Pileus Git - ~andy/linux/blob - drivers/net/bonding/bond_alb.c

   1 /*
   2  * Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms of the GNU General Public License as published by the
   6  * Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful, but
  10  * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  11  * or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License along
  15  * with this program; if not, write to the Free Software Foundation, Inc.,
  16  * 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
  17  *
  18  * The full GNU General Public License is included in this distribution in the
  19  * file called LICENSE.
  20  *
  21  */
  22
  23 #define 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 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp);
  88 static void rlb_src_unlink(struct bonding *bond, u32 index);
  89 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash,
  90                          u32 ip_dst_hash);
  91
  92 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
  93 {
  94         int i;
  95         u8 hash = 0;
  96
  97         for (i = 0; i < hash_size; i++) {
  98                 hash ^= hash_start[i];
  99         }
 100
 101         return hash;
 102 }
 103
 104 /*********************** tlb specific functions ***************************/
 105
 106 static inline void _lock_tx_hashtbl_bh(struct bonding *bond)
 107 {
 108         spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 109 }
 110
 111 static inline void _unlock_tx_hashtbl_bh(struct bonding *bond)
 112 {
 113         spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 114 }
 115
 116 static inline void _lock_tx_hashtbl(struct bonding *bond)
 117 {
 118         spin_lock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 119 }
 120
 121 static inline void _unlock_tx_hashtbl(struct bonding *bond)
 122 {
 123         spin_unlock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 124 }
 125
 126 /* Caller must hold tx_hashtbl lock */
 127 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
 128 {
 129         if (save_load) {
 130                 entry->load_history = 1 + entry->tx_bytes /
 131                                       BOND_TLB_REBALANCE_INTERVAL;
 132                 entry->tx_bytes = 0;
 133         }
 134
 135         entry->tx_slave = NULL;
 136         entry->next = TLB_NULL_INDEX;
 137         entry->prev = TLB_NULL_INDEX;
 138 }
 139
 140 static inline void tlb_init_slave(struct slave *slave)
 141 {
 142         SLAVE_TLB_INFO(slave).load = 0;
 143         SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
 144 }
 145
 146 /* Caller must hold bond lock for read, BH disabled */
 147 static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
 148                          int save_load)
 149 {
 150         struct tlb_client_info *tx_hash_table;
 151         u32 index;
 152
 153         /* clear slave from tx_hashtbl */
 154         tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
 155
 156         /* skip this if we've already freed the tx hash table */
 157         if (tx_hash_table) {
 158                 index = SLAVE_TLB_INFO(slave).head;
 159                 while (index != TLB_NULL_INDEX) {
 160                         u32 next_index = tx_hash_table[index].next;
 161                         tlb_init_table_entry(&tx_hash_table[index], save_load);
 162                         index = next_index;
 163                 }
 164         }
 165
 166         tlb_init_slave(slave);
 167 }
 168
 169 /* Caller must hold bond lock for read */
 170 static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
 171                          int save_load)
 172 {
 173         _lock_tx_hashtbl_bh(bond);
 174         __tlb_clear_slave(bond, slave, save_load);
 175         _unlock_tx_hashtbl_bh(bond);
 176 }
 177
 178 /* Must be called before starting the monitor timer */
 179 static int tlb_initialize(struct bonding *bond)
 180 {
 181         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 182         int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 183         struct tlb_client_info *new_hashtbl;
 184         int i;
 185
 186         new_hashtbl = kzalloc(size, GFP_KERNEL);
 187         if (!new_hashtbl)
 188                 return -1;
 189
 190         _lock_tx_hashtbl_bh(bond);
 191
 192         bond_info->tx_hashtbl = new_hashtbl;
 193
 194         for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
 195                 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
 196         }
 197
 198         _unlock_tx_hashtbl_bh(bond);
 199
 200         return 0;
 201 }
 202
 203 /* Must be called only after all slaves have been released */
 204 static void tlb_deinitialize(struct bonding *bond)
 205 {
 206         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 207
 208         _lock_tx_hashtbl_bh(bond);
 209
 210         kfree(bond_info->tx_hashtbl);
 211         bond_info->tx_hashtbl = NULL;
 212
 213         _unlock_tx_hashtbl_bh(bond);
 214 }
 215
 216 static long long compute_gap(struct slave *slave)
 217 {
 218         return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
 219                (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
 220 }
 221
 222 /* Caller must hold bond lock for read */
 223 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
 224 {
 225         struct slave *slave, *least_loaded;
 226         long long max_gap;
 227         int i;
 228
 229         least_loaded = NULL;
 230         max_gap = LLONG_MIN;
 231
 232         /* Find the slave with the largest gap */
 233         bond_for_each_slave(bond, slave, i) {
 234                 if (SLAVE_IS_OK(slave)) {
 235                         long long gap = compute_gap(slave);
 236
 237                         if (max_gap < gap) {
 238                                 least_loaded = slave;
 239                                 max_gap = gap;
 240                         }
 241                 }
 242         }
 243
 244         return least_loaded;
 245 }
 246
 247 static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
 248                                                 u32 skb_len)
 249 {
 250         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 251         struct tlb_client_info *hash_table;
 252         struct slave *assigned_slave;
 253
 254         hash_table = bond_info->tx_hashtbl;
 255         assigned_slave = hash_table[hash_index].tx_slave;
 256         if (!assigned_slave) {
 257                 assigned_slave = tlb_get_least_loaded_slave(bond);
 258
 259                 if (assigned_slave) {
 260                         struct tlb_slave_info *slave_info =
 261                                 &(SLAVE_TLB_INFO(assigned_slave));
 262                         u32 next_index = slave_info->head;
 263
 264                         hash_table[hash_index].tx_slave = assigned_slave;
 265                         hash_table[hash_index].next = next_index;
 266                         hash_table[hash_index].prev = TLB_NULL_INDEX;
 267
 268                         if (next_index != TLB_NULL_INDEX) {
 269                                 hash_table[next_index].prev = hash_index;
 270                         }
 271
 272                         slave_info->head = hash_index;
 273                         slave_info->load +=
 274                                 hash_table[hash_index].load_history;
 275                 }
 276         }
 277
 278         if (assigned_slave) {
 279                 hash_table[hash_index].tx_bytes += skb_len;
 280         }
 281
 282         return assigned_slave;
 283 }
 284
 285 /* Caller must hold bond lock for read */
 286 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
 287                                         u32 skb_len)
 288 {
 289         struct slave *tx_slave;
 290         /*
 291          * We don't need to disable softirq here, becase
 292          * tlb_choose_channel() is only called by bond_alb_xmit()
 293          * which already has softirq disabled.
 294          */
 295         _lock_tx_hashtbl(bond);
 296         tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
 297         _unlock_tx_hashtbl(bond);
 298         return tx_slave;
 299 }
 300
 301 /*********************** rlb specific functions ***************************/
 302 static inline void _lock_rx_hashtbl_bh(struct bonding *bond)
 303 {
 304         spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 305 }
 306
 307 static inline void _unlock_rx_hashtbl_bh(struct bonding *bond)
 308 {
 309         spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 310 }
 311
 312 static inline void _lock_rx_hashtbl(struct bonding *bond)
 313 {
 314         spin_lock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 315 }
 316
 317 static inline void _unlock_rx_hashtbl(struct bonding *bond)
 318 {
 319         spin_unlock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 320 }
 321
 322 /* when an ARP REPLY is received from a client update its info
 323  * in the rx_hashtbl
 324  */
 325 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 326 {
 327         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 328         struct rlb_client_info *client_info;
 329         u32 hash_index;
 330
 331         _lock_rx_hashtbl_bh(bond);
 332
 333         hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
 334         client_info = &(bond_info->rx_hashtbl[hash_index]);
 335
 336         if ((client_info->assigned) &&
 337             (client_info->ip_src == arp->ip_dst) &&
 338             (client_info->ip_dst == arp->ip_src) &&
 339             (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) {
 340                 /* update the clients MAC address */
 341                 memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
 342                 client_info->ntt = 1;
 343                 bond_info->rx_ntt = 1;
 344         }
 345
 346         _unlock_rx_hashtbl_bh(bond);
 347 }
 348
 349 static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
 350                         struct slave *slave)
 351 {
 352         struct arp_pkt *arp, _arp;
 353
 354         if (skb->protocol != cpu_to_be16(ETH_P_ARP))
 355                 goto out;
 356
 357         arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp);
 358         if (!arp)
 359                 goto out;
 360
 361         /* We received an ARP from arp->ip_src.
 362          * We might have used this IP address previously (on the bonding host
 363          * itself or on a system that is bridged together with the bond).
 364          * However, if arp->mac_src is different than what is stored in
 365          * rx_hashtbl, some other host is now using the IP and we must prevent
 366          * sending out client updates with this IP address and the old MAC
 367          * address.
 368          * Clean up all hash table entries that have this address as ip_src but
 369          * have a different mac_src.
 370          */
 371         rlb_purge_src_ip(bond, arp);
 372
 373         if (arp->op_code == htons(ARPOP_REPLY)) {
 374                 /* update rx hash table for this ARP */
 375                 rlb_update_entry_from_arp(bond, arp);
 376                 pr_debug("Server received an ARP Reply from client\n");
 377         }
 378 out:
 379         return RX_HANDLER_ANOTHER;
 380 }
 381
 382 /* Caller must hold bond lock for read */
 383 static struct slave *rlb_next_rx_slave(struct bonding *bond)
 384 {
 385         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 386         struct slave *rx_slave, *slave, *start_at;
 387         int i = 0;
 388
 389         if (bond_info->next_rx_slave) {
 390                 start_at = bond_info->next_rx_slave;
 391         } else {
 392                 start_at = bond->first_slave;
 393         }
 394
 395         rx_slave = NULL;
 396
 397         bond_for_each_slave_from(bond, slave, i, start_at) {
 398                 if (SLAVE_IS_OK(slave)) {
 399                         if (!rx_slave) {
 400                                 rx_slave = slave;
 401                         } else if (slave->speed > rx_slave->speed) {
 402                                 rx_slave = slave;
 403                         }
 404                 }
 405         }
 406
 407         if (rx_slave) {
 408                 bond_info->next_rx_slave = rx_slave->next;
 409         }
 410
 411         return rx_slave;
 412 }
 413
 414 /* teach the switch the mac of a disabled slave
 415  * on the primary for fault tolerance
 416  *
 417  * Caller must hold bond->curr_slave_lock for write or bond lock for write
 418  */
 419 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
 420 {
 421         if (!bond->curr_active_slave) {
 422                 return;
 423         }
 424
 425         if (!bond->alb_info.primary_is_promisc) {
 426                 if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1))
 427                         bond->alb_info.primary_is_promisc = 1;
 428                 else
 429                         bond->alb_info.primary_is_promisc = 0;
 430         }
 431
 432         bond->alb_info.rlb_promisc_timeout_counter = 0;
 433
 434         alb_send_learning_packets(bond->curr_active_slave, addr);
 435 }
 436
 437 /* slave being removed should not be active at this point
 438  *
 439  * Caller must hold bond lock for read
 440  */
 441 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 442 {
 443         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 444         struct rlb_client_info *rx_hash_table;
 445         u32 index, next_index;
 446
 447         /* clear slave from rx_hashtbl */
 448         _lock_rx_hashtbl_bh(bond);
 449
 450         rx_hash_table = bond_info->rx_hashtbl;
 451         index = bond_info->rx_hashtbl_used_head;
 452         for (; index != RLB_NULL_INDEX; index = next_index) {
 453                 next_index = rx_hash_table[index].used_next;
 454                 if (rx_hash_table[index].slave == slave) {
 455                         struct slave *assigned_slave = rlb_next_rx_slave(bond);
 456
 457                         if (assigned_slave) {
 458                                 rx_hash_table[index].slave = assigned_slave;
 459                                 if (!ether_addr_equal_64bits(rx_hash_table[index].mac_dst,
 460                                                              mac_bcast)) {
 461                                         bond_info->rx_hashtbl[index].ntt = 1;
 462                                         bond_info->rx_ntt = 1;
 463                                         /* A slave has been removed from the
 464                                          * table because it is either disabled
 465                                          * or being released. We must retry the
 466                                          * update to avoid clients from not
 467                                          * being updated & disconnecting when
 468                                          * there is stress
 469                                          */
 470                                         bond_info->rlb_update_retry_counter =
 471                                                 RLB_UPDATE_RETRY;
 472                                 }
 473                         } else {  /* there is no active slave */
 474                                 rx_hash_table[index].slave = NULL;
 475                         }
 476                 }
 477         }
 478
 479         _unlock_rx_hashtbl_bh(bond);
 480
 481         write_lock_bh(&bond->curr_slave_lock);
 482
 483         if (slave != bond->curr_active_slave) {
 484                 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 485         }
 486
 487         write_unlock_bh(&bond->curr_slave_lock);
 488 }
 489
 490 static void rlb_update_client(struct rlb_client_info *client_info)
 491 {
 492         int i;
 493
 494         if (!client_info->slave) {
 495                 return;
 496         }
 497
 498         for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 499                 struct sk_buff *skb;
 500
 501                 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 502                                  client_info->ip_dst,
 503                                  client_info->slave->dev,
 504                                  client_info->ip_src,
 505                                  client_info->mac_dst,
 506                                  client_info->slave->dev->dev_addr,
 507                                  client_info->mac_dst);
 508                 if (!skb) {
 509                         pr_err("%s: Error: failed to create an ARP packet\n",
 510                                client_info->slave->bond->dev->name);
 511                         continue;
 512                 }
 513
 514                 skb->dev = client_info->slave->dev;
 515
 516                 if (client_info->tag) {
 517                         skb = vlan_put_tag(skb, htons(ETH_P_8021Q), client_info->vlan_id);
 518                         if (!skb) {
 519                                 pr_err("%s: Error: failed to insert VLAN tag\n",
 520                                        client_info->slave->bond->dev->name);
 521                                 continue;
 522                         }
 523                 }
 524
 525                 arp_xmit(skb);
 526         }
 527 }
 528
 529 /* sends ARP REPLIES that update the clients that need updating */
 530 static void rlb_update_rx_clients(struct bonding *bond)
 531 {
 532         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 533         struct rlb_client_info *client_info;
 534         u32 hash_index;
 535
 536         _lock_rx_hashtbl_bh(bond);
 537
 538         hash_index = bond_info->rx_hashtbl_used_head;
 539         for (; hash_index != RLB_NULL_INDEX;
 540              hash_index = client_info->used_next) {
 541                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 542                 if (client_info->ntt) {
 543                         rlb_update_client(client_info);
 544                         if (bond_info->rlb_update_retry_counter == 0) {
 545                                 client_info->ntt = 0;
 546                         }
 547                 }
 548         }
 549
 550         /* do not update the entries again until this counter is zero so that
 551          * not to confuse the clients.
 552          */
 553         bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 554
 555         _unlock_rx_hashtbl_bh(bond);
 556 }
 557
 558 /* The slave was assigned a new mac address - update the clients */
 559 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 560 {
 561         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 562         struct rlb_client_info *client_info;
 563         int ntt = 0;
 564         u32 hash_index;
 565
 566         _lock_rx_hashtbl_bh(bond);
 567
 568         hash_index = bond_info->rx_hashtbl_used_head;
 569         for (; hash_index != RLB_NULL_INDEX;
 570              hash_index = client_info->used_next) {
 571                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 572
 573                 if ((client_info->slave == slave) &&
 574                     !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
 575                         client_info->ntt = 1;
 576                         ntt = 1;
 577                 }
 578         }
 579
 580         // update the team's flag only after the whole iteration
 581         if (ntt) {
 582                 bond_info->rx_ntt = 1;
 583                 //fasten the change
 584                 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 585         }
 586
 587         _unlock_rx_hashtbl_bh(bond);
 588 }
 589
 590 /* mark all clients using src_ip to be updated */
 591 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
 592 {
 593         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 594         struct rlb_client_info *client_info;
 595         u32 hash_index;
 596
 597         _lock_rx_hashtbl(bond);
 598
 599         hash_index = bond_info->rx_hashtbl_used_head;
 600         for (; hash_index != RLB_NULL_INDEX;
 601              hash_index = client_info->used_next) {
 602                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 603
 604                 if (!client_info->slave) {
 605                         pr_err("%s: Error: found a client with no channel in the client's hash table\n",
 606                                bond->dev->name);
 607                         continue;
 608                 }
 609                 /*update all clients using this src_ip, that are not assigned
 610                  * to the team's address (curr_active_slave) and have a known
 611                  * unicast mac address.
 612                  */
 613                 if ((client_info->ip_src == src_ip) &&
 614                     !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
 615                                              bond->dev->dev_addr) &&
 616                     !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
 617                         client_info->ntt = 1;
 618                         bond_info->rx_ntt = 1;
 619                 }
 620         }
 621
 622         _unlock_rx_hashtbl(bond);
 623 }
 624
 625 /* Caller must hold both bond and ptr locks for read */
 626 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
 627 {
 628         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 629         struct arp_pkt *arp = arp_pkt(skb);
 630         struct slave *assigned_slave;
 631         struct rlb_client_info *client_info;
 632         u32 hash_index = 0;
 633
 634         _lock_rx_hashtbl(bond);
 635
 636         hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
 637         client_info = &(bond_info->rx_hashtbl[hash_index]);
 638
 639         if (client_info->assigned) {
 640                 if ((client_info->ip_src == arp->ip_src) &&
 641                     (client_info->ip_dst == arp->ip_dst)) {
 642                         /* the entry is already assigned to this client */
 643                         if (!ether_addr_equal_64bits(arp->mac_dst, mac_bcast)) {
 644                                 /* update mac address from arp */
 645                                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 646                         }
 647                         memcpy(client_info->mac_src, arp->mac_src, ETH_ALEN);
 648
 649                         assigned_slave = client_info->slave;
 650                         if (assigned_slave) {
 651                                 _unlock_rx_hashtbl(bond);
 652                                 return assigned_slave;
 653                         }
 654                 } else {
 655                         /* the entry is already assigned to some other client,
 656                          * move the old client to primary (curr_active_slave) so
 657                          * that the new client can be assigned to this entry.
 658                          */
 659                         if (bond->curr_active_slave &&
 660                             client_info->slave != bond->curr_active_slave) {
 661                                 client_info->slave = bond->curr_active_slave;
 662                                 rlb_update_client(client_info);
 663                         }
 664                 }
 665         }
 666         /* assign a new slave */
 667         assigned_slave = rlb_next_rx_slave(bond);
 668
 669         if (assigned_slave) {
 670                 if (!(client_info->assigned &&
 671                       client_info->ip_src == arp->ip_src)) {
 672                         /* ip_src is going to be updated,
 673                          * fix the src hash list
 674                          */
 675                         u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
 676                                                     sizeof(arp->ip_src));
 677                         rlb_src_unlink(bond, hash_index);
 678                         rlb_src_link(bond, hash_src, hash_index);
 679                 }
 680
 681                 client_info->ip_src = arp->ip_src;
 682                 client_info->ip_dst = arp->ip_dst;
 683                 /* arp->mac_dst is broadcast for arp reqeusts.
 684                  * will be updated with clients actual unicast mac address
 685                  * upon receiving an arp reply.
 686                  */
 687                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 688                 memcpy(client_info->mac_src, arp->mac_src, ETH_ALEN);
 689                 client_info->slave = assigned_slave;
 690
 691                 if (!ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
 692                         client_info->ntt = 1;
 693                         bond->alb_info.rx_ntt = 1;
 694                 } else {
 695                         client_info->ntt = 0;
 696                 }
 697
 698                 if (bond_vlan_used(bond)) {
 699                         if (!vlan_get_tag(skb, &client_info->vlan_id))
 700                                 client_info->tag = 1;
 701                 }
 702
 703                 if (!client_info->assigned) {
 704                         u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
 705                         bond_info->rx_hashtbl_used_head = hash_index;
 706                         client_info->used_next = prev_tbl_head;
 707                         if (prev_tbl_head != RLB_NULL_INDEX) {
 708                                 bond_info->rx_hashtbl[prev_tbl_head].used_prev =
 709                                         hash_index;
 710                         }
 711                         client_info->assigned = 1;
 712                 }
 713         }
 714
 715         _unlock_rx_hashtbl(bond);
 716
 717         return assigned_slave;
 718 }
 719
 720 /* chooses (and returns) transmit channel for arp reply
 721  * does not choose channel for other arp types since they are
 722  * sent on the curr_active_slave
 723  */
 724 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 725 {
 726         struct arp_pkt *arp = arp_pkt(skb);
 727         struct slave *tx_slave = NULL;
 728
 729         /* Don't modify or load balance ARPs that do not originate locally
 730          * (e.g.,arrive via a bridge).
 731          */
 732         if (!bond_slave_has_mac(bond, arp->mac_src))
 733                 return NULL;
 734
 735         if (arp->op_code == htons(ARPOP_REPLY)) {
 736                 /* the arp must be sent on the selected
 737                 * rx channel
 738                 */
 739                 tx_slave = rlb_choose_channel(skb, bond);
 740                 if (tx_slave) {
 741                         memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
 742                 }
 743                 pr_debug("Server sent ARP Reply packet\n");
 744         } else if (arp->op_code == htons(ARPOP_REQUEST)) {
 745                 /* Create an entry in the rx_hashtbl for this client as a
 746                  * place holder.
 747                  * When the arp reply is received the entry will be updated
 748                  * with the correct unicast address of the client.
 749                  */
 750                 rlb_choose_channel(skb, bond);
 751
 752                 /* The ARP reply packets must be delayed so that
 753                  * they can cancel out the influence of the ARP request.
 754                  */
 755                 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 756
 757                 /* arp requests are broadcast and are sent on the primary
 758                  * the arp request will collapse all clients on the subnet to
 759                  * the primary slave. We must register these clients to be
 760                  * updated with their assigned mac.
 761                  */
 762                 rlb_req_update_subnet_clients(bond, arp->ip_src);
 763                 pr_debug("Server sent ARP Request packet\n");
 764         }
 765
 766         return tx_slave;
 767 }
 768
 769 /* Caller must hold bond lock for read */
 770 static void rlb_rebalance(struct bonding *bond)
 771 {
 772         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 773         struct slave *assigned_slave;
 774         struct rlb_client_info *client_info;
 775         int ntt;
 776         u32 hash_index;
 777
 778         _lock_rx_hashtbl_bh(bond);
 779
 780         ntt = 0;
 781         hash_index = bond_info->rx_hashtbl_used_head;
 782         for (; hash_index != RLB_NULL_INDEX;
 783              hash_index = client_info->used_next) {
 784                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 785                 assigned_slave = rlb_next_rx_slave(bond);
 786                 if (assigned_slave && (client_info->slave != assigned_slave)) {
 787                         client_info->slave = assigned_slave;
 788                         client_info->ntt = 1;
 789                         ntt = 1;
 790                 }
 791         }
 792
 793         /* update the team's flag only after the whole iteration */
 794         if (ntt) {
 795                 bond_info->rx_ntt = 1;
 796         }
 797         _unlock_rx_hashtbl_bh(bond);
 798 }
 799
 800 /* Caller must hold rx_hashtbl lock */
 801 static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
 802 {
 803         entry->used_next = RLB_NULL_INDEX;
 804         entry->used_prev = RLB_NULL_INDEX;
 805         entry->assigned = 0;
 806         entry->slave = NULL;
 807         entry->tag = 0;
 808 }
 809 static void rlb_init_table_entry_src(struct rlb_client_info *entry)
 810 {
 811         entry->src_first = RLB_NULL_INDEX;
 812         entry->src_prev = RLB_NULL_INDEX;
 813         entry->src_next = RLB_NULL_INDEX;
 814 }
 815
 816 static void rlb_init_table_entry(struct rlb_client_info *entry)
 817 {
 818         memset(entry, 0, sizeof(struct rlb_client_info));
 819         rlb_init_table_entry_dst(entry);
 820         rlb_init_table_entry_src(entry);
 821 }
 822
 823 static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
 824 {
 825         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 826         u32 next_index = bond_info->rx_hashtbl[index].used_next;
 827         u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
 828
 829         if (index == bond_info->rx_hashtbl_used_head)
 830                 bond_info->rx_hashtbl_used_head = next_index;
 831         if (prev_index != RLB_NULL_INDEX)
 832                 bond_info->rx_hashtbl[prev_index].used_next = next_index;
 833         if (next_index != RLB_NULL_INDEX)
 834                 bond_info->rx_hashtbl[next_index].used_prev = prev_index;
 835 }
 836
 837 /* unlink a rlb hash table entry from the src list */
 838 static void rlb_src_unlink(struct bonding *bond, u32 index)
 839 {
 840         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 841         u32 next_index = bond_info->rx_hashtbl[index].src_next;
 842         u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
 843
 844         bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
 845         bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
 846
 847         if (next_index != RLB_NULL_INDEX)
 848                 bond_info->rx_hashtbl[next_index].src_prev = prev_index;
 849
 850         if (prev_index == RLB_NULL_INDEX)
 851                 return;
 852
 853         /* is prev_index pointing to the head of this list? */
 854         if (bond_info->rx_hashtbl[prev_index].src_first == index)
 855                 bond_info->rx_hashtbl[prev_index].src_first = next_index;
 856         else
 857                 bond_info->rx_hashtbl[prev_index].src_next = next_index;
 858
 859 }
 860
 861 static void rlb_delete_table_entry(struct bonding *bond, u32 index)
 862 {
 863         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 864         struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 865
 866         rlb_delete_table_entry_dst(bond, index);
 867         rlb_init_table_entry_dst(entry);
 868
 869         rlb_src_unlink(bond, index);
 870 }
 871
 872 /* add the rx_hashtbl[ip_dst_hash] entry to the list
 873  * of entries with identical ip_src_hash
 874  */
 875 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
 876 {
 877         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 878         u32 next;
 879
 880         bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
 881         next = bond_info->rx_hashtbl[ip_src_hash].src_first;
 882         bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
 883         if (next != RLB_NULL_INDEX)
 884                 bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
 885         bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
 886 }
 887
 888 /* deletes all rx_hashtbl entries with  arp->ip_src if their mac_src does
 889  * not match arp->mac_src */
 890 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
 891 {
 892         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 893         u32 ip_src_hash = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
 894         u32 index;
 895
 896         _lock_rx_hashtbl_bh(bond);
 897
 898         index = bond_info->rx_hashtbl[ip_src_hash].src_first;
 899         while (index != RLB_NULL_INDEX) {
 900                 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 901                 u32 next_index = entry->src_next;
 902                 if (entry->ip_src == arp->ip_src &&
 903                     !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
 904                                 rlb_delete_table_entry(bond, index);
 905                 index = next_index;
 906         }
 907         _unlock_rx_hashtbl_bh(bond);
 908 }
 909
 910 static int rlb_initialize(struct bonding *bond)
 911 {
 912         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 913         struct rlb_client_info  *new_hashtbl;
 914         int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 915         int i;
 916
 917         new_hashtbl = kmalloc(size, GFP_KERNEL);
 918         if (!new_hashtbl)
 919                 return -1;
 920
 921         _lock_rx_hashtbl_bh(bond);
 922
 923         bond_info->rx_hashtbl = new_hashtbl;
 924
 925         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 926
 927         for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
 928                 rlb_init_table_entry(bond_info->rx_hashtbl + i);
 929         }
 930
 931         _unlock_rx_hashtbl_bh(bond);
 932
 933         /* register to receive ARPs */
 934         bond->recv_probe = rlb_arp_recv;
 935
 936         return 0;
 937 }
 938
 939 static void rlb_deinitialize(struct bonding *bond)
 940 {
 941         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 942
 943         _lock_rx_hashtbl_bh(bond);
 944
 945         kfree(bond_info->rx_hashtbl);
 946         bond_info->rx_hashtbl = NULL;
 947         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 948
 949         _unlock_rx_hashtbl_bh(bond);
 950 }
 951
 952 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 953 {
 954         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 955         u32 curr_index;
 956
 957         _lock_rx_hashtbl_bh(bond);
 958
 959         curr_index = bond_info->rx_hashtbl_used_head;
 960         while (curr_index != RLB_NULL_INDEX) {
 961                 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 962                 u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
 963
 964                 if (curr->tag && (curr->vlan_id == vlan_id))
 965                         rlb_delete_table_entry(bond, curr_index);
 966
 967                 curr_index = next_index;
 968         }
 969
 970         _unlock_rx_hashtbl_bh(bond);
 971 }
 972
 973 /*********************** tlb/rlb shared functions *********************/
 974
 975 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
 976 {
 977         struct bonding *bond = bond_get_bond_by_slave(slave);
 978         struct learning_pkt pkt;
 979         int size = sizeof(struct learning_pkt);
 980         int i;
 981
 982         memset(&pkt, 0, size);
 983         memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
 984         memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
 985         pkt.type = cpu_to_be16(ETH_P_LOOP);
 986
 987         for (i = 0; i < MAX_LP_BURST; i++) {
 988                 struct sk_buff *skb;
 989                 char *data;
 990
 991                 skb = dev_alloc_skb(size);
 992                 if (!skb) {
 993                         return;
 994                 }
 995
 996                 data = skb_put(skb, size);
 997                 memcpy(data, &pkt, size);
 998
 999                 skb_reset_mac_header(skb);
1000                 skb->network_header = skb->mac_header + ETH_HLEN;
1001                 skb->protocol = pkt.type;
1002                 skb->priority = TC_PRIO_CONTROL;
1003                 skb->dev = slave->dev;
1004
1005                 if (bond_vlan_used(bond)) {
1006                         struct vlan_entry *vlan;
1007
1008                         vlan = bond_next_vlan(bond,
1009                                               bond->alb_info.current_alb_vlan);
1010
1011                         bond->alb_info.current_alb_vlan = vlan;
1012                         if (!vlan) {
1013                                 kfree_skb(skb);
1014                                 continue;
1015                         }
1016
1017                         skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan->vlan_id);
1018                         if (!skb) {
1019                                 pr_err("%s: Error: failed to insert VLAN tag\n",
1020                                        bond->dev->name);
1021                                 continue;
1022                         }
1023                 }
1024
1025                 dev_queue_xmit(skb);
1026         }
1027 }
1028
1029 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[])
1030 {
1031         struct net_device *dev = slave->dev;
1032         struct sockaddr s_addr;
1033
1034         if (slave->bond->params.mode == BOND_MODE_TLB) {
1035                 memcpy(dev->dev_addr, addr, dev->addr_len);
1036                 return 0;
1037         }
1038
1039         /* for rlb each slave must have a unique hw mac addresses so that */
1040         /* each slave will receive packets destined to a different mac */
1041         memcpy(s_addr.sa_data, addr, dev->addr_len);
1042         s_addr.sa_family = dev->type;
1043         if (dev_set_mac_address(dev, &s_addr)) {
1044                 pr_err("%s: Error: dev_set_mac_address of dev %s failed!\n"
1045                        "ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
1046                        slave->bond->dev->name, dev->name);
1047                 return -EOPNOTSUPP;
1048         }
1049         return 0;
1050 }
1051
1052 /*
1053  * Swap MAC addresses between two slaves.
1054  *
1055  * Called with RTNL held, and no other locks.
1056  *
1057  */
1058
1059 static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1060 {
1061         u8 tmp_mac_addr[ETH_ALEN];
1062
1063         memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
1064         alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr);
1065         alb_set_slave_mac_addr(slave2, tmp_mac_addr);
1066
1067 }
1068
1069 /*
1070  * Send learning packets after MAC address swap.
1071  *
1072  * Called with RTNL and no other locks
1073  */
1074 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1075                                 struct slave *slave2)
1076 {
1077         int slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
1078         struct slave *disabled_slave = NULL;
1079
1080         ASSERT_RTNL();
1081
1082         /* fasten the change in the switch */
1083         if (SLAVE_IS_OK(slave1)) {
1084                 alb_send_learning_packets(slave1, slave1->dev->dev_addr);
1085                 if (bond->alb_info.rlb_enabled) {
1086                         /* inform the clients that the mac address
1087                          * has changed
1088                          */
1089                         rlb_req_update_slave_clients(bond, slave1);
1090                 }
1091         } else {
1092                 disabled_slave = slave1;
1093         }
1094
1095         if (SLAVE_IS_OK(slave2)) {
1096                 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
1097                 if (bond->alb_info.rlb_enabled) {
1098                         /* inform the clients that the mac address
1099                          * has changed
1100                          */
1101                         rlb_req_update_slave_clients(bond, slave2);
1102                 }
1103         } else {
1104                 disabled_slave = slave2;
1105         }
1106
1107         if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1108                 /* A disabled slave was assigned an active mac addr */
1109                 rlb_teach_disabled_mac_on_primary(bond,
1110                                                   disabled_slave->dev->dev_addr);
1111         }
1112 }
1113
1114 /**
1115  * alb_change_hw_addr_on_detach
1116  * @bond: bonding we're working on
1117  * @slave: the slave that was just detached
1118  *
1119  * We assume that @slave was already detached from the slave list.
1120  *
1121  * If @slave's permanent hw address is different both from its current
1122  * address and from @bond's address, then somewhere in the bond there's
1123  * a slave that has @slave's permanet address as its current address.
1124  * We'll make sure that that slave no longer uses @slave's permanent address.
1125  *
1126  * Caller must hold RTNL and no other locks
1127  */
1128 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1129 {
1130         int perm_curr_diff;
1131         int perm_bond_diff;
1132         struct slave *found_slave;
1133
1134         perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1135                                                   slave->dev->dev_addr);
1136         perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1137                                                   bond->dev->dev_addr);
1138
1139         if (perm_curr_diff && perm_bond_diff) {
1140                 found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1141
1142                 if (found_slave) {
1143                         /* locking: needs RTNL and nothing else */
1144                         alb_swap_mac_addr(slave, found_slave);
1145                         alb_fasten_mac_swap(bond, slave, found_slave);
1146                 }
1147         }
1148 }
1149
1150 /**
1151  * alb_handle_addr_collision_on_attach
1152  * @bond: bonding we're working on
1153  * @slave: the slave that was just attached
1154  *
1155  * checks uniqueness of slave's mac address and handles the case the
1156  * new slave uses the bonds mac address.
1157  *
1158  * If the permanent hw address of @slave is @bond's hw address, we need to
1159  * find a different hw address to give @slave, that isn't in use by any other
1160  * slave in the bond. This address must be, of course, one of the permanent
1161  * addresses of the other slaves.
1162  *
1163  * We go over the slave list, and for each slave there we compare its
1164  * permanent hw address with the current address of all the other slaves.
1165  * If no match was found, then we've found a slave with a permanent address
1166  * that isn't used by any other slave in the bond, so we can assign it to
1167  * @slave.
1168  *
1169  * assumption: this function is called before @slave is attached to the
1170  *             bond slave list.
1171  */
1172 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1173 {
1174         struct slave *tmp_slave1, *free_mac_slave = NULL;
1175         struct slave *has_bond_addr = bond->curr_active_slave;
1176         int i;
1177
1178         if (bond->slave_cnt == 0) {
1179                 /* this is the first slave */
1180                 return 0;
1181         }
1182
1183         /* if slave's mac address differs from bond's mac address
1184          * check uniqueness of slave's mac address against the other
1185          * slaves in the bond.
1186          */
1187         if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1188                 if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1189                         return 0;
1190
1191                 /* Try setting slave mac to bond address and fall-through
1192                    to code handling that situation below... */
1193                 alb_set_slave_mac_addr(slave, bond->dev->dev_addr);
1194         }
1195
1196         /* The slave's address is equal to the address of the bond.
1197          * Search for a spare address in the bond for this slave.
1198          */
1199         bond_for_each_slave(bond, tmp_slave1, i) {
1200                 if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1201                         /* no slave has tmp_slave1's perm addr
1202                          * as its curr addr
1203                          */
1204                         free_mac_slave = tmp_slave1;
1205                         break;
1206                 }
1207
1208                 if (!has_bond_addr) {
1209                         if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1210                                                     bond->dev->dev_addr)) {
1211
1212                                 has_bond_addr = tmp_slave1;
1213                         }
1214                 }
1215         }
1216
1217         if (free_mac_slave) {
1218                 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr);
1219
1220                 pr_warning("%s: Warning: the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1221                            bond->dev->name, slave->dev->name,
1222                            free_mac_slave->dev->name);
1223
1224         } else if (has_bond_addr) {
1225                 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",
1226                        bond->dev->name, slave->dev->name);
1227                 return -EFAULT;
1228         }
1229
1230         return 0;
1231 }
1232
1233 /**
1234  * alb_set_mac_address
1235  * @bond:
1236  * @addr:
1237  *
1238  * In TLB mode all slaves are configured to the bond's hw address, but set
1239  * their dev_addr field to different addresses (based on their permanent hw
1240  * addresses).
1241  *
1242  * For each slave, this function sets the interface to the new address and then
1243  * changes its dev_addr field to its previous value.
1244  *
1245  * Unwinding assumes bond's mac address has not yet changed.
1246  */
1247 static int alb_set_mac_address(struct bonding *bond, void *addr)
1248 {
1249         struct sockaddr sa;
1250         struct slave *slave, *stop_at;
1251         char tmp_addr[ETH_ALEN];
1252         int res;
1253         int i;
1254
1255         if (bond->alb_info.rlb_enabled) {
1256                 return 0;
1257         }
1258
1259         bond_for_each_slave(bond, slave, i) {
1260                 /* save net_device's current hw address */
1261                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1262
1263                 res = dev_set_mac_address(slave->dev, addr);
1264
1265                 /* restore net_device's hw address */
1266                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1267
1268                 if (res)
1269                         goto unwind;
1270         }
1271
1272         return 0;
1273
1274 unwind:
1275         memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1276         sa.sa_family = bond->dev->type;
1277
1278         /* unwind from head to the slave that failed */
1279         stop_at = slave;
1280         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
1281                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1282                 dev_set_mac_address(slave->dev, &sa);
1283                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1284         }
1285
1286         return res;
1287 }
1288
1289 /************************ exported alb funcions ************************/
1290
1291 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1292 {
1293         int res;
1294
1295         res = tlb_initialize(bond);
1296         if (res) {
1297                 return res;
1298         }
1299
1300         if (rlb_enabled) {
1301                 bond->alb_info.rlb_enabled = 1;
1302                 /* initialize rlb */
1303                 res = rlb_initialize(bond);
1304                 if (res) {
1305                         tlb_deinitialize(bond);
1306                         return res;
1307                 }
1308         } else {
1309                 bond->alb_info.rlb_enabled = 0;
1310         }
1311
1312         return 0;
1313 }
1314
1315 void bond_alb_deinitialize(struct bonding *bond)
1316 {
1317         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1318
1319         tlb_deinitialize(bond);
1320
1321         if (bond_info->rlb_enabled) {
1322                 rlb_deinitialize(bond);
1323         }
1324 }
1325
1326 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1327 {
1328         struct bonding *bond = netdev_priv(bond_dev);
1329         struct ethhdr *eth_data;
1330         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1331         struct slave *tx_slave = NULL;
1332         static const __be32 ip_bcast = htonl(0xffffffff);
1333         int hash_size = 0;
1334         int do_tx_balance = 1;
1335         u32 hash_index = 0;
1336         const u8 *hash_start = NULL;
1337         int res = 1;
1338         struct ipv6hdr *ip6hdr;
1339
1340         skb_reset_mac_header(skb);
1341         eth_data = eth_hdr(skb);
1342
1343         /* make sure that the curr_active_slave do not change during tx
1344          */
1345         read_lock(&bond->curr_slave_lock);
1346
1347         switch (ntohs(skb->protocol)) {
1348         case ETH_P_IP: {
1349                 const struct iphdr *iph = ip_hdr(skb);
1350
1351                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast) ||
1352                     (iph->daddr == ip_bcast) ||
1353                     (iph->protocol == IPPROTO_IGMP)) {
1354                         do_tx_balance = 0;
1355                         break;
1356                 }
1357                 hash_start = (char *)&(iph->daddr);
1358                 hash_size = sizeof(iph->daddr);
1359         }
1360                 break;
1361         case ETH_P_IPV6:
1362                 /* IPv6 doesn't really use broadcast mac address, but leave
1363                  * that here just in case.
1364                  */
1365                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast)) {
1366                         do_tx_balance = 0;
1367                         break;
1368                 }
1369
1370                 /* IPv6 uses all-nodes multicast as an equivalent to
1371                  * broadcasts in IPv4.
1372                  */
1373                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1374                         do_tx_balance = 0;
1375                         break;
1376                 }
1377
1378                 /* Additianally, DAD probes should not be tx-balanced as that
1379                  * will lead to false positives for duplicate addresses and
1380                  * prevent address configuration from working.
1381                  */
1382                 ip6hdr = ipv6_hdr(skb);
1383                 if (ipv6_addr_any(&ip6hdr->saddr)) {
1384                         do_tx_balance = 0;
1385                         break;
1386                 }
1387
1388                 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1389                 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1390                 break;
1391         case ETH_P_IPX:
1392                 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1393                         /* something is wrong with this packet */
1394                         do_tx_balance = 0;
1395                         break;
1396                 }
1397
1398                 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1399                         /* The only protocol worth balancing in
1400                          * this family since it has an "ARP" like
1401                          * mechanism
1402                          */
1403                         do_tx_balance = 0;
1404                         break;
1405                 }
1406
1407                 hash_start = (char*)eth_data->h_dest;
1408                 hash_size = ETH_ALEN;
1409                 break;
1410         case ETH_P_ARP:
1411                 do_tx_balance = 0;
1412                 if (bond_info->rlb_enabled) {
1413                         tx_slave = rlb_arp_xmit(skb, bond);
1414                 }
1415                 break;
1416         default:
1417                 do_tx_balance = 0;
1418                 break;
1419         }
1420
1421         if (do_tx_balance) {
1422                 hash_index = _simple_hash(hash_start, hash_size);
1423                 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1424         }
1425
1426         if (!tx_slave) {
1427                 /* unbalanced or unassigned, send through primary */
1428                 tx_slave = bond->curr_active_slave;
1429                 bond_info->unbalanced_load += skb->len;
1430         }
1431
1432         if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1433                 if (tx_slave != bond->curr_active_slave) {
1434                         memcpy(eth_data->h_source,
1435                                tx_slave->dev->dev_addr,
1436                                ETH_ALEN);
1437                 }
1438
1439                 res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1440         } else {
1441                 if (tx_slave) {
1442                         _lock_tx_hashtbl(bond);
1443                         __tlb_clear_slave(bond, tx_slave, 0);
1444                         _unlock_tx_hashtbl(bond);
1445                 }
1446         }
1447
1448         read_unlock(&bond->curr_slave_lock);
1449
1450         if (res) {
1451                 /* no suitable interface, frame not sent */
1452                 kfree_skb(skb);
1453         }
1454         return NETDEV_TX_OK;
1455 }
1456
1457 void bond_alb_monitor(struct work_struct *work)
1458 {
1459         struct bonding *bond = container_of(work, struct bonding,
1460                                             alb_work.work);
1461         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1462         struct slave *slave;
1463         int i;
1464
1465         read_lock(&bond->lock);
1466
1467         if (bond->slave_cnt == 0) {
1468                 bond_info->tx_rebalance_counter = 0;
1469                 bond_info->lp_counter = 0;
1470                 goto re_arm;
1471         }
1472
1473         bond_info->tx_rebalance_counter++;
1474         bond_info->lp_counter++;
1475
1476         /* send learning packets */
1477         if (bond_info->lp_counter >= BOND_ALB_LP_TICKS) {
1478                 /* change of curr_active_slave involves swapping of mac addresses.
1479                  * in order to avoid this swapping from happening while
1480                  * sending the learning packets, the curr_slave_lock must be held for
1481                  * read.
1482                  */
1483                 read_lock(&bond->curr_slave_lock);
1484
1485                 bond_for_each_slave(bond, slave, i) {
1486                         alb_send_learning_packets(slave, slave->dev->dev_addr);
1487                 }
1488
1489                 read_unlock(&bond->curr_slave_lock);
1490
1491                 bond_info->lp_counter = 0;
1492         }
1493
1494         /* rebalance tx traffic */
1495         if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1496
1497                 read_lock(&bond->curr_slave_lock);
1498
1499                 bond_for_each_slave(bond, slave, i) {
1500                         tlb_clear_slave(bond, slave, 1);
1501                         if (slave == bond->curr_active_slave) {
1502                                 SLAVE_TLB_INFO(slave).load =
1503                                         bond_info->unbalanced_load /
1504                                                 BOND_TLB_REBALANCE_INTERVAL;
1505                                 bond_info->unbalanced_load = 0;
1506                         }
1507                 }
1508
1509                 read_unlock(&bond->curr_slave_lock);
1510
1511                 bond_info->tx_rebalance_counter = 0;
1512         }
1513
1514         /* handle rlb stuff */
1515         if (bond_info->rlb_enabled) {
1516                 if (bond_info->primary_is_promisc &&
1517                     (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1518
1519                         /*
1520                          * dev_set_promiscuity requires rtnl and
1521                          * nothing else.  Avoid race with bond_close.
1522                          */
1523                         read_unlock(&bond->lock);
1524                         if (!rtnl_trylock()) {
1525                                 read_lock(&bond->lock);
1526                                 goto re_arm;
1527                         }
1528
1529                         bond_info->rlb_promisc_timeout_counter = 0;
1530
1531                         /* If the primary was set to promiscuous mode
1532                          * because a slave was disabled then
1533                          * it can now leave promiscuous mode.
1534                          */
1535                         dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1536                         bond_info->primary_is_promisc = 0;
1537
1538                         rtnl_unlock();
1539                         read_lock(&bond->lock);
1540                 }
1541
1542                 if (bond_info->rlb_rebalance) {
1543                         bond_info->rlb_rebalance = 0;
1544                         rlb_rebalance(bond);
1545                 }
1546
1547                 /* check if clients need updating */
1548                 if (bond_info->rx_ntt) {
1549                         if (bond_info->rlb_update_delay_counter) {
1550                                 --bond_info->rlb_update_delay_counter;
1551                         } else {
1552                                 rlb_update_rx_clients(bond);
1553                                 if (bond_info->rlb_update_retry_counter) {
1554                                         --bond_info->rlb_update_retry_counter;
1555                                 } else {
1556                                         bond_info->rx_ntt = 0;
1557                                 }
1558                         }
1559                 }
1560         }
1561
1562 re_arm:
1563         queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1564
1565         read_unlock(&bond->lock);
1566 }
1567
1568 /* assumption: called before the slave is attached to the bond
1569  * and not locked by the bond lock
1570  */
1571 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1572 {
1573         int res;
1574
1575         res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr);
1576         if (res) {
1577                 return res;
1578         }
1579
1580         res = alb_handle_addr_collision_on_attach(bond, slave);
1581         if (res) {
1582                 return res;
1583         }
1584
1585         tlb_init_slave(slave);
1586
1587         /* order a rebalance ASAP */
1588         bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1589
1590         if (bond->alb_info.rlb_enabled) {
1591                 bond->alb_info.rlb_rebalance = 1;
1592         }
1593
1594         return 0;
1595 }
1596
1597 /*
1598  * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1599  * if necessary.
1600  *
1601  * Caller must hold RTNL and no other locks
1602  */
1603 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1604 {
1605         if (bond->slave_cnt > 1) {
1606                 alb_change_hw_addr_on_detach(bond, slave);
1607         }
1608
1609         tlb_clear_slave(bond, slave, 0);
1610
1611         if (bond->alb_info.rlb_enabled) {
1612                 bond->alb_info.next_rx_slave = NULL;
1613                 rlb_clear_slave(bond, slave);
1614         }
1615 }
1616
1617 /* Caller must hold bond lock for read */
1618 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1619 {
1620         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1621
1622         if (link == BOND_LINK_DOWN) {
1623                 tlb_clear_slave(bond, slave, 0);
1624                 if (bond->alb_info.rlb_enabled) {
1625                         rlb_clear_slave(bond, slave);
1626                 }
1627         } else if (link == BOND_LINK_UP) {
1628                 /* order a rebalance ASAP */
1629                 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1630                 if (bond->alb_info.rlb_enabled) {
1631                         bond->alb_info.rlb_rebalance = 1;
1632                         /* If the updelay module parameter is smaller than the
1633                          * forwarding delay of the switch the rebalance will
1634                          * not work because the rebalance arp replies will
1635                          * not be forwarded to the clients..
1636                          */
1637                 }
1638         }
1639 }
1640
1641 /**
1642  * bond_alb_handle_active_change - assign new curr_active_slave
1643  * @bond: our bonding struct
1644  * @new_slave: new slave to assign
1645  *
1646  * Set the bond->curr_active_slave to @new_slave and handle
1647  * mac address swapping and promiscuity changes as needed.
1648  *
1649  * If new_slave is NULL, caller must hold curr_slave_lock or
1650  * bond->lock for write.
1651  *
1652  * If new_slave is not NULL, caller must hold RTNL, bond->lock for
1653  * read and curr_slave_lock for write.  Processing here may sleep, so
1654  * no other locks may be held.
1655  */
1656 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1657         __releases(&bond->curr_slave_lock)
1658         __releases(&bond->lock)
1659         __acquires(&bond->lock)
1660         __acquires(&bond->curr_slave_lock)
1661 {
1662         struct slave *swap_slave;
1663
1664         if (bond->curr_active_slave == new_slave) {
1665                 return;
1666         }
1667
1668         if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1669                 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1670                 bond->alb_info.primary_is_promisc = 0;
1671                 bond->alb_info.rlb_promisc_timeout_counter = 0;
1672         }
1673
1674         swap_slave = bond->curr_active_slave;
1675         bond->curr_active_slave = new_slave;
1676
1677         if (!new_slave || (bond->slave_cnt == 0)) {
1678                 return;
1679         }
1680
1681         /* set the new curr_active_slave to the bonds mac address
1682          * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1683          */
1684         if (!swap_slave)
1685                 swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1686
1687         /*
1688          * Arrange for swap_slave and new_slave to temporarily be
1689          * ignored so we can mess with their MAC addresses without
1690          * fear of interference from transmit activity.
1691          */
1692         if (swap_slave) {
1693                 tlb_clear_slave(bond, swap_slave, 1);
1694         }
1695         tlb_clear_slave(bond, new_slave, 1);
1696
1697         write_unlock_bh(&bond->curr_slave_lock);
1698         read_unlock(&bond->lock);
1699
1700         ASSERT_RTNL();
1701
1702         /* curr_active_slave must be set before calling alb_swap_mac_addr */
1703         if (swap_slave) {
1704                 /* swap mac address */
1705                 alb_swap_mac_addr(swap_slave, new_slave);
1706                 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1707                 read_lock(&bond->lock);
1708         } else {
1709                 /* set the new_slave to the bond mac address */
1710                 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr);
1711                 read_lock(&bond->lock);
1712                 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1713         }
1714
1715         write_lock_bh(&bond->curr_slave_lock);
1716 }
1717
1718 /*
1719  * Called with RTNL
1720  */
1721 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1722         __acquires(&bond->lock)
1723         __releases(&bond->lock)
1724 {
1725         struct bonding *bond = netdev_priv(bond_dev);
1726         struct sockaddr *sa = addr;
1727         struct slave *swap_slave;
1728         int res;
1729
1730         if (!is_valid_ether_addr(sa->sa_data)) {
1731                 return -EADDRNOTAVAIL;
1732         }
1733
1734         res = alb_set_mac_address(bond, addr);
1735         if (res) {
1736                 return res;
1737         }
1738
1739         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1740
1741         /* If there is no curr_active_slave there is nothing else to do.
1742          * Otherwise we'll need to pass the new address to it and handle
1743          * duplications.
1744          */
1745         if (!bond->curr_active_slave) {
1746                 return 0;
1747         }
1748
1749         swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1750
1751         if (swap_slave) {
1752                 alb_swap_mac_addr(swap_slave, bond->curr_active_slave);
1753                 alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave);
1754         } else {
1755                 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr);
1756
1757                 read_lock(&bond->lock);
1758                 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1759                 if (bond->alb_info.rlb_enabled) {
1760                         /* inform clients mac address has changed */
1761                         rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1762                 }
1763                 read_unlock(&bond->lock);
1764         }
1765
1766         return 0;
1767 }
1768
1769 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1770 {
1771         if (bond->alb_info.current_alb_vlan &&
1772             (bond->alb_info.current_alb_vlan->vlan_id == vlan_id)) {
1773                 bond->alb_info.current_alb_vlan = NULL;
1774         }
1775
1776         if (bond->alb_info.rlb_enabled) {
1777                 rlb_clear_vlan(bond, vlan_id);
1778         }
1779 }
1780