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, see <http://www.gnu.org/licenses/>.
  16  *
  17  * The full GNU General Public License is included in this distribution in the
  18  * file called LICENSE.
  19  *
  20  */
  21
  22 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  23
  24 #include <linux/skbuff.h>
  25 #include <linux/netdevice.h>
  26 #include <linux/etherdevice.h>
  27 #include <linux/pkt_sched.h>
  28 #include <linux/spinlock.h>
  29 #include <linux/slab.h>
  30 #include <linux/timer.h>
  31 #include <linux/ip.h>
  32 #include <linux/ipv6.h>
  33 #include <linux/if_arp.h>
  34 #include <linux/if_ether.h>
  35 #include <linux/if_bonding.h>
  36 #include <linux/if_vlan.h>
  37 #include <linux/in.h>
  38 #include <net/ipx.h>
  39 #include <net/arp.h>
  40 #include <net/ipv6.h>
  41 #include <asm/byteorder.h>
  42 #include "bonding.h"
  43 #include "bond_alb.h"
  44
  45
  46
  47 #ifndef __long_aligned
  48 #define __long_aligned __attribute__((aligned((sizeof(long)))))
  49 #endif
  50 static const u8 mac_bcast[ETH_ALEN] __long_aligned = {
  51         0xff, 0xff, 0xff, 0xff, 0xff, 0xff
  52 };
  53 static const u8 mac_v6_allmcast[ETH_ALEN] __long_aligned = {
  54         0x33, 0x33, 0x00, 0x00, 0x00, 0x01
  55 };
  56 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
  57
  58 #pragma pack(1)
  59 struct learning_pkt {
  60         u8 mac_dst[ETH_ALEN];
  61         u8 mac_src[ETH_ALEN];
  62         __be16 type;
  63         u8 padding[ETH_ZLEN - ETH_HLEN];
  64 };
  65
  66 struct arp_pkt {
  67         __be16  hw_addr_space;
  68         __be16  prot_addr_space;
  69         u8      hw_addr_len;
  70         u8      prot_addr_len;
  71         __be16  op_code;
  72         u8      mac_src[ETH_ALEN];      /* sender hardware address */
  73         __be32  ip_src;                 /* sender IP address */
  74         u8      mac_dst[ETH_ALEN];      /* target hardware address */
  75         __be32  ip_dst;                 /* target IP address */
  76 };
  77 #pragma pack()
  78
  79 static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb)
  80 {
  81         return (struct arp_pkt *)skb_network_header(skb);
  82 }
  83
  84 /* Forward declaration */
  85 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[]);
  86 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp);
  87 static void rlb_src_unlink(struct bonding *bond, u32 index);
  88 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash,
  89                          u32 ip_dst_hash);
  90
  91 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
  92 {
  93         int i;
  94         u8 hash = 0;
  95
  96         for (i = 0; i < hash_size; i++) {
  97                 hash ^= hash_start[i];
  98         }
  99
 100         return hash;
 101 }
 102
 103 /*********************** tlb specific functions ***************************/
 104
 105 static inline void _lock_tx_hashtbl_bh(struct bonding *bond)
 106 {
 107         spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 108 }
 109
 110 static inline void _unlock_tx_hashtbl_bh(struct bonding *bond)
 111 {
 112         spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 113 }
 114
 115 static inline void _lock_tx_hashtbl(struct bonding *bond)
 116 {
 117         spin_lock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 118 }
 119
 120 static inline void _unlock_tx_hashtbl(struct bonding *bond)
 121 {
 122         spin_unlock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 123 }
 124
 125 /* Caller must hold tx_hashtbl lock */
 126 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
 127 {
 128         if (save_load) {
 129                 entry->load_history = 1 + entry->tx_bytes /
 130                                       BOND_TLB_REBALANCE_INTERVAL;
 131                 entry->tx_bytes = 0;
 132         }
 133
 134         entry->tx_slave = NULL;
 135         entry->next = TLB_NULL_INDEX;
 136         entry->prev = TLB_NULL_INDEX;
 137 }
 138
 139 static inline void tlb_init_slave(struct slave *slave)
 140 {
 141         SLAVE_TLB_INFO(slave).load = 0;
 142         SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
 143 }
 144
 145 /* Caller must hold bond lock for read, BH disabled */
 146 static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
 147                          int save_load)
 148 {
 149         struct tlb_client_info *tx_hash_table;
 150         u32 index;
 151
 152         /* clear slave from tx_hashtbl */
 153         tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
 154
 155         /* skip this if we've already freed the tx hash table */
 156         if (tx_hash_table) {
 157                 index = SLAVE_TLB_INFO(slave).head;
 158                 while (index != TLB_NULL_INDEX) {
 159                         u32 next_index = tx_hash_table[index].next;
 160                         tlb_init_table_entry(&tx_hash_table[index], save_load);
 161                         index = next_index;
 162                 }
 163         }
 164
 165         tlb_init_slave(slave);
 166 }
 167
 168 /* Caller must hold bond lock for read */
 169 static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
 170                          int save_load)
 171 {
 172         _lock_tx_hashtbl_bh(bond);
 173         __tlb_clear_slave(bond, slave, save_load);
 174         _unlock_tx_hashtbl_bh(bond);
 175 }
 176
 177 /* Must be called before starting the monitor timer */
 178 static int tlb_initialize(struct bonding *bond)
 179 {
 180         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 181         int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 182         struct tlb_client_info *new_hashtbl;
 183         int i;
 184
 185         new_hashtbl = kzalloc(size, GFP_KERNEL);
 186         if (!new_hashtbl)
 187                 return -1;
 188
 189         _lock_tx_hashtbl_bh(bond);
 190
 191         bond_info->tx_hashtbl = new_hashtbl;
 192
 193         for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
 194                 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
 195         }
 196
 197         _unlock_tx_hashtbl_bh(bond);
 198
 199         return 0;
 200 }
 201
 202 /* Must be called only after all slaves have been released */
 203 static void tlb_deinitialize(struct bonding *bond)
 204 {
 205         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 206
 207         _lock_tx_hashtbl_bh(bond);
 208
 209         kfree(bond_info->tx_hashtbl);
 210         bond_info->tx_hashtbl = NULL;
 211
 212         _unlock_tx_hashtbl_bh(bond);
 213 }
 214
 215 static long long compute_gap(struct slave *slave)
 216 {
 217         return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
 218                (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
 219 }
 220
 221 /* Caller must hold bond lock for read */
 222 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
 223 {
 224         struct slave *slave, *least_loaded;
 225         struct list_head *iter;
 226         long long max_gap;
 227
 228         least_loaded = NULL;
 229         max_gap = LLONG_MIN;
 230
 231         /* Find the slave with the largest gap */
 232         bond_for_each_slave_rcu(bond, slave, iter) {
 233                 if (SLAVE_IS_OK(slave)) {
 234                         long long gap = compute_gap(slave);
 235
 236                         if (max_gap < gap) {
 237                                 least_loaded = slave;
 238                                 max_gap = gap;
 239                         }
 240                 }
 241         }
 242
 243         return least_loaded;
 244 }
 245
 246 static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
 247                                                 u32 skb_len)
 248 {
 249         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 250         struct tlb_client_info *hash_table;
 251         struct slave *assigned_slave;
 252
 253         hash_table = bond_info->tx_hashtbl;
 254         assigned_slave = hash_table[hash_index].tx_slave;
 255         if (!assigned_slave) {
 256                 assigned_slave = tlb_get_least_loaded_slave(bond);
 257
 258                 if (assigned_slave) {
 259                         struct tlb_slave_info *slave_info =
 260                                 &(SLAVE_TLB_INFO(assigned_slave));
 261                         u32 next_index = slave_info->head;
 262
 263                         hash_table[hash_index].tx_slave = assigned_slave;
 264                         hash_table[hash_index].next = next_index;
 265                         hash_table[hash_index].prev = TLB_NULL_INDEX;
 266
 267                         if (next_index != TLB_NULL_INDEX) {
 268                                 hash_table[next_index].prev = hash_index;
 269                         }
 270
 271                         slave_info->head = hash_index;
 272                         slave_info->load +=
 273                                 hash_table[hash_index].load_history;
 274                 }
 275         }
 276
 277         if (assigned_slave) {
 278                 hash_table[hash_index].tx_bytes += skb_len;
 279         }
 280
 281         return assigned_slave;
 282 }
 283
 284 /* Caller must hold bond lock for read */
 285 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
 286                                         u32 skb_len)
 287 {
 288         struct slave *tx_slave;
 289         /*
 290          * We don't need to disable softirq here, becase
 291          * tlb_choose_channel() is only called by bond_alb_xmit()
 292          * which already has softirq disabled.
 293          */
 294         _lock_tx_hashtbl(bond);
 295         tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
 296         _unlock_tx_hashtbl(bond);
 297         return tx_slave;
 298 }
 299
 300 /*********************** rlb specific functions ***************************/
 301 static inline void _lock_rx_hashtbl_bh(struct bonding *bond)
 302 {
 303         spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 304 }
 305
 306 static inline void _unlock_rx_hashtbl_bh(struct bonding *bond)
 307 {
 308         spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 309 }
 310
 311 static inline void _lock_rx_hashtbl(struct bonding *bond)
 312 {
 313         spin_lock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 314 }
 315
 316 static inline void _unlock_rx_hashtbl(struct bonding *bond)
 317 {
 318         spin_unlock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 319 }
 320
 321 /* when an ARP REPLY is received from a client update its info
 322  * in the rx_hashtbl
 323  */
 324 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 325 {
 326         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 327         struct rlb_client_info *client_info;
 328         u32 hash_index;
 329
 330         _lock_rx_hashtbl_bh(bond);
 331
 332         hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
 333         client_info = &(bond_info->rx_hashtbl[hash_index]);
 334
 335         if ((client_info->assigned) &&
 336             (client_info->ip_src == arp->ip_dst) &&
 337             (client_info->ip_dst == arp->ip_src) &&
 338             (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) {
 339                 /* update the clients MAC address */
 340                 memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
 341                 client_info->ntt = 1;
 342                 bond_info->rx_ntt = 1;
 343         }
 344
 345         _unlock_rx_hashtbl_bh(bond);
 346 }
 347
 348 static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
 349                         struct slave *slave)
 350 {
 351         struct arp_pkt *arp, _arp;
 352
 353         if (skb->protocol != cpu_to_be16(ETH_P_ARP))
 354                 goto out;
 355
 356         arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp);
 357         if (!arp)
 358                 goto out;
 359
 360         /* We received an ARP from arp->ip_src.
 361          * We might have used this IP address previously (on the bonding host
 362          * itself or on a system that is bridged together with the bond).
 363          * However, if arp->mac_src is different than what is stored in
 364          * rx_hashtbl, some other host is now using the IP and we must prevent
 365          * sending out client updates with this IP address and the old MAC
 366          * address.
 367          * Clean up all hash table entries that have this address as ip_src but
 368          * have a different mac_src.
 369          */
 370         rlb_purge_src_ip(bond, arp);
 371
 372         if (arp->op_code == htons(ARPOP_REPLY)) {
 373                 /* update rx hash table for this ARP */
 374                 rlb_update_entry_from_arp(bond, arp);
 375                 pr_debug("Server received an ARP Reply from client\n");
 376         }
 377 out:
 378         return RX_HANDLER_ANOTHER;
 379 }
 380
 381 /* Caller must hold bond lock for read */
 382 static struct slave *rlb_next_rx_slave(struct bonding *bond)
 383 {
 384         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 385         struct slave *before = NULL, *rx_slave = NULL, *slave;
 386         struct list_head *iter;
 387         bool found = false;
 388
 389         bond_for_each_slave(bond, slave, iter) {
 390                 if (!SLAVE_IS_OK(slave))
 391                         continue;
 392                 if (!found) {
 393                         if (!before || before->speed < slave->speed)
 394                                 before = slave;
 395                 } else {
 396                         if (!rx_slave || rx_slave->speed < slave->speed)
 397                                 rx_slave = slave;
 398                 }
 399                 if (slave == bond_info->rx_slave)
 400                         found = true;
 401         }
 402         /* we didn't find anything after the current or we have something
 403          * better before and up to the current slave
 404          */
 405         if (!rx_slave || (before && rx_slave->speed < before->speed))
 406                 rx_slave = before;
 407
 408         if (rx_slave)
 409                 bond_info->rx_slave = rx_slave;
 410
 411         return rx_slave;
 412 }
 413
 414 /* Caller must hold rcu_read_lock() for read */
 415 static struct slave *__rlb_next_rx_slave(struct bonding *bond)
 416 {
 417         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 418         struct slave *before = NULL, *rx_slave = NULL, *slave;
 419         struct list_head *iter;
 420         bool found = false;
 421
 422         bond_for_each_slave_rcu(bond, slave, iter) {
 423                 if (!SLAVE_IS_OK(slave))
 424                         continue;
 425                 if (!found) {
 426                         if (!before || before->speed < slave->speed)
 427                                 before = slave;
 428                 } else {
 429                         if (!rx_slave || rx_slave->speed < slave->speed)
 430                                 rx_slave = slave;
 431                 }
 432                 if (slave == bond_info->rx_slave)
 433                         found = true;
 434         }
 435         /* we didn't find anything after the current or we have something
 436          * better before and up to the current slave
 437          */
 438         if (!rx_slave || (before && rx_slave->speed < before->speed))
 439                 rx_slave = before;
 440
 441         if (rx_slave)
 442                 bond_info->rx_slave = rx_slave;
 443
 444         return rx_slave;
 445 }
 446
 447 /* teach the switch the mac of a disabled slave
 448  * on the primary for fault tolerance
 449  *
 450  * Caller must hold bond->curr_slave_lock for write or bond lock for write
 451  */
 452 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
 453 {
 454         if (!bond->curr_active_slave) {
 455                 return;
 456         }
 457
 458         if (!bond->alb_info.primary_is_promisc) {
 459                 if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1))
 460                         bond->alb_info.primary_is_promisc = 1;
 461                 else
 462                         bond->alb_info.primary_is_promisc = 0;
 463         }
 464
 465         bond->alb_info.rlb_promisc_timeout_counter = 0;
 466
 467         alb_send_learning_packets(bond->curr_active_slave, addr);
 468 }
 469
 470 /* slave being removed should not be active at this point
 471  *
 472  * Caller must hold bond lock for read
 473  */
 474 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 475 {
 476         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 477         struct rlb_client_info *rx_hash_table;
 478         u32 index, next_index;
 479
 480         /* clear slave from rx_hashtbl */
 481         _lock_rx_hashtbl_bh(bond);
 482
 483         rx_hash_table = bond_info->rx_hashtbl;
 484         index = bond_info->rx_hashtbl_used_head;
 485         for (; index != RLB_NULL_INDEX; index = next_index) {
 486                 next_index = rx_hash_table[index].used_next;
 487                 if (rx_hash_table[index].slave == slave) {
 488                         struct slave *assigned_slave = rlb_next_rx_slave(bond);
 489
 490                         if (assigned_slave) {
 491                                 rx_hash_table[index].slave = assigned_slave;
 492                                 if (!ether_addr_equal_64bits(rx_hash_table[index].mac_dst,
 493                                                              mac_bcast)) {
 494                                         bond_info->rx_hashtbl[index].ntt = 1;
 495                                         bond_info->rx_ntt = 1;
 496                                         /* A slave has been removed from the
 497                                          * table because it is either disabled
 498                                          * or being released. We must retry the
 499                                          * update to avoid clients from not
 500                                          * being updated & disconnecting when
 501                                          * there is stress
 502                                          */
 503                                         bond_info->rlb_update_retry_counter =
 504                                                 RLB_UPDATE_RETRY;
 505                                 }
 506                         } else {  /* there is no active slave */
 507                                 rx_hash_table[index].slave = NULL;
 508                         }
 509                 }
 510         }
 511
 512         _unlock_rx_hashtbl_bh(bond);
 513
 514         write_lock_bh(&bond->curr_slave_lock);
 515
 516         if (slave != bond->curr_active_slave) {
 517                 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 518         }
 519
 520         write_unlock_bh(&bond->curr_slave_lock);
 521 }
 522
 523 static void rlb_update_client(struct rlb_client_info *client_info)
 524 {
 525         int i;
 526
 527         if (!client_info->slave) {
 528                 return;
 529         }
 530
 531         for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 532                 struct sk_buff *skb;
 533
 534                 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 535                                  client_info->ip_dst,
 536                                  client_info->slave->dev,
 537                                  client_info->ip_src,
 538                                  client_info->mac_dst,
 539                                  client_info->slave->dev->dev_addr,
 540                                  client_info->mac_dst);
 541                 if (!skb) {
 542                         pr_err("%s: Error: failed to create an ARP packet\n",
 543                                client_info->slave->bond->dev->name);
 544                         continue;
 545                 }
 546
 547                 skb->dev = client_info->slave->dev;
 548
 549                 if (client_info->vlan_id) {
 550                         skb = vlan_put_tag(skb, htons(ETH_P_8021Q), client_info->vlan_id);
 551                         if (!skb) {
 552                                 pr_err("%s: Error: failed to insert VLAN tag\n",
 553                                        client_info->slave->bond->dev->name);
 554                                 continue;
 555                         }
 556                 }
 557
 558                 arp_xmit(skb);
 559         }
 560 }
 561
 562 /* sends ARP REPLIES that update the clients that need updating */
 563 static void rlb_update_rx_clients(struct bonding *bond)
 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_bh(bond);
 570
 571         hash_index = bond_info->rx_hashtbl_used_head;
 572         for (; hash_index != RLB_NULL_INDEX;
 573              hash_index = client_info->used_next) {
 574                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 575                 if (client_info->ntt) {
 576                         rlb_update_client(client_info);
 577                         if (bond_info->rlb_update_retry_counter == 0) {
 578                                 client_info->ntt = 0;
 579                         }
 580                 }
 581         }
 582
 583         /* do not update the entries again until this counter is zero so that
 584          * not to confuse the clients.
 585          */
 586         bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 587
 588         _unlock_rx_hashtbl_bh(bond);
 589 }
 590
 591 /* The slave was assigned a new mac address - update the clients */
 592 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 593 {
 594         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 595         struct rlb_client_info *client_info;
 596         int ntt = 0;
 597         u32 hash_index;
 598
 599         _lock_rx_hashtbl_bh(bond);
 600
 601         hash_index = bond_info->rx_hashtbl_used_head;
 602         for (; hash_index != RLB_NULL_INDEX;
 603              hash_index = client_info->used_next) {
 604                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 605
 606                 if ((client_info->slave == slave) &&
 607                     !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
 608                         client_info->ntt = 1;
 609                         ntt = 1;
 610                 }
 611         }
 612
 613         // update the team's flag only after the whole iteration
 614         if (ntt) {
 615                 bond_info->rx_ntt = 1;
 616                 //fasten the change
 617                 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 618         }
 619
 620         _unlock_rx_hashtbl_bh(bond);
 621 }
 622
 623 /* mark all clients using src_ip to be updated */
 624 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
 625 {
 626         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 627         struct rlb_client_info *client_info;
 628         u32 hash_index;
 629
 630         _lock_rx_hashtbl(bond);
 631
 632         hash_index = bond_info->rx_hashtbl_used_head;
 633         for (; hash_index != RLB_NULL_INDEX;
 634              hash_index = client_info->used_next) {
 635                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 636
 637                 if (!client_info->slave) {
 638                         pr_err("%s: Error: found a client with no channel in the client's hash table\n",
 639                                bond->dev->name);
 640                         continue;
 641                 }
 642                 /*update all clients using this src_ip, that are not assigned
 643                  * to the team's address (curr_active_slave) and have a known
 644                  * unicast mac address.
 645                  */
 646                 if ((client_info->ip_src == src_ip) &&
 647                     !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
 648                                              bond->dev->dev_addr) &&
 649                     !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
 650                         client_info->ntt = 1;
 651                         bond_info->rx_ntt = 1;
 652                 }
 653         }
 654
 655         _unlock_rx_hashtbl(bond);
 656 }
 657
 658 /* Caller must hold both bond and ptr locks for read */
 659 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
 660 {
 661         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 662         struct arp_pkt *arp = arp_pkt(skb);
 663         struct slave *assigned_slave, *curr_active_slave;
 664         struct rlb_client_info *client_info;
 665         u32 hash_index = 0;
 666
 667         _lock_rx_hashtbl(bond);
 668
 669         curr_active_slave = rcu_dereference(bond->curr_active_slave);
 670
 671         hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
 672         client_info = &(bond_info->rx_hashtbl[hash_index]);
 673
 674         if (client_info->assigned) {
 675                 if ((client_info->ip_src == arp->ip_src) &&
 676                     (client_info->ip_dst == arp->ip_dst)) {
 677                         /* the entry is already assigned to this client */
 678                         if (!ether_addr_equal_64bits(arp->mac_dst, mac_bcast)) {
 679                                 /* update mac address from arp */
 680                                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 681                         }
 682                         memcpy(client_info->mac_src, arp->mac_src, ETH_ALEN);
 683
 684                         assigned_slave = client_info->slave;
 685                         if (assigned_slave) {
 686                                 _unlock_rx_hashtbl(bond);
 687                                 return assigned_slave;
 688                         }
 689                 } else {
 690                         /* the entry is already assigned to some other client,
 691                          * move the old client to primary (curr_active_slave) so
 692                          * that the new client can be assigned to this entry.
 693                          */
 694                         if (bond->curr_active_slave &&
 695                             client_info->slave != curr_active_slave) {
 696                                 client_info->slave = curr_active_slave;
 697                                 rlb_update_client(client_info);
 698                         }
 699                 }
 700         }
 701         /* assign a new slave */
 702         assigned_slave = __rlb_next_rx_slave(bond);
 703
 704         if (assigned_slave) {
 705                 if (!(client_info->assigned &&
 706                       client_info->ip_src == arp->ip_src)) {
 707                         /* ip_src is going to be updated,
 708                          * fix the src hash list
 709                          */
 710                         u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
 711                                                     sizeof(arp->ip_src));
 712                         rlb_src_unlink(bond, hash_index);
 713                         rlb_src_link(bond, hash_src, hash_index);
 714                 }
 715
 716                 client_info->ip_src = arp->ip_src;
 717                 client_info->ip_dst = arp->ip_dst;
 718                 /* arp->mac_dst is broadcast for arp reqeusts.
 719                  * will be updated with clients actual unicast mac address
 720                  * upon receiving an arp reply.
 721                  */
 722                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 723                 memcpy(client_info->mac_src, arp->mac_src, ETH_ALEN);
 724                 client_info->slave = assigned_slave;
 725
 726                 if (!ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
 727                         client_info->ntt = 1;
 728                         bond->alb_info.rx_ntt = 1;
 729                 } else {
 730                         client_info->ntt = 0;
 731                 }
 732
 733                 if (!vlan_get_tag(skb, &client_info->vlan_id))
 734                         client_info->vlan_id = 0;
 735
 736                 if (!client_info->assigned) {
 737                         u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
 738                         bond_info->rx_hashtbl_used_head = hash_index;
 739                         client_info->used_next = prev_tbl_head;
 740                         if (prev_tbl_head != RLB_NULL_INDEX) {
 741                                 bond_info->rx_hashtbl[prev_tbl_head].used_prev =
 742                                         hash_index;
 743                         }
 744                         client_info->assigned = 1;
 745                 }
 746         }
 747
 748         _unlock_rx_hashtbl(bond);
 749
 750         return assigned_slave;
 751 }
 752
 753 /* chooses (and returns) transmit channel for arp reply
 754  * does not choose channel for other arp types since they are
 755  * sent on the curr_active_slave
 756  */
 757 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 758 {
 759         struct arp_pkt *arp = arp_pkt(skb);
 760         struct slave *tx_slave = NULL;
 761
 762         /* Don't modify or load balance ARPs that do not originate locally
 763          * (e.g.,arrive via a bridge).
 764          */
 765         if (!bond_slave_has_mac_rcu(bond, arp->mac_src))
 766                 return NULL;
 767
 768         if (arp->op_code == htons(ARPOP_REPLY)) {
 769                 /* the arp must be sent on the selected
 770                 * rx channel
 771                 */
 772                 tx_slave = rlb_choose_channel(skb, bond);
 773                 if (tx_slave) {
 774                         memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
 775                 }
 776                 pr_debug("Server sent ARP Reply packet\n");
 777         } else if (arp->op_code == htons(ARPOP_REQUEST)) {
 778                 /* Create an entry in the rx_hashtbl for this client as a
 779                  * place holder.
 780                  * When the arp reply is received the entry will be updated
 781                  * with the correct unicast address of the client.
 782                  */
 783                 rlb_choose_channel(skb, bond);
 784
 785                 /* The ARP reply packets must be delayed so that
 786                  * they can cancel out the influence of the ARP request.
 787                  */
 788                 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 789
 790                 /* arp requests are broadcast and are sent on the primary
 791                  * the arp request will collapse all clients on the subnet to
 792                  * the primary slave. We must register these clients to be
 793                  * updated with their assigned mac.
 794                  */
 795                 rlb_req_update_subnet_clients(bond, arp->ip_src);
 796                 pr_debug("Server sent ARP Request packet\n");
 797         }
 798
 799         return tx_slave;
 800 }
 801
 802 /* Caller must hold bond lock for read */
 803 static void rlb_rebalance(struct bonding *bond)
 804 {
 805         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 806         struct slave *assigned_slave;
 807         struct rlb_client_info *client_info;
 808         int ntt;
 809         u32 hash_index;
 810
 811         _lock_rx_hashtbl_bh(bond);
 812
 813         ntt = 0;
 814         hash_index = bond_info->rx_hashtbl_used_head;
 815         for (; hash_index != RLB_NULL_INDEX;
 816              hash_index = client_info->used_next) {
 817                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 818                 assigned_slave = rlb_next_rx_slave(bond);
 819                 if (assigned_slave && (client_info->slave != assigned_slave)) {
 820                         client_info->slave = assigned_slave;
 821                         client_info->ntt = 1;
 822                         ntt = 1;
 823                 }
 824         }
 825
 826         /* update the team's flag only after the whole iteration */
 827         if (ntt) {
 828                 bond_info->rx_ntt = 1;
 829         }
 830         _unlock_rx_hashtbl_bh(bond);
 831 }
 832
 833 /* Caller must hold rx_hashtbl lock */
 834 static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
 835 {
 836         entry->used_next = RLB_NULL_INDEX;
 837         entry->used_prev = RLB_NULL_INDEX;
 838         entry->assigned = 0;
 839         entry->slave = NULL;
 840         entry->vlan_id = 0;
 841 }
 842 static void rlb_init_table_entry_src(struct rlb_client_info *entry)
 843 {
 844         entry->src_first = RLB_NULL_INDEX;
 845         entry->src_prev = RLB_NULL_INDEX;
 846         entry->src_next = RLB_NULL_INDEX;
 847 }
 848
 849 static void rlb_init_table_entry(struct rlb_client_info *entry)
 850 {
 851         memset(entry, 0, sizeof(struct rlb_client_info));
 852         rlb_init_table_entry_dst(entry);
 853         rlb_init_table_entry_src(entry);
 854 }
 855
 856 static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
 857 {
 858         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 859         u32 next_index = bond_info->rx_hashtbl[index].used_next;
 860         u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
 861
 862         if (index == bond_info->rx_hashtbl_used_head)
 863                 bond_info->rx_hashtbl_used_head = next_index;
 864         if (prev_index != RLB_NULL_INDEX)
 865                 bond_info->rx_hashtbl[prev_index].used_next = next_index;
 866         if (next_index != RLB_NULL_INDEX)
 867                 bond_info->rx_hashtbl[next_index].used_prev = prev_index;
 868 }
 869
 870 /* unlink a rlb hash table entry from the src list */
 871 static void rlb_src_unlink(struct bonding *bond, u32 index)
 872 {
 873         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 874         u32 next_index = bond_info->rx_hashtbl[index].src_next;
 875         u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
 876
 877         bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
 878         bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
 879
 880         if (next_index != RLB_NULL_INDEX)
 881                 bond_info->rx_hashtbl[next_index].src_prev = prev_index;
 882
 883         if (prev_index == RLB_NULL_INDEX)
 884                 return;
 885
 886         /* is prev_index pointing to the head of this list? */
 887         if (bond_info->rx_hashtbl[prev_index].src_first == index)
 888                 bond_info->rx_hashtbl[prev_index].src_first = next_index;
 889         else
 890                 bond_info->rx_hashtbl[prev_index].src_next = next_index;
 891
 892 }
 893
 894 static void rlb_delete_table_entry(struct bonding *bond, u32 index)
 895 {
 896         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 897         struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 898
 899         rlb_delete_table_entry_dst(bond, index);
 900         rlb_init_table_entry_dst(entry);
 901
 902         rlb_src_unlink(bond, index);
 903 }
 904
 905 /* add the rx_hashtbl[ip_dst_hash] entry to the list
 906  * of entries with identical ip_src_hash
 907  */
 908 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
 909 {
 910         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 911         u32 next;
 912
 913         bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
 914         next = bond_info->rx_hashtbl[ip_src_hash].src_first;
 915         bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
 916         if (next != RLB_NULL_INDEX)
 917                 bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
 918         bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
 919 }
 920
 921 /* deletes all rx_hashtbl entries with  arp->ip_src if their mac_src does
 922  * not match arp->mac_src */
 923 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
 924 {
 925         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 926         u32 ip_src_hash = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
 927         u32 index;
 928
 929         _lock_rx_hashtbl_bh(bond);
 930
 931         index = bond_info->rx_hashtbl[ip_src_hash].src_first;
 932         while (index != RLB_NULL_INDEX) {
 933                 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 934                 u32 next_index = entry->src_next;
 935                 if (entry->ip_src == arp->ip_src &&
 936                     !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
 937                                 rlb_delete_table_entry(bond, index);
 938                 index = next_index;
 939         }
 940         _unlock_rx_hashtbl_bh(bond);
 941 }
 942
 943 static int rlb_initialize(struct bonding *bond)
 944 {
 945         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 946         struct rlb_client_info  *new_hashtbl;
 947         int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 948         int i;
 949
 950         new_hashtbl = kmalloc(size, GFP_KERNEL);
 951         if (!new_hashtbl)
 952                 return -1;
 953
 954         _lock_rx_hashtbl_bh(bond);
 955
 956         bond_info->rx_hashtbl = new_hashtbl;
 957
 958         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 959
 960         for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
 961                 rlb_init_table_entry(bond_info->rx_hashtbl + i);
 962         }
 963
 964         _unlock_rx_hashtbl_bh(bond);
 965
 966         /* register to receive ARPs */
 967         bond->recv_probe = rlb_arp_recv;
 968
 969         return 0;
 970 }
 971
 972 static void rlb_deinitialize(struct bonding *bond)
 973 {
 974         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 975
 976         _lock_rx_hashtbl_bh(bond);
 977
 978         kfree(bond_info->rx_hashtbl);
 979         bond_info->rx_hashtbl = NULL;
 980         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 981
 982         _unlock_rx_hashtbl_bh(bond);
 983 }
 984
 985 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 986 {
 987         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 988         u32 curr_index;
 989
 990         _lock_rx_hashtbl_bh(bond);
 991
 992         curr_index = bond_info->rx_hashtbl_used_head;
 993         while (curr_index != RLB_NULL_INDEX) {
 994                 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 995                 u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
 996
 997                 if (curr->vlan_id == vlan_id)
 998                         rlb_delete_table_entry(bond, curr_index);
 999
1000                 curr_index = next_index;
1001         }
1002
1003         _unlock_rx_hashtbl_bh(bond);
1004 }
1005
1006 /*********************** tlb/rlb shared functions *********************/
1007
1008 static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
1009                             u16 vid)
1010 {
1011         struct learning_pkt pkt;
1012         struct sk_buff *skb;
1013         int size = sizeof(struct learning_pkt);
1014         char *data;
1015
1016         memset(&pkt, 0, size);
1017         memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
1018         memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
1019         pkt.type = cpu_to_be16(ETH_P_LOOP);
1020
1021         skb = dev_alloc_skb(size);
1022         if (!skb)
1023                 return;
1024
1025         data = skb_put(skb, size);
1026         memcpy(data, &pkt, size);
1027
1028         skb_reset_mac_header(skb);
1029         skb->network_header = skb->mac_header + ETH_HLEN;
1030         skb->protocol = pkt.type;
1031         skb->priority = TC_PRIO_CONTROL;
1032         skb->dev = slave->dev;
1033
1034         if (vid) {
1035                 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vid);
1036                 if (!skb) {
1037                         pr_err("%s: Error: failed to insert VLAN tag\n",
1038                                slave->bond->dev->name);
1039                         return;
1040                 }
1041         }
1042
1043         dev_queue_xmit(skb);
1044 }
1045
1046
1047 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
1048 {
1049         struct bonding *bond = bond_get_bond_by_slave(slave);
1050         struct net_device *upper;
1051         struct list_head *iter;
1052
1053         /* send untagged */
1054         alb_send_lp_vid(slave, mac_addr, 0);
1055
1056         /* loop through vlans and send one packet for each */
1057         rcu_read_lock();
1058         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
1059                 if (upper->priv_flags & IFF_802_1Q_VLAN)
1060                         alb_send_lp_vid(slave, mac_addr,
1061                                         vlan_dev_vlan_id(upper));
1062         }
1063         rcu_read_unlock();
1064 }
1065
1066 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[])
1067 {
1068         struct net_device *dev = slave->dev;
1069         struct sockaddr s_addr;
1070
1071         if (slave->bond->params.mode == BOND_MODE_TLB) {
1072                 memcpy(dev->dev_addr, addr, dev->addr_len);
1073                 return 0;
1074         }
1075
1076         /* for rlb each slave must have a unique hw mac addresses so that */
1077         /* each slave will receive packets destined to a different mac */
1078         memcpy(s_addr.sa_data, addr, dev->addr_len);
1079         s_addr.sa_family = dev->type;
1080         if (dev_set_mac_address(dev, &s_addr)) {
1081                 pr_err("%s: Error: dev_set_mac_address of dev %s failed!\n"
1082                        "ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
1083                        slave->bond->dev->name, dev->name);
1084                 return -EOPNOTSUPP;
1085         }
1086         return 0;
1087 }
1088
1089 /*
1090  * Swap MAC addresses between two slaves.
1091  *
1092  * Called with RTNL held, and no other locks.
1093  *
1094  */
1095
1096 static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1097 {
1098         u8 tmp_mac_addr[ETH_ALEN];
1099
1100         memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
1101         alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr);
1102         alb_set_slave_mac_addr(slave2, tmp_mac_addr);
1103
1104 }
1105
1106 /*
1107  * Send learning packets after MAC address swap.
1108  *
1109  * Called with RTNL and no other locks
1110  */
1111 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1112                                 struct slave *slave2)
1113 {
1114         int slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
1115         struct slave *disabled_slave = NULL;
1116
1117         ASSERT_RTNL();
1118
1119         /* fasten the change in the switch */
1120         if (SLAVE_IS_OK(slave1)) {
1121                 alb_send_learning_packets(slave1, slave1->dev->dev_addr);
1122                 if (bond->alb_info.rlb_enabled) {
1123                         /* inform the clients that the mac address
1124                          * has changed
1125                          */
1126                         rlb_req_update_slave_clients(bond, slave1);
1127                 }
1128         } else {
1129                 disabled_slave = slave1;
1130         }
1131
1132         if (SLAVE_IS_OK(slave2)) {
1133                 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
1134                 if (bond->alb_info.rlb_enabled) {
1135                         /* inform the clients that the mac address
1136                          * has changed
1137                          */
1138                         rlb_req_update_slave_clients(bond, slave2);
1139                 }
1140         } else {
1141                 disabled_slave = slave2;
1142         }
1143
1144         if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1145                 /* A disabled slave was assigned an active mac addr */
1146                 rlb_teach_disabled_mac_on_primary(bond,
1147                                                   disabled_slave->dev->dev_addr);
1148         }
1149 }
1150
1151 /**
1152  * alb_change_hw_addr_on_detach
1153  * @bond: bonding we're working on
1154  * @slave: the slave that was just detached
1155  *
1156  * We assume that @slave was already detached from the slave list.
1157  *
1158  * If @slave's permanent hw address is different both from its current
1159  * address and from @bond's address, then somewhere in the bond there's
1160  * a slave that has @slave's permanet address as its current address.
1161  * We'll make sure that that slave no longer uses @slave's permanent address.
1162  *
1163  * Caller must hold RTNL and no other locks
1164  */
1165 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1166 {
1167         int perm_curr_diff;
1168         int perm_bond_diff;
1169         struct slave *found_slave;
1170
1171         perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1172                                                   slave->dev->dev_addr);
1173         perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1174                                                   bond->dev->dev_addr);
1175
1176         if (perm_curr_diff && perm_bond_diff) {
1177                 found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1178
1179                 if (found_slave) {
1180                         /* locking: needs RTNL and nothing else */
1181                         alb_swap_mac_addr(slave, found_slave);
1182                         alb_fasten_mac_swap(bond, slave, found_slave);
1183                 }
1184         }
1185 }
1186
1187 /**
1188  * alb_handle_addr_collision_on_attach
1189  * @bond: bonding we're working on
1190  * @slave: the slave that was just attached
1191  *
1192  * checks uniqueness of slave's mac address and handles the case the
1193  * new slave uses the bonds mac address.
1194  *
1195  * If the permanent hw address of @slave is @bond's hw address, we need to
1196  * find a different hw address to give @slave, that isn't in use by any other
1197  * slave in the bond. This address must be, of course, one of the permanent
1198  * addresses of the other slaves.
1199  *
1200  * We go over the slave list, and for each slave there we compare its
1201  * permanent hw address with the current address of all the other slaves.
1202  * If no match was found, then we've found a slave with a permanent address
1203  * that isn't used by any other slave in the bond, so we can assign it to
1204  * @slave.
1205  *
1206  * assumption: this function is called before @slave is attached to the
1207  *             bond slave list.
1208  */
1209 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1210 {
1211         struct slave *has_bond_addr = bond->curr_active_slave;
1212         struct slave *tmp_slave1, *free_mac_slave = NULL;
1213         struct list_head *iter;
1214
1215         if (!bond_has_slaves(bond)) {
1216                 /* this is the first slave */
1217                 return 0;
1218         }
1219
1220         /* if slave's mac address differs from bond's mac address
1221          * check uniqueness of slave's mac address against the other
1222          * slaves in the bond.
1223          */
1224         if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1225                 if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1226                         return 0;
1227
1228                 /* Try setting slave mac to bond address and fall-through
1229                    to code handling that situation below... */
1230                 alb_set_slave_mac_addr(slave, bond->dev->dev_addr);
1231         }
1232
1233         /* The slave's address is equal to the address of the bond.
1234          * Search for a spare address in the bond for this slave.
1235          */
1236         bond_for_each_slave(bond, tmp_slave1, iter) {
1237                 if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1238                         /* no slave has tmp_slave1's perm addr
1239                          * as its curr addr
1240                          */
1241                         free_mac_slave = tmp_slave1;
1242                         break;
1243                 }
1244
1245                 if (!has_bond_addr) {
1246                         if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1247                                                     bond->dev->dev_addr)) {
1248
1249                                 has_bond_addr = tmp_slave1;
1250                         }
1251                 }
1252         }
1253
1254         if (free_mac_slave) {
1255                 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr);
1256
1257                 pr_warning("%s: Warning: the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1258                            bond->dev->name, slave->dev->name,
1259                            free_mac_slave->dev->name);
1260
1261         } else if (has_bond_addr) {
1262                 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",
1263                        bond->dev->name, slave->dev->name);
1264                 return -EFAULT;
1265         }
1266
1267         return 0;
1268 }
1269
1270 /**
1271  * alb_set_mac_address
1272  * @bond:
1273  * @addr:
1274  *
1275  * In TLB mode all slaves are configured to the bond's hw address, but set
1276  * their dev_addr field to different addresses (based on their permanent hw
1277  * addresses).
1278  *
1279  * For each slave, this function sets the interface to the new address and then
1280  * changes its dev_addr field to its previous value.
1281  *
1282  * Unwinding assumes bond's mac address has not yet changed.
1283  */
1284 static int alb_set_mac_address(struct bonding *bond, void *addr)
1285 {
1286         struct slave *slave, *rollback_slave;
1287         struct list_head *iter;
1288         struct sockaddr sa;
1289         char tmp_addr[ETH_ALEN];
1290         int res;
1291
1292         if (bond->alb_info.rlb_enabled)
1293                 return 0;
1294
1295         bond_for_each_slave(bond, slave, iter) {
1296                 /* save net_device's current hw address */
1297                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1298
1299                 res = dev_set_mac_address(slave->dev, addr);
1300
1301                 /* restore net_device's hw address */
1302                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1303
1304                 if (res)
1305                         goto unwind;
1306         }
1307
1308         return 0;
1309
1310 unwind:
1311         memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1312         sa.sa_family = bond->dev->type;
1313
1314         /* unwind from head to the slave that failed */
1315         bond_for_each_slave(bond, rollback_slave, iter) {
1316                 if (rollback_slave == slave)
1317                         break;
1318                 memcpy(tmp_addr, rollback_slave->dev->dev_addr, ETH_ALEN);
1319                 dev_set_mac_address(rollback_slave->dev, &sa);
1320                 memcpy(rollback_slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1321         }
1322
1323         return res;
1324 }
1325
1326 /************************ exported alb funcions ************************/
1327
1328 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1329 {
1330         int res;
1331
1332         res = tlb_initialize(bond);
1333         if (res) {
1334                 return res;
1335         }
1336
1337         if (rlb_enabled) {
1338                 bond->alb_info.rlb_enabled = 1;
1339                 /* initialize rlb */
1340                 res = rlb_initialize(bond);
1341                 if (res) {
1342                         tlb_deinitialize(bond);
1343                         return res;
1344                 }
1345         } else {
1346                 bond->alb_info.rlb_enabled = 0;
1347         }
1348
1349         return 0;
1350 }
1351
1352 void bond_alb_deinitialize(struct bonding *bond)
1353 {
1354         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1355
1356         tlb_deinitialize(bond);
1357
1358         if (bond_info->rlb_enabled) {
1359                 rlb_deinitialize(bond);
1360         }
1361 }
1362
1363 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1364 {
1365         struct bonding *bond = netdev_priv(bond_dev);
1366         struct ethhdr *eth_data;
1367         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1368         struct slave *tx_slave = NULL;
1369         static const __be32 ip_bcast = htonl(0xffffffff);
1370         int hash_size = 0;
1371         int do_tx_balance = 1;
1372         u32 hash_index = 0;
1373         const u8 *hash_start = NULL;
1374         int res = 1;
1375         struct ipv6hdr *ip6hdr;
1376
1377         skb_reset_mac_header(skb);
1378         eth_data = eth_hdr(skb);
1379
1380         switch (ntohs(skb->protocol)) {
1381         case ETH_P_IP: {
1382                 const struct iphdr *iph = ip_hdr(skb);
1383
1384                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast) ||
1385                     (iph->daddr == ip_bcast) ||
1386                     (iph->protocol == IPPROTO_IGMP)) {
1387                         do_tx_balance = 0;
1388                         break;
1389                 }
1390                 hash_start = (char *)&(iph->daddr);
1391                 hash_size = sizeof(iph->daddr);
1392         }
1393                 break;
1394         case ETH_P_IPV6:
1395                 /* IPv6 doesn't really use broadcast mac address, but leave
1396                  * that here just in case.
1397                  */
1398                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast)) {
1399                         do_tx_balance = 0;
1400                         break;
1401                 }
1402
1403                 /* IPv6 uses all-nodes multicast as an equivalent to
1404                  * broadcasts in IPv4.
1405                  */
1406                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1407                         do_tx_balance = 0;
1408                         break;
1409                 }
1410
1411                 /* Additianally, DAD probes should not be tx-balanced as that
1412                  * will lead to false positives for duplicate addresses and
1413                  * prevent address configuration from working.
1414                  */
1415                 ip6hdr = ipv6_hdr(skb);
1416                 if (ipv6_addr_any(&ip6hdr->saddr)) {
1417                         do_tx_balance = 0;
1418                         break;
1419                 }
1420
1421                 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1422                 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1423                 break;
1424         case ETH_P_IPX:
1425                 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1426                         /* something is wrong with this packet */
1427                         do_tx_balance = 0;
1428                         break;
1429                 }
1430
1431                 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1432                         /* The only protocol worth balancing in
1433                          * this family since it has an "ARP" like
1434                          * mechanism
1435                          */
1436                         do_tx_balance = 0;
1437                         break;
1438                 }
1439
1440                 hash_start = (char*)eth_data->h_dest;
1441                 hash_size = ETH_ALEN;
1442                 break;
1443         case ETH_P_ARP:
1444                 do_tx_balance = 0;
1445                 if (bond_info->rlb_enabled) {
1446                         tx_slave = rlb_arp_xmit(skb, bond);
1447                 }
1448                 break;
1449         default:
1450                 do_tx_balance = 0;
1451                 break;
1452         }
1453
1454         if (do_tx_balance) {
1455                 hash_index = _simple_hash(hash_start, hash_size);
1456                 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1457         }
1458
1459         if (!tx_slave) {
1460                 /* unbalanced or unassigned, send through primary */
1461                 tx_slave = rcu_dereference(bond->curr_active_slave);
1462                 bond_info->unbalanced_load += skb->len;
1463         }
1464
1465         if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1466                 if (tx_slave != rcu_dereference(bond->curr_active_slave)) {
1467                         memcpy(eth_data->h_source,
1468                                tx_slave->dev->dev_addr,
1469                                ETH_ALEN);
1470                 }
1471
1472                 res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1473         } else {
1474                 if (tx_slave) {
1475                         _lock_tx_hashtbl(bond);
1476                         __tlb_clear_slave(bond, tx_slave, 0);
1477                         _unlock_tx_hashtbl(bond);
1478                 }
1479         }
1480
1481         if (res) {
1482                 /* no suitable interface, frame not sent */
1483                 kfree_skb(skb);
1484         }
1485
1486         return NETDEV_TX_OK;
1487 }
1488
1489 void bond_alb_monitor(struct work_struct *work)
1490 {
1491         struct bonding *bond = container_of(work, struct bonding,
1492                                             alb_work.work);
1493         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1494         struct list_head *iter;
1495         struct slave *slave;
1496
1497         read_lock(&bond->lock);
1498
1499         if (!bond_has_slaves(bond)) {
1500                 bond_info->tx_rebalance_counter = 0;
1501                 bond_info->lp_counter = 0;
1502                 goto re_arm;
1503         }
1504
1505         bond_info->tx_rebalance_counter++;
1506         bond_info->lp_counter++;
1507
1508         /* send learning packets */
1509         if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1510                 /* change of curr_active_slave involves swapping of mac addresses.
1511                  * in order to avoid this swapping from happening while
1512                  * sending the learning packets, the curr_slave_lock must be held for
1513                  * read.
1514                  */
1515                 read_lock(&bond->curr_slave_lock);
1516
1517                 bond_for_each_slave(bond, slave, iter)
1518                         alb_send_learning_packets(slave, slave->dev->dev_addr);
1519
1520                 read_unlock(&bond->curr_slave_lock);
1521
1522                 bond_info->lp_counter = 0;
1523         }
1524
1525         /* rebalance tx traffic */
1526         if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1527
1528                 read_lock(&bond->curr_slave_lock);
1529
1530                 bond_for_each_slave(bond, slave, iter) {
1531                         tlb_clear_slave(bond, slave, 1);
1532                         if (slave == bond->curr_active_slave) {
1533                                 SLAVE_TLB_INFO(slave).load =
1534                                         bond_info->unbalanced_load /
1535                                                 BOND_TLB_REBALANCE_INTERVAL;
1536                                 bond_info->unbalanced_load = 0;
1537                         }
1538                 }
1539
1540                 read_unlock(&bond->curr_slave_lock);
1541
1542                 bond_info->tx_rebalance_counter = 0;
1543         }
1544
1545         /* handle rlb stuff */
1546         if (bond_info->rlb_enabled) {
1547                 if (bond_info->primary_is_promisc &&
1548                     (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1549
1550                         /*
1551                          * dev_set_promiscuity requires rtnl and
1552                          * nothing else.  Avoid race with bond_close.
1553                          */
1554                         read_unlock(&bond->lock);
1555                         if (!rtnl_trylock()) {
1556                                 read_lock(&bond->lock);
1557                                 goto re_arm;
1558                         }
1559
1560                         bond_info->rlb_promisc_timeout_counter = 0;
1561
1562                         /* If the primary was set to promiscuous mode
1563                          * because a slave was disabled then
1564                          * it can now leave promiscuous mode.
1565                          */
1566                         dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1567                         bond_info->primary_is_promisc = 0;
1568
1569                         rtnl_unlock();
1570                         read_lock(&bond->lock);
1571                 }
1572
1573                 if (bond_info->rlb_rebalance) {
1574                         bond_info->rlb_rebalance = 0;
1575                         rlb_rebalance(bond);
1576                 }
1577
1578                 /* check if clients need updating */
1579                 if (bond_info->rx_ntt) {
1580                         if (bond_info->rlb_update_delay_counter) {
1581                                 --bond_info->rlb_update_delay_counter;
1582                         } else {
1583                                 rlb_update_rx_clients(bond);
1584                                 if (bond_info->rlb_update_retry_counter) {
1585                                         --bond_info->rlb_update_retry_counter;
1586                                 } else {
1587                                         bond_info->rx_ntt = 0;
1588                                 }
1589                         }
1590                 }
1591         }
1592
1593 re_arm:
1594         queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1595
1596         read_unlock(&bond->lock);
1597 }
1598
1599 /* assumption: called before the slave is attached to the bond
1600  * and not locked by the bond lock
1601  */
1602 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1603 {
1604         int res;
1605
1606         res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr);
1607         if (res) {
1608                 return res;
1609         }
1610
1611         res = alb_handle_addr_collision_on_attach(bond, slave);
1612         if (res) {
1613                 return res;
1614         }
1615
1616         tlb_init_slave(slave);
1617
1618         /* order a rebalance ASAP */
1619         bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1620
1621         if (bond->alb_info.rlb_enabled) {
1622                 bond->alb_info.rlb_rebalance = 1;
1623         }
1624
1625         return 0;
1626 }
1627
1628 /*
1629  * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1630  * if necessary.
1631  *
1632  * Caller must hold RTNL and no other locks
1633  */
1634 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1635 {
1636         if (bond_has_slaves(bond))
1637                 alb_change_hw_addr_on_detach(bond, slave);
1638
1639         tlb_clear_slave(bond, slave, 0);
1640
1641         if (bond->alb_info.rlb_enabled) {
1642                 bond->alb_info.rx_slave = NULL;
1643                 rlb_clear_slave(bond, slave);
1644         }
1645 }
1646
1647 /* Caller must hold bond lock for read */
1648 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1649 {
1650         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1651
1652         if (link == BOND_LINK_DOWN) {
1653                 tlb_clear_slave(bond, slave, 0);
1654                 if (bond->alb_info.rlb_enabled) {
1655                         rlb_clear_slave(bond, slave);
1656                 }
1657         } else if (link == BOND_LINK_UP) {
1658                 /* order a rebalance ASAP */
1659                 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1660                 if (bond->alb_info.rlb_enabled) {
1661                         bond->alb_info.rlb_rebalance = 1;
1662                         /* If the updelay module parameter is smaller than the
1663                          * forwarding delay of the switch the rebalance will
1664                          * not work because the rebalance arp replies will
1665                          * not be forwarded to the clients..
1666                          */
1667                 }
1668         }
1669 }
1670
1671 /**
1672  * bond_alb_handle_active_change - assign new curr_active_slave
1673  * @bond: our bonding struct
1674  * @new_slave: new slave to assign
1675  *
1676  * Set the bond->curr_active_slave to @new_slave and handle
1677  * mac address swapping and promiscuity changes as needed.
1678  *
1679  * If new_slave is NULL, caller must hold curr_slave_lock or
1680  * bond->lock for write.
1681  *
1682  * If new_slave is not NULL, caller must hold RTNL, bond->lock for
1683  * read and curr_slave_lock for write.  Processing here may sleep, so
1684  * no other locks may be held.
1685  */
1686 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1687         __releases(&bond->curr_slave_lock)
1688         __releases(&bond->lock)
1689         __acquires(&bond->lock)
1690         __acquires(&bond->curr_slave_lock)
1691 {
1692         struct slave *swap_slave;
1693
1694         if (bond->curr_active_slave == new_slave)
1695                 return;
1696
1697         if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1698                 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1699                 bond->alb_info.primary_is_promisc = 0;
1700                 bond->alb_info.rlb_promisc_timeout_counter = 0;
1701         }
1702
1703         swap_slave = bond->curr_active_slave;
1704         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1705
1706         if (!new_slave || !bond_has_slaves(bond))
1707                 return;
1708
1709         /* set the new curr_active_slave to the bonds mac address
1710          * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1711          */
1712         if (!swap_slave)
1713                 swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1714
1715         /*
1716          * Arrange for swap_slave and new_slave to temporarily be
1717          * ignored so we can mess with their MAC addresses without
1718          * fear of interference from transmit activity.
1719          */
1720         if (swap_slave)
1721                 tlb_clear_slave(bond, swap_slave, 1);
1722         tlb_clear_slave(bond, new_slave, 1);
1723
1724         write_unlock_bh(&bond->curr_slave_lock);
1725         read_unlock(&bond->lock);
1726
1727         ASSERT_RTNL();
1728
1729         /* in TLB mode, the slave might flip down/up with the old dev_addr,
1730          * and thus filter bond->dev_addr's packets, so force bond's mac
1731          */
1732         if (bond->params.mode == BOND_MODE_TLB) {
1733                 struct sockaddr sa;
1734                 u8 tmp_addr[ETH_ALEN];
1735
1736                 memcpy(tmp_addr, new_slave->dev->dev_addr, ETH_ALEN);
1737
1738                 memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1739                 sa.sa_family = bond->dev->type;
1740                 /* we don't care if it can't change its mac, best effort */
1741                 dev_set_mac_address(new_slave->dev, &sa);
1742
1743                 memcpy(new_slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1744         }
1745
1746         /* curr_active_slave must be set before calling alb_swap_mac_addr */
1747         if (swap_slave) {
1748                 /* swap mac address */
1749                 alb_swap_mac_addr(swap_slave, new_slave);
1750                 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1751                 read_lock(&bond->lock);
1752         } else {
1753                 /* set the new_slave to the bond mac address */
1754                 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr);
1755                 read_lock(&bond->lock);
1756                 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1757         }
1758
1759         write_lock_bh(&bond->curr_slave_lock);
1760 }
1761
1762 /*
1763  * Called with RTNL
1764  */
1765 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1766         __acquires(&bond->lock)
1767         __releases(&bond->lock)
1768 {
1769         struct bonding *bond = netdev_priv(bond_dev);
1770         struct sockaddr *sa = addr;
1771         struct slave *swap_slave;
1772         int res;
1773
1774         if (!is_valid_ether_addr(sa->sa_data)) {
1775                 return -EADDRNOTAVAIL;
1776         }
1777
1778         res = alb_set_mac_address(bond, addr);
1779         if (res) {
1780                 return res;
1781         }
1782
1783         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1784
1785         /* If there is no curr_active_slave there is nothing else to do.
1786          * Otherwise we'll need to pass the new address to it and handle
1787          * duplications.
1788          */
1789         if (!bond->curr_active_slave) {
1790                 return 0;
1791         }
1792
1793         swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1794
1795         if (swap_slave) {
1796                 alb_swap_mac_addr(swap_slave, bond->curr_active_slave);
1797                 alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave);
1798         } else {
1799                 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr);
1800
1801                 read_lock(&bond->lock);
1802                 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1803                 if (bond->alb_info.rlb_enabled) {
1804                         /* inform clients mac address has changed */
1805                         rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1806                 }
1807                 read_unlock(&bond->lock);
1808         }
1809
1810         return 0;
1811 }
1812
1813 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1814 {
1815         if (bond->alb_info.rlb_enabled) {
1816                 rlb_clear_vlan(bond, vlan_id);
1817         }
1818 }
1819