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 #include <linux/skbuff.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/etherdevice.h>
  25 #include <linux/pkt_sched.h>
  26 #include <linux/spinlock.h>
  27 #include <linux/slab.h>
  28 #include <linux/timer.h>
  29 #include <linux/ip.h>
  30 #include <linux/ipv6.h>
  31 #include <linux/if_arp.h>
  32 #include <linux/if_ether.h>
  33 #include <linux/if_bonding.h>
  34 #include <linux/if_vlan.h>
  35 #include <linux/in.h>
  36 #include <net/ipx.h>
  37 #include <net/arp.h>
  38 #include <net/ipv6.h>
  39 #include <asm/byteorder.h>
  40 #include <net/bonding.h>
  41 #include <net/bond_alb.h>
  42
  43
  44
  45 static const u8 mac_bcast[ETH_ALEN + 2] __long_aligned = {
  46         0xff, 0xff, 0xff, 0xff, 0xff, 0xff
  47 };
  48 static const u8 mac_v6_allmcast[ETH_ALEN + 2] __long_aligned = {
  49         0x33, 0x33, 0x00, 0x00, 0x00, 0x01
  50 };
  51 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
  52
  53 #pragma pack(1)
  54 struct learning_pkt {
  55         u8 mac_dst[ETH_ALEN];
  56         u8 mac_src[ETH_ALEN];
  57         __be16 type;
  58         u8 padding[ETH_ZLEN - ETH_HLEN];
  59 };
  60
  61 struct arp_pkt {
  62         __be16  hw_addr_space;
  63         __be16  prot_addr_space;
  64         u8      hw_addr_len;
  65         u8      prot_addr_len;
  66         __be16  op_code;
  67         u8      mac_src[ETH_ALEN];      /* sender hardware address */
  68         __be32  ip_src;                 /* sender IP address */
  69         u8      mac_dst[ETH_ALEN];      /* target hardware address */
  70         __be32  ip_dst;                 /* target IP address */
  71 };
  72 #pragma pack()
  73
  74 /* Forward declaration */
  75 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
  76                                       bool strict_match);
  77 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp);
  78 static void rlb_src_unlink(struct bonding *bond, u32 index);
  79 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash,
  80                          u32 ip_dst_hash);
  81
  82 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
  83 {
  84         int i;
  85         u8 hash = 0;
  86
  87         for (i = 0; i < hash_size; i++)
  88                 hash ^= hash_start[i];
  89
  90         return hash;
  91 }
  92
  93 /*********************** tlb specific functions ***************************/
  94
  95 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
  96 {
  97         if (save_load) {
  98                 entry->load_history = 1 + entry->tx_bytes /
  99                                       BOND_TLB_REBALANCE_INTERVAL;
 100                 entry->tx_bytes = 0;
 101         }
 102
 103         entry->tx_slave = NULL;
 104         entry->next = TLB_NULL_INDEX;
 105         entry->prev = TLB_NULL_INDEX;
 106 }
 107
 108 static inline void tlb_init_slave(struct slave *slave)
 109 {
 110         SLAVE_TLB_INFO(slave).load = 0;
 111         SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
 112 }
 113
 114 static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
 115                          int save_load)
 116 {
 117         struct tlb_client_info *tx_hash_table;
 118         u32 index;
 119
 120         /* clear slave from tx_hashtbl */
 121         tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
 122
 123         /* skip this if we've already freed the tx hash table */
 124         if (tx_hash_table) {
 125                 index = SLAVE_TLB_INFO(slave).head;
 126                 while (index != TLB_NULL_INDEX) {
 127                         u32 next_index = tx_hash_table[index].next;
 128                         tlb_init_table_entry(&tx_hash_table[index], save_load);
 129                         index = next_index;
 130                 }
 131         }
 132
 133         tlb_init_slave(slave);
 134 }
 135
 136 static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
 137                          int save_load)
 138 {
 139         spin_lock_bh(&bond->mode_lock);
 140         __tlb_clear_slave(bond, slave, save_load);
 141         spin_unlock_bh(&bond->mode_lock);
 142 }
 143
 144 /* Must be called before starting the monitor timer */
 145 static int tlb_initialize(struct bonding *bond)
 146 {
 147         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 148         int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 149         struct tlb_client_info *new_hashtbl;
 150         int i;
 151
 152         new_hashtbl = kzalloc(size, GFP_KERNEL);
 153         if (!new_hashtbl)
 154                 return -ENOMEM;
 155
 156         spin_lock_bh(&bond->mode_lock);
 157
 158         bond_info->tx_hashtbl = new_hashtbl;
 159
 160         for (i = 0; i < TLB_HASH_TABLE_SIZE; i++)
 161                 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
 162
 163         spin_unlock_bh(&bond->mode_lock);
 164
 165         return 0;
 166 }
 167
 168 /* Must be called only after all slaves have been released */
 169 static void tlb_deinitialize(struct bonding *bond)
 170 {
 171         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 172
 173         spin_lock_bh(&bond->mode_lock);
 174
 175         kfree(bond_info->tx_hashtbl);
 176         bond_info->tx_hashtbl = NULL;
 177
 178         spin_unlock_bh(&bond->mode_lock);
 179 }
 180
 181 static long long compute_gap(struct slave *slave)
 182 {
 183         return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
 184                (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
 185 }
 186
 187 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
 188 {
 189         struct slave *slave, *least_loaded;
 190         struct list_head *iter;
 191         long long max_gap;
 192
 193         least_loaded = NULL;
 194         max_gap = LLONG_MIN;
 195
 196         /* Find the slave with the largest gap */
 197         bond_for_each_slave_rcu(bond, slave, iter) {
 198                 if (bond_slave_can_tx(slave)) {
 199                         long long gap = compute_gap(slave);
 200
 201                         if (max_gap < gap) {
 202                                 least_loaded = slave;
 203                                 max_gap = gap;
 204                         }
 205                 }
 206         }
 207
 208         return least_loaded;
 209 }
 210
 211 static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
 212                                                 u32 skb_len)
 213 {
 214         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 215         struct tlb_client_info *hash_table;
 216         struct slave *assigned_slave;
 217
 218         hash_table = bond_info->tx_hashtbl;
 219         assigned_slave = hash_table[hash_index].tx_slave;
 220         if (!assigned_slave) {
 221                 assigned_slave = tlb_get_least_loaded_slave(bond);
 222
 223                 if (assigned_slave) {
 224                         struct tlb_slave_info *slave_info =
 225                                 &(SLAVE_TLB_INFO(assigned_slave));
 226                         u32 next_index = slave_info->head;
 227
 228                         hash_table[hash_index].tx_slave = assigned_slave;
 229                         hash_table[hash_index].next = next_index;
 230                         hash_table[hash_index].prev = TLB_NULL_INDEX;
 231
 232                         if (next_index != TLB_NULL_INDEX)
 233                                 hash_table[next_index].prev = hash_index;
 234
 235                         slave_info->head = hash_index;
 236                         slave_info->load +=
 237                                 hash_table[hash_index].load_history;
 238                 }
 239         }
 240
 241         if (assigned_slave)
 242                 hash_table[hash_index].tx_bytes += skb_len;
 243
 244         return assigned_slave;
 245 }
 246
 247 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
 248                                         u32 skb_len)
 249 {
 250         struct slave *tx_slave;
 251
 252         /* We don't need to disable softirq here, becase
 253          * tlb_choose_channel() is only called by bond_alb_xmit()
 254          * which already has softirq disabled.
 255          */
 256         spin_lock(&bond->mode_lock);
 257         tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
 258         spin_unlock(&bond->mode_lock);
 259
 260         return tx_slave;
 261 }
 262
 263 /*********************** rlb specific functions ***************************/
 264
 265 /* when an ARP REPLY is received from a client update its info
 266  * in the rx_hashtbl
 267  */
 268 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 269 {
 270         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 271         struct rlb_client_info *client_info;
 272         u32 hash_index;
 273
 274         spin_lock_bh(&bond->mode_lock);
 275
 276         hash_index = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
 277         client_info = &(bond_info->rx_hashtbl[hash_index]);
 278
 279         if ((client_info->assigned) &&
 280             (client_info->ip_src == arp->ip_dst) &&
 281             (client_info->ip_dst == arp->ip_src) &&
 282             (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) {
 283                 /* update the clients MAC address */
 284                 ether_addr_copy(client_info->mac_dst, arp->mac_src);
 285                 client_info->ntt = 1;
 286                 bond_info->rx_ntt = 1;
 287         }
 288
 289         spin_unlock_bh(&bond->mode_lock);
 290 }
 291
 292 static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
 293                         struct slave *slave)
 294 {
 295         struct arp_pkt *arp, _arp;
 296
 297         if (skb->protocol != cpu_to_be16(ETH_P_ARP))
 298                 goto out;
 299
 300         arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp);
 301         if (!arp)
 302                 goto out;
 303
 304         /* We received an ARP from arp->ip_src.
 305          * We might have used this IP address previously (on the bonding host
 306          * itself or on a system that is bridged together with the bond).
 307          * However, if arp->mac_src is different than what is stored in
 308          * rx_hashtbl, some other host is now using the IP and we must prevent
 309          * sending out client updates with this IP address and the old MAC
 310          * address.
 311          * Clean up all hash table entries that have this address as ip_src but
 312          * have a different mac_src.
 313          */
 314         rlb_purge_src_ip(bond, arp);
 315
 316         if (arp->op_code == htons(ARPOP_REPLY)) {
 317                 /* update rx hash table for this ARP */
 318                 rlb_update_entry_from_arp(bond, arp);
 319                 netdev_dbg(bond->dev, "Server received an ARP Reply from client\n");
 320         }
 321 out:
 322         return RX_HANDLER_ANOTHER;
 323 }
 324
 325 /* Caller must hold rcu_read_lock() */
 326 static struct slave *__rlb_next_rx_slave(struct bonding *bond)
 327 {
 328         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 329         struct slave *before = NULL, *rx_slave = NULL, *slave;
 330         struct list_head *iter;
 331         bool found = false;
 332
 333         bond_for_each_slave_rcu(bond, slave, iter) {
 334                 if (!bond_slave_can_tx(slave))
 335                         continue;
 336                 if (!found) {
 337                         if (!before || before->speed < slave->speed)
 338                                 before = slave;
 339                 } else {
 340                         if (!rx_slave || rx_slave->speed < slave->speed)
 341                                 rx_slave = slave;
 342                 }
 343                 if (slave == bond_info->rx_slave)
 344                         found = true;
 345         }
 346         /* we didn't find anything after the current or we have something
 347          * better before and up to the current slave
 348          */
 349         if (!rx_slave || (before && rx_slave->speed < before->speed))
 350                 rx_slave = before;
 351
 352         if (rx_slave)
 353                 bond_info->rx_slave = rx_slave;
 354
 355         return rx_slave;
 356 }
 357
 358 /* Caller must hold RTNL, rcu_read_lock is obtained only to silence checkers */
 359 static struct slave *rlb_next_rx_slave(struct bonding *bond)
 360 {
 361         struct slave *rx_slave;
 362
 363         ASSERT_RTNL();
 364
 365         rcu_read_lock();
 366         rx_slave = __rlb_next_rx_slave(bond);
 367         rcu_read_unlock();
 368
 369         return rx_slave;
 370 }
 371
 372 /* teach the switch the mac of a disabled slave
 373  * on the primary for fault tolerance
 374  *
 375  * Caller must hold RTNL
 376  */
 377 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
 378 {
 379         struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 380
 381         if (!curr_active)
 382                 return;
 383
 384         if (!bond->alb_info.primary_is_promisc) {
 385                 if (!dev_set_promiscuity(curr_active->dev, 1))
 386                         bond->alb_info.primary_is_promisc = 1;
 387                 else
 388                         bond->alb_info.primary_is_promisc = 0;
 389         }
 390
 391         bond->alb_info.rlb_promisc_timeout_counter = 0;
 392
 393         alb_send_learning_packets(curr_active, addr, true);
 394 }
 395
 396 /* slave being removed should not be active at this point
 397  *
 398  * Caller must hold rtnl.
 399  */
 400 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 401 {
 402         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 403         struct rlb_client_info *rx_hash_table;
 404         u32 index, next_index;
 405
 406         /* clear slave from rx_hashtbl */
 407         spin_lock_bh(&bond->mode_lock);
 408
 409         rx_hash_table = bond_info->rx_hashtbl;
 410         index = bond_info->rx_hashtbl_used_head;
 411         for (; index != RLB_NULL_INDEX; index = next_index) {
 412                 next_index = rx_hash_table[index].used_next;
 413                 if (rx_hash_table[index].slave == slave) {
 414                         struct slave *assigned_slave = rlb_next_rx_slave(bond);
 415
 416                         if (assigned_slave) {
 417                                 rx_hash_table[index].slave = assigned_slave;
 418                                 if (!ether_addr_equal_64bits(rx_hash_table[index].mac_dst,
 419                                                              mac_bcast)) {
 420                                         bond_info->rx_hashtbl[index].ntt = 1;
 421                                         bond_info->rx_ntt = 1;
 422                                         /* A slave has been removed from the
 423                                          * table because it is either disabled
 424                                          * or being released. We must retry the
 425                                          * update to avoid clients from not
 426                                          * being updated & disconnecting when
 427                                          * there is stress
 428                                          */
 429                                         bond_info->rlb_update_retry_counter =
 430                                                 RLB_UPDATE_RETRY;
 431                                 }
 432                         } else {  /* there is no active slave */
 433                                 rx_hash_table[index].slave = NULL;
 434                         }
 435                 }
 436         }
 437
 438         spin_unlock_bh(&bond->mode_lock);
 439
 440         if (slave != rtnl_dereference(bond->curr_active_slave))
 441                 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 442 }
 443
 444 static void rlb_update_client(struct rlb_client_info *client_info)
 445 {
 446         int i;
 447
 448         if (!client_info->slave || !is_valid_ether_addr(client_info->mac_dst))
 449                 return;
 450
 451         for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 452                 struct sk_buff *skb;
 453
 454                 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 455                                  client_info->ip_dst,
 456                                  client_info->slave->dev,
 457                                  client_info->ip_src,
 458                                  client_info->mac_dst,
 459                                  client_info->slave->dev->dev_addr,
 460                                  client_info->mac_dst);
 461                 if (!skb) {
 462                         netdev_err(client_info->slave->bond->dev,
 463                                    "failed to create an ARP packet\n");
 464                         continue;
 465                 }
 466
 467                 skb->dev = client_info->slave->dev;
 468
 469                 if (client_info->vlan_id) {
 470                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 471                                                client_info->vlan_id);
 472                 }
 473
 474                 arp_xmit(skb);
 475         }
 476 }
 477
 478 /* sends ARP REPLIES that update the clients that need updating */
 479 static void rlb_update_rx_clients(struct bonding *bond)
 480 {
 481         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 482         struct rlb_client_info *client_info;
 483         u32 hash_index;
 484
 485         spin_lock_bh(&bond->mode_lock);
 486
 487         hash_index = bond_info->rx_hashtbl_used_head;
 488         for (; hash_index != RLB_NULL_INDEX;
 489              hash_index = client_info->used_next) {
 490                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 491                 if (client_info->ntt) {
 492                         rlb_update_client(client_info);
 493                         if (bond_info->rlb_update_retry_counter == 0)
 494                                 client_info->ntt = 0;
 495                 }
 496         }
 497
 498         /* do not update the entries again until this counter is zero so that
 499          * not to confuse the clients.
 500          */
 501         bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 502
 503         spin_unlock_bh(&bond->mode_lock);
 504 }
 505
 506 /* The slave was assigned a new mac address - update the clients */
 507 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 508 {
 509         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 510         struct rlb_client_info *client_info;
 511         int ntt = 0;
 512         u32 hash_index;
 513
 514         spin_lock_bh(&bond->mode_lock);
 515
 516         hash_index = bond_info->rx_hashtbl_used_head;
 517         for (; hash_index != RLB_NULL_INDEX;
 518              hash_index = client_info->used_next) {
 519                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 520
 521                 if ((client_info->slave == slave) &&
 522                     !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
 523                         client_info->ntt = 1;
 524                         ntt = 1;
 525                 }
 526         }
 527
 528         /* update the team's flag only after the whole iteration */
 529         if (ntt) {
 530                 bond_info->rx_ntt = 1;
 531                 /* fasten the change */
 532                 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 533         }
 534
 535         spin_unlock_bh(&bond->mode_lock);
 536 }
 537
 538 /* mark all clients using src_ip to be updated */
 539 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
 540 {
 541         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 542         struct rlb_client_info *client_info;
 543         u32 hash_index;
 544
 545         spin_lock(&bond->mode_lock);
 546
 547         hash_index = bond_info->rx_hashtbl_used_head;
 548         for (; hash_index != RLB_NULL_INDEX;
 549              hash_index = client_info->used_next) {
 550                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 551
 552                 if (!client_info->slave) {
 553                         netdev_err(bond->dev, "found a client with no channel in the client's hash table\n");
 554                         continue;
 555                 }
 556                 /* update all clients using this src_ip, that are not assigned
 557                  * to the team's address (curr_active_slave) and have a known
 558                  * unicast mac address.
 559                  */
 560                 if ((client_info->ip_src == src_ip) &&
 561                     !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
 562                                              bond->dev->dev_addr) &&
 563                     !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
 564                         client_info->ntt = 1;
 565                         bond_info->rx_ntt = 1;
 566                 }
 567         }
 568
 569         spin_unlock(&bond->mode_lock);
 570 }
 571
 572 static struct slave *rlb_choose_channel(struct sk_buff *skb,
 573                                         struct bonding *bond,
 574                                         const struct arp_pkt *arp)
 575 {
 576         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 577         struct slave *assigned_slave, *curr_active_slave;
 578         struct rlb_client_info *client_info;
 579         u32 hash_index = 0;
 580
 581         spin_lock(&bond->mode_lock);
 582
 583         curr_active_slave = rcu_dereference(bond->curr_active_slave);
 584
 585         hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
 586         client_info = &(bond_info->rx_hashtbl[hash_index]);
 587
 588         if (client_info->assigned) {
 589                 if ((client_info->ip_src == arp->ip_src) &&
 590                     (client_info->ip_dst == arp->ip_dst)) {
 591                         /* the entry is already assigned to this client */
 592                         if (!ether_addr_equal_64bits(arp->mac_dst, mac_bcast)) {
 593                                 /* update mac address from arp */
 594                                 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
 595                         }
 596                         ether_addr_copy(client_info->mac_src, arp->mac_src);
 597
 598                         assigned_slave = client_info->slave;
 599                         if (assigned_slave) {
 600                                 spin_unlock(&bond->mode_lock);
 601                                 return assigned_slave;
 602                         }
 603                 } else {
 604                         /* the entry is already assigned to some other client,
 605                          * move the old client to primary (curr_active_slave) so
 606                          * that the new client can be assigned to this entry.
 607                          */
 608                         if (curr_active_slave &&
 609                             client_info->slave != curr_active_slave) {
 610                                 client_info->slave = curr_active_slave;
 611                                 rlb_update_client(client_info);
 612                         }
 613                 }
 614         }
 615         /* assign a new slave */
 616         assigned_slave = __rlb_next_rx_slave(bond);
 617
 618         if (assigned_slave) {
 619                 if (!(client_info->assigned &&
 620                       client_info->ip_src == arp->ip_src)) {
 621                         /* ip_src is going to be updated,
 622                          * fix the src hash list
 623                          */
 624                         u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
 625                                                     sizeof(arp->ip_src));
 626                         rlb_src_unlink(bond, hash_index);
 627                         rlb_src_link(bond, hash_src, hash_index);
 628                 }
 629
 630                 client_info->ip_src = arp->ip_src;
 631                 client_info->ip_dst = arp->ip_dst;
 632                 /* arp->mac_dst is broadcast for arp reqeusts.
 633                  * will be updated with clients actual unicast mac address
 634                  * upon receiving an arp reply.
 635                  */
 636                 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
 637                 ether_addr_copy(client_info->mac_src, arp->mac_src);
 638                 client_info->slave = assigned_slave;
 639
 640                 if (!ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
 641                         client_info->ntt = 1;
 642                         bond->alb_info.rx_ntt = 1;
 643                 } else {
 644                         client_info->ntt = 0;
 645                 }
 646
 647                 if (vlan_get_tag(skb, &client_info->vlan_id))
 648                         client_info->vlan_id = 0;
 649
 650                 if (!client_info->assigned) {
 651                         u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
 652                         bond_info->rx_hashtbl_used_head = hash_index;
 653                         client_info->used_next = prev_tbl_head;
 654                         if (prev_tbl_head != RLB_NULL_INDEX) {
 655                                 bond_info->rx_hashtbl[prev_tbl_head].used_prev =
 656                                         hash_index;
 657                         }
 658                         client_info->assigned = 1;
 659                 }
 660         }
 661
 662         spin_unlock(&bond->mode_lock);
 663
 664         return assigned_slave;
 665 }
 666
 667 /* chooses (and returns) transmit channel for arp reply
 668  * does not choose channel for other arp types since they are
 669  * sent on the curr_active_slave
 670  */
 671 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 672 {
 673         struct slave *tx_slave = NULL;
 674         struct arp_pkt *arp;
 675
 676         if (!pskb_network_may_pull(skb, sizeof(*arp)))
 677                 return NULL;
 678         arp = (struct arp_pkt *)skb_network_header(skb);
 679
 680         /* Don't modify or load balance ARPs that do not originate locally
 681          * (e.g.,arrive via a bridge).
 682          */
 683         if (!bond_slave_has_mac_rx(bond, arp->mac_src))
 684                 return NULL;
 685
 686         if (arp->op_code == htons(ARPOP_REPLY)) {
 687                 /* the arp must be sent on the selected rx channel */
 688                 tx_slave = rlb_choose_channel(skb, bond, arp);
 689                 if (tx_slave)
 690                         bond_hw_addr_copy(arp->mac_src, tx_slave->dev->dev_addr,
 691                                           tx_slave->dev->addr_len);
 692                 netdev_dbg(bond->dev, "Server sent ARP Reply packet\n");
 693         } else if (arp->op_code == htons(ARPOP_REQUEST)) {
 694                 /* Create an entry in the rx_hashtbl for this client as a
 695                  * place holder.
 696                  * When the arp reply is received the entry will be updated
 697                  * with the correct unicast address of the client.
 698                  */
 699                 rlb_choose_channel(skb, bond, arp);
 700
 701                 /* The ARP reply packets must be delayed so that
 702                  * they can cancel out the influence of the ARP request.
 703                  */
 704                 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 705
 706                 /* arp requests are broadcast and are sent on the primary
 707                  * the arp request will collapse all clients on the subnet to
 708                  * the primary slave. We must register these clients to be
 709                  * updated with their assigned mac.
 710                  */
 711                 rlb_req_update_subnet_clients(bond, arp->ip_src);
 712                 netdev_dbg(bond->dev, "Server sent ARP Request packet\n");
 713         }
 714
 715         return tx_slave;
 716 }
 717
 718 static void rlb_rebalance(struct bonding *bond)
 719 {
 720         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 721         struct slave *assigned_slave;
 722         struct rlb_client_info *client_info;
 723         int ntt;
 724         u32 hash_index;
 725
 726         spin_lock_bh(&bond->mode_lock);
 727
 728         ntt = 0;
 729         hash_index = bond_info->rx_hashtbl_used_head;
 730         for (; hash_index != RLB_NULL_INDEX;
 731              hash_index = client_info->used_next) {
 732                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 733                 assigned_slave = __rlb_next_rx_slave(bond);
 734                 if (assigned_slave && (client_info->slave != assigned_slave)) {
 735                         client_info->slave = assigned_slave;
 736                         client_info->ntt = 1;
 737                         ntt = 1;
 738                 }
 739         }
 740
 741         /* update the team's flag only after the whole iteration */
 742         if (ntt)
 743                 bond_info->rx_ntt = 1;
 744         spin_unlock_bh(&bond->mode_lock);
 745 }
 746
 747 /* Caller must hold mode_lock */
 748 static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
 749 {
 750         entry->used_next = RLB_NULL_INDEX;
 751         entry->used_prev = RLB_NULL_INDEX;
 752         entry->assigned = 0;
 753         entry->slave = NULL;
 754         entry->vlan_id = 0;
 755 }
 756 static void rlb_init_table_entry_src(struct rlb_client_info *entry)
 757 {
 758         entry->src_first = RLB_NULL_INDEX;
 759         entry->src_prev = RLB_NULL_INDEX;
 760         entry->src_next = RLB_NULL_INDEX;
 761 }
 762
 763 static void rlb_init_table_entry(struct rlb_client_info *entry)
 764 {
 765         memset(entry, 0, sizeof(struct rlb_client_info));
 766         rlb_init_table_entry_dst(entry);
 767         rlb_init_table_entry_src(entry);
 768 }
 769
 770 static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
 771 {
 772         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 773         u32 next_index = bond_info->rx_hashtbl[index].used_next;
 774         u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
 775
 776         if (index == bond_info->rx_hashtbl_used_head)
 777                 bond_info->rx_hashtbl_used_head = next_index;
 778         if (prev_index != RLB_NULL_INDEX)
 779                 bond_info->rx_hashtbl[prev_index].used_next = next_index;
 780         if (next_index != RLB_NULL_INDEX)
 781                 bond_info->rx_hashtbl[next_index].used_prev = prev_index;
 782 }
 783
 784 /* unlink a rlb hash table entry from the src list */
 785 static void rlb_src_unlink(struct bonding *bond, u32 index)
 786 {
 787         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 788         u32 next_index = bond_info->rx_hashtbl[index].src_next;
 789         u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
 790
 791         bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
 792         bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
 793
 794         if (next_index != RLB_NULL_INDEX)
 795                 bond_info->rx_hashtbl[next_index].src_prev = prev_index;
 796
 797         if (prev_index == RLB_NULL_INDEX)
 798                 return;
 799
 800         /* is prev_index pointing to the head of this list? */
 801         if (bond_info->rx_hashtbl[prev_index].src_first == index)
 802                 bond_info->rx_hashtbl[prev_index].src_first = next_index;
 803         else
 804                 bond_info->rx_hashtbl[prev_index].src_next = next_index;
 805
 806 }
 807
 808 static void rlb_delete_table_entry(struct bonding *bond, u32 index)
 809 {
 810         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 811         struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 812
 813         rlb_delete_table_entry_dst(bond, index);
 814         rlb_init_table_entry_dst(entry);
 815
 816         rlb_src_unlink(bond, index);
 817 }
 818
 819 /* add the rx_hashtbl[ip_dst_hash] entry to the list
 820  * of entries with identical ip_src_hash
 821  */
 822 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
 823 {
 824         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 825         u32 next;
 826
 827         bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
 828         next = bond_info->rx_hashtbl[ip_src_hash].src_first;
 829         bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
 830         if (next != RLB_NULL_INDEX)
 831                 bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
 832         bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
 833 }
 834
 835 /* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does
 836  * not match arp->mac_src
 837  */
 838 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
 839 {
 840         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 841         u32 ip_src_hash = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
 842         u32 index;
 843
 844         spin_lock_bh(&bond->mode_lock);
 845
 846         index = bond_info->rx_hashtbl[ip_src_hash].src_first;
 847         while (index != RLB_NULL_INDEX) {
 848                 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 849                 u32 next_index = entry->src_next;
 850                 if (entry->ip_src == arp->ip_src &&
 851                     !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
 852                                 rlb_delete_table_entry(bond, index);
 853                 index = next_index;
 854         }
 855         spin_unlock_bh(&bond->mode_lock);
 856 }
 857
 858 static int rlb_initialize(struct bonding *bond)
 859 {
 860         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 861         struct rlb_client_info  *new_hashtbl;
 862         int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 863         int i;
 864
 865         new_hashtbl = kmalloc(size, GFP_KERNEL);
 866         if (!new_hashtbl)
 867                 return -1;
 868
 869         spin_lock_bh(&bond->mode_lock);
 870
 871         bond_info->rx_hashtbl = new_hashtbl;
 872
 873         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 874
 875         for (i = 0; i < RLB_HASH_TABLE_SIZE; i++)
 876                 rlb_init_table_entry(bond_info->rx_hashtbl + i);
 877
 878         spin_unlock_bh(&bond->mode_lock);
 879
 880         /* register to receive ARPs */
 881         bond->recv_probe = rlb_arp_recv;
 882
 883         return 0;
 884 }
 885
 886 static void rlb_deinitialize(struct bonding *bond)
 887 {
 888         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 889
 890         spin_lock_bh(&bond->mode_lock);
 891
 892         kfree(bond_info->rx_hashtbl);
 893         bond_info->rx_hashtbl = NULL;
 894         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 895
 896         spin_unlock_bh(&bond->mode_lock);
 897 }
 898
 899 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 900 {
 901         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 902         u32 curr_index;
 903
 904         spin_lock_bh(&bond->mode_lock);
 905
 906         curr_index = bond_info->rx_hashtbl_used_head;
 907         while (curr_index != RLB_NULL_INDEX) {
 908                 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 909                 u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
 910
 911                 if (curr->vlan_id == vlan_id)
 912                         rlb_delete_table_entry(bond, curr_index);
 913
 914                 curr_index = next_index;
 915         }
 916
 917         spin_unlock_bh(&bond->mode_lock);
 918 }
 919
 920 /*********************** tlb/rlb shared functions *********************/
 921
 922 static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
 923                             __be16 vlan_proto, u16 vid)
 924 {
 925         struct learning_pkt pkt;
 926         struct sk_buff *skb;
 927         int size = sizeof(struct learning_pkt);
 928
 929         memset(&pkt, 0, size);
 930         ether_addr_copy(pkt.mac_dst, mac_addr);
 931         ether_addr_copy(pkt.mac_src, mac_addr);
 932         pkt.type = cpu_to_be16(ETH_P_LOOPBACK);
 933
 934         skb = dev_alloc_skb(size);
 935         if (!skb)
 936                 return;
 937
 938         skb_put_data(skb, &pkt, size);
 939
 940         skb_reset_mac_header(skb);
 941         skb->network_header = skb->mac_header + ETH_HLEN;
 942         skb->protocol = pkt.type;
 943         skb->priority = TC_PRIO_CONTROL;
 944         skb->dev = slave->dev;
 945
 946         netdev_dbg(slave->bond->dev,
 947                    "Send learning packet: dev %s mac %pM vlan %d\n",
 948                    slave->dev->name, mac_addr, vid);
 949
 950         if (vid)
 951                 __vlan_hwaccel_put_tag(skb, vlan_proto, vid);
 952
 953         dev_queue_xmit(skb);
 954 }
 955
 956 struct alb_walk_data {
 957         struct bonding *bond;
 958         struct slave *slave;
 959         u8 *mac_addr;
 960         bool strict_match;
 961 };
 962
 963 static int alb_upper_dev_walk(struct net_device *upper, void *_data)
 964 {
 965         struct alb_walk_data *data = _data;
 966         bool strict_match = data->strict_match;
 967         struct bonding *bond = data->bond;
 968         struct slave *slave = data->slave;
 969         u8 *mac_addr = data->mac_addr;
 970         struct bond_vlan_tag *tags;
 971
 972         if (is_vlan_dev(upper) &&
 973             bond->nest_level == vlan_get_encap_level(upper) - 1) {
 974                 if (upper->addr_assign_type == NET_ADDR_STOLEN) {
 975                         alb_send_lp_vid(slave, mac_addr,
 976                                         vlan_dev_vlan_proto(upper),
 977                                         vlan_dev_vlan_id(upper));
 978                 } else {
 979                         alb_send_lp_vid(slave, upper->dev_addr,
 980                                         vlan_dev_vlan_proto(upper),
 981                                         vlan_dev_vlan_id(upper));
 982                 }
 983         }
 984
 985         /* If this is a macvlan device, then only send updates
 986          * when strict_match is turned off.
 987          */
 988         if (netif_is_macvlan(upper) && !strict_match) {
 989                 tags = bond_verify_device_path(bond->dev, upper, 0);
 990                 if (IS_ERR_OR_NULL(tags))
 991                         BUG();
 992                 alb_send_lp_vid(slave, upper->dev_addr,
 993                                 tags[0].vlan_proto, tags[0].vlan_id);
 994                 kfree(tags);
 995         }
 996
 997         return 0;
 998 }
 999
1000 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
1001                                       bool strict_match)
1002 {
1003         struct bonding *bond = bond_get_bond_by_slave(slave);
1004         struct alb_walk_data data = {
1005                 .strict_match = strict_match,
1006                 .mac_addr = mac_addr,
1007                 .slave = slave,
1008                 .bond = bond,
1009         };
1010
1011         /* send untagged */
1012         alb_send_lp_vid(slave, mac_addr, 0, 0);
1013
1014         /* loop through all devices and see if we need to send a packet
1015          * for that device.
1016          */
1017         rcu_read_lock();
1018         netdev_walk_all_upper_dev_rcu(bond->dev, alb_upper_dev_walk, &data);
1019         rcu_read_unlock();
1020 }
1021
1022 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[],
1023                                   unsigned int len)
1024 {
1025         struct net_device *dev = slave->dev;
1026         struct sockaddr_storage ss;
1027
1028         if (BOND_MODE(slave->bond) == BOND_MODE_TLB) {
1029                 memcpy(dev->dev_addr, addr, len);
1030                 return 0;
1031         }
1032
1033         /* for rlb each slave must have a unique hw mac addresses so that
1034          * each slave will receive packets destined to a different mac
1035          */
1036         memcpy(ss.__data, addr, len);
1037         ss.ss_family = dev->type;
1038         if (dev_set_mac_address(dev, (struct sockaddr *)&ss)) {
1039                 netdev_err(slave->bond->dev, "dev_set_mac_address of dev %s failed! ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
1040                            dev->name);
1041                 return -EOPNOTSUPP;
1042         }
1043         return 0;
1044 }
1045
1046 /* Swap MAC addresses between two slaves.
1047  *
1048  * Called with RTNL held, and no other locks.
1049  */
1050 static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1051 {
1052         u8 tmp_mac_addr[MAX_ADDR_LEN];
1053
1054         bond_hw_addr_copy(tmp_mac_addr, slave1->dev->dev_addr,
1055                           slave1->dev->addr_len);
1056         alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr,
1057                                slave2->dev->addr_len);
1058         alb_set_slave_mac_addr(slave2, tmp_mac_addr,
1059                                slave1->dev->addr_len);
1060
1061 }
1062
1063 /* Send learning packets after MAC address swap.
1064  *
1065  * Called with RTNL and no other locks
1066  */
1067 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1068                                 struct slave *slave2)
1069 {
1070         int slaves_state_differ = (bond_slave_can_tx(slave1) != bond_slave_can_tx(slave2));
1071         struct slave *disabled_slave = NULL;
1072
1073         ASSERT_RTNL();
1074
1075         /* fasten the change in the switch */
1076         if (bond_slave_can_tx(slave1)) {
1077                 alb_send_learning_packets(slave1, slave1->dev->dev_addr, false);
1078                 if (bond->alb_info.rlb_enabled) {
1079                         /* inform the clients that the mac address
1080                          * has changed
1081                          */
1082                         rlb_req_update_slave_clients(bond, slave1);
1083                 }
1084         } else {
1085                 disabled_slave = slave1;
1086         }
1087
1088         if (bond_slave_can_tx(slave2)) {
1089                 alb_send_learning_packets(slave2, slave2->dev->dev_addr, false);
1090                 if (bond->alb_info.rlb_enabled) {
1091                         /* inform the clients that the mac address
1092                          * has changed
1093                          */
1094                         rlb_req_update_slave_clients(bond, slave2);
1095                 }
1096         } else {
1097                 disabled_slave = slave2;
1098         }
1099
1100         if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1101                 /* A disabled slave was assigned an active mac addr */
1102                 rlb_teach_disabled_mac_on_primary(bond,
1103                                                   disabled_slave->dev->dev_addr);
1104         }
1105 }
1106
1107 /**
1108  * alb_change_hw_addr_on_detach
1109  * @bond: bonding we're working on
1110  * @slave: the slave that was just detached
1111  *
1112  * We assume that @slave was already detached from the slave list.
1113  *
1114  * If @slave's permanent hw address is different both from its current
1115  * address and from @bond's address, then somewhere in the bond there's
1116  * a slave that has @slave's permanet address as its current address.
1117  * We'll make sure that that slave no longer uses @slave's permanent address.
1118  *
1119  * Caller must hold RTNL and no other locks
1120  */
1121 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1122 {
1123         int perm_curr_diff;
1124         int perm_bond_diff;
1125         struct slave *found_slave;
1126
1127         perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1128                                                   slave->dev->dev_addr);
1129         perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1130                                                   bond->dev->dev_addr);
1131
1132         if (perm_curr_diff && perm_bond_diff) {
1133                 found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1134
1135                 if (found_slave) {
1136                         alb_swap_mac_addr(slave, found_slave);
1137                         alb_fasten_mac_swap(bond, slave, found_slave);
1138                 }
1139         }
1140 }
1141
1142 /**
1143  * alb_handle_addr_collision_on_attach
1144  * @bond: bonding we're working on
1145  * @slave: the slave that was just attached
1146  *
1147  * checks uniqueness of slave's mac address and handles the case the
1148  * new slave uses the bonds mac address.
1149  *
1150  * If the permanent hw address of @slave is @bond's hw address, we need to
1151  * find a different hw address to give @slave, that isn't in use by any other
1152  * slave in the bond. This address must be, of course, one of the permanent
1153  * addresses of the other slaves.
1154  *
1155  * We go over the slave list, and for each slave there we compare its
1156  * permanent hw address with the current address of all the other slaves.
1157  * If no match was found, then we've found a slave with a permanent address
1158  * that isn't used by any other slave in the bond, so we can assign it to
1159  * @slave.
1160  *
1161  * assumption: this function is called before @slave is attached to the
1162  *             bond slave list.
1163  */
1164 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1165 {
1166         struct slave *has_bond_addr = rcu_access_pointer(bond->curr_active_slave);
1167         struct slave *tmp_slave1, *free_mac_slave = NULL;
1168         struct list_head *iter;
1169
1170         if (!bond_has_slaves(bond)) {
1171                 /* this is the first slave */
1172                 return 0;
1173         }
1174
1175         /* if slave's mac address differs from bond's mac address
1176          * check uniqueness of slave's mac address against the other
1177          * slaves in the bond.
1178          */
1179         if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1180                 if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1181                         return 0;
1182
1183                 /* Try setting slave mac to bond address and fall-through
1184                  * to code handling that situation below...
1185                  */
1186                 alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1187                                        bond->dev->addr_len);
1188         }
1189
1190         /* The slave's address is equal to the address of the bond.
1191          * Search for a spare address in the bond for this slave.
1192          */
1193         bond_for_each_slave(bond, tmp_slave1, iter) {
1194                 if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1195                         /* no slave has tmp_slave1's perm addr
1196                          * as its curr addr
1197                          */
1198                         free_mac_slave = tmp_slave1;
1199                         break;
1200                 }
1201
1202                 if (!has_bond_addr) {
1203                         if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1204                                                     bond->dev->dev_addr)) {
1205
1206                                 has_bond_addr = tmp_slave1;
1207                         }
1208                 }
1209         }
1210
1211         if (free_mac_slave) {
1212                 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1213                                        free_mac_slave->dev->addr_len);
1214
1215                 netdev_warn(bond->dev, "the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1216                             slave->dev->name, free_mac_slave->dev->name);
1217
1218         } else if (has_bond_addr) {
1219                 netdev_err(bond->dev, "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",
1220                            slave->dev->name);
1221                 return -EFAULT;
1222         }
1223
1224         return 0;
1225 }
1226
1227 /**
1228  * alb_set_mac_address
1229  * @bond:
1230  * @addr:
1231  *
1232  * In TLB mode all slaves are configured to the bond's hw address, but set
1233  * their dev_addr field to different addresses (based on their permanent hw
1234  * addresses).
1235  *
1236  * For each slave, this function sets the interface to the new address and then
1237  * changes its dev_addr field to its previous value.
1238  *
1239  * Unwinding assumes bond's mac address has not yet changed.
1240  */
1241 static int alb_set_mac_address(struct bonding *bond, void *addr)
1242 {
1243         struct slave *slave, *rollback_slave;
1244         struct list_head *iter;
1245         struct sockaddr_storage ss;
1246         char tmp_addr[MAX_ADDR_LEN];
1247         int res;
1248
1249         if (bond->alb_info.rlb_enabled)
1250                 return 0;
1251
1252         bond_for_each_slave(bond, slave, iter) {
1253                 /* save net_device's current hw address */
1254                 bond_hw_addr_copy(tmp_addr, slave->dev->dev_addr,
1255                                   slave->dev->addr_len);
1256
1257                 res = dev_set_mac_address(slave->dev, addr);
1258
1259                 /* restore net_device's hw address */
1260                 bond_hw_addr_copy(slave->dev->dev_addr, tmp_addr,
1261                                   slave->dev->addr_len);
1262
1263                 if (res)
1264                         goto unwind;
1265         }
1266
1267         return 0;
1268
1269 unwind:
1270         memcpy(ss.__data, bond->dev->dev_addr, bond->dev->addr_len);
1271         ss.ss_family = bond->dev->type;
1272
1273         /* unwind from head to the slave that failed */
1274         bond_for_each_slave(bond, rollback_slave, iter) {
1275                 if (rollback_slave == slave)
1276                         break;
1277                 bond_hw_addr_copy(tmp_addr, rollback_slave->dev->dev_addr,
1278                                   rollback_slave->dev->addr_len);
1279                 dev_set_mac_address(rollback_slave->dev,
1280                                     (struct sockaddr *)&ss);
1281                 bond_hw_addr_copy(rollback_slave->dev->dev_addr, tmp_addr,
1282                                   rollback_slave->dev->addr_len);
1283         }
1284
1285         return res;
1286 }
1287
1288 /************************ exported alb funcions ************************/
1289
1290 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1291 {
1292         int res;
1293
1294         res = tlb_initialize(bond);
1295         if (res)
1296                 return res;
1297
1298         if (rlb_enabled) {
1299                 bond->alb_info.rlb_enabled = 1;
1300                 res = rlb_initialize(bond);
1301                 if (res) {
1302                         tlb_deinitialize(bond);
1303                         return res;
1304                 }
1305         } else {
1306                 bond->alb_info.rlb_enabled = 0;
1307         }
1308
1309         return 0;
1310 }
1311
1312 void bond_alb_deinitialize(struct bonding *bond)
1313 {
1314         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1315
1316         tlb_deinitialize(bond);
1317
1318         if (bond_info->rlb_enabled)
1319                 rlb_deinitialize(bond);
1320 }
1321
1322 static int bond_do_alb_xmit(struct sk_buff *skb, struct bonding *bond,
1323                             struct slave *tx_slave)
1324 {
1325         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1326         struct ethhdr *eth_data = eth_hdr(skb);
1327
1328         if (!tx_slave) {
1329                 /* unbalanced or unassigned, send through primary */
1330                 tx_slave = rcu_dereference(bond->curr_active_slave);
1331                 if (bond->params.tlb_dynamic_lb)
1332                         bond_info->unbalanced_load += skb->len;
1333         }
1334
1335         if (tx_slave && bond_slave_can_tx(tx_slave)) {
1336                 if (tx_slave != rcu_access_pointer(bond->curr_active_slave)) {
1337                         ether_addr_copy(eth_data->h_source,
1338                                         tx_slave->dev->dev_addr);
1339                 }
1340
1341                 bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1342                 goto out;
1343         }
1344
1345         if (tx_slave && bond->params.tlb_dynamic_lb) {
1346                 spin_lock(&bond->mode_lock);
1347                 __tlb_clear_slave(bond, tx_slave, 0);
1348                 spin_unlock(&bond->mode_lock);
1349         }
1350
1351         /* no suitable interface, frame not sent */
1352         bond_tx_drop(bond->dev, skb);
1353 out:
1354         return NETDEV_TX_OK;
1355 }
1356
1357 int bond_tlb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1358 {
1359         struct bonding *bond = netdev_priv(bond_dev);
1360         struct ethhdr *eth_data;
1361         struct slave *tx_slave = NULL;
1362         u32 hash_index;
1363
1364         skb_reset_mac_header(skb);
1365         eth_data = eth_hdr(skb);
1366
1367         /* Do not TX balance any multicast or broadcast */
1368         if (!is_multicast_ether_addr(eth_data->h_dest)) {
1369                 switch (skb->protocol) {
1370                 case htons(ETH_P_IP):
1371                 case htons(ETH_P_IPX):
1372                     /* In case of IPX, it will falback to L2 hash */
1373                 case htons(ETH_P_IPV6):
1374                         hash_index = bond_xmit_hash(bond, skb);
1375                         if (bond->params.tlb_dynamic_lb) {
1376                                 tx_slave = tlb_choose_channel(bond,
1377                                                               hash_index & 0xFF,
1378                                                               skb->len);
1379                         } else {
1380                                 struct bond_up_slave *slaves;
1381                                 unsigned int count;
1382
1383                                 slaves = rcu_dereference(bond->slave_arr);
1384                                 count = slaves ? ACCESS_ONCE(slaves->count) : 0;
1385                                 if (likely(count))
1386                                         tx_slave = slaves->arr[hash_index %
1387                                                                count];
1388                         }
1389                         break;
1390                 }
1391         }
1392         return bond_do_alb_xmit(skb, bond, tx_slave);
1393 }
1394
1395 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1396 {
1397         struct bonding *bond = netdev_priv(bond_dev);
1398         struct ethhdr *eth_data;
1399         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1400         struct slave *tx_slave = NULL;
1401         static const __be32 ip_bcast = htonl(0xffffffff);
1402         int hash_size = 0;
1403         bool do_tx_balance = true;
1404         u32 hash_index = 0;
1405         const u8 *hash_start = NULL;
1406
1407         skb_reset_mac_header(skb);
1408         eth_data = eth_hdr(skb);
1409
1410         switch (ntohs(skb->protocol)) {
1411         case ETH_P_IP: {
1412                 const struct iphdr *iph;
1413
1414                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast) ||
1415                     (!pskb_network_may_pull(skb, sizeof(*iph)))) {
1416                         do_tx_balance = false;
1417                         break;
1418                 }
1419                 iph = ip_hdr(skb);
1420                 if (iph->daddr == ip_bcast || iph->protocol == IPPROTO_IGMP) {
1421                         do_tx_balance = false;
1422                         break;
1423                 }
1424                 hash_start = (char *)&(iph->daddr);
1425                 hash_size = sizeof(iph->daddr);
1426                 break;
1427         }
1428         case ETH_P_IPV6: {
1429                 const struct ipv6hdr *ip6hdr;
1430
1431                 /* IPv6 doesn't really use broadcast mac address, but leave
1432                  * that here just in case.
1433                  */
1434                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast)) {
1435                         do_tx_balance = false;
1436                         break;
1437                 }
1438
1439                 /* IPv6 uses all-nodes multicast as an equivalent to
1440                  * broadcasts in IPv4.
1441                  */
1442                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1443                         do_tx_balance = false;
1444                         break;
1445                 }
1446
1447                 if (!pskb_network_may_pull(skb, sizeof(*ip6hdr))) {
1448                         do_tx_balance = false;
1449                         break;
1450                 }
1451                 /* Additionally, DAD probes should not be tx-balanced as that
1452                  * will lead to false positives for duplicate addresses and
1453                  * prevent address configuration from working.
1454                  */
1455                 ip6hdr = ipv6_hdr(skb);
1456                 if (ipv6_addr_any(&ip6hdr->saddr)) {
1457                         do_tx_balance = false;
1458                         break;
1459                 }
1460
1461                 hash_start = (char *)&ip6hdr->daddr;
1462                 hash_size = sizeof(ip6hdr->daddr);
1463                 break;
1464         }
1465         case ETH_P_IPX: {
1466                 const struct ipxhdr *ipxhdr;
1467
1468                 if (pskb_network_may_pull(skb, sizeof(*ipxhdr))) {
1469                         do_tx_balance = false;
1470                         break;
1471                 }
1472                 ipxhdr = (struct ipxhdr *)skb_network_header(skb);
1473
1474                 if (ipxhdr->ipx_checksum != IPX_NO_CHECKSUM) {
1475                         /* something is wrong with this packet */
1476                         do_tx_balance = false;
1477                         break;
1478                 }
1479
1480                 if (ipxhdr->ipx_type != IPX_TYPE_NCP) {
1481                         /* The only protocol worth balancing in
1482                          * this family since it has an "ARP" like
1483                          * mechanism
1484                          */
1485                         do_tx_balance = false;
1486                         break;
1487                 }
1488
1489                 eth_data = eth_hdr(skb);
1490                 hash_start = (char *)eth_data->h_dest;
1491                 hash_size = ETH_ALEN;
1492                 break;
1493         }
1494         case ETH_P_ARP:
1495                 do_tx_balance = false;
1496                 if (bond_info->rlb_enabled)
1497                         tx_slave = rlb_arp_xmit(skb, bond);
1498                 break;
1499         default:
1500                 do_tx_balance = false;
1501                 break;
1502         }
1503
1504         if (do_tx_balance) {
1505                 hash_index = _simple_hash(hash_start, hash_size);
1506                 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1507         }
1508
1509         return bond_do_alb_xmit(skb, bond, tx_slave);
1510 }
1511
1512 void bond_alb_monitor(struct work_struct *work)
1513 {
1514         struct bonding *bond = container_of(work, struct bonding,
1515                                             alb_work.work);
1516         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1517         struct list_head *iter;
1518         struct slave *slave;
1519
1520         if (!bond_has_slaves(bond)) {
1521                 bond_info->tx_rebalance_counter = 0;
1522                 bond_info->lp_counter = 0;
1523                 goto re_arm;
1524         }
1525
1526         rcu_read_lock();
1527
1528         bond_info->tx_rebalance_counter++;
1529         bond_info->lp_counter++;
1530
1531         /* send learning packets */
1532         if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1533                 bool strict_match;
1534
1535                 bond_for_each_slave_rcu(bond, slave, iter) {
1536                         /* If updating current_active, use all currently
1537                          * user mac addreses (!strict_match).  Otherwise, only
1538                          * use mac of the slave device.
1539                          * In RLB mode, we always use strict matches.
1540                          */
1541                         strict_match = (slave != rcu_access_pointer(bond->curr_active_slave) ||
1542                                         bond_info->rlb_enabled);
1543                         alb_send_learning_packets(slave, slave->dev->dev_addr,
1544                                                   strict_match);
1545                 }
1546                 bond_info->lp_counter = 0;
1547         }
1548
1549         /* rebalance tx traffic */
1550         if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1551                 bond_for_each_slave_rcu(bond, slave, iter) {
1552                         tlb_clear_slave(bond, slave, 1);
1553                         if (slave == rcu_access_pointer(bond->curr_active_slave)) {
1554                                 SLAVE_TLB_INFO(slave).load =
1555                                         bond_info->unbalanced_load /
1556                                                 BOND_TLB_REBALANCE_INTERVAL;
1557                                 bond_info->unbalanced_load = 0;
1558                         }
1559                 }
1560                 bond_info->tx_rebalance_counter = 0;
1561         }
1562
1563         if (bond_info->rlb_enabled) {
1564                 if (bond_info->primary_is_promisc &&
1565                     (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1566
1567                         /* dev_set_promiscuity requires rtnl and
1568                          * nothing else.  Avoid race with bond_close.
1569                          */
1570                         rcu_read_unlock();
1571                         if (!rtnl_trylock())
1572                                 goto re_arm;
1573
1574                         bond_info->rlb_promisc_timeout_counter = 0;
1575
1576                         /* If the primary was set to promiscuous mode
1577                          * because a slave was disabled then
1578                          * it can now leave promiscuous mode.
1579                          */
1580                         dev_set_promiscuity(rtnl_dereference(bond->curr_active_slave)->dev,
1581                                             -1);
1582                         bond_info->primary_is_promisc = 0;
1583
1584                         rtnl_unlock();
1585                         rcu_read_lock();
1586                 }
1587
1588                 if (bond_info->rlb_rebalance) {
1589                         bond_info->rlb_rebalance = 0;
1590                         rlb_rebalance(bond);
1591                 }
1592
1593                 /* check if clients need updating */
1594                 if (bond_info->rx_ntt) {
1595                         if (bond_info->rlb_update_delay_counter) {
1596                                 --bond_info->rlb_update_delay_counter;
1597                         } else {
1598                                 rlb_update_rx_clients(bond);
1599                                 if (bond_info->rlb_update_retry_counter)
1600                                         --bond_info->rlb_update_retry_counter;
1601                                 else
1602                                         bond_info->rx_ntt = 0;
1603                         }
1604                 }
1605         }
1606         rcu_read_unlock();
1607 re_arm:
1608         queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1609 }
1610
1611 /* assumption: called before the slave is attached to the bond
1612  * and not locked by the bond lock
1613  */
1614 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1615 {
1616         int res;
1617
1618         res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1619                                      slave->dev->addr_len);
1620         if (res)
1621                 return res;
1622
1623         res = alb_handle_addr_collision_on_attach(bond, slave);
1624         if (res)
1625                 return res;
1626
1627         tlb_init_slave(slave);
1628
1629         /* order a rebalance ASAP */
1630         bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1631
1632         if (bond->alb_info.rlb_enabled)
1633                 bond->alb_info.rlb_rebalance = 1;
1634
1635         return 0;
1636 }
1637
1638 /* Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1639  * if necessary.
1640  *
1641  * Caller must hold RTNL and no other locks
1642  */
1643 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1644 {
1645         if (bond_has_slaves(bond))
1646                 alb_change_hw_addr_on_detach(bond, slave);
1647
1648         tlb_clear_slave(bond, slave, 0);
1649
1650         if (bond->alb_info.rlb_enabled) {
1651                 bond->alb_info.rx_slave = NULL;
1652                 rlb_clear_slave(bond, slave);
1653         }
1654
1655 }
1656
1657 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1658 {
1659         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1660
1661         if (link == BOND_LINK_DOWN) {
1662                 tlb_clear_slave(bond, slave, 0);
1663                 if (bond->alb_info.rlb_enabled)
1664                         rlb_clear_slave(bond, slave);
1665         } else if (link == BOND_LINK_UP) {
1666                 /* order a rebalance ASAP */
1667                 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1668                 if (bond->alb_info.rlb_enabled) {
1669                         bond->alb_info.rlb_rebalance = 1;
1670                         /* If the updelay module parameter is smaller than the
1671                          * forwarding delay of the switch the rebalance will
1672                          * not work because the rebalance arp replies will
1673                          * not be forwarded to the clients..
1674                          */
1675                 }
1676         }
1677
1678         if (bond_is_nondyn_tlb(bond)) {
1679                 if (bond_update_slave_arr(bond, NULL))
1680                         pr_err("Failed to build slave-array for TLB mode.\n");
1681         }
1682 }
1683
1684 /**
1685  * bond_alb_handle_active_change - assign new curr_active_slave
1686  * @bond: our bonding struct
1687  * @new_slave: new slave to assign
1688  *
1689  * Set the bond->curr_active_slave to @new_slave and handle
1690  * mac address swapping and promiscuity changes as needed.
1691  *
1692  * Caller must hold RTNL
1693  */
1694 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1695 {
1696         struct slave *swap_slave;
1697         struct slave *curr_active;
1698
1699         curr_active = rtnl_dereference(bond->curr_active_slave);
1700         if (curr_active == new_slave)
1701                 return;
1702
1703         if (curr_active && bond->alb_info.primary_is_promisc) {
1704                 dev_set_promiscuity(curr_active->dev, -1);
1705                 bond->alb_info.primary_is_promisc = 0;
1706                 bond->alb_info.rlb_promisc_timeout_counter = 0;
1707         }
1708
1709         swap_slave = curr_active;
1710         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1711
1712         if (!new_slave || !bond_has_slaves(bond))
1713                 return;
1714
1715         /* set the new curr_active_slave to the bonds mac address
1716          * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1717          */
1718         if (!swap_slave)
1719                 swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1720
1721         /* Arrange for swap_slave and new_slave to temporarily be
1722          * ignored so we can mess with their MAC addresses without
1723          * fear of interference from transmit activity.
1724          */
1725         if (swap_slave)
1726                 tlb_clear_slave(bond, swap_slave, 1);
1727         tlb_clear_slave(bond, new_slave, 1);
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_MODE(bond) == BOND_MODE_TLB) {
1733                 struct sockaddr_storage ss;
1734                 u8 tmp_addr[MAX_ADDR_LEN];
1735
1736                 bond_hw_addr_copy(tmp_addr, new_slave->dev->dev_addr,
1737                                   new_slave->dev->addr_len);
1738
1739                 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
1740                                   bond->dev->addr_len);
1741                 ss.ss_family = bond->dev->type;
1742                 /* we don't care if it can't change its mac, best effort */
1743                 dev_set_mac_address(new_slave->dev, (struct sockaddr *)&ss);
1744
1745                 bond_hw_addr_copy(new_slave->dev->dev_addr, tmp_addr,
1746                                   new_slave->dev->addr_len);
1747         }
1748
1749         /* curr_active_slave must be set before calling alb_swap_mac_addr */
1750         if (swap_slave) {
1751                 /* swap mac address */
1752                 alb_swap_mac_addr(swap_slave, new_slave);
1753                 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1754         } else {
1755                 /* set the new_slave to the bond mac address */
1756                 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1757                                        bond->dev->addr_len);
1758                 alb_send_learning_packets(new_slave, bond->dev->dev_addr,
1759                                           false);
1760         }
1761 }
1762
1763 /* Called with RTNL */
1764 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1765 {
1766         struct bonding *bond = netdev_priv(bond_dev);
1767         struct sockaddr_storage *ss = addr;
1768         struct slave *curr_active;
1769         struct slave *swap_slave;
1770         int res;
1771
1772         if (!is_valid_ether_addr(ss->__data))
1773                 return -EADDRNOTAVAIL;
1774
1775         res = alb_set_mac_address(bond, addr);
1776         if (res)
1777                 return res;
1778
1779         bond_hw_addr_copy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
1780
1781         /* If there is no curr_active_slave there is nothing else to do.
1782          * Otherwise we'll need to pass the new address to it and handle
1783          * duplications.
1784          */
1785         curr_active = rtnl_dereference(bond->curr_active_slave);
1786         if (!curr_active)
1787                 return 0;
1788
1789         swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1790
1791         if (swap_slave) {
1792                 alb_swap_mac_addr(swap_slave, curr_active);
1793                 alb_fasten_mac_swap(bond, swap_slave, curr_active);
1794         } else {
1795                 alb_set_slave_mac_addr(curr_active, bond_dev->dev_addr,
1796                                        bond_dev->addr_len);
1797
1798                 alb_send_learning_packets(curr_active,
1799                                           bond_dev->dev_addr, false);
1800                 if (bond->alb_info.rlb_enabled) {
1801                         /* inform clients mac address has changed */
1802                         rlb_req_update_slave_clients(bond, curr_active);
1803                 }
1804         }
1805
1806         return 0;
1807 }
1808
1809 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1810 {
1811         if (bond->alb_info.rlb_enabled)
1812                 rlb_clear_vlan(bond, vlan_id);
1813 }
1814