drivers/net/bonding/bond_alb.c

   1 /*
   2  * Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms of the GNU General Public License as published by the
   6  * Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful, but
  10  * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  11  * or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License along
  15  * with this program; if not, write to the Free Software Foundation, Inc.,
  16  * 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
  17  *
  18  * The full GNU General Public License is included in this distribution in the
  19  * file called LICENSE.
  20  *
  21  */
  22
  23 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  24
  25 #include <linux/skbuff.h>
  26 #include <linux/netdevice.h>
  27 #include <linux/etherdevice.h>
  28 #include <linux/pkt_sched.h>
  29 #include <linux/spinlock.h>
  30 #include <linux/slab.h>
  31 #include <linux/timer.h>
  32 #include <linux/ip.h>
  33 #include <linux/ipv6.h>
  34 #include <linux/if_arp.h>
  35 #include <linux/if_ether.h>
  36 #include <linux/if_bonding.h>
  37 #include <linux/if_vlan.h>
  38 #include <linux/in.h>
  39 #include <net/ipx.h>
  40 #include <net/arp.h>
  41 #include <net/ipv6.h>
  42 #include <asm/byteorder.h>
  43 #include "bonding.h"
  44 #include "bond_alb.h"
  45
  46
  47
  48 #ifndef __long_aligned
  49 #define __long_aligned __attribute__((aligned((sizeof(long)))))
  50 #endif
  51 static const u8 mac_bcast[ETH_ALEN] __long_aligned = {
  52         0xff, 0xff, 0xff, 0xff, 0xff, 0xff
  53 };
  54 static const u8 mac_v6_allmcast[ETH_ALEN] __long_aligned = {
  55         0x33, 0x33, 0x00, 0x00, 0x00, 0x01
  56 };
  57 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
  58
  59 #pragma pack(1)
  60 struct learning_pkt {
  61         u8 mac_dst[ETH_ALEN];
  62         u8 mac_src[ETH_ALEN];
  63         __be16 type;
  64         u8 padding[ETH_ZLEN - ETH_HLEN];
  65 };
  66
  67 struct arp_pkt {
  68         __be16  hw_addr_space;
  69         __be16  prot_addr_space;
  70         u8      hw_addr_len;
  71         u8      prot_addr_len;
  72         __be16  op_code;
  73         u8      mac_src[ETH_ALEN];      /* sender hardware address */
  74         __be32  ip_src;                 /* sender IP address */
  75         u8      mac_dst[ETH_ALEN];      /* target hardware address */
  76         __be32  ip_dst;                 /* target IP address */
  77 };
  78 #pragma pack()
  79
  80 static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb)
  81 {
  82         return (struct arp_pkt *)skb_network_header(skb);
  83 }
  84
  85 /* Forward declaration */
  86 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[]);
  87
  88 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
  89 {
  90         int i;
  91         u8 hash = 0;
  92
  93         for (i = 0; i < hash_size; i++) {
  94                 hash ^= hash_start[i];
  95         }
  96
  97         return hash;
  98 }
  99
 100 /*********************** tlb specific functions ***************************/
 101
 102 static inline void _lock_tx_hashtbl(struct bonding *bond)
 103 {
 104         spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 105 }
 106
 107 static inline void _unlock_tx_hashtbl(struct bonding *bond)
 108 {
 109         spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 110 }
 111
 112 /* Caller must hold tx_hashtbl lock */
 113 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
 114 {
 115         if (save_load) {
 116                 entry->load_history = 1 + entry->tx_bytes /
 117                                       BOND_TLB_REBALANCE_INTERVAL;
 118                 entry->tx_bytes = 0;
 119         }
 120
 121         entry->tx_slave = NULL;
 122         entry->next = TLB_NULL_INDEX;
 123         entry->prev = TLB_NULL_INDEX;
 124 }
 125
 126 static inline void tlb_init_slave(struct slave *slave)
 127 {
 128         SLAVE_TLB_INFO(slave).load = 0;
 129         SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
 130 }
 131
 132 /* Caller must hold bond lock for read */
 133 static void tlb_clear_slave(struct bonding *bond, struct slave *slave, int save_load)
 134 {
 135         struct tlb_client_info *tx_hash_table;
 136         u32 index;
 137
 138         _lock_tx_hashtbl(bond);
 139
 140         /* clear slave from tx_hashtbl */
 141         tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
 142
 143         /* skip this if we've already freed the tx hash table */
 144         if (tx_hash_table) {
 145                 index = SLAVE_TLB_INFO(slave).head;
 146                 while (index != TLB_NULL_INDEX) {
 147                         u32 next_index = tx_hash_table[index].next;
 148                         tlb_init_table_entry(&tx_hash_table[index], save_load);
 149                         index = next_index;
 150                 }
 151         }
 152
 153         tlb_init_slave(slave);
 154
 155         _unlock_tx_hashtbl(bond);
 156 }
 157
 158 /* Must be called before starting the monitor timer */
 159 static int tlb_initialize(struct bonding *bond)
 160 {
 161         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 162         int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 163         struct tlb_client_info *new_hashtbl;
 164         int i;
 165
 166         spin_lock_init(&(bond_info->tx_hashtbl_lock));
 167
 168         new_hashtbl = kzalloc(size, GFP_KERNEL);
 169         if (!new_hashtbl) {
 170                 pr_err("%s: Error: Failed to allocate TLB hash table\n",
 171                        bond->dev->name);
 172                 return -1;
 173         }
 174         _lock_tx_hashtbl(bond);
 175
 176         bond_info->tx_hashtbl = new_hashtbl;
 177
 178         for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
 179                 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
 180         }
 181
 182         _unlock_tx_hashtbl(bond);
 183
 184         return 0;
 185 }
 186
 187 /* Must be called only after all slaves have been released */
 188 static void tlb_deinitialize(struct bonding *bond)
 189 {
 190         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 191
 192         _lock_tx_hashtbl(bond);
 193
 194         kfree(bond_info->tx_hashtbl);
 195         bond_info->tx_hashtbl = NULL;
 196
 197         _unlock_tx_hashtbl(bond);
 198 }
 199
 200 static long long compute_gap(struct slave *slave)
 201 {
 202         return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
 203                (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
 204 }
 205
 206 /* Caller must hold bond lock for read */
 207 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
 208 {
 209         struct slave *slave, *least_loaded;
 210         long long max_gap;
 211         int i;
 212
 213         least_loaded = NULL;
 214         max_gap = LLONG_MIN;
 215
 216         /* Find the slave with the largest gap */
 217         bond_for_each_slave(bond, slave, i) {
 218                 if (SLAVE_IS_OK(slave)) {
 219                         long long gap = compute_gap(slave);
 220
 221                         if (max_gap < gap) {
 222                                 least_loaded = slave;
 223                                 max_gap = gap;
 224                         }
 225                 }
 226         }
 227
 228         return least_loaded;
 229 }
 230
 231 /* Caller must hold bond lock for read */
 232 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index, u32 skb_len)
 233 {
 234         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 235         struct tlb_client_info *hash_table;
 236         struct slave *assigned_slave;
 237
 238         _lock_tx_hashtbl(bond);
 239
 240         hash_table = bond_info->tx_hashtbl;
 241         assigned_slave = hash_table[hash_index].tx_slave;
 242         if (!assigned_slave) {
 243                 assigned_slave = tlb_get_least_loaded_slave(bond);
 244
 245                 if (assigned_slave) {
 246                         struct tlb_slave_info *slave_info =
 247                                 &(SLAVE_TLB_INFO(assigned_slave));
 248                         u32 next_index = slave_info->head;
 249
 250                         hash_table[hash_index].tx_slave = assigned_slave;
 251                         hash_table[hash_index].next = next_index;
 252                         hash_table[hash_index].prev = TLB_NULL_INDEX;
 253
 254                         if (next_index != TLB_NULL_INDEX) {
 255                                 hash_table[next_index].prev = hash_index;
 256                         }
 257
 258                         slave_info->head = hash_index;
 259                         slave_info->load +=
 260                                 hash_table[hash_index].load_history;
 261                 }
 262         }
 263
 264         if (assigned_slave) {
 265                 hash_table[hash_index].tx_bytes += skb_len;
 266         }
 267
 268         _unlock_tx_hashtbl(bond);
 269
 270         return assigned_slave;
 271 }
 272
 273 /*********************** rlb specific functions ***************************/
 274 static inline void _lock_rx_hashtbl(struct bonding *bond)
 275 {
 276         spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 277 }
 278
 279 static inline void _unlock_rx_hashtbl(struct bonding *bond)
 280 {
 281         spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 282 }
 283
 284 /* when an ARP REPLY is received from a client update its info
 285  * in the rx_hashtbl
 286  */
 287 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 288 {
 289         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 290         struct rlb_client_info *client_info;
 291         u32 hash_index;
 292
 293         _lock_rx_hashtbl(bond);
 294
 295         hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
 296         client_info = &(bond_info->rx_hashtbl[hash_index]);
 297
 298         if ((client_info->assigned) &&
 299             (client_info->ip_src == arp->ip_dst) &&
 300             (client_info->ip_dst == arp->ip_src) &&
 301             (compare_ether_addr_64bits(client_info->mac_dst, arp->mac_src))) {
 302                 /* update the clients MAC address */
 303                 memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
 304                 client_info->ntt = 1;
 305                 bond_info->rx_ntt = 1;
 306         }
 307
 308         _unlock_rx_hashtbl(bond);
 309 }
 310
 311 static int rlb_arp_recv(struct sk_buff *skb, struct net_device *bond_dev, struct packet_type *ptype, struct net_device *orig_dev)
 312 {
 313         struct bonding *bond;
 314         struct arp_pkt *arp = (struct arp_pkt *)skb->data;
 315         int res = NET_RX_DROP;
 316
 317         while (bond_dev->priv_flags & IFF_802_1Q_VLAN)
 318                 bond_dev = vlan_dev_real_dev(bond_dev);
 319
 320         if (!(bond_dev->priv_flags & IFF_BONDING) ||
 321             !(bond_dev->flags & IFF_MASTER))
 322                 goto out;
 323
 324         if (!arp) {
 325                 pr_debug("Packet has no ARP data\n");
 326                 goto out;
 327         }
 328
 329         skb = skb_share_check(skb, GFP_ATOMIC);
 330         if (!skb)
 331                 goto out;
 332
 333         if (!pskb_may_pull(skb, arp_hdr_len(bond_dev)))
 334                 goto out;
 335
 336         if (skb->len < sizeof(struct arp_pkt)) {
 337                 pr_debug("Packet is too small to be an ARP\n");
 338                 goto out;
 339         }
 340
 341         if (arp->op_code == htons(ARPOP_REPLY)) {
 342                 /* update rx hash table for this ARP */
 343                 bond = netdev_priv(bond_dev);
 344                 rlb_update_entry_from_arp(bond, arp);
 345                 pr_debug("Server received an ARP Reply from client\n");
 346         }
 347
 348         res = NET_RX_SUCCESS;
 349
 350 out:
 351         dev_kfree_skb(skb);
 352
 353         return res;
 354 }
 355
 356 /* Caller must hold bond lock for read */
 357 static struct slave *rlb_next_rx_slave(struct bonding *bond)
 358 {
 359         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 360         struct slave *rx_slave, *slave, *start_at;
 361         int i = 0;
 362
 363         if (bond_info->next_rx_slave) {
 364                 start_at = bond_info->next_rx_slave;
 365         } else {
 366                 start_at = bond->first_slave;
 367         }
 368
 369         rx_slave = NULL;
 370
 371         bond_for_each_slave_from(bond, slave, i, start_at) {
 372                 if (SLAVE_IS_OK(slave)) {
 373                         if (!rx_slave) {
 374                                 rx_slave = slave;
 375                         } else if (slave->speed > rx_slave->speed) {
 376                                 rx_slave = slave;
 377                         }
 378                 }
 379         }
 380
 381         if (rx_slave) {
 382                 bond_info->next_rx_slave = rx_slave->next;
 383         }
 384
 385         return rx_slave;
 386 }
 387
 388 /* teach the switch the mac of a disabled slave
 389  * on the primary for fault tolerance
 390  *
 391  * Caller must hold bond->curr_slave_lock for write or bond lock for write
 392  */
 393 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
 394 {
 395         if (!bond->curr_active_slave) {
 396                 return;
 397         }
 398
 399         if (!bond->alb_info.primary_is_promisc) {
 400                 if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1))
 401                         bond->alb_info.primary_is_promisc = 1;
 402                 else
 403                         bond->alb_info.primary_is_promisc = 0;
 404         }
 405
 406         bond->alb_info.rlb_promisc_timeout_counter = 0;
 407
 408         alb_send_learning_packets(bond->curr_active_slave, addr);
 409 }
 410
 411 /* slave being removed should not be active at this point
 412  *
 413  * Caller must hold bond lock for read
 414  */
 415 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 416 {
 417         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 418         struct rlb_client_info *rx_hash_table;
 419         u32 index, next_index;
 420
 421         /* clear slave from rx_hashtbl */
 422         _lock_rx_hashtbl(bond);
 423
 424         rx_hash_table = bond_info->rx_hashtbl;
 425         index = bond_info->rx_hashtbl_head;
 426         for (; index != RLB_NULL_INDEX; index = next_index) {
 427                 next_index = rx_hash_table[index].next;
 428                 if (rx_hash_table[index].slave == slave) {
 429                         struct slave *assigned_slave = rlb_next_rx_slave(bond);
 430
 431                         if (assigned_slave) {
 432                                 rx_hash_table[index].slave = assigned_slave;
 433                                 if (compare_ether_addr_64bits(rx_hash_table[index].mac_dst,
 434                                                               mac_bcast)) {
 435                                         bond_info->rx_hashtbl[index].ntt = 1;
 436                                         bond_info->rx_ntt = 1;
 437                                         /* A slave has been removed from the
 438                                          * table because it is either disabled
 439                                          * or being released. We must retry the
 440                                          * update to avoid clients from not
 441                                          * being updated & disconnecting when
 442                                          * there is stress
 443                                          */
 444                                         bond_info->rlb_update_retry_counter =
 445                                                 RLB_UPDATE_RETRY;
 446                                 }
 447                         } else {  /* there is no active slave */
 448                                 rx_hash_table[index].slave = NULL;
 449                         }
 450                 }
 451         }
 452
 453         _unlock_rx_hashtbl(bond);
 454
 455         write_lock_bh(&bond->curr_slave_lock);
 456
 457         if (slave != bond->curr_active_slave) {
 458                 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 459         }
 460
 461         write_unlock_bh(&bond->curr_slave_lock);
 462 }
 463
 464 static void rlb_update_client(struct rlb_client_info *client_info)
 465 {
 466         int i;
 467
 468         if (!client_info->slave) {
 469                 return;
 470         }
 471
 472         for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 473                 struct sk_buff *skb;
 474
 475                 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 476                                  client_info->ip_dst,
 477                                  client_info->slave->dev,
 478                                  client_info->ip_src,
 479                                  client_info->mac_dst,
 480                                  client_info->slave->dev->dev_addr,
 481                                  client_info->mac_dst);
 482                 if (!skb) {
 483                         pr_err("%s: Error: failed to create an ARP packet\n",
 484                                client_info->slave->dev->master->name);
 485                         continue;
 486                 }
 487
 488                 skb->dev = client_info->slave->dev;
 489
 490                 if (client_info->tag) {
 491                         skb = vlan_put_tag(skb, client_info->vlan_id);
 492                         if (!skb) {
 493                                 pr_err("%s: Error: failed to insert VLAN tag\n",
 494                                        client_info->slave->dev->master->name);
 495                                 continue;
 496                         }
 497                 }
 498
 499                 arp_xmit(skb);
 500         }
 501 }
 502
 503 /* sends ARP REPLIES that update the clients that need updating */
 504 static void rlb_update_rx_clients(struct bonding *bond)
 505 {
 506         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 507         struct rlb_client_info *client_info;
 508         u32 hash_index;
 509
 510         _lock_rx_hashtbl(bond);
 511
 512         hash_index = bond_info->rx_hashtbl_head;
 513         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 514                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 515                 if (client_info->ntt) {
 516                         rlb_update_client(client_info);
 517                         if (bond_info->rlb_update_retry_counter == 0) {
 518                                 client_info->ntt = 0;
 519                         }
 520                 }
 521         }
 522
 523         /* do not update the entries again until this counter is zero so that
 524          * not to confuse the clients.
 525          */
 526         bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 527
 528         _unlock_rx_hashtbl(bond);
 529 }
 530
 531 /* The slave was assigned a new mac address - update the clients */
 532 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 533 {
 534         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 535         struct rlb_client_info *client_info;
 536         int ntt = 0;
 537         u32 hash_index;
 538
 539         _lock_rx_hashtbl(bond);
 540
 541         hash_index = bond_info->rx_hashtbl_head;
 542         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 543                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 544
 545                 if ((client_info->slave == slave) &&
 546                     compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
 547                         client_info->ntt = 1;
 548                         ntt = 1;
 549                 }
 550         }
 551
 552         // update the team's flag only after the whole iteration
 553         if (ntt) {
 554                 bond_info->rx_ntt = 1;
 555                 //fasten the change
 556                 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 557         }
 558
 559         _unlock_rx_hashtbl(bond);
 560 }
 561
 562 /* mark all clients using src_ip to be updated */
 563 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
 564 {
 565         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 566         struct rlb_client_info *client_info;
 567         u32 hash_index;
 568
 569         _lock_rx_hashtbl(bond);
 570
 571         hash_index = bond_info->rx_hashtbl_head;
 572         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 573                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 574
 575                 if (!client_info->slave) {
 576                         pr_err("%s: Error: found a client with no channel in the client's hash table\n",
 577                                bond->dev->name);
 578                         continue;
 579                 }
 580                 /*update all clients using this src_ip, that are not assigned
 581                  * to the team's address (curr_active_slave) and have a known
 582                  * unicast mac address.
 583                  */
 584                 if ((client_info->ip_src == src_ip) &&
 585                     compare_ether_addr_64bits(client_info->slave->dev->dev_addr,
 586                            bond->dev->dev_addr) &&
 587                     compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
 588                         client_info->ntt = 1;
 589                         bond_info->rx_ntt = 1;
 590                 }
 591         }
 592
 593         _unlock_rx_hashtbl(bond);
 594 }
 595
 596 /* Caller must hold both bond and ptr locks for read */
 597 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
 598 {
 599         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 600         struct arp_pkt *arp = arp_pkt(skb);
 601         struct slave *assigned_slave;
 602         struct rlb_client_info *client_info;
 603         u32 hash_index = 0;
 604
 605         _lock_rx_hashtbl(bond);
 606
 607         hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
 608         client_info = &(bond_info->rx_hashtbl[hash_index]);
 609
 610         if (client_info->assigned) {
 611                 if ((client_info->ip_src == arp->ip_src) &&
 612                     (client_info->ip_dst == arp->ip_dst)) {
 613                         /* the entry is already assigned to this client */
 614                         if (compare_ether_addr_64bits(arp->mac_dst, mac_bcast)) {
 615                                 /* update mac address from arp */
 616                                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 617                         }
 618
 619                         assigned_slave = client_info->slave;
 620                         if (assigned_slave) {
 621                                 _unlock_rx_hashtbl(bond);
 622                                 return assigned_slave;
 623                         }
 624                 } else {
 625                         /* the entry is already assigned to some other client,
 626                          * move the old client to primary (curr_active_slave) so
 627                          * that the new client can be assigned to this entry.
 628                          */
 629                         if (bond->curr_active_slave &&
 630                             client_info->slave != bond->curr_active_slave) {
 631                                 client_info->slave = bond->curr_active_slave;
 632                                 rlb_update_client(client_info);
 633                         }
 634                 }
 635         }
 636         /* assign a new slave */
 637         assigned_slave = rlb_next_rx_slave(bond);
 638
 639         if (assigned_slave) {
 640                 client_info->ip_src = arp->ip_src;
 641                 client_info->ip_dst = arp->ip_dst;
 642                 /* arp->mac_dst is broadcast for arp reqeusts.
 643                  * will be updated with clients actual unicast mac address
 644                  * upon receiving an arp reply.
 645                  */
 646                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 647                 client_info->slave = assigned_slave;
 648
 649                 if (compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
 650                         client_info->ntt = 1;
 651                         bond->alb_info.rx_ntt = 1;
 652                 } else {
 653                         client_info->ntt = 0;
 654                 }
 655
 656                 if (bond->vlgrp) {
 657                         if (!vlan_get_tag(skb, &client_info->vlan_id))
 658                                 client_info->tag = 1;
 659                 }
 660
 661                 if (!client_info->assigned) {
 662                         u32 prev_tbl_head = bond_info->rx_hashtbl_head;
 663                         bond_info->rx_hashtbl_head = hash_index;
 664                         client_info->next = prev_tbl_head;
 665                         if (prev_tbl_head != RLB_NULL_INDEX) {
 666                                 bond_info->rx_hashtbl[prev_tbl_head].prev =
 667                                         hash_index;
 668                         }
 669                         client_info->assigned = 1;
 670                 }
 671         }
 672
 673         _unlock_rx_hashtbl(bond);
 674
 675         return assigned_slave;
 676 }
 677
 678 /* chooses (and returns) transmit channel for arp reply
 679  * does not choose channel for other arp types since they are
 680  * sent on the curr_active_slave
 681  */
 682 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 683 {
 684         struct arp_pkt *arp = arp_pkt(skb);
 685         struct slave *tx_slave = NULL;
 686
 687         if (arp->op_code == htons(ARPOP_REPLY)) {
 688                 /* the arp must be sent on the selected
 689                 * rx channel
 690                 */
 691                 tx_slave = rlb_choose_channel(skb, bond);
 692                 if (tx_slave) {
 693                         memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
 694                 }
 695                 pr_debug("Server sent ARP Reply packet\n");
 696         } else if (arp->op_code == htons(ARPOP_REQUEST)) {
 697                 /* Create an entry in the rx_hashtbl for this client as a
 698                  * place holder.
 699                  * When the arp reply is received the entry will be updated
 700                  * with the correct unicast address of the client.
 701                  */
 702                 rlb_choose_channel(skb, bond);
 703
 704                 /* The ARP reply packets must be delayed so that
 705                  * they can cancel out the influence of the ARP request.
 706                  */
 707                 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 708
 709                 /* arp requests are broadcast and are sent on the primary
 710                  * the arp request will collapse all clients on the subnet to
 711                  * the primary slave. We must register these clients to be
 712                  * updated with their assigned mac.
 713                  */
 714                 rlb_req_update_subnet_clients(bond, arp->ip_src);
 715                 pr_debug("Server sent ARP Request packet\n");
 716         }
 717
 718         return tx_slave;
 719 }
 720
 721 /* Caller must hold bond lock for read */
 722 static void rlb_rebalance(struct bonding *bond)
 723 {
 724         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 725         struct slave *assigned_slave;
 726         struct rlb_client_info *client_info;
 727         int ntt;
 728         u32 hash_index;
 729
 730         _lock_rx_hashtbl(bond);
 731
 732         ntt = 0;
 733         hash_index = bond_info->rx_hashtbl_head;
 734         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 735                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 736                 assigned_slave = rlb_next_rx_slave(bond);
 737                 if (assigned_slave && (client_info->slave != assigned_slave)) {
 738                         client_info->slave = assigned_slave;
 739                         client_info->ntt = 1;
 740                         ntt = 1;
 741                 }
 742         }
 743
 744         /* update the team's flag only after the whole iteration */
 745         if (ntt) {
 746                 bond_info->rx_ntt = 1;
 747         }
 748         _unlock_rx_hashtbl(bond);
 749 }
 750
 751 /* Caller must hold rx_hashtbl lock */
 752 static void rlb_init_table_entry(struct rlb_client_info *entry)
 753 {
 754         memset(entry, 0, sizeof(struct rlb_client_info));
 755         entry->next = RLB_NULL_INDEX;
 756         entry->prev = RLB_NULL_INDEX;
 757 }
 758
 759 static int rlb_initialize(struct bonding *bond)
 760 {
 761         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 762         struct packet_type *pk_type = &(BOND_ALB_INFO(bond).rlb_pkt_type);
 763         struct rlb_client_info  *new_hashtbl;
 764         int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 765         int i;
 766
 767         spin_lock_init(&(bond_info->rx_hashtbl_lock));
 768
 769         new_hashtbl = kmalloc(size, GFP_KERNEL);
 770         if (!new_hashtbl) {
 771                 pr_err("%s: Error: Failed to allocate RLB hash table\n",
 772                        bond->dev->name);
 773                 return -1;
 774         }
 775         _lock_rx_hashtbl(bond);
 776
 777         bond_info->rx_hashtbl = new_hashtbl;
 778
 779         bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
 780
 781         for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
 782                 rlb_init_table_entry(bond_info->rx_hashtbl + i);
 783         }
 784
 785         _unlock_rx_hashtbl(bond);
 786
 787         /*initialize packet type*/
 788         pk_type->type = cpu_to_be16(ETH_P_ARP);
 789         pk_type->dev = bond->dev;
 790         pk_type->func = rlb_arp_recv;
 791
 792         /* register to receive ARPs */
 793         dev_add_pack(pk_type);
 794
 795         return 0;
 796 }
 797
 798 static void rlb_deinitialize(struct bonding *bond)
 799 {
 800         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 801
 802         dev_remove_pack(&(bond_info->rlb_pkt_type));
 803
 804         _lock_rx_hashtbl(bond);
 805
 806         kfree(bond_info->rx_hashtbl);
 807         bond_info->rx_hashtbl = NULL;
 808         bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
 809
 810         _unlock_rx_hashtbl(bond);
 811 }
 812
 813 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 814 {
 815         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 816         u32 curr_index;
 817
 818         _lock_rx_hashtbl(bond);
 819
 820         curr_index = bond_info->rx_hashtbl_head;
 821         while (curr_index != RLB_NULL_INDEX) {
 822                 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 823                 u32 next_index = bond_info->rx_hashtbl[curr_index].next;
 824                 u32 prev_index = bond_info->rx_hashtbl[curr_index].prev;
 825
 826                 if (curr->tag && (curr->vlan_id == vlan_id)) {
 827                         if (curr_index == bond_info->rx_hashtbl_head) {
 828                                 bond_info->rx_hashtbl_head = next_index;
 829                         }
 830                         if (prev_index != RLB_NULL_INDEX) {
 831                                 bond_info->rx_hashtbl[prev_index].next = next_index;
 832                         }
 833                         if (next_index != RLB_NULL_INDEX) {
 834                                 bond_info->rx_hashtbl[next_index].prev = prev_index;
 835                         }
 836
 837                         rlb_init_table_entry(curr);
 838                 }
 839
 840                 curr_index = next_index;
 841         }
 842
 843         _unlock_rx_hashtbl(bond);
 844 }
 845
 846 /*********************** tlb/rlb shared functions *********************/
 847
 848 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
 849 {
 850         struct bonding *bond = bond_get_bond_by_slave(slave);
 851         struct learning_pkt pkt;
 852         int size = sizeof(struct learning_pkt);
 853         int i;
 854
 855         memset(&pkt, 0, size);
 856         memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
 857         memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
 858         pkt.type = cpu_to_be16(ETH_P_LOOP);
 859
 860         for (i = 0; i < MAX_LP_BURST; i++) {
 861                 struct sk_buff *skb;
 862                 char *data;
 863
 864                 skb = dev_alloc_skb(size);
 865                 if (!skb) {
 866                         return;
 867                 }
 868
 869                 data = skb_put(skb, size);
 870                 memcpy(data, &pkt, size);
 871
 872                 skb_reset_mac_header(skb);
 873                 skb->network_header = skb->mac_header + ETH_HLEN;
 874                 skb->protocol = pkt.type;
 875                 skb->priority = TC_PRIO_CONTROL;
 876                 skb->dev = slave->dev;
 877
 878                 if (bond->vlgrp) {
 879                         struct vlan_entry *vlan;
 880
 881                         vlan = bond_next_vlan(bond,
 882                                               bond->alb_info.current_alb_vlan);
 883
 884                         bond->alb_info.current_alb_vlan = vlan;
 885                         if (!vlan) {
 886                                 kfree_skb(skb);
 887                                 continue;
 888                         }
 889
 890                         skb = vlan_put_tag(skb, vlan->vlan_id);
 891                         if (!skb) {
 892                                 pr_err("%s: Error: failed to insert VLAN tag\n",
 893                                        bond->dev->name);
 894                                 continue;
 895                         }
 896                 }
 897
 898                 dev_queue_xmit(skb);
 899         }
 900 }
 901
 902 /* hw is a boolean parameter that determines whether we should try and
 903  * set the hw address of the device as well as the hw address of the
 904  * net_device
 905  */
 906 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[], int hw)
 907 {
 908         struct net_device *dev = slave->dev;
 909         struct sockaddr s_addr;
 910
 911         if (!hw) {
 912                 memcpy(dev->dev_addr, addr, dev->addr_len);
 913                 return 0;
 914         }
 915
 916         /* for rlb each slave must have a unique hw mac addresses so that */
 917         /* each slave will receive packets destined to a different mac */
 918         memcpy(s_addr.sa_data, addr, dev->addr_len);
 919         s_addr.sa_family = dev->type;
 920         if (dev_set_mac_address(dev, &s_addr)) {
 921                 pr_err("%s: Error: dev_set_mac_address of dev %s failed!\n"
 922                        "ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
 923                        dev->master->name, dev->name);
 924                 return -EOPNOTSUPP;
 925         }
 926         return 0;
 927 }
 928
 929 /*
 930  * Swap MAC addresses between two slaves.
 931  *
 932  * Called with RTNL held, and no other locks.
 933  *
 934  */
 935
 936 static void alb_swap_mac_addr(struct bonding *bond, struct slave *slave1, struct slave *slave2)
 937 {
 938         u8 tmp_mac_addr[ETH_ALEN];
 939
 940         memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
 941         alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr, bond->alb_info.rlb_enabled);
 942         alb_set_slave_mac_addr(slave2, tmp_mac_addr, bond->alb_info.rlb_enabled);
 943
 944 }
 945
 946 /*
 947  * Send learning packets after MAC address swap.
 948  *
 949  * Called with RTNL and no other locks
 950  */
 951 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
 952                                 struct slave *slave2)
 953 {
 954         int slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
 955         struct slave *disabled_slave = NULL;
 956
 957         ASSERT_RTNL();
 958
 959         /* fasten the change in the switch */
 960         if (SLAVE_IS_OK(slave1)) {
 961                 alb_send_learning_packets(slave1, slave1->dev->dev_addr);
 962                 if (bond->alb_info.rlb_enabled) {
 963                         /* inform the clients that the mac address
 964                          * has changed
 965                          */
 966                         rlb_req_update_slave_clients(bond, slave1);
 967                 }
 968         } else {
 969                 disabled_slave = slave1;
 970         }
 971
 972         if (SLAVE_IS_OK(slave2)) {
 973                 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
 974                 if (bond->alb_info.rlb_enabled) {
 975                         /* inform the clients that the mac address
 976                          * has changed
 977                          */
 978                         rlb_req_update_slave_clients(bond, slave2);
 979                 }
 980         } else {
 981                 disabled_slave = slave2;
 982         }
 983
 984         if (bond->alb_info.rlb_enabled && slaves_state_differ) {
 985                 /* A disabled slave was assigned an active mac addr */
 986                 rlb_teach_disabled_mac_on_primary(bond,
 987                                                   disabled_slave->dev->dev_addr);
 988         }
 989 }
 990
 991 /**
 992  * alb_change_hw_addr_on_detach
 993  * @bond: bonding we're working on
 994  * @slave: the slave that was just detached
 995  *
 996  * We assume that @slave was already detached from the slave list.
 997  *
 998  * If @slave's permanent hw address is different both from its current
 999  * address and from @bond's address, then somewhere in the bond there's
1000  * a slave that has @slave's permanet address as its current address.
1001  * We'll make sure that that slave no longer uses @slave's permanent address.
1002  *
1003  * Caller must hold RTNL and no other locks
1004  */
1005 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1006 {
1007         int perm_curr_diff;
1008         int perm_bond_diff;
1009
1010         perm_curr_diff = compare_ether_addr_64bits(slave->perm_hwaddr,
1011                                                    slave->dev->dev_addr);
1012         perm_bond_diff = compare_ether_addr_64bits(slave->perm_hwaddr,
1013                                                    bond->dev->dev_addr);
1014
1015         if (perm_curr_diff && perm_bond_diff) {
1016                 struct slave *tmp_slave;
1017                 int i, found = 0;
1018
1019                 bond_for_each_slave(bond, tmp_slave, i) {
1020                         if (!compare_ether_addr_64bits(slave->perm_hwaddr,
1021                                                        tmp_slave->dev->dev_addr)) {
1022                                 found = 1;
1023                                 break;
1024                         }
1025                 }
1026
1027                 if (found) {
1028                         /* locking: needs RTNL and nothing else */
1029                         alb_swap_mac_addr(bond, slave, tmp_slave);
1030                         alb_fasten_mac_swap(bond, slave, tmp_slave);
1031                 }
1032         }
1033 }
1034
1035 /**
1036  * alb_handle_addr_collision_on_attach
1037  * @bond: bonding we're working on
1038  * @slave: the slave that was just attached
1039  *
1040  * checks uniqueness of slave's mac address and handles the case the
1041  * new slave uses the bonds mac address.
1042  *
1043  * If the permanent hw address of @slave is @bond's hw address, we need to
1044  * find a different hw address to give @slave, that isn't in use by any other
1045  * slave in the bond. This address must be, of course, one of the permanent
1046  * addresses of the other slaves.
1047  *
1048  * We go over the slave list, and for each slave there we compare its
1049  * permanent hw address with the current address of all the other slaves.
1050  * If no match was found, then we've found a slave with a permanent address
1051  * that isn't used by any other slave in the bond, so we can assign it to
1052  * @slave.
1053  *
1054  * assumption: this function is called before @slave is attached to the
1055  *             bond slave list.
1056  *
1057  * caller must hold the bond lock for write since the mac addresses are compared
1058  * and may be swapped.
1059  */
1060 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1061 {
1062         struct slave *tmp_slave1, *tmp_slave2, *free_mac_slave;
1063         struct slave *has_bond_addr = bond->curr_active_slave;
1064         int i, j, found = 0;
1065
1066         if (bond->slave_cnt == 0) {
1067                 /* this is the first slave */
1068                 return 0;
1069         }
1070
1071         /* if slave's mac address differs from bond's mac address
1072          * check uniqueness of slave's mac address against the other
1073          * slaves in the bond.
1074          */
1075         if (compare_ether_addr_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1076                 bond_for_each_slave(bond, tmp_slave1, i) {
1077                         if (!compare_ether_addr_64bits(tmp_slave1->dev->dev_addr,
1078                                                        slave->dev->dev_addr)) {
1079                                 found = 1;
1080                                 break;
1081                         }
1082                 }
1083
1084                 if (!found)
1085                         return 0;
1086
1087                 /* Try setting slave mac to bond address and fall-through
1088                    to code handling that situation below... */
1089                 alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1090                                        bond->alb_info.rlb_enabled);
1091         }
1092
1093         /* The slave's address is equal to the address of the bond.
1094          * Search for a spare address in the bond for this slave.
1095          */
1096         free_mac_slave = NULL;
1097
1098         bond_for_each_slave(bond, tmp_slave1, i) {
1099                 found = 0;
1100                 bond_for_each_slave(bond, tmp_slave2, j) {
1101                         if (!compare_ether_addr_64bits(tmp_slave1->perm_hwaddr,
1102                                                        tmp_slave2->dev->dev_addr)) {
1103                                 found = 1;
1104                                 break;
1105                         }
1106                 }
1107
1108                 if (!found) {
1109                         /* no slave has tmp_slave1's perm addr
1110                          * as its curr addr
1111                          */
1112                         free_mac_slave = tmp_slave1;
1113                         break;
1114                 }
1115
1116                 if (!has_bond_addr) {
1117                         if (!compare_ether_addr_64bits(tmp_slave1->dev->dev_addr,
1118                                                        bond->dev->dev_addr)) {
1119
1120                                 has_bond_addr = tmp_slave1;
1121                         }
1122                 }
1123         }
1124
1125         if (free_mac_slave) {
1126                 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1127                                        bond->alb_info.rlb_enabled);
1128
1129                 pr_warning("%s: Warning: the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1130                            bond->dev->name, slave->dev->name,
1131                            free_mac_slave->dev->name);
1132
1133         } else if (has_bond_addr) {
1134                 pr_err("%s: Error: the hw address of slave %s is in use by the bond; couldn't find a slave with a free hw address to give it (this should not have happened)\n",
1135                        bond->dev->name, slave->dev->name);
1136                 return -EFAULT;
1137         }
1138
1139         return 0;
1140 }
1141
1142 /**
1143  * alb_set_mac_address
1144  * @bond:
1145  * @addr:
1146  *
1147  * In TLB mode all slaves are configured to the bond's hw address, but set
1148  * their dev_addr field to different addresses (based on their permanent hw
1149  * addresses).
1150  *
1151  * For each slave, this function sets the interface to the new address and then
1152  * changes its dev_addr field to its previous value.
1153  *
1154  * Unwinding assumes bond's mac address has not yet changed.
1155  */
1156 static int alb_set_mac_address(struct bonding *bond, void *addr)
1157 {
1158         struct sockaddr sa;
1159         struct slave *slave, *stop_at;
1160         char tmp_addr[ETH_ALEN];
1161         int res;
1162         int i;
1163
1164         if (bond->alb_info.rlb_enabled) {
1165                 return 0;
1166         }
1167
1168         bond_for_each_slave(bond, slave, i) {
1169                 /* save net_device's current hw address */
1170                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1171
1172                 res = dev_set_mac_address(slave->dev, addr);
1173
1174                 /* restore net_device's hw address */
1175                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1176
1177                 if (res)
1178                         goto unwind;
1179         }
1180
1181         return 0;
1182
1183 unwind:
1184         memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1185         sa.sa_family = bond->dev->type;
1186
1187         /* unwind from head to the slave that failed */
1188         stop_at = slave;
1189         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
1190                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1191                 dev_set_mac_address(slave->dev, &sa);
1192                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1193         }
1194
1195         return res;
1196 }
1197
1198 /************************ exported alb funcions ************************/
1199
1200 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1201 {
1202         int res;
1203
1204         res = tlb_initialize(bond);
1205         if (res) {
1206                 return res;
1207         }
1208
1209         if (rlb_enabled) {
1210                 bond->alb_info.rlb_enabled = 1;
1211                 /* initialize rlb */
1212                 res = rlb_initialize(bond);
1213                 if (res) {
1214                         tlb_deinitialize(bond);
1215                         return res;
1216                 }
1217         } else {
1218                 bond->alb_info.rlb_enabled = 0;
1219         }
1220
1221         return 0;
1222 }
1223
1224 void bond_alb_deinitialize(struct bonding *bond)
1225 {
1226         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1227
1228         tlb_deinitialize(bond);
1229
1230         if (bond_info->rlb_enabled) {
1231                 rlb_deinitialize(bond);
1232         }
1233 }
1234
1235 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1236 {
1237         struct bonding *bond = netdev_priv(bond_dev);
1238         struct ethhdr *eth_data;
1239         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1240         struct slave *tx_slave = NULL;
1241         static const __be32 ip_bcast = htonl(0xffffffff);
1242         int hash_size = 0;
1243         int do_tx_balance = 1;
1244         u32 hash_index = 0;
1245         const u8 *hash_start = NULL;
1246         int res = 1;
1247         struct ipv6hdr *ip6hdr;
1248
1249         skb_reset_mac_header(skb);
1250         eth_data = eth_hdr(skb);
1251
1252         /* make sure that the curr_active_slave and the slaves list do
1253          * not change during tx
1254          */
1255         read_lock(&bond->lock);
1256         read_lock(&bond->curr_slave_lock);
1257
1258         if (!BOND_IS_OK(bond)) {
1259                 goto out;
1260         }
1261
1262         switch (ntohs(skb->protocol)) {
1263         case ETH_P_IP: {
1264                 const struct iphdr *iph = ip_hdr(skb);
1265
1266                 if (!compare_ether_addr_64bits(eth_data->h_dest, mac_bcast) ||
1267                     (iph->daddr == ip_bcast) ||
1268                     (iph->protocol == IPPROTO_IGMP)) {
1269                         do_tx_balance = 0;
1270                         break;
1271                 }
1272                 hash_start = (char *)&(iph->daddr);
1273                 hash_size = sizeof(iph->daddr);
1274         }
1275                 break;
1276         case ETH_P_IPV6:
1277                 /* IPv6 doesn't really use broadcast mac address, but leave
1278                  * that here just in case.
1279                  */
1280                 if (!compare_ether_addr_64bits(eth_data->h_dest, mac_bcast)) {
1281                         do_tx_balance = 0;
1282                         break;
1283                 }
1284
1285                 /* IPv6 uses all-nodes multicast as an equivalent to
1286                  * broadcasts in IPv4.
1287                  */
1288                 if (!compare_ether_addr_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1289                         do_tx_balance = 0;
1290                         break;
1291                 }
1292
1293                 /* Additianally, DAD probes should not be tx-balanced as that
1294                  * will lead to false positives for duplicate addresses and
1295                  * prevent address configuration from working.
1296                  */
1297                 ip6hdr = ipv6_hdr(skb);
1298                 if (ipv6_addr_any(&ip6hdr->saddr)) {
1299                         do_tx_balance = 0;
1300                         break;
1301                 }
1302
1303                 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1304                 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1305                 break;
1306         case ETH_P_IPX:
1307                 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1308                         /* something is wrong with this packet */
1309                         do_tx_balance = 0;
1310                         break;
1311                 }
1312
1313                 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1314                         /* The only protocol worth balancing in
1315                          * this family since it has an "ARP" like
1316                          * mechanism
1317                          */
1318                         do_tx_balance = 0;
1319                         break;
1320                 }
1321
1322                 hash_start = (char*)eth_data->h_dest;
1323                 hash_size = ETH_ALEN;
1324                 break;
1325         case ETH_P_ARP:
1326                 do_tx_balance = 0;
1327                 if (bond_info->rlb_enabled) {
1328                         tx_slave = rlb_arp_xmit(skb, bond);
1329                 }
1330                 break;
1331         default:
1332                 do_tx_balance = 0;
1333                 break;
1334         }
1335
1336         if (do_tx_balance) {
1337                 hash_index = _simple_hash(hash_start, hash_size);
1338                 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1339         }
1340
1341         if (!tx_slave) {
1342                 /* unbalanced or unassigned, send through primary */
1343                 tx_slave = bond->curr_active_slave;
1344                 bond_info->unbalanced_load += skb->len;
1345         }
1346
1347         if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1348                 if (tx_slave != bond->curr_active_slave) {
1349                         memcpy(eth_data->h_source,
1350                                tx_slave->dev->dev_addr,
1351                                ETH_ALEN);
1352                 }
1353
1354                 res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1355         } else {
1356                 if (tx_slave) {
1357                         tlb_clear_slave(bond, tx_slave, 0);
1358                 }
1359         }
1360
1361 out:
1362         if (res) {
1363                 /* no suitable interface, frame not sent */
1364                 dev_kfree_skb(skb);
1365         }
1366         read_unlock(&bond->curr_slave_lock);
1367         read_unlock(&bond->lock);
1368         return NETDEV_TX_OK;
1369 }
1370
1371 void bond_alb_monitor(struct work_struct *work)
1372 {
1373         struct bonding *bond = container_of(work, struct bonding,
1374                                             alb_work.work);
1375         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1376         struct slave *slave;
1377         int i;
1378
1379         read_lock(&bond->lock);
1380
1381         if (bond->kill_timers) {
1382                 goto out;
1383         }
1384
1385         if (bond->slave_cnt == 0) {
1386                 bond_info->tx_rebalance_counter = 0;
1387                 bond_info->lp_counter = 0;
1388                 goto re_arm;
1389         }
1390
1391         bond_info->tx_rebalance_counter++;
1392         bond_info->lp_counter++;
1393
1394         /* send learning packets */
1395         if (bond_info->lp_counter >= BOND_ALB_LP_TICKS) {
1396                 /* change of curr_active_slave involves swapping of mac addresses.
1397                  * in order to avoid this swapping from happening while
1398                  * sending the learning packets, the curr_slave_lock must be held for
1399                  * read.
1400                  */
1401                 read_lock(&bond->curr_slave_lock);
1402
1403                 bond_for_each_slave(bond, slave, i) {
1404                         alb_send_learning_packets(slave, slave->dev->dev_addr);
1405                 }
1406
1407                 read_unlock(&bond->curr_slave_lock);
1408
1409                 bond_info->lp_counter = 0;
1410         }
1411
1412         /* rebalance tx traffic */
1413         if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1414
1415                 read_lock(&bond->curr_slave_lock);
1416
1417                 bond_for_each_slave(bond, slave, i) {
1418                         tlb_clear_slave(bond, slave, 1);
1419                         if (slave == bond->curr_active_slave) {
1420                                 SLAVE_TLB_INFO(slave).load =
1421                                         bond_info->unbalanced_load /
1422                                                 BOND_TLB_REBALANCE_INTERVAL;
1423                                 bond_info->unbalanced_load = 0;
1424                         }
1425                 }
1426
1427                 read_unlock(&bond->curr_slave_lock);
1428
1429                 bond_info->tx_rebalance_counter = 0;
1430         }
1431
1432         /* handle rlb stuff */
1433         if (bond_info->rlb_enabled) {
1434                 if (bond_info->primary_is_promisc &&
1435                     (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1436
1437                         /*
1438                          * dev_set_promiscuity requires rtnl and
1439                          * nothing else.
1440                          */
1441                         read_unlock(&bond->lock);
1442                         rtnl_lock();
1443
1444                         bond_info->rlb_promisc_timeout_counter = 0;
1445
1446                         /* If the primary was set to promiscuous mode
1447                          * because a slave was disabled then
1448                          * it can now leave promiscuous mode.
1449                          */
1450                         dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1451                         bond_info->primary_is_promisc = 0;
1452
1453                         rtnl_unlock();
1454                         read_lock(&bond->lock);
1455                 }
1456
1457                 if (bond_info->rlb_rebalance) {
1458                         bond_info->rlb_rebalance = 0;
1459                         rlb_rebalance(bond);
1460                 }
1461
1462                 /* check if clients need updating */
1463                 if (bond_info->rx_ntt) {
1464                         if (bond_info->rlb_update_delay_counter) {
1465                                 --bond_info->rlb_update_delay_counter;
1466                         } else {
1467                                 rlb_update_rx_clients(bond);
1468                                 if (bond_info->rlb_update_retry_counter) {
1469                                         --bond_info->rlb_update_retry_counter;
1470                                 } else {
1471                                         bond_info->rx_ntt = 0;
1472                                 }
1473                         }
1474                 }
1475         }
1476
1477 re_arm:
1478         queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1479 out:
1480         read_unlock(&bond->lock);
1481 }
1482
1483 /* assumption: called before the slave is attached to the bond
1484  * and not locked by the bond lock
1485  */
1486 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1487 {
1488         int res;
1489
1490         res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1491                                      bond->alb_info.rlb_enabled);
1492         if (res) {
1493                 return res;
1494         }
1495
1496         /* caller must hold the bond lock for write since the mac addresses
1497          * are compared and may be swapped.
1498          */
1499         read_lock(&bond->lock);
1500
1501         res = alb_handle_addr_collision_on_attach(bond, slave);
1502
1503         read_unlock(&bond->lock);
1504
1505         if (res) {
1506                 return res;
1507         }
1508
1509         tlb_init_slave(slave);
1510
1511         /* order a rebalance ASAP */
1512         bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1513
1514         if (bond->alb_info.rlb_enabled) {
1515                 bond->alb_info.rlb_rebalance = 1;
1516         }
1517
1518         return 0;
1519 }
1520
1521 /*
1522  * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1523  * if necessary.
1524  *
1525  * Caller must hold RTNL and no other locks
1526  */
1527 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1528 {
1529         if (bond->slave_cnt > 1) {
1530                 alb_change_hw_addr_on_detach(bond, slave);
1531         }
1532
1533         tlb_clear_slave(bond, slave, 0);
1534
1535         if (bond->alb_info.rlb_enabled) {
1536                 bond->alb_info.next_rx_slave = NULL;
1537                 rlb_clear_slave(bond, slave);
1538         }
1539 }
1540
1541 /* Caller must hold bond lock for read */
1542 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1543 {
1544         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1545
1546         if (link == BOND_LINK_DOWN) {
1547                 tlb_clear_slave(bond, slave, 0);
1548                 if (bond->alb_info.rlb_enabled) {
1549                         rlb_clear_slave(bond, slave);
1550                 }
1551         } else if (link == BOND_LINK_UP) {
1552                 /* order a rebalance ASAP */
1553                 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1554                 if (bond->alb_info.rlb_enabled) {
1555                         bond->alb_info.rlb_rebalance = 1;
1556                         /* If the updelay module parameter is smaller than the
1557                          * forwarding delay of the switch the rebalance will
1558                          * not work because the rebalance arp replies will
1559                          * not be forwarded to the clients..
1560                          */
1561                 }
1562         }
1563 }
1564
1565 /**
1566  * bond_alb_handle_active_change - assign new curr_active_slave
1567  * @bond: our bonding struct
1568  * @new_slave: new slave to assign
1569  *
1570  * Set the bond->curr_active_slave to @new_slave and handle
1571  * mac address swapping and promiscuity changes as needed.
1572  *
1573  * If new_slave is NULL, caller must hold curr_slave_lock or
1574  * bond->lock for write.
1575  *
1576  * If new_slave is not NULL, caller must hold RTNL, bond->lock for
1577  * read and curr_slave_lock for write.  Processing here may sleep, so
1578  * no other locks may be held.
1579  */
1580 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1581         __releases(&bond->curr_slave_lock)
1582         __releases(&bond->lock)
1583         __acquires(&bond->lock)
1584         __acquires(&bond->curr_slave_lock)
1585 {
1586         struct slave *swap_slave;
1587         int i;
1588
1589         if (bond->curr_active_slave == new_slave) {
1590                 return;
1591         }
1592
1593         if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1594                 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1595                 bond->alb_info.primary_is_promisc = 0;
1596                 bond->alb_info.rlb_promisc_timeout_counter = 0;
1597         }
1598
1599         swap_slave = bond->curr_active_slave;
1600         bond->curr_active_slave = new_slave;
1601
1602         if (!new_slave || (bond->slave_cnt == 0)) {
1603                 return;
1604         }
1605
1606         /* set the new curr_active_slave to the bonds mac address
1607          * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1608          */
1609         if (!swap_slave) {
1610                 struct slave *tmp_slave;
1611                 /* find slave that is holding the bond's mac address */
1612                 bond_for_each_slave(bond, tmp_slave, i) {
1613                         if (!compare_ether_addr_64bits(tmp_slave->dev->dev_addr,
1614                                                        bond->dev->dev_addr)) {
1615                                 swap_slave = tmp_slave;
1616                                 break;
1617                         }
1618                 }
1619         }
1620
1621         /*
1622          * Arrange for swap_slave and new_slave to temporarily be
1623          * ignored so we can mess with their MAC addresses without
1624          * fear of interference from transmit activity.
1625          */
1626         if (swap_slave) {
1627                 tlb_clear_slave(bond, swap_slave, 1);
1628         }
1629         tlb_clear_slave(bond, new_slave, 1);
1630
1631         write_unlock_bh(&bond->curr_slave_lock);
1632         read_unlock(&bond->lock);
1633
1634         ASSERT_RTNL();
1635
1636         /* curr_active_slave must be set before calling alb_swap_mac_addr */
1637         if (swap_slave) {
1638                 /* swap mac address */
1639                 alb_swap_mac_addr(bond, swap_slave, new_slave);
1640         } else {
1641                 /* set the new_slave to the bond mac address */
1642                 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1643                                        bond->alb_info.rlb_enabled);
1644         }
1645
1646         if (swap_slave) {
1647                 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1648                 read_lock(&bond->lock);
1649         } else {
1650                 read_lock(&bond->lock);
1651                 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1652         }
1653
1654         write_lock_bh(&bond->curr_slave_lock);
1655 }
1656
1657 /*
1658  * Called with RTNL
1659  */
1660 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1661         __acquires(&bond->lock)
1662         __releases(&bond->lock)
1663 {
1664         struct bonding *bond = netdev_priv(bond_dev);
1665         struct sockaddr *sa = addr;
1666         struct slave *slave, *swap_slave;
1667         int res;
1668         int i;
1669
1670         if (!is_valid_ether_addr(sa->sa_data)) {
1671                 return -EADDRNOTAVAIL;
1672         }
1673
1674         res = alb_set_mac_address(bond, addr);
1675         if (res) {
1676                 return res;
1677         }
1678
1679         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1680
1681         /* If there is no curr_active_slave there is nothing else to do.
1682          * Otherwise we'll need to pass the new address to it and handle
1683          * duplications.
1684          */
1685         if (!bond->curr_active_slave) {
1686                 return 0;
1687         }
1688
1689         swap_slave = NULL;
1690
1691         bond_for_each_slave(bond, slave, i) {
1692                 if (!compare_ether_addr_64bits(slave->dev->dev_addr,
1693                                                bond_dev->dev_addr)) {
1694                         swap_slave = slave;
1695                         break;
1696                 }
1697         }
1698
1699         if (swap_slave) {
1700                 alb_swap_mac_addr(bond, swap_slave, bond->curr_active_slave);
1701                 alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave);
1702         } else {
1703                 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr,
1704                                        bond->alb_info.rlb_enabled);
1705
1706                 read_lock(&bond->lock);
1707                 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1708                 if (bond->alb_info.rlb_enabled) {
1709                         /* inform clients mac address has changed */
1710                         rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1711                 }
1712                 read_unlock(&bond->lock);
1713         }
1714
1715         return 0;
1716 }
1717
1718 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1719 {
1720         if (bond->alb_info.current_alb_vlan &&
1721             (bond->alb_info.current_alb_vlan->vlan_id == vlan_id)) {
1722                 bond->alb_info.current_alb_vlan = NULL;
1723         }
1724
1725         if (bond->alb_info.rlb_enabled) {
1726                 rlb_clear_vlan(bond, vlan_id);
1727         }
1728 }
1729