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 static const u8 mac_v6_allmcast[ETH_ALEN + 2] __long_aligned = {
  44         0x33, 0x33, 0x00, 0x00, 0x00, 0x01
  45 };
  46 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
  47
  48 #pragma pack(1)
  49 struct learning_pkt {
  50         u8 mac_dst[ETH_ALEN];
  51         u8 mac_src[ETH_ALEN];
  52         __be16 type;
  53         u8 padding[ETH_ZLEN - ETH_HLEN];
  54 };
  55
  56 struct arp_pkt {
  57         __be16  hw_addr_space;
  58         __be16  prot_addr_space;
  59         u8      hw_addr_len;
  60         u8      prot_addr_len;
  61         __be16  op_code;
  62         u8      mac_src[ETH_ALEN];      /* sender hardware address */
  63         __be32  ip_src;                 /* sender IP address */
  64         u8      mac_dst[ETH_ALEN];      /* target hardware address */
  65         __be32  ip_dst;                 /* target IP address */
  66 };
  67 #pragma pack()
  68
  69 static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb)
  70 {
  71         return (struct arp_pkt *)skb_network_header(skb);
  72 }
  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 (is_valid_ether_addr(rx_hash_table[index].mac_dst)) {
 419                                         bond_info->rx_hashtbl[index].ntt = 1;
 420                                         bond_info->rx_ntt = 1;
 421                                         /* A slave has been removed from the
 422                                          * table because it is either disabled
 423                                          * or being released. We must retry the
 424                                          * update to avoid clients from not
 425                                          * being updated & disconnecting when
 426                                          * there is stress
 427                                          */
 428                                         bond_info->rlb_update_retry_counter =
 429                                                 RLB_UPDATE_RETRY;
 430                                 }
 431                         } else {  /* there is no active slave */
 432                                 rx_hash_table[index].slave = NULL;
 433                         }
 434                 }
 435         }
 436
 437         spin_unlock_bh(&bond->mode_lock);
 438
 439         if (slave != rtnl_dereference(bond->curr_active_slave))
 440                 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 441 }
 442
 443 static void rlb_update_client(struct rlb_client_info *client_info)
 444 {
 445         int i;
 446
 447         if (!client_info->slave || !is_valid_ether_addr(client_info->mac_dst))
 448                 return;
 449
 450         for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 451                 struct sk_buff *skb;
 452
 453                 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 454                                  client_info->ip_dst,
 455                                  client_info->slave->dev,
 456                                  client_info->ip_src,
 457                                  client_info->mac_dst,
 458                                  client_info->slave->dev->dev_addr,
 459                                  client_info->mac_dst);
 460                 if (!skb) {
 461                         netdev_err(client_info->slave->bond->dev,
 462                                    "failed to create an ARP packet\n");
 463                         continue;
 464                 }
 465
 466                 skb->dev = client_info->slave->dev;
 467
 468                 if (client_info->vlan_id) {
 469                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 470                                                client_info->vlan_id);
 471                 }
 472
 473                 arp_xmit(skb);
 474         }
 475 }
 476
 477 /* sends ARP REPLIES that update the clients that need updating */
 478 static void rlb_update_rx_clients(struct bonding *bond)
 479 {
 480         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 481         struct rlb_client_info *client_info;
 482         u32 hash_index;
 483
 484         spin_lock_bh(&bond->mode_lock);
 485
 486         hash_index = bond_info->rx_hashtbl_used_head;
 487         for (; hash_index != RLB_NULL_INDEX;
 488              hash_index = client_info->used_next) {
 489                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 490                 if (client_info->ntt) {
 491                         rlb_update_client(client_info);
 492                         if (bond_info->rlb_update_retry_counter == 0)
 493                                 client_info->ntt = 0;
 494                 }
 495         }
 496
 497         /* do not update the entries again until this counter is zero so that
 498          * not to confuse the clients.
 499          */
 500         bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 501
 502         spin_unlock_bh(&bond->mode_lock);
 503 }
 504
 505 /* The slave was assigned a new mac address - update the clients */
 506 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 507 {
 508         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 509         struct rlb_client_info *client_info;
 510         int ntt = 0;
 511         u32 hash_index;
 512
 513         spin_lock_bh(&bond->mode_lock);
 514
 515         hash_index = bond_info->rx_hashtbl_used_head;
 516         for (; hash_index != RLB_NULL_INDEX;
 517              hash_index = client_info->used_next) {
 518                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 519
 520                 if ((client_info->slave == slave) &&
 521                     is_valid_ether_addr(client_info->mac_dst)) {
 522                         client_info->ntt = 1;
 523                         ntt = 1;
 524                 }
 525         }
 526
 527         /* update the team's flag only after the whole iteration */
 528         if (ntt) {
 529                 bond_info->rx_ntt = 1;
 530                 /* fasten the change */
 531                 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 532         }
 533
 534         spin_unlock_bh(&bond->mode_lock);
 535 }
 536
 537 /* mark all clients using src_ip to be updated */
 538 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
 539 {
 540         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 541         struct rlb_client_info *client_info;
 542         u32 hash_index;
 543
 544         spin_lock(&bond->mode_lock);
 545
 546         hash_index = bond_info->rx_hashtbl_used_head;
 547         for (; hash_index != RLB_NULL_INDEX;
 548              hash_index = client_info->used_next) {
 549                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 550
 551                 if (!client_info->slave) {
 552                         netdev_err(bond->dev, "found a client with no channel in the client's hash table\n");
 553                         continue;
 554                 }
 555                 /* update all clients using this src_ip, that are not assigned
 556                  * to the team's address (curr_active_slave) and have a known
 557                  * unicast mac address.
 558                  */
 559                 if ((client_info->ip_src == src_ip) &&
 560                     !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
 561                                              bond->dev->dev_addr) &&
 562                     is_valid_ether_addr(client_info->mac_dst)) {
 563                         client_info->ntt = 1;
 564                         bond_info->rx_ntt = 1;
 565                 }
 566         }
 567
 568         spin_unlock(&bond->mode_lock);
 569 }
 570
 571 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
 572 {
 573         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 574         struct arp_pkt *arp = arp_pkt(skb);
 575         struct slave *assigned_slave, *curr_active_slave;
 576         struct rlb_client_info *client_info;
 577         u32 hash_index = 0;
 578
 579         spin_lock(&bond->mode_lock);
 580
 581         curr_active_slave = rcu_dereference(bond->curr_active_slave);
 582
 583         hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
 584         client_info = &(bond_info->rx_hashtbl[hash_index]);
 585
 586         if (client_info->assigned) {
 587                 if ((client_info->ip_src == arp->ip_src) &&
 588                     (client_info->ip_dst == arp->ip_dst)) {
 589                         /* the entry is already assigned to this client */
 590                         if (!is_broadcast_ether_addr(arp->mac_dst)) {
 591                                 /* update mac address from arp */
 592                                 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
 593                         }
 594                         ether_addr_copy(client_info->mac_src, arp->mac_src);
 595
 596                         assigned_slave = client_info->slave;
 597                         if (assigned_slave) {
 598                                 spin_unlock(&bond->mode_lock);
 599                                 return assigned_slave;
 600                         }
 601                 } else {
 602                         /* the entry is already assigned to some other client,
 603                          * move the old client to primary (curr_active_slave) so
 604                          * that the new client can be assigned to this entry.
 605                          */
 606                         if (curr_active_slave &&
 607                             client_info->slave != curr_active_slave) {
 608                                 client_info->slave = curr_active_slave;
 609                                 rlb_update_client(client_info);
 610                         }
 611                 }
 612         }
 613         /* assign a new slave */
 614         assigned_slave = __rlb_next_rx_slave(bond);
 615
 616         if (assigned_slave) {
 617                 if (!(client_info->assigned &&
 618                       client_info->ip_src == arp->ip_src)) {
 619                         /* ip_src is going to be updated,
 620                          * fix the src hash list
 621                          */
 622                         u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
 623                                                     sizeof(arp->ip_src));
 624                         rlb_src_unlink(bond, hash_index);
 625                         rlb_src_link(bond, hash_src, hash_index);
 626                 }
 627
 628                 client_info->ip_src = arp->ip_src;
 629                 client_info->ip_dst = arp->ip_dst;
 630                 /* arp->mac_dst is broadcast for arp reqeusts.
 631                  * will be updated with clients actual unicast mac address
 632                  * upon receiving an arp reply.
 633                  */
 634                 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
 635                 ether_addr_copy(client_info->mac_src, arp->mac_src);
 636                 client_info->slave = assigned_slave;
 637
 638                 if (is_valid_ether_addr(client_info->mac_dst)) {
 639                         client_info->ntt = 1;
 640                         bond->alb_info.rx_ntt = 1;
 641                 } else {
 642                         client_info->ntt = 0;
 643                 }
 644
 645                 if (vlan_get_tag(skb, &client_info->vlan_id))
 646                         client_info->vlan_id = 0;
 647
 648                 if (!client_info->assigned) {
 649                         u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
 650                         bond_info->rx_hashtbl_used_head = hash_index;
 651                         client_info->used_next = prev_tbl_head;
 652                         if (prev_tbl_head != RLB_NULL_INDEX) {
 653                                 bond_info->rx_hashtbl[prev_tbl_head].used_prev =
 654                                         hash_index;
 655                         }
 656                         client_info->assigned = 1;
 657                 }
 658         }
 659
 660         spin_unlock(&bond->mode_lock);
 661
 662         return assigned_slave;
 663 }
 664
 665 /* chooses (and returns) transmit channel for arp reply
 666  * does not choose channel for other arp types since they are
 667  * sent on the curr_active_slave
 668  */
 669 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 670 {
 671         struct arp_pkt *arp = arp_pkt(skb);
 672         struct slave *tx_slave = NULL;
 673
 674         /* Don't modify or load balance ARPs that do not originate locally
 675          * (e.g.,arrive via a bridge).
 676          */
 677         if (!bond_slave_has_mac_rx(bond, arp->mac_src))
 678                 return NULL;
 679
 680         if (arp->op_code == htons(ARPOP_REPLY)) {
 681                 /* the arp must be sent on the selected rx channel */
 682                 tx_slave = rlb_choose_channel(skb, bond);
 683                 if (tx_slave)
 684                         bond_hw_addr_copy(arp->mac_src, tx_slave->dev->dev_addr,
 685                                           tx_slave->dev->addr_len);
 686                 netdev_dbg(bond->dev, "Server sent ARP Reply packet\n");
 687         } else if (arp->op_code == htons(ARPOP_REQUEST)) {
 688                 /* Create an entry in the rx_hashtbl for this client as a
 689                  * place holder.
 690                  * When the arp reply is received the entry will be updated
 691                  * with the correct unicast address of the client.
 692                  */
 693                 rlb_choose_channel(skb, bond);
 694
 695                 /* The ARP reply packets must be delayed so that
 696                  * they can cancel out the influence of the ARP request.
 697                  */
 698                 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 699
 700                 /* arp requests are broadcast and are sent on the primary
 701                  * the arp request will collapse all clients on the subnet to
 702                  * the primary slave. We must register these clients to be
 703                  * updated with their assigned mac.
 704                  */
 705                 rlb_req_update_subnet_clients(bond, arp->ip_src);
 706                 netdev_dbg(bond->dev, "Server sent ARP Request packet\n");
 707         }
 708
 709         return tx_slave;
 710 }
 711
 712 static void rlb_rebalance(struct bonding *bond)
 713 {
 714         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 715         struct slave *assigned_slave;
 716         struct rlb_client_info *client_info;
 717         int ntt;
 718         u32 hash_index;
 719
 720         spin_lock_bh(&bond->mode_lock);
 721
 722         ntt = 0;
 723         hash_index = bond_info->rx_hashtbl_used_head;
 724         for (; hash_index != RLB_NULL_INDEX;
 725              hash_index = client_info->used_next) {
 726                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 727                 assigned_slave = __rlb_next_rx_slave(bond);
 728                 if (assigned_slave && (client_info->slave != assigned_slave)) {
 729                         client_info->slave = assigned_slave;
 730                         if (!is_zero_ether_addr(client_info->mac_dst)) {
 731                                 client_info->ntt = 1;
 732                                 ntt = 1;
 733                         }
 734                 }
 735         }
 736
 737         /* update the team's flag only after the whole iteration */
 738         if (ntt)
 739                 bond_info->rx_ntt = 1;
 740         spin_unlock_bh(&bond->mode_lock);
 741 }
 742
 743 /* Caller must hold mode_lock */
 744 static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
 745 {
 746         entry->used_next = RLB_NULL_INDEX;
 747         entry->used_prev = RLB_NULL_INDEX;
 748         entry->assigned = 0;
 749         entry->slave = NULL;
 750         entry->vlan_id = 0;
 751 }
 752 static void rlb_init_table_entry_src(struct rlb_client_info *entry)
 753 {
 754         entry->src_first = RLB_NULL_INDEX;
 755         entry->src_prev = RLB_NULL_INDEX;
 756         entry->src_next = RLB_NULL_INDEX;
 757 }
 758
 759 static void rlb_init_table_entry(struct rlb_client_info *entry)
 760 {
 761         memset(entry, 0, sizeof(struct rlb_client_info));
 762         rlb_init_table_entry_dst(entry);
 763         rlb_init_table_entry_src(entry);
 764 }
 765
 766 static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
 767 {
 768         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 769         u32 next_index = bond_info->rx_hashtbl[index].used_next;
 770         u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
 771
 772         if (index == bond_info->rx_hashtbl_used_head)
 773                 bond_info->rx_hashtbl_used_head = next_index;
 774         if (prev_index != RLB_NULL_INDEX)
 775                 bond_info->rx_hashtbl[prev_index].used_next = next_index;
 776         if (next_index != RLB_NULL_INDEX)
 777                 bond_info->rx_hashtbl[next_index].used_prev = prev_index;
 778 }
 779
 780 /* unlink a rlb hash table entry from the src list */
 781 static void rlb_src_unlink(struct bonding *bond, u32 index)
 782 {
 783         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 784         u32 next_index = bond_info->rx_hashtbl[index].src_next;
 785         u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
 786
 787         bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
 788         bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
 789
 790         if (next_index != RLB_NULL_INDEX)
 791                 bond_info->rx_hashtbl[next_index].src_prev = prev_index;
 792
 793         if (prev_index == RLB_NULL_INDEX)
 794                 return;
 795
 796         /* is prev_index pointing to the head of this list? */
 797         if (bond_info->rx_hashtbl[prev_index].src_first == index)
 798                 bond_info->rx_hashtbl[prev_index].src_first = next_index;
 799         else
 800                 bond_info->rx_hashtbl[prev_index].src_next = next_index;
 801
 802 }
 803
 804 static void rlb_delete_table_entry(struct bonding *bond, u32 index)
 805 {
 806         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 807         struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 808
 809         rlb_delete_table_entry_dst(bond, index);
 810         rlb_init_table_entry_dst(entry);
 811
 812         rlb_src_unlink(bond, index);
 813 }
 814
 815 /* add the rx_hashtbl[ip_dst_hash] entry to the list
 816  * of entries with identical ip_src_hash
 817  */
 818 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
 819 {
 820         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 821         u32 next;
 822
 823         bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
 824         next = bond_info->rx_hashtbl[ip_src_hash].src_first;
 825         bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
 826         if (next != RLB_NULL_INDEX)
 827                 bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
 828         bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
 829 }
 830
 831 /* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does
 832  * not match arp->mac_src
 833  */
 834 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
 835 {
 836         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 837         u32 ip_src_hash = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
 838         u32 index;
 839
 840         spin_lock_bh(&bond->mode_lock);
 841
 842         index = bond_info->rx_hashtbl[ip_src_hash].src_first;
 843         while (index != RLB_NULL_INDEX) {
 844                 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 845                 u32 next_index = entry->src_next;
 846                 if (entry->ip_src == arp->ip_src &&
 847                     !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
 848                                 rlb_delete_table_entry(bond, index);
 849                 index = next_index;
 850         }
 851         spin_unlock_bh(&bond->mode_lock);
 852 }
 853
 854 static int rlb_initialize(struct bonding *bond)
 855 {
 856         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 857         struct rlb_client_info  *new_hashtbl;
 858         int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 859         int i;
 860
 861         new_hashtbl = kmalloc(size, GFP_KERNEL);
 862         if (!new_hashtbl)
 863                 return -1;
 864
 865         spin_lock_bh(&bond->mode_lock);
 866
 867         bond_info->rx_hashtbl = new_hashtbl;
 868
 869         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 870
 871         for (i = 0; i < RLB_HASH_TABLE_SIZE; i++)
 872                 rlb_init_table_entry(bond_info->rx_hashtbl + i);
 873
 874         spin_unlock_bh(&bond->mode_lock);
 875
 876         /* register to receive ARPs */
 877         bond->recv_probe = rlb_arp_recv;
 878
 879         return 0;
 880 }
 881
 882 static void rlb_deinitialize(struct bonding *bond)
 883 {
 884         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 885
 886         spin_lock_bh(&bond->mode_lock);
 887
 888         kfree(bond_info->rx_hashtbl);
 889         bond_info->rx_hashtbl = NULL;
 890         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 891
 892         spin_unlock_bh(&bond->mode_lock);
 893 }
 894
 895 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 896 {
 897         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 898         u32 curr_index;
 899
 900         spin_lock_bh(&bond->mode_lock);
 901
 902         curr_index = bond_info->rx_hashtbl_used_head;
 903         while (curr_index != RLB_NULL_INDEX) {
 904                 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 905                 u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
 906
 907                 if (curr->vlan_id == vlan_id)
 908                         rlb_delete_table_entry(bond, curr_index);
 909
 910                 curr_index = next_index;
 911         }
 912
 913         spin_unlock_bh(&bond->mode_lock);
 914 }
 915
 916 /*********************** tlb/rlb shared functions *********************/
 917
 918 static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
 919                             __be16 vlan_proto, u16 vid)
 920 {
 921         struct learning_pkt pkt;
 922         struct sk_buff *skb;
 923         int size = sizeof(struct learning_pkt);
 924
 925         memset(&pkt, 0, size);
 926         ether_addr_copy(pkt.mac_dst, mac_addr);
 927         ether_addr_copy(pkt.mac_src, mac_addr);
 928         pkt.type = cpu_to_be16(ETH_P_LOOPBACK);
 929
 930         skb = dev_alloc_skb(size);
 931         if (!skb)
 932                 return;
 933
 934         skb_put_data(skb, &pkt, size);
 935
 936         skb_reset_mac_header(skb);
 937         skb->network_header = skb->mac_header + ETH_HLEN;
 938         skb->protocol = pkt.type;
 939         skb->priority = TC_PRIO_CONTROL;
 940         skb->dev = slave->dev;
 941
 942         netdev_dbg(slave->bond->dev,
 943                    "Send learning packet: dev %s mac %pM vlan %d\n",
 944                    slave->dev->name, mac_addr, vid);
 945
 946         if (vid)
 947                 __vlan_hwaccel_put_tag(skb, vlan_proto, vid);
 948
 949         dev_queue_xmit(skb);
 950 }
 951
 952 struct alb_walk_data {
 953         struct bonding *bond;
 954         struct slave *slave;
 955         u8 *mac_addr;
 956         bool strict_match;
 957 };
 958
 959 static int alb_upper_dev_walk(struct net_device *upper, void *_data)
 960 {
 961         struct alb_walk_data *data = _data;
 962         bool strict_match = data->strict_match;
 963         struct bonding *bond = data->bond;
 964         struct slave *slave = data->slave;
 965         u8 *mac_addr = data->mac_addr;
 966         struct bond_vlan_tag *tags;
 967
 968         if (is_vlan_dev(upper) &&
 969             bond->nest_level == vlan_get_encap_level(upper) - 1) {
 970                 if (upper->addr_assign_type == NET_ADDR_STOLEN) {
 971                         alb_send_lp_vid(slave, mac_addr,
 972                                         vlan_dev_vlan_proto(upper),
 973                                         vlan_dev_vlan_id(upper));
 974                 } else {
 975                         alb_send_lp_vid(slave, upper->dev_addr,
 976                                         vlan_dev_vlan_proto(upper),
 977                                         vlan_dev_vlan_id(upper));
 978                 }
 979         }
 980
 981         /* If this is a macvlan device, then only send updates
 982          * when strict_match is turned off.
 983          */
 984         if (netif_is_macvlan(upper) && !strict_match) {
 985                 tags = bond_verify_device_path(bond->dev, upper, 0);
 986                 if (IS_ERR_OR_NULL(tags))
 987                         BUG();
 988                 alb_send_lp_vid(slave, upper->dev_addr,
 989                                 tags[0].vlan_proto, tags[0].vlan_id);
 990                 kfree(tags);
 991         }
 992
 993         return 0;
 994 }
 995
 996 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
 997                                       bool strict_match)
 998 {
 999         struct bonding *bond = bond_get_bond_by_slave(slave);
1000         struct alb_walk_data data = {
1001                 .strict_match = strict_match,
1002                 .mac_addr = mac_addr,
1003                 .slave = slave,
1004                 .bond = bond,
1005         };
1006
1007         /* send untagged */
1008         alb_send_lp_vid(slave, mac_addr, 0, 0);
1009
1010         /* loop through all devices and see if we need to send a packet
1011          * for that device.
1012          */
1013         rcu_read_lock();
1014         netdev_walk_all_upper_dev_rcu(bond->dev, alb_upper_dev_walk, &data);
1015         rcu_read_unlock();
1016 }
1017
1018 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[],
1019                                   unsigned int len)
1020 {
1021         struct net_device *dev = slave->dev;
1022         struct sockaddr_storage ss;
1023
1024         if (BOND_MODE(slave->bond) == BOND_MODE_TLB) {
1025                 memcpy(dev->dev_addr, addr, len);
1026                 return 0;
1027         }
1028
1029         /* for rlb each slave must have a unique hw mac addresses so that
1030          * each slave will receive packets destined to a different mac
1031          */
1032         memcpy(ss.__data, addr, len);
1033         ss.ss_family = dev->type;
1034         if (dev_set_mac_address(dev, (struct sockaddr *)&ss, NULL)) {
1035                 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",
1036                            dev->name);
1037                 return -EOPNOTSUPP;
1038         }
1039         return 0;
1040 }
1041
1042 /* Swap MAC addresses between two slaves.
1043  *
1044  * Called with RTNL held, and no other locks.
1045  */
1046 static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1047 {
1048         u8 tmp_mac_addr[MAX_ADDR_LEN];
1049
1050         bond_hw_addr_copy(tmp_mac_addr, slave1->dev->dev_addr,
1051                           slave1->dev->addr_len);
1052         alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr,
1053                                slave2->dev->addr_len);
1054         alb_set_slave_mac_addr(slave2, tmp_mac_addr,
1055                                slave1->dev->addr_len);
1056
1057 }
1058
1059 /* Send learning packets after MAC address swap.
1060  *
1061  * Called with RTNL and no other locks
1062  */
1063 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1064                                 struct slave *slave2)
1065 {
1066         int slaves_state_differ = (bond_slave_can_tx(slave1) != bond_slave_can_tx(slave2));
1067         struct slave *disabled_slave = NULL;
1068
1069         ASSERT_RTNL();
1070
1071         /* fasten the change in the switch */
1072         if (bond_slave_can_tx(slave1)) {
1073                 alb_send_learning_packets(slave1, slave1->dev->dev_addr, false);
1074                 if (bond->alb_info.rlb_enabled) {
1075                         /* inform the clients that the mac address
1076                          * has changed
1077                          */
1078                         rlb_req_update_slave_clients(bond, slave1);
1079                 }
1080         } else {
1081                 disabled_slave = slave1;
1082         }
1083
1084         if (bond_slave_can_tx(slave2)) {
1085                 alb_send_learning_packets(slave2, slave2->dev->dev_addr, false);
1086                 if (bond->alb_info.rlb_enabled) {
1087                         /* inform the clients that the mac address
1088                          * has changed
1089                          */
1090                         rlb_req_update_slave_clients(bond, slave2);
1091                 }
1092         } else {
1093                 disabled_slave = slave2;
1094         }
1095
1096         if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1097                 /* A disabled slave was assigned an active mac addr */
1098                 rlb_teach_disabled_mac_on_primary(bond,
1099                                                   disabled_slave->dev->dev_addr);
1100         }
1101 }
1102
1103 /**
1104  * alb_change_hw_addr_on_detach
1105  * @bond: bonding we're working on
1106  * @slave: the slave that was just detached
1107  *
1108  * We assume that @slave was already detached from the slave list.
1109  *
1110  * If @slave's permanent hw address is different both from its current
1111  * address and from @bond's address, then somewhere in the bond there's
1112  * a slave that has @slave's permanet address as its current address.
1113  * We'll make sure that that slave no longer uses @slave's permanent address.
1114  *
1115  * Caller must hold RTNL and no other locks
1116  */
1117 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1118 {
1119         int perm_curr_diff;
1120         int perm_bond_diff;
1121         struct slave *found_slave;
1122
1123         perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1124                                                   slave->dev->dev_addr);
1125         perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1126                                                   bond->dev->dev_addr);
1127
1128         if (perm_curr_diff && perm_bond_diff) {
1129                 found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1130
1131                 if (found_slave) {
1132                         alb_swap_mac_addr(slave, found_slave);
1133                         alb_fasten_mac_swap(bond, slave, found_slave);
1134                 }
1135         }
1136 }
1137
1138 /**
1139  * alb_handle_addr_collision_on_attach
1140  * @bond: bonding we're working on
1141  * @slave: the slave that was just attached
1142  *
1143  * checks uniqueness of slave's mac address and handles the case the
1144  * new slave uses the bonds mac address.
1145  *
1146  * If the permanent hw address of @slave is @bond's hw address, we need to
1147  * find a different hw address to give @slave, that isn't in use by any other
1148  * slave in the bond. This address must be, of course, one of the permanent
1149  * addresses of the other slaves.
1150  *
1151  * We go over the slave list, and for each slave there we compare its
1152  * permanent hw address with the current address of all the other slaves.
1153  * If no match was found, then we've found a slave with a permanent address
1154  * that isn't used by any other slave in the bond, so we can assign it to
1155  * @slave.
1156  *
1157  * assumption: this function is called before @slave is attached to the
1158  *             bond slave list.
1159  */
1160 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1161 {
1162         struct slave *has_bond_addr = rcu_access_pointer(bond->curr_active_slave);
1163         struct slave *tmp_slave1, *free_mac_slave = NULL;
1164         struct list_head *iter;
1165
1166         if (!bond_has_slaves(bond)) {
1167                 /* this is the first slave */
1168                 return 0;
1169         }
1170
1171         /* if slave's mac address differs from bond's mac address
1172          * check uniqueness of slave's mac address against the other
1173          * slaves in the bond.
1174          */
1175         if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1176                 if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1177                         return 0;
1178
1179                 /* Try setting slave mac to bond address and fall-through
1180                  * to code handling that situation below...
1181                  */
1182                 alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1183                                        bond->dev->addr_len);
1184         }
1185
1186         /* The slave's address is equal to the address of the bond.
1187          * Search for a spare address in the bond for this slave.
1188          */
1189         bond_for_each_slave(bond, tmp_slave1, iter) {
1190                 if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1191                         /* no slave has tmp_slave1's perm addr
1192                          * as its curr addr
1193                          */
1194                         free_mac_slave = tmp_slave1;
1195                         break;
1196                 }
1197
1198                 if (!has_bond_addr) {
1199                         if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1200                                                     bond->dev->dev_addr)) {
1201
1202                                 has_bond_addr = tmp_slave1;
1203                         }
1204                 }
1205         }
1206
1207         if (free_mac_slave) {
1208                 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1209                                        free_mac_slave->dev->addr_len);
1210
1211                 netdev_warn(bond->dev, "the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1212                             slave->dev->name, free_mac_slave->dev->name);
1213
1214         } else if (has_bond_addr) {
1215                 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",
1216                            slave->dev->name);
1217                 return -EFAULT;
1218         }
1219
1220         return 0;
1221 }
1222
1223 /**
1224  * alb_set_mac_address
1225  * @bond:
1226  * @addr:
1227  *
1228  * In TLB mode all slaves are configured to the bond's hw address, but set
1229  * their dev_addr field to different addresses (based on their permanent hw
1230  * addresses).
1231  *
1232  * For each slave, this function sets the interface to the new address and then
1233  * changes its dev_addr field to its previous value.
1234  *
1235  * Unwinding assumes bond's mac address has not yet changed.
1236  */
1237 static int alb_set_mac_address(struct bonding *bond, void *addr)
1238 {
1239         struct slave *slave, *rollback_slave;
1240         struct list_head *iter;
1241         struct sockaddr_storage ss;
1242         char tmp_addr[MAX_ADDR_LEN];
1243         int res;
1244
1245         if (bond->alb_info.rlb_enabled)
1246                 return 0;
1247
1248         bond_for_each_slave(bond, slave, iter) {
1249                 /* save net_device's current hw address */
1250                 bond_hw_addr_copy(tmp_addr, slave->dev->dev_addr,
1251                                   slave->dev->addr_len);
1252
1253                 res = dev_set_mac_address(slave->dev, addr, NULL);
1254
1255                 /* restore net_device's hw address */
1256                 bond_hw_addr_copy(slave->dev->dev_addr, tmp_addr,
1257                                   slave->dev->addr_len);
1258
1259                 if (res)
1260                         goto unwind;
1261         }
1262
1263         return 0;
1264
1265 unwind:
1266         memcpy(ss.__data, bond->dev->dev_addr, bond->dev->addr_len);
1267         ss.ss_family = bond->dev->type;
1268
1269         /* unwind from head to the slave that failed */
1270         bond_for_each_slave(bond, rollback_slave, iter) {
1271                 if (rollback_slave == slave)
1272                         break;
1273                 bond_hw_addr_copy(tmp_addr, rollback_slave->dev->dev_addr,
1274                                   rollback_slave->dev->addr_len);
1275                 dev_set_mac_address(rollback_slave->dev,
1276                                     (struct sockaddr *)&ss, NULL);
1277                 bond_hw_addr_copy(rollback_slave->dev->dev_addr, tmp_addr,
1278                                   rollback_slave->dev->addr_len);
1279         }
1280
1281         return res;
1282 }
1283
1284 /************************ exported alb funcions ************************/
1285
1286 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1287 {
1288         int res;
1289
1290         res = tlb_initialize(bond);
1291         if (res)
1292                 return res;
1293
1294         if (rlb_enabled) {
1295                 bond->alb_info.rlb_enabled = 1;
1296                 res = rlb_initialize(bond);
1297                 if (res) {
1298                         tlb_deinitialize(bond);
1299                         return res;
1300                 }
1301         } else {
1302                 bond->alb_info.rlb_enabled = 0;
1303         }
1304
1305         return 0;
1306 }
1307
1308 void bond_alb_deinitialize(struct bonding *bond)
1309 {
1310         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1311
1312         tlb_deinitialize(bond);
1313
1314         if (bond_info->rlb_enabled)
1315                 rlb_deinitialize(bond);
1316 }
1317
1318 static netdev_tx_t bond_do_alb_xmit(struct sk_buff *skb, struct bonding *bond,
1319                                     struct slave *tx_slave)
1320 {
1321         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1322         struct ethhdr *eth_data = eth_hdr(skb);
1323
1324         if (!tx_slave) {
1325                 /* unbalanced or unassigned, send through primary */
1326                 tx_slave = rcu_dereference(bond->curr_active_slave);
1327                 if (bond->params.tlb_dynamic_lb)
1328                         bond_info->unbalanced_load += skb->len;
1329         }
1330
1331         if (tx_slave && bond_slave_can_tx(tx_slave)) {
1332                 if (tx_slave != rcu_access_pointer(bond->curr_active_slave)) {
1333                         ether_addr_copy(eth_data->h_source,
1334                                         tx_slave->dev->dev_addr);
1335                 }
1336
1337                 bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1338                 goto out;
1339         }
1340
1341         if (tx_slave && bond->params.tlb_dynamic_lb) {
1342                 spin_lock(&bond->mode_lock);
1343                 __tlb_clear_slave(bond, tx_slave, 0);
1344                 spin_unlock(&bond->mode_lock);
1345         }
1346
1347         /* no suitable interface, frame not sent */
1348         bond_tx_drop(bond->dev, skb);
1349 out:
1350         return NETDEV_TX_OK;
1351 }
1352
1353 netdev_tx_t bond_tlb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1354 {
1355         struct bonding *bond = netdev_priv(bond_dev);
1356         struct ethhdr *eth_data;
1357         struct slave *tx_slave = NULL;
1358         u32 hash_index;
1359
1360         skb_reset_mac_header(skb);
1361         eth_data = eth_hdr(skb);
1362
1363         /* Do not TX balance any multicast or broadcast */
1364         if (!is_multicast_ether_addr(eth_data->h_dest)) {
1365                 switch (skb->protocol) {
1366                 case htons(ETH_P_IP):
1367                 case htons(ETH_P_IPX):
1368                     /* In case of IPX, it will falback to L2 hash */
1369                 case htons(ETH_P_IPV6):
1370                         hash_index = bond_xmit_hash(bond, skb);
1371                         if (bond->params.tlb_dynamic_lb) {
1372                                 tx_slave = tlb_choose_channel(bond,
1373                                                               hash_index & 0xFF,
1374                                                               skb->len);
1375                         } else {
1376                                 struct bond_up_slave *slaves;
1377                                 unsigned int count;
1378
1379                                 slaves = rcu_dereference(bond->slave_arr);
1380                                 count = slaves ? READ_ONCE(slaves->count) : 0;
1381                                 if (likely(count))
1382                                         tx_slave = slaves->arr[hash_index %
1383                                                                count];
1384                         }
1385                         break;
1386                 }
1387         }
1388         return bond_do_alb_xmit(skb, bond, tx_slave);
1389 }
1390
1391 netdev_tx_t bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1392 {
1393         struct bonding *bond = netdev_priv(bond_dev);
1394         struct ethhdr *eth_data;
1395         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1396         struct slave *tx_slave = NULL;
1397         static const __be32 ip_bcast = htonl(0xffffffff);
1398         int hash_size = 0;
1399         bool do_tx_balance = true;
1400         u32 hash_index = 0;
1401         const u8 *hash_start = NULL;
1402         struct ipv6hdr *ip6hdr;
1403
1404         skb_reset_mac_header(skb);
1405         eth_data = eth_hdr(skb);
1406
1407         switch (ntohs(skb->protocol)) {
1408         case ETH_P_IP: {
1409                 const struct iphdr *iph = ip_hdr(skb);
1410
1411                 if (is_broadcast_ether_addr(eth_data->h_dest) ||
1412                     iph->daddr == ip_bcast ||
1413                     iph->protocol == IPPROTO_IGMP) {
1414                         do_tx_balance = false;
1415                         break;
1416                 }
1417                 hash_start = (char *)&(iph->daddr);
1418                 hash_size = sizeof(iph->daddr);
1419         }
1420                 break;
1421         case ETH_P_IPV6:
1422                 /* IPv6 doesn't really use broadcast mac address, but leave
1423                  * that here just in case.
1424                  */
1425                 if (is_broadcast_ether_addr(eth_data->h_dest)) {
1426                         do_tx_balance = false;
1427                         break;
1428                 }
1429
1430                 /* IPv6 uses all-nodes multicast as an equivalent to
1431                  * broadcasts in IPv4.
1432                  */
1433                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1434                         do_tx_balance = false;
1435                         break;
1436                 }
1437
1438                 /* Additianally, DAD probes should not be tx-balanced as that
1439                  * will lead to false positives for duplicate addresses and
1440                  * prevent address configuration from working.
1441                  */
1442                 ip6hdr = ipv6_hdr(skb);
1443                 if (ipv6_addr_any(&ip6hdr->saddr)) {
1444                         do_tx_balance = false;
1445                         break;
1446                 }
1447
1448                 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1449                 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1450                 break;
1451         case ETH_P_IPX:
1452                 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1453                         /* something is wrong with this packet */
1454                         do_tx_balance = false;
1455                         break;
1456                 }
1457
1458                 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1459                         /* The only protocol worth balancing in
1460                          * this family since it has an "ARP" like
1461                          * mechanism
1462                          */
1463                         do_tx_balance = false;
1464                         break;
1465                 }
1466
1467                 hash_start = (char *)eth_data->h_dest;
1468                 hash_size = ETH_ALEN;
1469                 break;
1470         case ETH_P_ARP:
1471                 do_tx_balance = false;
1472                 if (bond_info->rlb_enabled)
1473                         tx_slave = rlb_arp_xmit(skb, bond);
1474                 break;
1475         default:
1476                 do_tx_balance = false;
1477                 break;
1478         }
1479
1480         if (do_tx_balance) {
1481                 if (bond->params.tlb_dynamic_lb) {
1482                         hash_index = _simple_hash(hash_start, hash_size);
1483                         tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1484                 } else {
1485                         /*
1486                          * do_tx_balance means we are free to select the tx_slave
1487                          * So we do exactly what tlb would do for hash selection
1488                          */
1489
1490                         struct bond_up_slave *slaves;
1491                         unsigned int count;
1492
1493                         slaves = rcu_dereference(bond->slave_arr);
1494                         count = slaves ? READ_ONCE(slaves->count) : 0;
1495                         if (likely(count))
1496                                 tx_slave = slaves->arr[bond_xmit_hash(bond, skb) %
1497                                                        count];
1498                 }
1499         }
1500
1501         return bond_do_alb_xmit(skb, bond, tx_slave);
1502 }
1503
1504 void bond_alb_monitor(struct work_struct *work)
1505 {
1506         struct bonding *bond = container_of(work, struct bonding,
1507                                             alb_work.work);
1508         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1509         struct list_head *iter;
1510         struct slave *slave;
1511
1512         if (!bond_has_slaves(bond)) {
1513                 bond_info->tx_rebalance_counter = 0;
1514                 bond_info->lp_counter = 0;
1515                 goto re_arm;
1516         }
1517
1518         rcu_read_lock();
1519
1520         bond_info->tx_rebalance_counter++;
1521         bond_info->lp_counter++;
1522
1523         /* send learning packets */
1524         if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1525                 bool strict_match;
1526
1527                 bond_for_each_slave_rcu(bond, slave, iter) {
1528                         /* If updating current_active, use all currently
1529                          * user mac addreses (!strict_match).  Otherwise, only
1530                          * use mac of the slave device.
1531                          * In RLB mode, we always use strict matches.
1532                          */
1533                         strict_match = (slave != rcu_access_pointer(bond->curr_active_slave) ||
1534                                         bond_info->rlb_enabled);
1535                         alb_send_learning_packets(slave, slave->dev->dev_addr,
1536                                                   strict_match);
1537                 }
1538                 bond_info->lp_counter = 0;
1539         }
1540
1541         /* rebalance tx traffic */
1542         if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1543                 bond_for_each_slave_rcu(bond, slave, iter) {
1544                         tlb_clear_slave(bond, slave, 1);
1545                         if (slave == rcu_access_pointer(bond->curr_active_slave)) {
1546                                 SLAVE_TLB_INFO(slave).load =
1547                                         bond_info->unbalanced_load /
1548                                                 BOND_TLB_REBALANCE_INTERVAL;
1549                                 bond_info->unbalanced_load = 0;
1550                         }
1551                 }
1552                 bond_info->tx_rebalance_counter = 0;
1553         }
1554
1555         if (bond_info->rlb_enabled) {
1556                 if (bond_info->primary_is_promisc &&
1557                     (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1558
1559                         /* dev_set_promiscuity requires rtnl and
1560                          * nothing else.  Avoid race with bond_close.
1561                          */
1562                         rcu_read_unlock();
1563                         if (!rtnl_trylock())
1564                                 goto re_arm;
1565
1566                         bond_info->rlb_promisc_timeout_counter = 0;
1567
1568                         /* If the primary was set to promiscuous mode
1569                          * because a slave was disabled then
1570                          * it can now leave promiscuous mode.
1571                          */
1572                         dev_set_promiscuity(rtnl_dereference(bond->curr_active_slave)->dev,
1573                                             -1);
1574                         bond_info->primary_is_promisc = 0;
1575
1576                         rtnl_unlock();
1577                         rcu_read_lock();
1578                 }
1579
1580                 if (bond_info->rlb_rebalance) {
1581                         bond_info->rlb_rebalance = 0;
1582                         rlb_rebalance(bond);
1583                 }
1584
1585                 /* check if clients need updating */
1586                 if (bond_info->rx_ntt) {
1587                         if (bond_info->rlb_update_delay_counter) {
1588                                 --bond_info->rlb_update_delay_counter;
1589                         } else {
1590                                 rlb_update_rx_clients(bond);
1591                                 if (bond_info->rlb_update_retry_counter)
1592                                         --bond_info->rlb_update_retry_counter;
1593                                 else
1594                                         bond_info->rx_ntt = 0;
1595                         }
1596                 }
1597         }
1598         rcu_read_unlock();
1599 re_arm:
1600         queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1601 }
1602
1603 /* assumption: called before the slave is attached to the bond
1604  * and not locked by the bond lock
1605  */
1606 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1607 {
1608         int res;
1609
1610         res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1611                                      slave->dev->addr_len);
1612         if (res)
1613                 return res;
1614
1615         res = alb_handle_addr_collision_on_attach(bond, slave);
1616         if (res)
1617                 return res;
1618
1619         tlb_init_slave(slave);
1620
1621         /* order a rebalance ASAP */
1622         bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1623
1624         if (bond->alb_info.rlb_enabled)
1625                 bond->alb_info.rlb_rebalance = 1;
1626
1627         return 0;
1628 }
1629
1630 /* Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1631  * if necessary.
1632  *
1633  * Caller must hold RTNL and no other locks
1634  */
1635 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1636 {
1637         if (bond_has_slaves(bond))
1638                 alb_change_hw_addr_on_detach(bond, slave);
1639
1640         tlb_clear_slave(bond, slave, 0);
1641
1642         if (bond->alb_info.rlb_enabled) {
1643                 bond->alb_info.rx_slave = NULL;
1644                 rlb_clear_slave(bond, slave);
1645         }
1646
1647 }
1648
1649 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1650 {
1651         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1652
1653         if (link == BOND_LINK_DOWN) {
1654                 tlb_clear_slave(bond, slave, 0);
1655                 if (bond->alb_info.rlb_enabled)
1656                         rlb_clear_slave(bond, slave);
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         if (bond_is_nondyn_tlb(bond)) {
1671                 if (bond_update_slave_arr(bond, NULL))
1672                         pr_err("Failed to build slave-array for TLB mode.\n");
1673         }
1674 }
1675
1676 /**
1677  * bond_alb_handle_active_change - assign new curr_active_slave
1678  * @bond: our bonding struct
1679  * @new_slave: new slave to assign
1680  *
1681  * Set the bond->curr_active_slave to @new_slave and handle
1682  * mac address swapping and promiscuity changes as needed.
1683  *
1684  * Caller must hold RTNL
1685  */
1686 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1687 {
1688         struct slave *swap_slave;
1689         struct slave *curr_active;
1690
1691         curr_active = rtnl_dereference(bond->curr_active_slave);
1692         if (curr_active == new_slave)
1693                 return;
1694
1695         if (curr_active && bond->alb_info.primary_is_promisc) {
1696                 dev_set_promiscuity(curr_active->dev, -1);
1697                 bond->alb_info.primary_is_promisc = 0;
1698                 bond->alb_info.rlb_promisc_timeout_counter = 0;
1699         }
1700
1701         swap_slave = curr_active;
1702         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1703
1704         if (!new_slave || !bond_has_slaves(bond))
1705                 return;
1706
1707         /* set the new curr_active_slave to the bonds mac address
1708          * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1709          */
1710         if (!swap_slave)
1711                 swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1712
1713         /* Arrange for swap_slave and new_slave to temporarily be
1714          * ignored so we can mess with their MAC addresses without
1715          * fear of interference from transmit activity.
1716          */
1717         if (swap_slave)
1718                 tlb_clear_slave(bond, swap_slave, 1);
1719         tlb_clear_slave(bond, new_slave, 1);
1720
1721         /* in TLB mode, the slave might flip down/up with the old dev_addr,
1722          * and thus filter bond->dev_addr's packets, so force bond's mac
1723          */
1724         if (BOND_MODE(bond) == BOND_MODE_TLB) {
1725                 struct sockaddr_storage ss;
1726                 u8 tmp_addr[MAX_ADDR_LEN];
1727
1728                 bond_hw_addr_copy(tmp_addr, new_slave->dev->dev_addr,
1729                                   new_slave->dev->addr_len);
1730
1731                 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
1732                                   bond->dev->addr_len);
1733                 ss.ss_family = bond->dev->type;
1734                 /* we don't care if it can't change its mac, best effort */
1735                 dev_set_mac_address(new_slave->dev, (struct sockaddr *)&ss,
1736                                     NULL);
1737
1738                 bond_hw_addr_copy(new_slave->dev->dev_addr, tmp_addr,
1739                                   new_slave->dev->addr_len);
1740         }
1741
1742         /* curr_active_slave must be set before calling alb_swap_mac_addr */
1743         if (swap_slave) {
1744                 /* swap mac address */
1745                 alb_swap_mac_addr(swap_slave, new_slave);
1746                 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1747         } else {
1748                 /* set the new_slave to the bond mac address */
1749                 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1750                                        bond->dev->addr_len);
1751                 alb_send_learning_packets(new_slave, bond->dev->dev_addr,
1752                                           false);
1753         }
1754 }
1755
1756 /* Called with RTNL */
1757 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1758 {
1759         struct bonding *bond = netdev_priv(bond_dev);
1760         struct sockaddr_storage *ss = addr;
1761         struct slave *curr_active;
1762         struct slave *swap_slave;
1763         int res;
1764
1765         if (!is_valid_ether_addr(ss->__data))
1766                 return -EADDRNOTAVAIL;
1767
1768         res = alb_set_mac_address(bond, addr);
1769         if (res)
1770                 return res;
1771
1772         bond_hw_addr_copy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
1773
1774         /* If there is no curr_active_slave there is nothing else to do.
1775          * Otherwise we'll need to pass the new address to it and handle
1776          * duplications.
1777          */
1778         curr_active = rtnl_dereference(bond->curr_active_slave);
1779         if (!curr_active)
1780                 return 0;
1781
1782         swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1783
1784         if (swap_slave) {
1785                 alb_swap_mac_addr(swap_slave, curr_active);
1786                 alb_fasten_mac_swap(bond, swap_slave, curr_active);
1787         } else {
1788                 alb_set_slave_mac_addr(curr_active, bond_dev->dev_addr,
1789                                        bond_dev->addr_len);
1790
1791                 alb_send_learning_packets(curr_active,
1792                                           bond_dev->dev_addr, false);
1793                 if (bond->alb_info.rlb_enabled) {
1794                         /* inform clients mac address has changed */
1795                         rlb_req_update_slave_clients(bond, curr_active);
1796                 }
1797         }
1798
1799         return 0;
1800 }
1801
1802 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1803 {
1804         if (bond->alb_info.rlb_enabled)
1805                 rlb_clear_vlan(bond, vlan_id);
1806 }
1807