmulik.hen at intel dot com>
  33  *      - Code cleanup and style changes
  34  *
  35  * 2003/12/30 - Amir Noam <amir.noam at intel dot com>
  36  *      - Fixed: Cannot remove and re-enslave the original active slave.
  37  *
  38  * 2004/01/14 - Shmulik Hen <shmulik.hen at intel dot com>
  39  *      - Add capability to tag self generated packets in ALB/TLB modes.
  40  */
  41
  42 //#define BONDING_DEBUG 1
  43
  44 #include <linux/skbuff.h>
  45 #include <linux/netdevice.h>
  46 #include <linux/etherdevice.h>
  47 #include <linux/pkt_sched.h>
  48 #include <linux/spinlock.h>
  49 #include <linux/slab.h>
  50 #include <linux/timer.h>
  51 #include <linux/ip.h>
  52 #include <linux/ipv6.h>
  53 #include <linux/if_arp.h>
  54 #include <linux/if_ether.h>
  55 #include <linux/if_bonding.h>
  56 #include <linux/if_vlan.h>
  57 #include <linux/in.h>
  58 #include <net/ipx.h>
  59 #include <net/arp.h>
  60 #include <asm/byteorder.h>
  61 #include "bonding.h"
  62 #include "bond_alb.h"
  63
  64
  65 #define ALB_TIMER_TICKS_PER_SEC     10  /* should be a divisor of HZ */
  66 #define BOND_TLB_REBALANCE_INTERVAL 10  /* In seconds, periodic re-balancing.
  67                                          * Used for division - never set
  68                                          * to zero !!!
  69                                          */
  70 #define BOND_ALB_LP_INTERVAL        1   /* In seconds, periodic send of
  71                                          * learning packets to the switch
  72                                          */
  73
  74 #define BOND_TLB_REBALANCE_TICKS (BOND_TLB_REBALANCE_INTERVAL \
  75                                   * ALB_TIMER_TICKS_PER_SEC)
  76
  77 #define BOND_ALB_LP_TICKS (BOND_ALB_LP_INTERVAL \
  78                            * ALB_TIMER_TICKS_PER_SEC)
  79
  80 #define TLB_HASH_TABLE_SIZE 256 /* The size of the clients hash table.
  81                                  * Note that this value MUST NOT be smaller
  82                                  * because the key hash table is BYTE wide !
  83                                  */
  84
  85
  86 #define TLB_NULL_INDEX          0xffffffff
  87 #define MAX_LP_BURST            3
  88
  89 /* rlb defs */
  90 #define RLB_HASH_TABLE_SIZE     256
  91 #define RLB_NULL_INDEX          0xffffffff
  92 #define RLB_UPDATE_DELAY        2*ALB_TIMER_TICKS_PER_SEC /* 2 seconds */
  93 #define RLB_ARP_BURST_SIZE      2
  94 #define RLB_UPDATE_RETRY        3       /* 3-ticks - must be smaller than the rlb
  95                                          * rebalance interval (5 min).
  96                                          */
  97 /* RLB_PROMISC_TIMEOUT = 10 sec equals the time that the current slave is
  98  * promiscuous after failover
  99  */
 100 #define RLB_PROMISC_TIMEOUT     10*ALB_TIMER_TICKS_PER_SEC
 101
 102 static const u8 mac_bcast[ETH_ALEN] = {0xff,0xff,0xff,0xff,0xff,0xff};
 103 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
 104
 105 #pragma pack(1)
 106 struct learning_pkt {
 107         u8 mac_dst[ETH_ALEN];
 108         u8 mac_src[ETH_ALEN];
 109         u16 type;
 110         u8 padding[ETH_ZLEN - ETH_HLEN];
 111 };
 112
 113 struct arp_pkt {
 114         u16     hw_addr_space;
 115         u16     prot_addr_space;
 116         u8      hw_addr_len;
 117         u8      prot_addr_len;
 118         u16     op_code;
 119         u8      mac_src[ETH_ALEN];      /* sender hardware address */
 120         u32     ip_src;                 /* sender IP address */
 121         u8      mac_dst[ETH_ALEN];      /* target hardware address */
 122         u32     ip_dst;                 /* target IP address */
 123 };
 124 #pragma pack()
 125
 126 /* Forward declaration */
 127 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[]);
 128
 129 static inline u8 _simple_hash(u8 *hash_start, int hash_size)
 130 {
 131         int i;
 132         u8 hash = 0;
 133
 134         for (i = 0; i < hash_size; i++) {
 135                 hash ^= hash_start[i];
 136         }
 137
 138         return hash;
 139 }
 140
 141 /*********************** tlb specific functions ***************************/
 142
 143 static inline void _lock_tx_hashtbl(struct bonding *bond)
 144 {
 145         spin_lock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 146 }
 147
 148 static inline void _unlock_tx_hashtbl(struct bonding *bond)
 149 {
 150         spin_unlock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 151 }
 152
 153 /* Caller must hold tx_hashtbl lock */
 154 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
 155 {
 156         if (save_load) {
 157                 entry->load_history = 1 + entry->tx_bytes /
 158                                       BOND_TLB_REBALANCE_INTERVAL;
 159                 entry->tx_bytes = 0;
 160         }
 161
 162         entry->tx_slave = NULL;
 163         entry->next = TLB_NULL_INDEX;
 164         entry->prev = TLB_NULL_INDEX;
 165 }
 166
 167 static inline void tlb_init_slave(struct slave *slave)
 168 {
 169         SLAVE_TLB_INFO(slave).load = 0;
 170         SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
 171 }
 172
 173 /* Caller must hold bond lock for read */
 174 static void tlb_clear_slave(struct bonding *bond, struct slave *slave, int save_load)
 175 {
 176         struct tlb_client_info *tx_hash_table;
 177         u32 index;
 178
 179         _lock_tx_hashtbl(bond);
 180
 181         /* clear slave from tx_hashtbl */
 182         tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
 183
 184         index = SLAVE_TLB_INFO(slave).head;
 185         while (index != TLB_NULL_INDEX) {
 186                 u32 next_index = tx_hash_table[index].next;
 187                 tlb_init_table_entry(&tx_hash_table[index], save_load);
 188                 index = next_index;
 189         }
 190
 191         _unlock_tx_hashtbl(bond);
 192
 193         tlb_init_slave(slave);
 194 }
 195
 196 /* Must be called before starting the monitor timer */
 197 static int tlb_initialize(struct bonding *bond)
 198 {
 199         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 200         int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 201         struct tlb_client_info *new_hashtbl;
 202         int i;
 203
 204         spin_lock_init(&(bond_info->tx_hashtbl_lock));
 205
 206         new_hashtbl = kmalloc(size, GFP_KERNEL);
 207         if (!new_hashtbl) {
 208                 printk(KERN_ERR DRV_NAME
 209                        ": %s: Error: Failed to allocate TLB hash table\n",
 210                        bond->dev->name);
 211                 return -1;
 212         }
 213         _lock_tx_hashtbl(bond);
 214
 215         bond_info->tx_hashtbl = new_hashtbl;
 216
 217         memset(bond_info->tx_hashtbl, 0, size);
 218
 219         for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
 220                 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 1);
 221         }
 222
 223         _unlock_tx_hashtbl(bond);
 224
 225         return 0;
 226 }
 227
 228 /* Must be called only after all slaves have been released */
 229 static void tlb_deinitialize(struct bonding *bond)
 230 {
 231         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 232
 233         _lock_tx_hashtbl(bond);
 234
 235         kfree(bond_info->tx_hashtbl);
 236         bond_info->tx_hashtbl = NULL;
 237
 238         _unlock_tx_hashtbl(bond);
 239 }
 240
 241 /* Caller must hold bond lock for read */
 242 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
 243 {
 244         struct slave *slave, *least_loaded;
 245         s64 max_gap;
 246         int i, found = 0;
 247
 248         /* Find the first enabled slave */
 249         bond_for_each_slave(bond, slave, i) {
 250                 if (SLAVE_IS_OK(slave)) {
 251                         found = 1;
 252                         break;
 253                 }
 254         }
 255
 256         if (!found) {
 257                 return NULL;
 258         }
 259
 260         least_loaded = slave;
 261         max_gap = (s64)(slave->speed << 20) - /* Convert to Megabit per sec */
 262                         (s64)(SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
 263
 264         /* Find the slave with the largest gap */
 265         bond_for_each_slave_from(bond, slave, i, least_loaded) {
 266                 if (SLAVE_IS_OK(slave)) {
 267                         s64 gap = (s64)(slave->speed << 20) -
 268                                         (s64)(SLAVE_TLB_INFO(slave).load << 3);
 269                         if (max_gap < gap) {
 270                                 least_loaded = slave;
 271                                 max_gap = gap;
 272                         }
 273                 }
 274         }
 275
 276         return least_loaded;
 277 }
 278
 279 /* Caller must hold bond lock for read */
 280 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index, u32 skb_len)
 281 {
 282         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 283         struct tlb_client_info *hash_table;
 284         struct slave *assigned_slave;
 285
 286         _lock_tx_hashtbl(bond);
 287
 288         hash_table = bond_info->tx_hashtbl;
 289         assigned_slave = hash_table[hash_index].tx_slave;
 290         if (!assigned_slave) {
 291                 assigned_slave = tlb_get_least_loaded_slave(bond);
 292
 293                 if (assigned_slave) {
 294                         struct tlb_slave_info *slave_info =
 295                                 &(SLAVE_TLB_INFO(assigned_slave));
 296                         u32 next_index = slave_info->head;
 297
 298                         hash_table[hash_index].tx_slave = assigned_slave;
 299                         hash_table[hash_index].next = next_index;
 300                         hash_table[hash_index].prev = TLB_NULL_INDEX;
 301
 302                         if (next_index != TLB_NULL_INDEX) {
 303                                 hash_table[next_index].prev = hash_index;
 304                         }
 305
 306                         slave_info->head = hash_index;
 307                         slave_info->load +=
 308                                 hash_table[hash_index].load_history;
 309                 }
 310         }
 311
 312         if (assigned_slave) {
 313                 hash_table[hash_index].tx_bytes += skb_len;
 314         }
 315
 316         _unlock_tx_hashtbl(bond);
 317
 318         return assigned_slave;
 319 }
 320
 321 /*********************** rlb specific functions ***************************/
 322 static inline void _lock_rx_hashtbl(struct bonding *bond)
 323 {
 324         spin_lock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 325 }
 326
 327 static inline void _unlock_rx_hashtbl(struct bonding *bond)
 328 {
 329         spin_unlock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 330 }
 331
 332 /* when an ARP REPLY is received from a client update its info
 333  * in the rx_hashtbl
 334  */
 335 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 336 {
 337         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 338         struct rlb_client_info *client_info;
 339         u32 hash_index;
 340
 341         _lock_rx_hashtbl(bond);
 342
 343         hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
 344         client_info = &(bond_info->rx_hashtbl[hash_index]);
 345
 346         if ((client_info->assigned) &&
 347             (client_info->ip_src == arp->ip_dst) &&
 348             (client_info->ip_dst == arp->ip_src)) {
 349                 /* update the clients MAC address */
 350                 memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
 351                 client_info->ntt = 1;
 352                 bond_info->rx_ntt = 1;
 353         }
 354
 355         _unlock_rx_hashtbl(bond);
 356 }
 357
 358 static int rlb_arp_recv(struct sk_buff *skb, struct net_device *bond_dev, struct packet_type *ptype, struct net_device *orig_dev)
 359 {
 360         struct bonding *bond = bond_dev->priv;
 361         struct arp_pkt *arp = (struct arp_pkt *)skb->data;
 362         int res = NET_RX_DROP;
 363
 364         if (!(bond_dev->flags & IFF_MASTER))
 365                 goto out;
 366
 367         if (!arp) {
 368                 dprintk("Packet has no ARP data\n");
 369                 goto out;
 370         }
 371
 372         if (skb->len < sizeof(struct arp_pkt)) {
 373                 dprintk("Packet is too small to be an ARP\n");
 374                 goto out;
 375         }
 376
 377         if (arp->op_code == htons(ARPOP_REPLY)) {
 378                 /* update rx hash table for this ARP */
 379                 rlb_update_entry_from_arp(bond, arp);
 380                 dprintk("Server received an ARP Reply from client\n");
 381         }
 382
 383         res = NET_RX_SUCCESS;
 384
 385 out:
 386         dev_kfree_skb(skb);
 387
 388         return res;
 389 }
 390
 391 /* Caller must hold bond lock for read */
 392 static struct slave *rlb_next_rx_slave(struct bonding *bond)
 393 {
 394         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 395         struct slave *rx_slave, *slave, *start_at;
 396         int i = 0;
 397
 398         if (bond_info->next_rx_slave) {
 399                 start_at = bond_info->next_rx_slave;
 400         } else {
 401                 start_at = bond->first_slave;
 402         }
 403
 404         rx_slave = NULL;
 405
 406         bond_for_each_slave_from(bond, slave, i, start_at) {
 407                 if (SLAVE_IS_OK(slave)) {
 408                         if (!rx_slave) {
 409                                 rx_slave = slave;
 410                         } else if (slave->speed > rx_slave->speed) {
 411                                 rx_slave = slave;
 412                         }
 413                 }
 414         }
 415
 416         if (rx_slave) {
 417                 bond_info->next_rx_slave = rx_slave->next;
 418         }
 419
 420         return rx_slave;
 421 }
 422
 423 /* teach the switch the mac of a disabled slave
 424  * on the primary for fault tolerance
 425  *
 426  * Caller must hold bond->curr_slave_lock for write or bond lock for write
 427  */
 428 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
 429 {
 430         if (!bond->curr_active_slave) {
 431                 return;
 432         }
 433
 434         if (!bond->alb_info.primary_is_promisc) {
 435                 bond->alb_info.primary_is_promisc = 1;
 436                 dev_set_promiscuity(bond->curr_active_slave->dev, 1);
 437         }
 438
 439         bond->alb_info.rlb_promisc_timeout_counter = 0;
 440
 441         alb_send_learning_packets(bond->curr_active_slave, addr);
 442 }
 443
 444 /* slave being removed should not be active at this point
 445  *
 446  * Caller must hold bond lock for read
 447  */
 448 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 449 {
 450         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 451         struct rlb_client_info *rx_hash_table;
 452         u32 index, next_index;
 453
 454         /* clear slave from rx_hashtbl */
 455         _lock_rx_hashtbl(bond);
 456
 457         rx_hash_table = bond_info->rx_hashtbl;
 458         index = bond_info->rx_hashtbl_head;
 459         for (; index != RLB_NULL_INDEX; index = next_index) {
 460                 next_index = rx_hash_table[index].next;
 461                 if (rx_hash_table[index].slave == slave) {
 462                         struct slave *assigned_slave = rlb_next_rx_slave(bond);
 463
 464                         if (assigned_slave) {
 465                                 rx_hash_table[index].slave = assigned_slave;
 466                                 if (memcmp(rx_hash_table[index].mac_dst,
 467                                            mac_bcast, ETH_ALEN)) {
 468                                         bond_info->rx_hashtbl[index].ntt = 1;
 469                                         bond_info->rx_ntt = 1;
 470                                         /* A slave has been removed from the
 471                                          * table because it is either disabled
 472                                          * or being released. We must retry the
 473                                          * update to avoid clients from not
 474                                          * being updated & disconnecting when
 475                                          * there is stress
 476                                          */
 477                                         bond_info->rlb_update_retry_counter =
 478                                                 RLB_UPDATE_RETRY;
 479                                 }
 480                         } else {  /* there is no active slave */
 481                                 rx_hash_table[index].slave = NULL;
 482                         }
 483                 }
 484         }
 485
 486         _unlock_rx_hashtbl(bond);
 487
 488         write_lock(&bond->curr_slave_lock);
 489
 490         if (slave != bond->curr_active_slave) {
 491                 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 492         }
 493
 494         write_unlock(&bond->curr_slave_lock);
 495 }
 496
 497 static void rlb_update_client(struct rlb_client_info *client_info)
 498 {
 499         int i;
 500
 501         if (!client_info->slave) {
 502                 return;
 503         }
 504
 505         for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 506                 struct sk_buff *skb;
 507
 508                 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 509                                  client_info->ip_dst,
 510                                  client_info->slave->dev,
 511                                  client_info->ip_src,
 512                                  client_info->mac_dst,
 513                                  client_info->slave->dev->dev_addr,
 514                                  client_info->mac_dst);
 515                 if (!skb) {
 516                         printk(KERN_ERR DRV_NAME
 517                                ": %s: Error: failed to create an ARP packet\n",
 518                                client_info->slave->dev->master->name);
 519                         continue;
 520                 }
 521
 522                 skb->dev = client_info->slave->dev;
 523
 524                 if (client_info->tag) {
 525                         skb = vlan_put_tag(skb, client_info->vlan_id);
 526                         if (!skb) {
 527                                 printk(KERN_ERR DRV_NAME
 528                                        ": %s: Error: failed to insert VLAN tag\n",
 529                                        client_info->slave->dev->master->name);
 530                                 continue;
 531                         }
 532                 }
 533
 534                 arp_xmit(skb);
 535         }
 536 }
 537
 538 /* sends ARP REPLIES that update the clients that need updating */
 539 static void rlb_update_rx_clients(struct bonding *bond)
 540 {
 541         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 542         struct rlb_client_info *client_info;
 543         u32 hash_index;
 544
 545         _lock_rx_hashtbl(bond);
 546
 547         hash_index = bond_info->rx_hashtbl_head;
 548         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 549                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 550                 if (client_info->ntt) {
 551                         rlb_update_client(client_info);
 552                         if (bond_info->rlb_update_retry_counter == 0) {
 553                                 client_info->ntt = 0;
 554                         }
 555                 }
 556         }
 557
 558         /* do not update the entries again untill this counter is zero so that
 559          * not to confuse the clients.
 560          */
 561         bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 562
 563         _unlock_rx_hashtbl(bond);
 564 }
 565
 566 /* The slave was assigned a new mac address - update the clients */
 567 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 568 {
 569         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 570         struct rlb_client_info *client_info;
 571         int ntt = 0;
 572         u32 hash_index;
 573
 574         _lock_rx_hashtbl(bond);
 575
 576         hash_index = bond_info->rx_hashtbl_head;
 577         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 578                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 579
 580                 if ((client_info->slave == slave) &&
 581                     memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
 582                         client_info->ntt = 1;
 583                         ntt = 1;
 584                 }
 585         }
 586
 587         // update the team's flag only after the whole iteration
 588         if (ntt) {
 589                 bond_info->rx_ntt = 1;
 590                 //fasten the change
 591                 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 592         }
 593
 594         _unlock_rx_hashtbl(bond);
 595 }
 596
 597 /* mark all clients using src_ip to be updated */
 598 static void rlb_req_update_subnet_clients(struct bonding *bond, u32 src_ip)
 599 {
 600         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 601         struct rlb_client_info *client_info;
 602         u32 hash_index;
 603
 604         _lock_rx_hashtbl(bond);
 605
 606         hash_index = bond_info->rx_hashtbl_head;
 607         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 608                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 609
 610                 if (!client_info->slave) {
 611                         printk(KERN_ERR DRV_NAME
 612                                ": %s: Error: found a client with no channel in "
 613                                "the client's hash table\n",
 614                                bond->dev->name);
 615                         continue;
 616                 }
 617                 /*update all clients using this src_ip, that are not assigned
 618                  * to the team's address (curr_active_slave) and have a known
 619                  * unicast mac address.
 620                  */
 621                 if ((client_info->ip_src == src_ip) &&
 622                     memcmp(client_info->slave->dev->dev_addr,
 623                            bond->dev->dev_addr, ETH_ALEN) &&
 624                     memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
 625                         client_info->ntt = 1;
 626                         bond_info->rx_ntt = 1;
 627                 }
 628         }
 629
 630         _unlock_rx_hashtbl(bond);
 631 }
 632
 633 /* Caller must hold both bond and ptr locks for read */
 634 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
 635 {
 636         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 637         struct arp_pkt *arp = (struct arp_pkt *)skb->nh.raw;
 638         struct slave *assigned_slave;
 639         struct rlb_client_info *client_info;
 640         u32 hash_index = 0;
 641
 642         _lock_rx_hashtbl(bond);
 643
 644         hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_src));
 645         client_info = &(bond_info->rx_hashtbl[hash_index]);
 646
 647         if (client_info->assigned) {
 648                 if ((client_info->ip_src == arp->ip_src) &&
 649                     (client_info->ip_dst == arp->ip_dst)) {
 650                         /* the entry is already assigned to this client */
 651                         if (memcmp(arp->mac_dst, mac_bcast, ETH_ALEN)) {
 652                                 /* update mac address from arp */
 653                                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 654                         }
 655
 656                         assigned_slave = client_info->slave;
 657                         if (assigned_slave) {
 658                                 _unlock_rx_hashtbl(bond);
 659                                 return assigned_slave;
 660                         }
 661                 } else {
 662                         /* the entry is already assigned to some other client,
 663                          * move the old client to primary (curr_active_slave) so
 664                          * that the new client can be assigned to this entry.
 665                          */
 666                         if (bond->curr_active_slave &&
 667                             client_info->slave != bond->curr_active_slave) {
 668                                 client_info->slave = bond->curr_active_slave;
 669                                 rlb_update_client(client_info);
 670                         }
 671                 }
 672         }
 673         /* assign a new slave */
 674         assigned_slave = rlb_next_rx_slave(bond);
 675
 676         if (assigned_slave) {
 677                 client_info->ip_src = arp->ip_src;
 678                 client_info->ip_dst = arp->ip_dst;
 679                 /* arp->mac_dst is broadcast for arp reqeusts.
 680                  * will be updated with clients actual unicast mac address
 681                  * upon receiving an arp reply.
 682                  */
 683                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 684                 client_info->slave = assigned_slave;
 685
 686                 if (memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
 687                         client_info->ntt = 1;
 688                         bond->alb_info.rx_ntt = 1;
 689                 } else {
 690                         client_info->ntt = 0;
 691                 }
 692
 693                 if (!list_empty(&bond->vlan_list)) {
 694                         unsigned short vlan_id;
 695                         int res = vlan_get_tag(skb, &vlan_id);
 696                         if (!res) {
 697                                 client_info->tag = 1;
 698                                 client_info->vlan_id = vlan_id;
 699                         }
 700                 }
 701
 702                 if (!client_info->assigned) {
 703                         u32 prev_tbl_head = bond_info->rx_hashtbl_head;
 704                         bond_info->rx_hashtbl_head = hash_index;
 705                         client_info->next = prev_tbl_head;
 706                         if (prev_tbl_head != RLB_NULL_INDEX) {
 707                                 bond_info->rx_hashtbl[prev_tbl_head].prev =
 708                                         hash_index;
 709                         }
 710                         client_info->assigned = 1;
 711                 }
 712         }
 713
 714         _unlock_rx_hashtbl(bond);
 715
 716         return assigned_slave;
 717 }
 718
 719 /* chooses (and returns) transmit channel for arp reply
 720  * does not choose channel for other arp types since they are
 721  * sent on the curr_active_slave
 722  */
 723 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 724 {
 725         struct arp_pkt *arp = (struct arp_pkt *)skb->nh.raw;
 726         struct slave *tx_slave = NULL;
 727
 728         if (arp->op_code == __constant_htons(ARPOP_REPLY)) {
 729                 /* the arp must be sent on the selected
 730                 * rx channel
 731                 */
 732                 tx_slave = rlb_choose_channel(skb, bond);
 733                 if (tx_slave) {
 734                         memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
 735                 }
 736                 dprintk("Server sent ARP Reply packet\n");
 737         } else if (arp->op_code == __constant_htons(ARPOP_REQUEST)) {
 738                 /* Create an entry in the rx_hashtbl for this client as a
 739                  * place holder.
 740                  * When the arp reply is received the entry will be updated
 741                  * with the correct unicast address of the client.
 742                  */
 743                 rlb_choose_channel(skb, bond);
 744
 745                 /* The ARP relpy packets must be delayed so that
 746                  * they can cancel out the influence of the ARP request.
 747                  */
 748                 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 749
 750                 /* arp requests are broadcast and are sent on the primary
 751                  * the arp request will collapse all clients on the subnet to
 752                  * the primary slave. We must register these clients to be
 753                  * updated with their assigned mac.
 754                  */
 755                 rlb_req_update_subnet_clients(bond, arp->ip_src);
 756                 dprintk("Server sent ARP Request packet\n");
 757         }
 758
 759         return tx_slave;
 760 }
 761
 762 /* Caller must hold bond lock for read */
 763 static void rlb_rebalance(struct bonding *bond)
 764 {
 765         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 766         struct slave *assigned_slave;
 767         struct rlb_client_info *client_info;
 768         int ntt;
 769         u32 hash_index;
 770
 771         _lock_rx_hashtbl(bond);
 772
 773         ntt = 0;
 774         hash_index = bond_info->rx_hashtbl_head;
 775         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 776                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 777                 assigned_slave = rlb_next_rx_slave(bond);
 778                 if (assigned_slave && (client_info->slave != assigned_slave)) {
 779                         client_info->slave = assigned_slave;
 780                         client_info->ntt = 1;
 781                         ntt = 1;
 782                 }
 783         }
 784
 785         /* update the team's flag only after the whole iteration */
 786         if (ntt) {
 787                 bond_info->rx_ntt = 1;
 788         }
 789         _unlock_rx_hashtbl(bond);
 790 }
 791
 792 /* Caller must hold rx_hashtbl lock */
 793 static void rlb_init_table_entry(struct rlb_client_info *entry)
 794 {
 795         memset(entry, 0, sizeof(struct rlb_client_info));
 796         entry->next = RLB_NULL_INDEX;
 797         entry->prev = RLB_NULL_INDEX;
 798 }
 799
 800 static int rlb_initialize(struct bonding *bond)
 801 {
 802         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 803         struct packet_type *pk_type = &(BOND_ALB_INFO(bond).rlb_pkt_type);
 804         struct rlb_client_info  *new_hashtbl;
 805         int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 806         int i;
 807
 808         spin_lock_init(&(bond_info->rx_hashtbl_lock));
 809
 810         new_hashtbl = kmalloc(size, GFP_KERNEL);
 811         if (!new_hashtbl) {
 812                 printk(KERN_ERR DRV_NAME
 813                        ": %s: Error: Failed to allocate RLB hash table\n",
 814                        bond->dev->name);
 815                 return -1;
 816         }
 817         _lock_rx_hashtbl(bond);
 818
 819         bond_info->rx_hashtbl = new_hashtbl;
 820
 821         bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
 822
 823         for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
 824                 rlb_init_table_entry(bond_info->rx_hashtbl + i);
 825         }
 826
 827         _unlock_rx_hashtbl(bond);
 828
 829         /*initialize packet type*/
 830         pk_type->type = __constant_htons(ETH_P_ARP);
 831         pk_type->dev = bond->dev;
 832         pk_type->func = rlb_arp_recv;
 833
 834         /* register to receive ARPs */
 835         dev_add_pack(pk_type);
 836
 837         return 0;
 838 }
 839
 840 static void rlb_deinitialize(struct bonding *bond)
 841 {
 842         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 843
 844         dev_remove_pack(&(bond_info->rlb_pkt_type));
 845
 846         _lock_rx_hashtbl(bond);
 847
 848         kfree(bond_info->rx_hashtbl);
 849         bond_info->rx_hashtbl = NULL;
 850         bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
 851
 852         _unlock_rx_hashtbl(bond);
 853 }
 854
 855 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 856 {
 857         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 858         u32 curr_index;
 859
 860         _lock_rx_hashtbl(bond);
 861
 862         curr_index = bond_info->rx_hashtbl_head;
 863         while (curr_index != RLB_NULL_INDEX) {
 864                 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 865                 u32 next_index = bond_info->rx_hashtbl[curr_index].next;
 866                 u32 prev_index = bond_info->rx_hashtbl[curr_index].prev;
 867
 868                 if (curr->tag && (curr->vlan_id == vlan_id)) {
 869                         if (curr_index == bond_info->rx_hashtbl_head) {
 870                                 bond_info->rx_hashtbl_head = next_index;
 871                         }
 872                         if (prev_index != RLB_NULL_INDEX) {
 873                                 bond_info->rx_hashtbl[prev_index].next = next_index;
 874                         }
 875                         if (next_index != RLB_NULL_INDEX) {
 876                                 bond_info->rx_hashtbl[next_index].prev = prev_index;
 877                         }
 878
 879                         rlb_init_table_entry(curr);
 880                 }
 881
 882                 curr_index = next_index;
 883         }
 884
 885         _unlock_rx_hashtbl(bond);
 886 }
 887
 888 /*********************** tlb/rlb shared functions *********************/
 889
 890 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
 891 {
 892         struct bonding *bond = bond_get_bond_by_slave(slave);
 893         struct learning_pkt pkt;
 894         int size = sizeof(struct learning_pkt);
 895         int i;
 896
 897         memset(&pkt, 0, size);
 898         memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
 899         memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
 900         pkt.type = __constant_htons(ETH_P_LOOP);
 901
 902         for (i = 0; i < MAX_LP_BURST; i++) {
 903                 struct sk_buff *skb;
 904                 char *data;
 905
 906                 skb = dev_alloc_skb(size);
 907                 if (!skb) {
 908                         return;
 909                 }
 910
 911                 data = skb_put(skb, size);
 912                 memcpy(data, &pkt, size);
 913
 914                 skb->mac.raw = data;
 915                 skb->nh.raw = data + ETH_HLEN;
 916                 skb->protocol = pkt.type;
 917                 skb->priority = TC_PRIO_CONTROL;
 918                 skb->dev = slave->dev;
 919
 920                 if (!list_empty(&bond->vlan_list)) {
 921                         struct vlan_entry *vlan;
 922
 923                         vlan = bond_next_vlan(bond,
 924                                               bond->alb_info.current_alb_vlan);
 925
 926                         bond->alb_info.current_alb_vlan = vlan;
 927                         if (!vlan) {
 928                                 kfree_skb(skb);
 929                                 continue;
 930                         }
 931
 932                         skb = vlan_put_tag(skb, vlan->vlan_id);
 933                         if (!skb) {
 934                                 printk(KERN_ERR DRV_NAME
 935                                        ": %s: Error: failed to insert VLAN tag\n",
 936                                        bond->dev->name);
 937                                 continue;
 938                         }
 939                 }
 940
 941                 dev_queue_xmit(skb);
 942         }
 943 }
 944
 945 /* hw is a boolean parameter that determines whether we should try and
 946  * set the hw address of the device as well as the hw address of the
 947  * net_device
 948  */
 949 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[], int hw)
 950 {
 951         struct net_device *dev = slave->dev;
 952         struct sockaddr s_addr;
 953
 954         if (!hw) {
 955                 memcpy(dev->dev_addr, addr, dev->addr_len);
 956                 return 0;
 957         }
 958
 959         /* for rlb each slave must have a unique hw mac addresses so that */
 960         /* each slave will receive packets destined to a different mac */
 961         memcpy(s_addr.sa_data, addr, dev->addr_len);
 962         s_addr.sa_family = dev->type;
 963         if (dev_set_mac_address(dev, &s_addr)) {
 964                 printk(KERN_ERR DRV_NAME
 965                        ": %s: Error: dev_set_mac_address of dev %s failed! ALB "
 966                        "mode requires that the base driver support setting "
 967                        "the hw address also when the network device's "
 968                        "interface is open\n",
 969                        dev->master->name, dev->name);
 970                 return -EOPNOTSUPP;
 971         }
 972         return 0;
 973 }
 974
 975 /* Caller must hold bond lock for write or curr_slave_lock for write*/
 976 static void alb_swap_mac_addr(struct bonding *bond, struct slave *slave1, struct slave *slave2)
 977 {
 978         struct slave *disabled_slave = NULL;
 979         u8 tmp_mac_addr[ETH_ALEN];
 980         int slaves_state_differ;
 981
 982         slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
 983
 984         memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
 985         alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr, bond->alb_info.rlb_enabled);
 986         alb_set_slave_mac_addr(slave2, tmp_mac_addr, bond->alb_info.rlb_enabled);
 987
 988         /* fasten the change in the switch */
 989         if (SLAVE_IS_OK(slave1)) {
 990                 alb_send_learning_packets(slave1, slave1->dev->dev_addr);
 991                 if (bond->alb_info.rlb_enabled) {
 992                         /* inform the clients that the mac address
 993                          * has changed
 994                          */
 995                         rlb_req_update_slave_clients(bond, slave1);
 996                 }
 997         } else {
 998                 disabled_slave = slave1;
 999         }
1000
1001         if (SLAVE_IS_OK(slave2)) {
1002                 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
1003                 if (bond->alb_info.rlb_enabled) {
1004                         /* inform the clients that the mac address
1005                          * has changed
1006                          */
1007                         rlb_req_update_slave_clients(bond, slave2);
1008                 }
1009         } else {
1010                 disabled_slave = slave2;
1011         }
1012
1013         if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1014                 /* A disabled slave was assigned an active mac addr */
1015                 rlb_teach_disabled_mac_on_primary(bond,
1016                                                   disabled_slave->dev->dev_addr);
1017         }
1018 }
1019
1020 /**
1021  * alb_change_hw_addr_on_detach
1022  * @bond: bonding we're working on
1023  * @slave: the slave that was just detached
1024  *
1025  * We assume that @slave was already detached from the slave list.
1026  *
1027  * If @slave's permanent hw address is different both from its current
1028  * address and from @bond's address, then somewhere in the bond there's
1029  * a slave that has @slave's permanet address as its current address.
1030  * We'll make sure that that slave no longer uses @slave's permanent address.
1031  *
1032  * Caller must hold bond lock
1033  */
1034 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1035 {
1036         int perm_curr_diff;
1037         int perm_bond_diff;
1038
1039         perm_curr_diff = memcmp(slave->perm_hwaddr,
1040                                 slave->dev->dev_addr,
1041                                 ETH_ALEN);
1042         perm_bond_diff = memcmp(slave->perm_hwaddr,
1043                                 bond->dev->dev_addr,
1044                                 ETH_ALEN);
1045
1046         if (perm_curr_diff && perm_bond_diff) {
1047                 struct slave *tmp_slave;
1048                 int i, found = 0;
1049
1050                 bond_for_each_slave(bond, tmp_slave, i) {
1051                         if (!memcmp(slave->perm_hwaddr,
1052                                     tmp_slave->dev->dev_addr,
1053                                     ETH_ALEN)) {
1054                                 found = 1;
1055                                 break;
1056                         }
1057                 }
1058
1059                 if (found) {
1060                         alb_swap_mac_addr(bond, slave, tmp_slave);
1061                 }
1062         }
1063 }
1064
1065 /**
1066  * alb_handle_addr_collision_on_attach
1067  * @bond: bonding we're working on
1068  * @slave: the slave that was just attached
1069  *
1070  * checks uniqueness of slave's mac address and handles the case the
1071  * new slave uses the bonds mac address.
1072  *
1073  * If the permanent hw address of @slave is @bond's hw address, we need to
1074  * find a different hw address to give @slave, that isn't in use by any other
1075  * slave in the bond. This address must be, of course, one of the premanent
1076  * addresses of the other slaves.
1077  *
1078  * We go over the slave list, and for each slave there we compare its
1079  * permanent hw address with the current address of all the other slaves.
1080  * If no match was found, then we've found a slave with a permanent address
1081  * that isn't used by any other slave in the bond, so we can assign it to
1082  * @slave.
1083  *
1084  * assumption: this function is called before @slave is attached to the
1085  *             bond slave list.
1086  *
1087  * caller must hold the bond lock for write since the mac addresses are compared
1088  * and may be swapped.
1089  */
1090 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1091 {
1092         struct slave *tmp_slave1, *tmp_slave2, *free_mac_slave;
1093         struct slave *has_bond_addr = bond->curr_active_slave;
1094         int i, j, found = 0;
1095
1096         if (bond->slave_cnt == 0) {
1097                 /* this is the first slave */
1098                 return 0;
1099         }
1100
1101         /* if slave's mac address differs from bond's mac address
1102          * check uniqueness of slave's mac address against the other
1103          * slaves in the bond.
1104          */
1105         if (memcmp(slave->perm_hwaddr, bond->dev->dev_addr, ETH_ALEN)) {
1106                 bond_for_each_slave(bond, tmp_slave1, i) {
1107                         if (!memcmp(tmp_slave1->dev->dev_addr, slave->dev->dev_addr,
1108                                     ETH_ALEN)) {
1109                                 found = 1;
1110                                 break;
1111                         }
1112                 }
1113
1114                 if (!found)
1115                         return 0;
1116
1117                 /* Try setting slave mac to bond address and fall-through
1118                    to code handling that situation below... */
1119                 alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1120                                        bond->alb_info.rlb_enabled);
1121         }
1122
1123         /* The slave's address is equal to the address of the bond.
1124          * Search for a spare address in the bond for this slave.
1125          */
1126         free_mac_slave = NULL;
1127
1128         bond_for_each_slave(bond, tmp_slave1, i) {
1129                 found = 0;
1130                 bond_for_each_slave(bond, tmp_slave2, j) {
1131                         if (!memcmp(tmp_slave1->perm_hwaddr,
1132                                     tmp_slave2->dev->dev_addr,
1133                                     ETH_ALEN)) {
1134                                 found = 1;
1135                                 break;
1136                         }
1137                 }
1138
1139                 if (!found) {
1140                         /* no slave has tmp_slave1's perm addr
1141                          * as its curr addr
1142                          */
1143                         free_mac_slave = tmp_slave1;
1144                         break;
1145                 }
1146
1147                 if (!has_bond_addr) {
1148                         if (!memcmp(tmp_slave1->dev->dev_addr,
1149                                     bond->dev->dev_addr,
1150                                     ETH_ALEN)) {
1151
1152                                 has_bond_addr = tmp_slave1;
1153                         }
1154                 }
1155         }
1156
1157         if (free_mac_slave) {
1158                 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1159                                        bond->alb_info.rlb_enabled);
1160
1161                 printk(KERN_WARNING DRV_NAME
1162                        ": %s: Warning: the hw address of slave %s is in use by "
1163                        "the bond; giving it the hw address of %s\n",
1164                        bond->dev->name, slave->dev->name, free_mac_slave->dev->name);
1165
1166         } else if (has_bond_addr) {
1167                 printk(KERN_ERR DRV_NAME
1168                        ": %s: Error: the hw address of slave %s is in use by the "
1169                        "bond; couldn't find a slave with a free hw address to "
1170                        "give it (this should not have happened)\n",
1171                        bond->dev->name, slave->dev->name);
1172                 return -EFAULT;
1173         }
1174
1175         return 0;
1176 }
1177
1178 /**
1179  * alb_set_mac_address
1180  * @bond:
1181  * @addr:
1182  *
1183  * In TLB mode all slaves are configured to the bond's hw address, but set
1184  * their dev_addr field to different addresses (based on their permanent hw
1185  * addresses).
1186  *
1187  * For each slave, this function sets the interface to the new address and then
1188  * changes its dev_addr field to its previous value.
1189  *
1190  * Unwinding assumes bond's mac address has not yet changed.
1191  */
1192 static int alb_set_mac_address(struct bonding *bond, void *addr)
1193 {
1194         struct sockaddr sa;
1195         struct slave *slave, *stop_at;
1196         char tmp_addr[ETH_ALEN];
1197         int res;
1198         int i;
1199
1200         if (bond->alb_info.rlb_enabled) {
1201                 return 0;
1202         }
1203
1204         bond_for_each_slave(bond, slave, i) {
1205                 if (slave->dev->set_mac_address == NULL) {
1206                         res = -EOPNOTSUPP;
1207                         goto unwind;
1208                 }
1209
1210                 /* save net_device's current hw address */
1211                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1212
1213                 res = dev_set_mac_address(slave->dev, addr);
1214
1215                 /* restore net_device's hw address */
1216                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1217
1218                 if (res) {
1219                         goto unwind;
1220                 }
1221         }
1222
1223         return 0;
1224
1225 unwind:
1226         memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1227         sa.sa_family = bond->dev->type;
1228
1229         /* unwind from head to the slave that failed */
1230         stop_at = slave;
1231         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
1232                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1233                 dev_set_mac_address(slave->dev, &sa);
1234                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1235         }
1236
1237         return res;
1238 }
1239
1240 /************************ exported alb funcions ************************/
1241
1242 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1243 {
1244         int res;
1245
1246         res = tlb_initialize(bond);
1247         if (res) {
1248                 return res;
1249         }
1250
1251         if (rlb_enabled) {
1252                 bond->alb_info.rlb_enabled = 1;
1253                 /* initialize rlb */
1254                 res = rlb_initialize(bond);
1255                 if (res) {
1256                         tlb_deinitialize(bond);
1257                         return res;
1258                 }
1259         } else {
1260                 bond->alb_info.rlb_enabled = 0;
1261         }
1262
1263         return 0;
1264 }
1265
1266 void bond_alb_deinitialize(struct bonding *bond)
1267 {
1268         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1269
1270         tlb_deinitialize(bond);
1271
1272         if (bond_info->rlb_enabled) {
1273                 rlb_deinitialize(bond);
1274         }
1275 }
1276
1277 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1278 {
1279         struct bonding *bond = bond_dev->priv;
1280         struct ethhdr *eth_data;
1281         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1282         struct slave *tx_slave = NULL;
1283         static u32 ip_bcast = 0xffffffff;
1284         int hash_size = 0;
1285         int do_tx_balance = 1;
1286         u32 hash_index = 0;
1287         u8 *hash_start = NULL;
1288         int res = 1;
1289
1290         skb->mac.raw = (unsigned char *)skb->data;
1291         eth_data = eth_hdr(skb);
1292
1293         /* make sure that the curr_active_slave and the slaves list do
1294          * not change during tx
1295          */
1296         read_lock(&bond->lock);
1297         read_lock(&bond->curr_slave_lock);
1298
1299         if (!BOND_IS_OK(bond)) {
1300                 goto out;
1301         }
1302
1303         switch (ntohs(skb->protocol)) {
1304         case ETH_P_IP:
1305                 if ((memcmp(eth_data->h_dest, mac_bcast, ETH_ALEN) == 0) ||
1306                     (skb->nh.iph->daddr == ip_bcast) ||
1307                     (skb->nh.iph->protocol == IPPROTO_IGMP)) {
1308                         do_tx_balance = 0;
1309                         break;
1310                 }
1311                 hash_start = (char*)&(skb->nh.iph->daddr);
1312                 hash_size = sizeof(skb->nh.iph->daddr);
1313                 break;
1314         case ETH_P_IPV6:
1315                 if (memcmp(eth_data->h_dest, mac_bcast, ETH_ALEN) == 0) {
1316                         do_tx_balance = 0;
1317                         break;
1318                 }
1319
1320                 hash_start = (char*)&(skb->nh.ipv6h->daddr);
1321                 hash_size = sizeof(skb->nh.ipv6h->daddr);
1322                 break;
1323         case ETH_P_IPX:
1324                 if (ipx_hdr(skb)->ipx_checksum !=
1325                     __constant_htons(IPX_NO_CHECKSUM)) {
1326                         /* something is wrong with this packet */
1327                         do_tx_balance = 0;
1328                         break;
1329                 }
1330
1331                 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1332                         /* The only protocol worth balancing in
1333                          * this family since it has an "ARP" like
1334                          * mechanism
1335                          */
1336                         do_tx_balance = 0;
1337                         break;
1338                 }
1339
1340                 hash_start = (char*)eth_data->h_dest;
1341                 hash_size = ETH_ALEN;
1342                 break;
1343         case ETH_P_ARP:
1344                 do_tx_balance = 0;
1345                 if (bond_info->rlb_enabled) {
1346                         tx_slave = rlb_arp_xmit(skb, bond);
1347                 }
1348                 break;
1349         default:
1350                 do_tx_balance = 0;
1351                 break;
1352         }
1353
1354         if (do_tx_balance) {
1355                 hash_index = _simple_hash(hash_start, hash_size);
1356                 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1357         }
1358
1359         if (!tx_slave) {
1360                 /* unbalanced or unassigned, send through primary */
1361                 tx_slave = bond->curr_active_slave;
1362                 bond_info->unbalanced_load += skb->len;
1363         }
1364
1365         if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1366                 if (tx_slave != bond->curr_active_slave) {
1367                         memcpy(eth_data->h_source,
1368                                tx_slave->dev->dev_addr,
1369                                ETH_ALEN);
1370                 }
1371
1372                 res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1373         } else {
1374                 if (tx_slave) {
1375                         tlb_clear_slave(bond, tx_slave, 0);
1376                 }
1377         }
1378
1379 out:
1380         if (res) {
1381                 /* no suitable interface, frame not sent */
1382                 dev_kfree_skb(skb);
1383         }
1384         read_unlock(&bond->curr_slave_lock);
1385         read_unlock(&bond->lock);
1386         return 0;
1387 }
1388
1389 void bond_alb_monitor(struct bonding *bond)
1390 {
1391         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1392         struct slave *slave;
1393         int i;
1394
1395         read_lock(&bond->lock);
1396
1397         if (bond->kill_timers) {
1398                 goto out;
1399         }
1400
1401         if (bond->slave_cnt == 0) {
1402                 bond_info->tx_rebalance_counter = 0;
1403                 bond_info->lp_counter = 0;
1404                 goto re_arm;
1405         }
1406
1407         bond_info->tx_rebalance_counter++;
1408         bond_info->lp_counter++;
1409
1410         /* send learning packets */
1411         if (bond_info->lp_counter >= BOND_ALB_LP_TICKS) {
1412                 /* change of curr_active_slave involves swapping of mac addresses.
1413                  * in order to avoid this swapping from happening while
1414                  * sending the learning packets, the curr_slave_lock must be held for
1415                  * read.
1416                  */
1417                 read_lock(&bond->curr_slave_lock);
1418
1419                 bond_for_each_slave(bond, slave, i) {
1420                         alb_send_learning_packets(slave, slave->dev->dev_addr);
1421                 }
1422
1423                 read_unlock(&bond->curr_slave_lock);
1424
1425                 bond_info->lp_counter = 0;
1426         }
1427
1428         /* rebalance tx traffic */
1429         if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1430
1431                 read_lock(&bond->curr_slave_lock);
1432
1433                 bond_for_each_slave(bond, slave, i) {
1434                         tlb_clear_slave(bond, slave, 1);
1435                         if (slave == bond->curr_active_slave) {
1436                                 SLAVE_TLB_INFO(slave).load =
1437                                         bond_info->unbalanced_load /
1438                                                 BOND_TLB_REBALANCE_INTERVAL;
1439                                 bond_info->unbalanced_load = 0;
1440                         }
1441                 }
1442
1443                 read_unlock(&bond->curr_slave_lock);
1444
1445                 bond_info->tx_rebalance_counter = 0;
1446         }
1447
1448         /* handle rlb stuff */
1449         if (bond_info->rlb_enabled) {
1450                 /* the following code changes the promiscuity of the
1451                  * the curr_active_slave. It needs to be locked with a
1452                  * write lock to protect from other code that also
1453                  * sets the promiscuity.
1454                  */
1455                 write_lock(&bond->curr_slave_lock);
1456
1457                 if (bond_info->primary_is_promisc &&
1458                     (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1459
1460                         bond_info->rlb_promisc_timeout_counter = 0;
1461
1462                         /* If the primary was set to promiscuous mode
1463                          * because a slave was disabled then
1464                          * it can now leave promiscuous mode.
1465                          */
1466                         dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1467                         bond_info->primary_is_promisc = 0;
1468                 }
1469
1470                 write_unlock(&bond->curr_slave_lock);
1471
1472                 if (bond_info->rlb_rebalance) {
1473                         bond_info->rlb_rebalance = 0;
1474                         rlb_rebalance(bond);
1475                 }
1476
1477                 /* check if clients need updating */
1478                 if (bond_info->rx_ntt) {
1479                         if (bond_info->rlb_update_delay_counter) {
1480                                 --bond_info->rlb_update_delay_counter;
1481                         } else {
1482                                 rlb_update_rx_clients(bond);
1483                                 if (bond_info->rlb_update_retry_counter) {
1484                                         --bond_info->rlb_update_retry_counter;
1485                                 } else {
1486                                         bond_info->rx_ntt = 0;
1487                                 }
1488                         }
1489                 }
1490         }
1491
1492 re_arm:
1493         mod_timer(&(bond_info->alb_timer), jiffies + alb_delta_in_ticks);
1494 out:
1495         read_unlock(&bond->lock);
1496 }
1497
1498 /* assumption: called before the slave is attached to the bond
1499  * and not locked by the bond lock
1500  */
1501 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1502 {
1503         int res;
1504
1505         res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1506                                      bond->alb_info.rlb_enabled);
1507         if (res) {
1508                 return res;
1509         }
1510
1511         /* caller must hold the bond lock for write since the mac addresses
1512          * are compared and may be swapped.
1513          */
1514         write_lock_bh(&bond->lock);
1515
1516         res = alb_handle_addr_collision_on_attach(bond, slave);
1517
1518         write_unlock_bh(&bond->lock);
1519
1520         if (res) {
1521                 return res;
1522         }
1523
1524         tlb_init_slave(slave);
1525
1526         /* order a rebalance ASAP */
1527         bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1528
1529         if (bond->alb_info.rlb_enabled) {
1530                 bond->alb_info.rlb_rebalance = 1;
1531         }
1532
1533         return 0;
1534 }
1535
1536 /* Caller must hold bond lock for write */
1537 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1538 {
1539         if (bond->slave_cnt > 1) {
1540                 alb_change_hw_addr_on_detach(bond, slave);
1541         }
1542
1543         tlb_clear_slave(bond, slave, 0);
1544
1545         if (bond->alb_info.rlb_enabled) {
1546                 bond->alb_info.next_rx_slave = NULL;
1547                 rlb_clear_slave(bond, slave);
1548         }
1549 }
1550
1551 /* Caller must hold bond lock for read */
1552 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1553 {
1554         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1555
1556         if (link == BOND_LINK_DOWN) {
1557                 tlb_clear_slave(bond, slave, 0);
1558                 if (bond->alb_info.rlb_enabled) {
1559                         rlb_clear_slave(bond, slave);
1560                 }
1561         } else if (link == BOND_LINK_UP) {
1562                 /* order a rebalance ASAP */
1563                 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1564                 if (bond->alb_info.rlb_enabled) {
1565                         bond->alb_info.rlb_rebalance = 1;
1566                         /* If the updelay module parameter is smaller than the
1567                          * forwarding delay of the switch the rebalance will
1568                          * not work because the rebalance arp replies will
1569                          * not be forwarded to the clients..
1570                          */
1571                 }
1572         }
1573 }
1574
1575 /**
1576  * bond_alb_handle_active_change - assign new curr_active_slave
1577  * @bond: our bonding struct
1578  * @new_slave: new slave to assign
1579  *
1580  * Set the bond->curr_active_slave to @new_slave and handle
1581  * mac address swapping and promiscuity changes as needed.
1582  *
1583  * Caller must hold bond curr_slave_lock for write (or bond lock for write)
1584  */
1585 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1586 {
1587         struct slave *swap_slave;
1588         int i;
1589
1590         if (bond->curr_active_slave == new_slave) {
1591                 return;
1592         }
1593
1594         if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1595                 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1596                 bond->alb_info.primary_is_promisc = 0;
1597                 bond->alb_info.rlb_promisc_timeout_counter = 0;
1598         }
1599
1600         swap_slave = bond->curr_active_slave;
1601         bond->curr_active_slave = new_slave;
1602
1603         if (!new_slave || (bond->slave_cnt == 0)) {
1604                 return;
1605         }
1606
1607         /* set the new curr_active_slave to the bonds mac address
1608          * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1609          */
1610         if (!swap_slave) {
1611                 struct slave *tmp_slave;
1612                 /* find slave that is holding the bond's mac address */
1613                 bond_for_each_slave(bond, tmp_slave, i) {
1614                         if (!memcmp(tmp_slave->dev->dev_addr,
1615                                     bond->dev->dev_addr, ETH_ALEN)) {
1616                                 swap_slave = tmp_slave;
1617                                 break;
1618                         }
1619                 }
1620         }
1621
1622         /* curr_active_slave must be set before calling alb_swap_mac_addr */
1623         if (swap_slave) {
1624                 /* swap mac address */
1625                 alb_swap_mac_addr(bond, swap_slave, new_slave);
1626         } else {
1627                 /* set the new_slave to the bond mac address */
1628                 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1629                                        bond->alb_info.rlb_enabled);
1630                 /* fasten bond mac on new current slave */
1631                 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1632         }
1633 }
1634
1635 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1636 {
1637         struct bonding *bond = bond_dev->priv;
1638         struct sockaddr *sa = addr;
1639         struct slave *slave, *swap_slave;
1640         int res;
1641         int i;
1642
1643         if (!is_valid_ether_addr(sa->sa_data)) {
1644                 return -EADDRNOTAVAIL;
1645         }
1646
1647         res = alb_set_mac_address(bond, addr);
1648         if (res) {
1649                 return res;
1650         }
1651
1652         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1653
1654         /* If there is no curr_active_slave there is nothing else to do.
1655          * Otherwise we'll need to pass the new address to it and handle
1656          * duplications.
1657          */
1658         if (!bond->curr_active_slave) {
1659                 return 0;
1660         }
1661
1662         swap_slave = NULL;
1663
1664         bond_for_each_slave(bond, slave, i) {
1665                 if (!memcmp(slave->dev->dev_addr, bond_dev->dev_addr, ETH_ALEN)) {
1666                         swap_slave = slave;
1667                         break;
1668                 }
1669         }
1670
1671         if (swap_slave) {
1672                 alb_swap_mac_addr(bond, swap_slave, bond->curr_active_slave);
1673         } else {
1674                 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr,
1675                                        bond->alb_info.rlb_enabled);
1676
1677                 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1678                 if (bond->alb_info.rlb_enabled) {
1679                         /* inform clients mac address has changed */
1680                         rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1681                 }
1682         }
1683
1684         return 0;
1685 }
1686
1687 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1688 {
1689         if (bond->alb_info.current_alb_vlan &&
1690             (bond->alb_info.current_alb_vlan->vlan_id == vlan_id)) {
1691                 bond->alb_info.current_alb_vlan = NULL;
1692         }
1693
1694         if (bond->alb_info.rlb_enabled) {
1695                 rlb_clear_vlan(bond, vlan_id);
1696         }
1697 }
1698