drivers/net/bonding/bond_alb.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved.
   4  */
   5
   6 #include <linux/skbuff.h>
   7 #include <linux/netdevice.h>
   8 #include <linux/etherdevice.h>
   9 #include <linux/pkt_sched.h>
  10 #include <linux/spinlock.h>
  11 #include <linux/slab.h>
  12 #include <linux/timer.h>
  13 #include <linux/ip.h>
  14 #include <linux/ipv6.h>
  15 #include <linux/if_arp.h>
  16 #include <linux/if_ether.h>
  17 #include <linux/if_bonding.h>
  18 #include <linux/if_vlan.h>
  19 #include <linux/in.h>
  20 #include <net/arp.h>
  21 #include <net/ipv6.h>
  22 #include <net/ndisc.h>
  23 #include <asm/byteorder.h>
  24 #include <net/bonding.h>
  25 #include <net/bond_alb.h>
  26
  27 static const u8 mac_v6_allmcast[ETH_ALEN + 2] __long_aligned = {
  28         0x33, 0x33, 0x00, 0x00, 0x00, 0x01
  29 };
  30 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
  31
  32 #pragma pack(1)
  33 struct learning_pkt {
  34         u8 mac_dst[ETH_ALEN];
  35         u8 mac_src[ETH_ALEN];
  36         __be16 type;
  37         u8 padding[ETH_ZLEN - ETH_HLEN];
  38 };
  39
  40 struct arp_pkt {
  41         __be16  hw_addr_space;
  42         __be16  prot_addr_space;
  43         u8      hw_addr_len;
  44         u8      prot_addr_len;
  45         __be16  op_code;
  46         u8      mac_src[ETH_ALEN];      /* sender hardware address */
  47         __be32  ip_src;                 /* sender IP address */
  48         u8      mac_dst[ETH_ALEN];      /* target hardware address */
  49         __be32  ip_dst;                 /* target IP address */
  50 };
  51 #pragma pack()
  52
  53 /* Forward declaration */
  54 static void alb_send_learning_packets(struct slave *slave, const u8 mac_addr[],
  55                                       bool strict_match);
  56 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp);
  57 static void rlb_src_unlink(struct bonding *bond, u32 index);
  58 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash,
  59                          u32 ip_dst_hash);
  60
  61 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
  62 {
  63         int i;
  64         u8 hash = 0;
  65
  66         for (i = 0; i < hash_size; i++)
  67                 hash ^= hash_start[i];
  68
  69         return hash;
  70 }
  71
  72 /*********************** tlb specific functions ***************************/
  73
  74 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
  75 {
  76         if (save_load) {
  77                 entry->load_history = 1 + entry->tx_bytes /
  78                                       BOND_TLB_REBALANCE_INTERVAL;
  79                 entry->tx_bytes = 0;
  80         }
  81
  82         entry->tx_slave = NULL;
  83         entry->next = TLB_NULL_INDEX;
  84         entry->prev = TLB_NULL_INDEX;
  85 }
  86
  87 static inline void tlb_init_slave(struct slave *slave)
  88 {
  89         SLAVE_TLB_INFO(slave).load = 0;
  90         SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
  91 }
  92
  93 static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
  94                          int save_load)
  95 {
  96         struct tlb_client_info *tx_hash_table;
  97         u32 index;
  98
  99         /* clear slave from tx_hashtbl */
 100         tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
 101
 102         /* skip this if we've already freed the tx hash table */
 103         if (tx_hash_table) {
 104                 index = SLAVE_TLB_INFO(slave).head;
 105                 while (index != TLB_NULL_INDEX) {
 106                         u32 next_index = tx_hash_table[index].next;
 107
 108                         tlb_init_table_entry(&tx_hash_table[index], save_load);
 109                         index = next_index;
 110                 }
 111         }
 112
 113         tlb_init_slave(slave);
 114 }
 115
 116 static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
 117                          int save_load)
 118 {
 119         spin_lock_bh(&bond->mode_lock);
 120         __tlb_clear_slave(bond, slave, save_load);
 121         spin_unlock_bh(&bond->mode_lock);
 122 }
 123
 124 /* Must be called before starting the monitor timer */
 125 static int tlb_initialize(struct bonding *bond)
 126 {
 127         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 128         int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 129         struct tlb_client_info *new_hashtbl;
 130         int i;
 131
 132         new_hashtbl = kzalloc(size, GFP_KERNEL);
 133         if (!new_hashtbl)
 134                 return -ENOMEM;
 135
 136         spin_lock_bh(&bond->mode_lock);
 137
 138         bond_info->tx_hashtbl = new_hashtbl;
 139
 140         for (i = 0; i < TLB_HASH_TABLE_SIZE; i++)
 141                 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
 142
 143         spin_unlock_bh(&bond->mode_lock);
 144
 145         return 0;
 146 }
 147
 148 /* Must be called only after all slaves have been released */
 149 static void tlb_deinitialize(struct bonding *bond)
 150 {
 151         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 152
 153         spin_lock_bh(&bond->mode_lock);
 154
 155         kfree(bond_info->tx_hashtbl);
 156         bond_info->tx_hashtbl = NULL;
 157
 158         spin_unlock_bh(&bond->mode_lock);
 159 }
 160
 161 static long long compute_gap(struct slave *slave)
 162 {
 163         return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
 164                (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
 165 }
 166
 167 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
 168 {
 169         struct slave *slave, *least_loaded;
 170         struct list_head *iter;
 171         long long max_gap;
 172
 173         least_loaded = NULL;
 174         max_gap = LLONG_MIN;
 175
 176         /* Find the slave with the largest gap */
 177         bond_for_each_slave_rcu(bond, slave, iter) {
 178                 if (bond_slave_can_tx(slave)) {
 179                         long long gap = compute_gap(slave);
 180
 181                         if (max_gap < gap) {
 182                                 least_loaded = slave;
 183                                 max_gap = gap;
 184                         }
 185                 }
 186         }
 187
 188         return least_loaded;
 189 }
 190
 191 static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
 192                                                 u32 skb_len)
 193 {
 194         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 195         struct tlb_client_info *hash_table;
 196         struct slave *assigned_slave;
 197
 198         hash_table = bond_info->tx_hashtbl;
 199         assigned_slave = hash_table[hash_index].tx_slave;
 200         if (!assigned_slave) {
 201                 assigned_slave = tlb_get_least_loaded_slave(bond);
 202
 203                 if (assigned_slave) {
 204                         struct tlb_slave_info *slave_info =
 205                                 &(SLAVE_TLB_INFO(assigned_slave));
 206                         u32 next_index = slave_info->head;
 207
 208                         hash_table[hash_index].tx_slave = assigned_slave;
 209                         hash_table[hash_index].next = next_index;
 210                         hash_table[hash_index].prev = TLB_NULL_INDEX;
 211
 212                         if (next_index != TLB_NULL_INDEX)
 213                                 hash_table[next_index].prev = hash_index;
 214
 215                         slave_info->head = hash_index;
 216                         slave_info->load +=
 217                                 hash_table[hash_index].load_history;
 218                 }
 219         }
 220
 221         if (assigned_slave)
 222                 hash_table[hash_index].tx_bytes += skb_len;
 223
 224         return assigned_slave;
 225 }
 226
 227 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
 228                                         u32 skb_len)
 229 {
 230         struct slave *tx_slave;
 231
 232         /* We don't need to disable softirq here, because
 233          * tlb_choose_channel() is only called by bond_alb_xmit()
 234          * which already has softirq disabled.
 235          */
 236         spin_lock(&bond->mode_lock);
 237         tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
 238         spin_unlock(&bond->mode_lock);
 239
 240         return tx_slave;
 241 }
 242
 243 /*********************** rlb specific functions ***************************/
 244
 245 /* when an ARP REPLY is received from a client update its info
 246  * in the rx_hashtbl
 247  */
 248 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 249 {
 250         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 251         struct rlb_client_info *client_info;
 252         u32 hash_index;
 253
 254         spin_lock_bh(&bond->mode_lock);
 255
 256         hash_index = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
 257         client_info = &(bond_info->rx_hashtbl[hash_index]);
 258
 259         if ((client_info->assigned) &&
 260             (client_info->ip_src == arp->ip_dst) &&
 261             (client_info->ip_dst == arp->ip_src) &&
 262             (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) {
 263                 /* update the clients MAC address */
 264                 ether_addr_copy(client_info->mac_dst, arp->mac_src);
 265                 client_info->ntt = 1;
 266                 bond_info->rx_ntt = 1;
 267         }
 268
 269         spin_unlock_bh(&bond->mode_lock);
 270 }
 271
 272 static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
 273                         struct slave *slave)
 274 {
 275         struct arp_pkt *arp, _arp;
 276
 277         if (skb->protocol != cpu_to_be16(ETH_P_ARP))
 278                 goto out;
 279
 280         arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp);
 281         if (!arp)
 282                 goto out;
 283
 284         /* We received an ARP from arp->ip_src.
 285          * We might have used this IP address previously (on the bonding host
 286          * itself or on a system that is bridged together with the bond).
 287          * However, if arp->mac_src is different than what is stored in
 288          * rx_hashtbl, some other host is now using the IP and we must prevent
 289          * sending out client updates with this IP address and the old MAC
 290          * address.
 291          * Clean up all hash table entries that have this address as ip_src but
 292          * have a different mac_src.
 293          */
 294         rlb_purge_src_ip(bond, arp);
 295
 296         if (arp->op_code == htons(ARPOP_REPLY)) {
 297                 /* update rx hash table for this ARP */
 298                 rlb_update_entry_from_arp(bond, arp);
 299                 slave_dbg(bond->dev, slave->dev, "Server received an ARP Reply from client\n");
 300         }
 301 out:
 302         return RX_HANDLER_ANOTHER;
 303 }
 304
 305 /* Caller must hold rcu_read_lock() */
 306 static struct slave *__rlb_next_rx_slave(struct bonding *bond)
 307 {
 308         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 309         struct slave *before = NULL, *rx_slave = NULL, *slave;
 310         struct list_head *iter;
 311         bool found = false;
 312
 313         bond_for_each_slave_rcu(bond, slave, iter) {
 314                 if (!bond_slave_can_tx(slave))
 315                         continue;
 316                 if (!found) {
 317                         if (!before || before->speed < slave->speed)
 318                                 before = slave;
 319                 } else {
 320                         if (!rx_slave || rx_slave->speed < slave->speed)
 321                                 rx_slave = slave;
 322                 }
 323                 if (slave == bond_info->rx_slave)
 324                         found = true;
 325         }
 326         /* we didn't find anything after the current or we have something
 327          * better before and up to the current slave
 328          */
 329         if (!rx_slave || (before && rx_slave->speed < before->speed))
 330                 rx_slave = before;
 331
 332         if (rx_slave)
 333                 bond_info->rx_slave = rx_slave;
 334
 335         return rx_slave;
 336 }
 337
 338 /* Caller must hold RTNL, rcu_read_lock is obtained only to silence checkers */
 339 static struct slave *rlb_next_rx_slave(struct bonding *bond)
 340 {
 341         struct slave *rx_slave;
 342
 343         ASSERT_RTNL();
 344
 345         rcu_read_lock();
 346         rx_slave = __rlb_next_rx_slave(bond);
 347         rcu_read_unlock();
 348
 349         return rx_slave;
 350 }
 351
 352 /* teach the switch the mac of a disabled slave
 353  * on the primary for fault tolerance
 354  *
 355  * Caller must hold RTNL
 356  */
 357 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond,
 358                                               const u8 addr[])
 359 {
 360         struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 361
 362         if (!curr_active)
 363                 return;
 364
 365         if (!bond->alb_info.primary_is_promisc) {
 366                 if (!dev_set_promiscuity(curr_active->dev, 1))
 367                         bond->alb_info.primary_is_promisc = 1;
 368                 else
 369                         bond->alb_info.primary_is_promisc = 0;
 370         }
 371
 372         bond->alb_info.rlb_promisc_timeout_counter = 0;
 373
 374         alb_send_learning_packets(curr_active, addr, true);
 375 }
 376
 377 /* slave being removed should not be active at this point
 378  *
 379  * Caller must hold rtnl.
 380  */
 381 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 382 {
 383         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 384         struct rlb_client_info *rx_hash_table;
 385         u32 index, next_index;
 386
 387         /* clear slave from rx_hashtbl */
 388         spin_lock_bh(&bond->mode_lock);
 389
 390         rx_hash_table = bond_info->rx_hashtbl;
 391         index = bond_info->rx_hashtbl_used_head;
 392         for (; index != RLB_NULL_INDEX; index = next_index) {
 393                 next_index = rx_hash_table[index].used_next;
 394                 if (rx_hash_table[index].slave == slave) {
 395                         struct slave *assigned_slave = rlb_next_rx_slave(bond);
 396
 397                         if (assigned_slave) {
 398                                 rx_hash_table[index].slave = assigned_slave;
 399                                 if (is_valid_ether_addr(rx_hash_table[index].mac_dst)) {
 400                                         bond_info->rx_hashtbl[index].ntt = 1;
 401                                         bond_info->rx_ntt = 1;
 402                                         /* A slave has been removed from the
 403                                          * table because it is either disabled
 404                                          * or being released. We must retry the
 405                                          * update to avoid clients from not
 406                                          * being updated & disconnecting when
 407                                          * there is stress
 408                                          */
 409                                         bond_info->rlb_update_retry_counter =
 410                                                 RLB_UPDATE_RETRY;
 411                                 }
 412                         } else {  /* there is no active slave */
 413                                 rx_hash_table[index].slave = NULL;
 414                         }
 415                 }
 416         }
 417
 418         spin_unlock_bh(&bond->mode_lock);
 419
 420         if (slave != rtnl_dereference(bond->curr_active_slave))
 421                 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 422 }
 423
 424 static void rlb_update_client(struct rlb_client_info *client_info)
 425 {
 426         int i;
 427
 428         if (!client_info->slave || !is_valid_ether_addr(client_info->mac_dst))
 429                 return;
 430
 431         for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 432                 struct sk_buff *skb;
 433
 434                 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 435                                  client_info->ip_dst,
 436                                  client_info->slave->dev,
 437                                  client_info->ip_src,
 438                                  client_info->mac_dst,
 439                                  client_info->slave->dev->dev_addr,
 440                                  client_info->mac_dst);
 441                 if (!skb) {
 442                         slave_err(client_info->slave->bond->dev,
 443                                   client_info->slave->dev,
 444                                   "failed to create an ARP packet\n");
 445                         continue;
 446                 }
 447
 448                 skb->dev = client_info->slave->dev;
 449
 450                 if (client_info->vlan_id) {
 451                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 452                                                client_info->vlan_id);
 453                 }
 454
 455                 arp_xmit(skb);
 456         }
 457 }
 458
 459 /* sends ARP REPLIES that update the clients that need updating */
 460 static void rlb_update_rx_clients(struct bonding *bond)
 461 {
 462         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 463         struct rlb_client_info *client_info;
 464         u32 hash_index;
 465
 466         spin_lock_bh(&bond->mode_lock);
 467
 468         hash_index = bond_info->rx_hashtbl_used_head;
 469         for (; hash_index != RLB_NULL_INDEX;
 470              hash_index = client_info->used_next) {
 471                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 472                 if (client_info->ntt) {
 473                         rlb_update_client(client_info);
 474                         if (bond_info->rlb_update_retry_counter == 0)
 475                                 client_info->ntt = 0;
 476                 }
 477         }
 478
 479         /* do not update the entries again until this counter is zero so that
 480          * not to confuse the clients.
 481          */
 482         bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 483
 484         spin_unlock_bh(&bond->mode_lock);
 485 }
 486
 487 /* The slave was assigned a new mac address - update the clients */
 488 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 489 {
 490         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 491         struct rlb_client_info *client_info;
 492         int ntt = 0;
 493         u32 hash_index;
 494
 495         spin_lock_bh(&bond->mode_lock);
 496
 497         hash_index = bond_info->rx_hashtbl_used_head;
 498         for (; hash_index != RLB_NULL_INDEX;
 499              hash_index = client_info->used_next) {
 500                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 501
 502                 if ((client_info->slave == slave) &&
 503                     is_valid_ether_addr(client_info->mac_dst)) {
 504                         client_info->ntt = 1;
 505                         ntt = 1;
 506                 }
 507         }
 508
 509         /* update the team's flag only after the whole iteration */
 510         if (ntt) {
 511                 bond_info->rx_ntt = 1;
 512                 /* fasten the change */
 513                 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 514         }
 515
 516         spin_unlock_bh(&bond->mode_lock);
 517 }
 518
 519 /* mark all clients using src_ip to be updated */
 520 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
 521 {
 522         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 523         struct rlb_client_info *client_info;
 524         u32 hash_index;
 525
 526         spin_lock(&bond->mode_lock);
 527
 528         hash_index = bond_info->rx_hashtbl_used_head;
 529         for (; hash_index != RLB_NULL_INDEX;
 530              hash_index = client_info->used_next) {
 531                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 532
 533                 if (!client_info->slave) {
 534                         netdev_err(bond->dev, "found a client with no channel in the client's hash table\n");
 535                         continue;
 536                 }
 537                 /* update all clients using this src_ip, that are not assigned
 538                  * to the team's address (curr_active_slave) and have a known
 539                  * unicast mac address.
 540                  */
 541                 if ((client_info->ip_src == src_ip) &&
 542                     !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
 543                                              bond->dev->dev_addr) &&
 544                     is_valid_ether_addr(client_info->mac_dst)) {
 545                         client_info->ntt = 1;
 546                         bond_info->rx_ntt = 1;
 547                 }
 548         }
 549
 550         spin_unlock(&bond->mode_lock);
 551 }
 552
 553 static struct slave *rlb_choose_channel(struct sk_buff *skb,
 554                                         struct bonding *bond,
 555                                         const struct arp_pkt *arp)
 556 {
 557         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 558         struct slave *assigned_slave, *curr_active_slave;
 559         struct rlb_client_info *client_info;
 560         u32 hash_index = 0;
 561
 562         spin_lock(&bond->mode_lock);
 563
 564         curr_active_slave = rcu_dereference(bond->curr_active_slave);
 565
 566         hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
 567         client_info = &(bond_info->rx_hashtbl[hash_index]);
 568
 569         if (client_info->assigned) {
 570                 if ((client_info->ip_src == arp->ip_src) &&
 571                     (client_info->ip_dst == arp->ip_dst)) {
 572                         /* the entry is already assigned to this client */
 573                         if (!is_broadcast_ether_addr(arp->mac_dst)) {
 574                                 /* update mac address from arp */
 575                                 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
 576                         }
 577                         ether_addr_copy(client_info->mac_src, arp->mac_src);
 578
 579                         assigned_slave = client_info->slave;
 580                         if (assigned_slave) {
 581                                 spin_unlock(&bond->mode_lock);
 582                                 return assigned_slave;
 583                         }
 584                 } else {
 585                         /* the entry is already assigned to some other client,
 586                          * move the old client to primary (curr_active_slave) so
 587                          * that the new client can be assigned to this entry.
 588                          */
 589                         if (curr_active_slave &&
 590                             client_info->slave != curr_active_slave) {
 591                                 client_info->slave = curr_active_slave;
 592                                 rlb_update_client(client_info);
 593                         }
 594                 }
 595         }
 596         /* assign a new slave */
 597         assigned_slave = __rlb_next_rx_slave(bond);
 598
 599         if (assigned_slave) {
 600                 if (!(client_info->assigned &&
 601                       client_info->ip_src == arp->ip_src)) {
 602                         /* ip_src is going to be updated,
 603                          * fix the src hash list
 604                          */
 605                         u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
 606                                                     sizeof(arp->ip_src));
 607                         rlb_src_unlink(bond, hash_index);
 608                         rlb_src_link(bond, hash_src, hash_index);
 609                 }
 610
 611                 client_info->ip_src = arp->ip_src;
 612                 client_info->ip_dst = arp->ip_dst;
 613                 /* arp->mac_dst is broadcast for arp requests.
 614                  * will be updated with clients actual unicast mac address
 615                  * upon receiving an arp reply.
 616                  */
 617                 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
 618                 ether_addr_copy(client_info->mac_src, arp->mac_src);
 619                 client_info->slave = assigned_slave;
 620
 621                 if (is_valid_ether_addr(client_info->mac_dst)) {
 622                         client_info->ntt = 1;
 623                         bond->alb_info.rx_ntt = 1;
 624                 } else {
 625                         client_info->ntt = 0;
 626                 }
 627
 628                 if (vlan_get_tag(skb, &client_info->vlan_id))
 629                         client_info->vlan_id = 0;
 630
 631                 if (!client_info->assigned) {
 632                         u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
 633
 634                         bond_info->rx_hashtbl_used_head = hash_index;
 635                         client_info->used_next = prev_tbl_head;
 636                         if (prev_tbl_head != RLB_NULL_INDEX) {
 637                                 bond_info->rx_hashtbl[prev_tbl_head].used_prev =
 638                                         hash_index;
 639                         }
 640                         client_info->assigned = 1;
 641                 }
 642         }
 643
 644         spin_unlock(&bond->mode_lock);
 645
 646         return assigned_slave;
 647 }
 648
 649 /* chooses (and returns) transmit channel for arp reply
 650  * does not choose channel for other arp types since they are
 651  * sent on the curr_active_slave
 652  */
 653 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 654 {
 655         struct slave *tx_slave = NULL;
 656         struct arp_pkt *arp;
 657
 658         if (!pskb_network_may_pull(skb, sizeof(*arp)))
 659                 return NULL;
 660         arp = (struct arp_pkt *)skb_network_header(skb);
 661
 662         /* Don't modify or load balance ARPs that do not originate locally
 663          * (e.g.,arrive via a bridge).
 664          */
 665         if (!bond_slave_has_mac_rx(bond, arp->mac_src))
 666                 return NULL;
 667
 668         if (arp->op_code == htons(ARPOP_REPLY)) {
 669                 /* the arp must be sent on the selected rx channel */
 670                 tx_slave = rlb_choose_channel(skb, bond, arp);
 671                 if (tx_slave)
 672                         bond_hw_addr_copy(arp->mac_src, tx_slave->dev->dev_addr,
 673                                           tx_slave->dev->addr_len);
 674                 netdev_dbg(bond->dev, "(slave %s): Server sent ARP Reply packet\n",
 675                            tx_slave ? tx_slave->dev->name : "NULL");
 676         } else if (arp->op_code == htons(ARPOP_REQUEST)) {
 677                 /* Create an entry in the rx_hashtbl for this client as a
 678                  * place holder.
 679                  * When the arp reply is received the entry will be updated
 680                  * with the correct unicast address of the client.
 681                  */
 682                 tx_slave = rlb_choose_channel(skb, bond, arp);
 683
 684                 /* The ARP reply packets must be delayed so that
 685                  * they can cancel out the influence of the ARP request.
 686                  */
 687                 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 688
 689                 /* arp requests are broadcast and are sent on the primary
 690                  * the arp request will collapse all clients on the subnet to
 691                  * the primary slave. We must register these clients to be
 692                  * updated with their assigned mac.
 693                  */
 694                 rlb_req_update_subnet_clients(bond, arp->ip_src);
 695                 netdev_dbg(bond->dev, "(slave %s): Server sent ARP Request packet\n",
 696                            tx_slave ? tx_slave->dev->name : "NULL");
 697         }
 698
 699         return tx_slave;
 700 }
 701
 702 static void rlb_rebalance(struct bonding *bond)
 703 {
 704         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 705         struct slave *assigned_slave;
 706         struct rlb_client_info *client_info;
 707         int ntt;
 708         u32 hash_index;
 709
 710         spin_lock_bh(&bond->mode_lock);
 711
 712         ntt = 0;
 713         hash_index = bond_info->rx_hashtbl_used_head;
 714         for (; hash_index != RLB_NULL_INDEX;
 715              hash_index = client_info->used_next) {
 716                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 717                 assigned_slave = __rlb_next_rx_slave(bond);
 718                 if (assigned_slave && (client_info->slave != assigned_slave)) {
 719                         client_info->slave = assigned_slave;
 720                         if (!is_zero_ether_addr(client_info->mac_dst)) {
 721                                 client_info->ntt = 1;
 722                                 ntt = 1;
 723                         }
 724                 }
 725         }
 726
 727         /* update the team's flag only after the whole iteration */
 728         if (ntt)
 729                 bond_info->rx_ntt = 1;
 730         spin_unlock_bh(&bond->mode_lock);
 731 }
 732
 733 /* Caller must hold mode_lock */
 734 static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
 735 {
 736         entry->used_next = RLB_NULL_INDEX;
 737         entry->used_prev = RLB_NULL_INDEX;
 738         entry->assigned = 0;
 739         entry->slave = NULL;
 740         entry->vlan_id = 0;
 741 }
 742 static void rlb_init_table_entry_src(struct rlb_client_info *entry)
 743 {
 744         entry->src_first = RLB_NULL_INDEX;
 745         entry->src_prev = RLB_NULL_INDEX;
 746         entry->src_next = RLB_NULL_INDEX;
 747 }
 748
 749 static void rlb_init_table_entry(struct rlb_client_info *entry)
 750 {
 751         memset(entry, 0, sizeof(struct rlb_client_info));
 752         rlb_init_table_entry_dst(entry);
 753         rlb_init_table_entry_src(entry);
 754 }
 755
 756 static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
 757 {
 758         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 759         u32 next_index = bond_info->rx_hashtbl[index].used_next;
 760         u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
 761
 762         if (index == bond_info->rx_hashtbl_used_head)
 763                 bond_info->rx_hashtbl_used_head = next_index;
 764         if (prev_index != RLB_NULL_INDEX)
 765                 bond_info->rx_hashtbl[prev_index].used_next = next_index;
 766         if (next_index != RLB_NULL_INDEX)
 767                 bond_info->rx_hashtbl[next_index].used_prev = prev_index;
 768 }
 769
 770 /* unlink a rlb hash table entry from the src list */
 771 static void rlb_src_unlink(struct bonding *bond, u32 index)
 772 {
 773         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 774         u32 next_index = bond_info->rx_hashtbl[index].src_next;
 775         u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
 776
 777         bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
 778         bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
 779
 780         if (next_index != RLB_NULL_INDEX)
 781                 bond_info->rx_hashtbl[next_index].src_prev = prev_index;
 782
 783         if (prev_index == RLB_NULL_INDEX)
 784                 return;
 785
 786         /* is prev_index pointing to the head of this list? */
 787         if (bond_info->rx_hashtbl[prev_index].src_first == index)
 788                 bond_info->rx_hashtbl[prev_index].src_first = next_index;
 789         else
 790                 bond_info->rx_hashtbl[prev_index].src_next = next_index;
 791
 792 }
 793
 794 static void rlb_delete_table_entry(struct bonding *bond, u32 index)
 795 {
 796         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 797         struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 798
 799         rlb_delete_table_entry_dst(bond, index);
 800         rlb_init_table_entry_dst(entry);
 801
 802         rlb_src_unlink(bond, index);
 803 }
 804
 805 /* add the rx_hashtbl[ip_dst_hash] entry to the list
 806  * of entries with identical ip_src_hash
 807  */
 808 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
 809 {
 810         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 811         u32 next;
 812
 813         bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
 814         next = bond_info->rx_hashtbl[ip_src_hash].src_first;
 815         bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
 816         if (next != RLB_NULL_INDEX)
 817                 bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
 818         bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
 819 }
 820
 821 /* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does
 822  * not match arp->mac_src
 823  */
 824 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
 825 {
 826         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 827         u32 ip_src_hash = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
 828         u32 index;
 829
 830         spin_lock_bh(&bond->mode_lock);
 831
 832         index = bond_info->rx_hashtbl[ip_src_hash].src_first;
 833         while (index != RLB_NULL_INDEX) {
 834                 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 835                 u32 next_index = entry->src_next;
 836
 837                 if (entry->ip_src == arp->ip_src &&
 838                     !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
 839                         rlb_delete_table_entry(bond, index);
 840                 index = next_index;
 841         }
 842         spin_unlock_bh(&bond->mode_lock);
 843 }
 844
 845 static int rlb_initialize(struct bonding *bond)
 846 {
 847         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 848         struct rlb_client_info  *new_hashtbl;
 849         int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 850         int i;
 851
 852         new_hashtbl = kmalloc(size, GFP_KERNEL);
 853         if (!new_hashtbl)
 854                 return -1;
 855
 856         spin_lock_bh(&bond->mode_lock);
 857
 858         bond_info->rx_hashtbl = new_hashtbl;
 859
 860         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 861
 862         for (i = 0; i < RLB_HASH_TABLE_SIZE; i++)
 863                 rlb_init_table_entry(bond_info->rx_hashtbl + i);
 864
 865         spin_unlock_bh(&bond->mode_lock);
 866
 867         /* register to receive ARPs */
 868         bond->recv_probe = rlb_arp_recv;
 869
 870         return 0;
 871 }
 872
 873 static void rlb_deinitialize(struct bonding *bond)
 874 {
 875         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 876
 877         spin_lock_bh(&bond->mode_lock);
 878
 879         kfree(bond_info->rx_hashtbl);
 880         bond_info->rx_hashtbl = NULL;
 881         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 882
 883         spin_unlock_bh(&bond->mode_lock);
 884 }
 885
 886 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 887 {
 888         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 889         u32 curr_index;
 890
 891         spin_lock_bh(&bond->mode_lock);
 892
 893         curr_index = bond_info->rx_hashtbl_used_head;
 894         while (curr_index != RLB_NULL_INDEX) {
 895                 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 896                 u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
 897
 898                 if (curr->vlan_id == vlan_id)
 899                         rlb_delete_table_entry(bond, curr_index);
 900
 901                 curr_index = next_index;
 902         }
 903
 904         spin_unlock_bh(&bond->mode_lock);
 905 }
 906
 907 /*********************** tlb/rlb shared functions *********************/
 908
 909 static void alb_send_lp_vid(struct slave *slave, const u8 mac_addr[],
 910                             __be16 vlan_proto, u16 vid)
 911 {
 912         struct learning_pkt pkt;
 913         struct sk_buff *skb;
 914         int size = sizeof(struct learning_pkt);
 915
 916         memset(&pkt, 0, size);
 917         ether_addr_copy(pkt.mac_dst, mac_addr);
 918         ether_addr_copy(pkt.mac_src, mac_addr);
 919         pkt.type = cpu_to_be16(ETH_P_LOOPBACK);
 920
 921         skb = dev_alloc_skb(size);
 922         if (!skb)
 923                 return;
 924
 925         skb_put_data(skb, &pkt, size);
 926
 927         skb_reset_mac_header(skb);
 928         skb->network_header = skb->mac_header + ETH_HLEN;
 929         skb->protocol = pkt.type;
 930         skb->priority = TC_PRIO_CONTROL;
 931         skb->dev = slave->dev;
 932
 933         slave_dbg(slave->bond->dev, slave->dev,
 934                   "Send learning packet: mac %pM vlan %d\n", mac_addr, vid);
 935
 936         if (vid)
 937                 __vlan_hwaccel_put_tag(skb, vlan_proto, vid);
 938
 939         dev_queue_xmit(skb);
 940 }
 941
 942 struct alb_walk_data {
 943         struct bonding *bond;
 944         struct slave *slave;
 945         const u8 *mac_addr;
 946         bool strict_match;
 947 };
 948
 949 static int alb_upper_dev_walk(struct net_device *upper,
 950                               struct netdev_nested_priv *priv)
 951 {
 952         struct alb_walk_data *data = (struct alb_walk_data *)priv->data;
 953         bool strict_match = data->strict_match;
 954         const u8 *mac_addr = data->mac_addr;
 955         struct bonding *bond = data->bond;
 956         struct slave *slave = data->slave;
 957         struct bond_vlan_tag *tags;
 958
 959         if (is_vlan_dev(upper) &&
 960             bond->dev->lower_level == upper->lower_level - 1) {
 961                 if (upper->addr_assign_type == NET_ADDR_STOLEN) {
 962                         alb_send_lp_vid(slave, mac_addr,
 963                                         vlan_dev_vlan_proto(upper),
 964                                         vlan_dev_vlan_id(upper));
 965                 } else {
 966                         alb_send_lp_vid(slave, upper->dev_addr,
 967                                         vlan_dev_vlan_proto(upper),
 968                                         vlan_dev_vlan_id(upper));
 969                 }
 970         }
 971
 972         /* If this is a macvlan device, then only send updates
 973          * when strict_match is turned off.
 974          */
 975         if (netif_is_macvlan(upper) && !strict_match) {
 976                 tags = bond_verify_device_path(bond->dev, upper, 0);
 977                 if (IS_ERR_OR_NULL(tags))
 978                         BUG();
 979                 alb_send_lp_vid(slave, upper->dev_addr,
 980                                 tags[0].vlan_proto, tags[0].vlan_id);
 981                 kfree(tags);
 982         }
 983
 984         return 0;
 985 }
 986
 987 static void alb_send_learning_packets(struct slave *slave, const u8 mac_addr[],
 988                                       bool strict_match)
 989 {
 990         struct bonding *bond = bond_get_bond_by_slave(slave);
 991         struct netdev_nested_priv priv;
 992         struct alb_walk_data data = {
 993                 .strict_match = strict_match,
 994                 .mac_addr = mac_addr,
 995                 .slave = slave,
 996                 .bond = bond,
 997         };
 998
 999         priv.data = (void *)&data;
1000         /* send untagged */
1001         alb_send_lp_vid(slave, mac_addr, 0, 0);
1002
1003         /* loop through all devices and see if we need to send a packet
1004          * for that device.
1005          */
1006         rcu_read_lock();
1007         netdev_walk_all_upper_dev_rcu(bond->dev, alb_upper_dev_walk, &priv);
1008         rcu_read_unlock();
1009 }
1010
1011 static int alb_set_slave_mac_addr(struct slave *slave, const u8 addr[],
1012                                   unsigned int len)
1013 {
1014         struct net_device *dev = slave->dev;
1015         struct sockaddr_storage ss;
1016
1017         if (BOND_MODE(slave->bond) == BOND_MODE_TLB) {
1018                 __dev_addr_set(dev, addr, len);
1019                 return 0;
1020         }
1021
1022         /* for rlb each slave must have a unique hw mac addresses so that
1023          * each slave will receive packets destined to a different mac
1024          */
1025         memcpy(ss.__data, addr, len);
1026         ss.ss_family = dev->type;
1027         if (dev_set_mac_address(dev, (struct sockaddr *)&ss, NULL)) {
1028                 slave_err(slave->bond->dev, dev, "dev_set_mac_address on slave failed! ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n");
1029                 return -EOPNOTSUPP;
1030         }
1031         return 0;
1032 }
1033
1034 /* Swap MAC addresses between two slaves.
1035  *
1036  * Called with RTNL held, and no other locks.
1037  */
1038 static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1039 {
1040         u8 tmp_mac_addr[MAX_ADDR_LEN];
1041
1042         bond_hw_addr_copy(tmp_mac_addr, slave1->dev->dev_addr,
1043                           slave1->dev->addr_len);
1044         alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr,
1045                                slave2->dev->addr_len);
1046         alb_set_slave_mac_addr(slave2, tmp_mac_addr,
1047                                slave1->dev->addr_len);
1048
1049 }
1050
1051 /* Send learning packets after MAC address swap.
1052  *
1053  * Called with RTNL and no other locks
1054  */
1055 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1056                                 struct slave *slave2)
1057 {
1058         int slaves_state_differ = (bond_slave_can_tx(slave1) != bond_slave_can_tx(slave2));
1059         struct slave *disabled_slave = NULL;
1060
1061         ASSERT_RTNL();
1062
1063         /* fasten the change in the switch */
1064         if (bond_slave_can_tx(slave1)) {
1065                 alb_send_learning_packets(slave1, slave1->dev->dev_addr, false);
1066                 if (bond->alb_info.rlb_enabled) {
1067                         /* inform the clients that the mac address
1068                          * has changed
1069                          */
1070                         rlb_req_update_slave_clients(bond, slave1);
1071                 }
1072         } else {
1073                 disabled_slave = slave1;
1074         }
1075
1076         if (bond_slave_can_tx(slave2)) {
1077                 alb_send_learning_packets(slave2, slave2->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, slave2);
1083                 }
1084         } else {
1085                 disabled_slave = slave2;
1086         }
1087
1088         if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1089                 /* A disabled slave was assigned an active mac addr */
1090                 rlb_teach_disabled_mac_on_primary(bond,
1091                                                   disabled_slave->dev->dev_addr);
1092         }
1093 }
1094
1095 /**
1096  * alb_change_hw_addr_on_detach
1097  * @bond: bonding we're working on
1098  * @slave: the slave that was just detached
1099  *
1100  * We assume that @slave was already detached from the slave list.
1101  *
1102  * If @slave's permanent hw address is different both from its current
1103  * address and from @bond's address, then somewhere in the bond there's
1104  * a slave that has @slave's permanet address as its current address.
1105  * We'll make sure that slave no longer uses @slave's permanent address.
1106  *
1107  * Caller must hold RTNL and no other locks
1108  */
1109 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1110 {
1111         int perm_curr_diff;
1112         int perm_bond_diff;
1113         struct slave *found_slave;
1114
1115         perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1116                                                   slave->dev->dev_addr);
1117         perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1118                                                   bond->dev->dev_addr);
1119
1120         if (perm_curr_diff && perm_bond_diff) {
1121                 found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1122
1123                 if (found_slave) {
1124                         alb_swap_mac_addr(slave, found_slave);
1125                         alb_fasten_mac_swap(bond, slave, found_slave);
1126                 }
1127         }
1128 }
1129
1130 /**
1131  * alb_handle_addr_collision_on_attach
1132  * @bond: bonding we're working on
1133  * @slave: the slave that was just attached
1134  *
1135  * checks uniqueness of slave's mac address and handles the case the
1136  * new slave uses the bonds mac address.
1137  *
1138  * If the permanent hw address of @slave is @bond's hw address, we need to
1139  * find a different hw address to give @slave, that isn't in use by any other
1140  * slave in the bond. This address must be, of course, one of the permanent
1141  * addresses of the other slaves.
1142  *
1143  * We go over the slave list, and for each slave there we compare its
1144  * permanent hw address with the current address of all the other slaves.
1145  * If no match was found, then we've found a slave with a permanent address
1146  * that isn't used by any other slave in the bond, so we can assign it to
1147  * @slave.
1148  *
1149  * assumption: this function is called before @slave is attached to the
1150  *             bond slave list.
1151  */
1152 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1153 {
1154         struct slave *has_bond_addr = rcu_access_pointer(bond->curr_active_slave);
1155         struct slave *tmp_slave1, *free_mac_slave = NULL;
1156         struct list_head *iter;
1157
1158         if (!bond_has_slaves(bond)) {
1159                 /* this is the first slave */
1160                 return 0;
1161         }
1162
1163         /* if slave's mac address differs from bond's mac address
1164          * check uniqueness of slave's mac address against the other
1165          * slaves in the bond.
1166          */
1167         if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1168                 if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1169                         return 0;
1170
1171                 /* Try setting slave mac to bond address and fall-through
1172                  * to code handling that situation below...
1173                  */
1174                 alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1175                                        bond->dev->addr_len);
1176         }
1177
1178         /* The slave's address is equal to the address of the bond.
1179          * Search for a spare address in the bond for this slave.
1180          */
1181         bond_for_each_slave(bond, tmp_slave1, iter) {
1182                 if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1183                         /* no slave has tmp_slave1's perm addr
1184                          * as its curr addr
1185                          */
1186                         free_mac_slave = tmp_slave1;
1187                         break;
1188                 }
1189
1190                 if (!has_bond_addr) {
1191                         if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1192                                                     bond->dev->dev_addr)) {
1193
1194                                 has_bond_addr = tmp_slave1;
1195                         }
1196                 }
1197         }
1198
1199         if (free_mac_slave) {
1200                 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1201                                        free_mac_slave->dev->addr_len);
1202
1203                 slave_warn(bond->dev, slave->dev, "the slave hw address is in use by the bond; giving it the hw address of %s\n",
1204                            free_mac_slave->dev->name);
1205
1206         } else if (has_bond_addr) {
1207                 slave_err(bond->dev, slave->dev, "the slave hw address 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");
1208                 return -EFAULT;
1209         }
1210
1211         return 0;
1212 }
1213
1214 /**
1215  * alb_set_mac_address
1216  * @bond: bonding we're working on
1217  * @addr: MAC address to set
1218  *
1219  * In TLB mode all slaves are configured to the bond's hw address, but set
1220  * their dev_addr field to different addresses (based on their permanent hw
1221  * addresses).
1222  *
1223  * For each slave, this function sets the interface to the new address and then
1224  * changes its dev_addr field to its previous value.
1225  *
1226  * Unwinding assumes bond's mac address has not yet changed.
1227  */
1228 static int alb_set_mac_address(struct bonding *bond, void *addr)
1229 {
1230         struct slave *slave, *rollback_slave;
1231         struct list_head *iter;
1232         struct sockaddr_storage ss;
1233         char tmp_addr[MAX_ADDR_LEN];
1234         int res;
1235
1236         if (bond->alb_info.rlb_enabled)
1237                 return 0;
1238
1239         bond_for_each_slave(bond, slave, iter) {
1240                 /* save net_device's current hw address */
1241                 bond_hw_addr_copy(tmp_addr, slave->dev->dev_addr,
1242                                   slave->dev->addr_len);
1243
1244                 res = dev_set_mac_address(slave->dev, addr, NULL);
1245
1246                 /* restore net_device's hw address */
1247                 dev_addr_set(slave->dev, tmp_addr);
1248
1249                 if (res)
1250                         goto unwind;
1251         }
1252
1253         return 0;
1254
1255 unwind:
1256         memcpy(ss.__data, bond->dev->dev_addr, bond->dev->addr_len);
1257         ss.ss_family = bond->dev->type;
1258
1259         /* unwind from head to the slave that failed */
1260         bond_for_each_slave(bond, rollback_slave, iter) {
1261                 if (rollback_slave == slave)
1262                         break;
1263                 bond_hw_addr_copy(tmp_addr, rollback_slave->dev->dev_addr,
1264                                   rollback_slave->dev->addr_len);
1265                 dev_set_mac_address(rollback_slave->dev,
1266                                     (struct sockaddr *)&ss, NULL);
1267                 dev_addr_set(rollback_slave->dev, tmp_addr);
1268         }
1269
1270         return res;
1271 }
1272
1273 /* determine if the packet is NA or NS */
1274 static bool alb_determine_nd(struct sk_buff *skb, struct bonding *bond)
1275 {
1276         struct ipv6hdr *ip6hdr;
1277         struct icmp6hdr *hdr;
1278
1279         if (!pskb_network_may_pull(skb, sizeof(*ip6hdr)))
1280                 return true;
1281
1282         ip6hdr = ipv6_hdr(skb);
1283         if (ip6hdr->nexthdr != IPPROTO_ICMPV6)
1284                 return false;
1285
1286         if (!pskb_network_may_pull(skb, sizeof(*ip6hdr) + sizeof(*hdr)))
1287                 return true;
1288
1289         hdr = icmp6_hdr(skb);
1290         return hdr->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT ||
1291                 hdr->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION;
1292 }
1293
1294 /************************ exported alb functions ************************/
1295
1296 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1297 {
1298         int res;
1299
1300         res = tlb_initialize(bond);
1301         if (res)
1302                 return res;
1303
1304         if (rlb_enabled) {
1305                 res = rlb_initialize(bond);
1306                 if (res) {
1307                         tlb_deinitialize(bond);
1308                         return res;
1309                 }
1310                 bond->alb_info.rlb_enabled = 1;
1311         } else {
1312                 bond->alb_info.rlb_enabled = 0;
1313         }
1314
1315         return 0;
1316 }
1317
1318 void bond_alb_deinitialize(struct bonding *bond)
1319 {
1320         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1321
1322         tlb_deinitialize(bond);
1323
1324         if (bond_info->rlb_enabled)
1325                 rlb_deinitialize(bond);
1326 }
1327
1328 static netdev_tx_t bond_do_alb_xmit(struct sk_buff *skb, struct bonding *bond,
1329                                     struct slave *tx_slave)
1330 {
1331         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1332         struct ethhdr *eth_data = eth_hdr(skb);
1333
1334         if (!tx_slave) {
1335                 /* unbalanced or unassigned, send through primary */
1336                 tx_slave = rcu_dereference(bond->curr_active_slave);
1337                 if (bond->params.tlb_dynamic_lb)
1338                         bond_info->unbalanced_load += skb->len;
1339         }
1340
1341         if (tx_slave && bond_slave_can_tx(tx_slave)) {
1342                 if (tx_slave != rcu_access_pointer(bond->curr_active_slave)) {
1343                         ether_addr_copy(eth_data->h_source,
1344                                         tx_slave->dev->dev_addr);
1345                 }
1346
1347                 return bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1348         }
1349
1350         if (tx_slave && bond->params.tlb_dynamic_lb) {
1351                 spin_lock(&bond->mode_lock);
1352                 __tlb_clear_slave(bond, tx_slave, 0);
1353                 spin_unlock(&bond->mode_lock);
1354         }
1355
1356         /* no suitable interface, frame not sent */
1357         return bond_tx_drop(bond->dev, skb);
1358 }
1359
1360 struct slave *bond_xmit_tlb_slave_get(struct bonding *bond,
1361                                       struct sk_buff *skb)
1362 {
1363         struct slave *tx_slave = NULL;
1364         struct ethhdr *eth_data;
1365         u32 hash_index;
1366
1367         skb_reset_mac_header(skb);
1368         eth_data = eth_hdr(skb);
1369
1370         /* Do not TX balance any multicast or broadcast */
1371         if (!is_multicast_ether_addr(eth_data->h_dest)) {
1372                 switch (skb->protocol) {
1373                 case htons(ETH_P_IPV6):
1374                         if (alb_determine_nd(skb, bond))
1375                                 break;
1376                         fallthrough;
1377                 case htons(ETH_P_IP):
1378                         hash_index = bond_xmit_hash(bond, skb);
1379                         if (bond->params.tlb_dynamic_lb) {
1380                                 tx_slave = tlb_choose_channel(bond,
1381                                                               hash_index & 0xFF,
1382                                                               skb->len);
1383                         } else {
1384                                 struct bond_up_slave *slaves;
1385                                 unsigned int count;
1386
1387                                 slaves = rcu_dereference(bond->usable_slaves);
1388                                 count = slaves ? READ_ONCE(slaves->count) : 0;
1389                                 if (likely(count))
1390                                         tx_slave = slaves->arr[hash_index %
1391                                                                count];
1392                         }
1393                         break;
1394                 }
1395         }
1396         return tx_slave;
1397 }
1398
1399 netdev_tx_t bond_tlb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1400 {
1401         struct bonding *bond = netdev_priv(bond_dev);
1402         struct slave *tx_slave;
1403
1404         tx_slave = bond_xmit_tlb_slave_get(bond, skb);
1405         return bond_do_alb_xmit(skb, bond, tx_slave);
1406 }
1407
1408 struct slave *bond_xmit_alb_slave_get(struct bonding *bond,
1409                                       struct sk_buff *skb)
1410 {
1411         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1412         static const __be32 ip_bcast = htonl(0xffffffff);
1413         struct slave *tx_slave = NULL;
1414         const u8 *hash_start = NULL;
1415         bool do_tx_balance = true;
1416         struct ethhdr *eth_data;
1417         u32 hash_index = 0;
1418         int hash_size = 0;
1419
1420         skb_reset_mac_header(skb);
1421         eth_data = eth_hdr(skb);
1422
1423         switch (ntohs(skb->protocol)) {
1424         case ETH_P_IP: {
1425                 const struct iphdr *iph;
1426
1427                 if (is_broadcast_ether_addr(eth_data->h_dest) ||
1428                     !pskb_network_may_pull(skb, sizeof(*iph))) {
1429                         do_tx_balance = false;
1430                         break;
1431                 }
1432                 iph = ip_hdr(skb);
1433                 if (iph->daddr == ip_bcast || iph->protocol == IPPROTO_IGMP) {
1434                         do_tx_balance = false;
1435                         break;
1436                 }
1437                 hash_start = (char *)&(iph->daddr);
1438                 hash_size = sizeof(iph->daddr);
1439                 break;
1440         }
1441         case ETH_P_IPV6: {
1442                 const struct ipv6hdr *ip6hdr;
1443
1444                 /* IPv6 doesn't really use broadcast mac address, but leave
1445                  * that here just in case.
1446                  */
1447                 if (is_broadcast_ether_addr(eth_data->h_dest)) {
1448                         do_tx_balance = false;
1449                         break;
1450                 }
1451
1452                 /* IPv6 uses all-nodes multicast as an equivalent to
1453                  * broadcasts in IPv4.
1454                  */
1455                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1456                         do_tx_balance = false;
1457                         break;
1458                 }
1459
1460                 if (alb_determine_nd(skb, bond)) {
1461                         do_tx_balance = false;
1462                         break;
1463                 }
1464
1465                 /* The IPv6 header is pulled by alb_determine_nd */
1466                 /* Additionally, DAD probes should not be tx-balanced as that
1467                  * will lead to false positives for duplicate addresses and
1468                  * prevent address configuration from working.
1469                  */
1470                 ip6hdr = ipv6_hdr(skb);
1471                 if (ipv6_addr_any(&ip6hdr->saddr)) {
1472                         do_tx_balance = false;
1473                         break;
1474                 }
1475
1476                 hash_start = (char *)&ip6hdr->daddr;
1477                 hash_size = sizeof(ip6hdr->daddr);
1478                 break;
1479         }
1480         case ETH_P_ARP:
1481                 do_tx_balance = false;
1482                 if (bond_info->rlb_enabled)
1483                         tx_slave = rlb_arp_xmit(skb, bond);
1484                 break;
1485         default:
1486                 do_tx_balance = false;
1487                 break;
1488         }
1489
1490         if (do_tx_balance) {
1491                 if (bond->params.tlb_dynamic_lb) {
1492                         hash_index = _simple_hash(hash_start, hash_size);
1493                         tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1494                 } else {
1495                         /*
1496                          * do_tx_balance means we are free to select the tx_slave
1497                          * So we do exactly what tlb would do for hash selection
1498                          */
1499
1500                         struct bond_up_slave *slaves;
1501                         unsigned int count;
1502
1503                         slaves = rcu_dereference(bond->usable_slaves);
1504                         count = slaves ? READ_ONCE(slaves->count) : 0;
1505                         if (likely(count))
1506                                 tx_slave = slaves->arr[bond_xmit_hash(bond, skb) %
1507                                                        count];
1508                 }
1509         }
1510         return tx_slave;
1511 }
1512
1513 netdev_tx_t bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1514 {
1515         struct bonding *bond = netdev_priv(bond_dev);
1516         struct slave *tx_slave = NULL;
1517
1518         tx_slave = bond_xmit_alb_slave_get(bond, skb);
1519         return bond_do_alb_xmit(skb, bond, tx_slave);
1520 }
1521
1522 void bond_alb_monitor(struct work_struct *work)
1523 {
1524         struct bonding *bond = container_of(work, struct bonding,
1525                                             alb_work.work);
1526         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1527         struct list_head *iter;
1528         struct slave *slave;
1529
1530         if (!bond_has_slaves(bond)) {
1531                 atomic_set(&bond_info->tx_rebalance_counter, 0);
1532                 bond_info->lp_counter = 0;
1533                 goto re_arm;
1534         }
1535
1536         rcu_read_lock();
1537
1538         atomic_inc(&bond_info->tx_rebalance_counter);
1539         bond_info->lp_counter++;
1540
1541         /* send learning packets */
1542         if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1543                 bool strict_match;
1544
1545                 bond_for_each_slave_rcu(bond, slave, iter) {
1546                         /* If updating current_active, use all currently
1547                          * user mac addresses (!strict_match).  Otherwise, only
1548                          * use mac of the slave device.
1549                          * In RLB mode, we always use strict matches.
1550                          */
1551                         strict_match = (slave != rcu_access_pointer(bond->curr_active_slave) ||
1552                                         bond_info->rlb_enabled);
1553                         alb_send_learning_packets(slave, slave->dev->dev_addr,
1554                                                   strict_match);
1555                 }
1556                 bond_info->lp_counter = 0;
1557         }
1558
1559         /* rebalance tx traffic */
1560         if (atomic_read(&bond_info->tx_rebalance_counter) >= BOND_TLB_REBALANCE_TICKS) {
1561                 bond_for_each_slave_rcu(bond, slave, iter) {
1562                         tlb_clear_slave(bond, slave, 1);
1563                         if (slave == rcu_access_pointer(bond->curr_active_slave)) {
1564                                 SLAVE_TLB_INFO(slave).load =
1565                                         bond_info->unbalanced_load /
1566                                                 BOND_TLB_REBALANCE_INTERVAL;
1567                                 bond_info->unbalanced_load = 0;
1568                         }
1569                 }
1570                 atomic_set(&bond_info->tx_rebalance_counter, 0);
1571         }
1572
1573         if (bond_info->rlb_enabled) {
1574                 if (bond_info->primary_is_promisc &&
1575                     (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1576
1577                         /* dev_set_promiscuity requires rtnl and
1578                          * nothing else.  Avoid race with bond_close.
1579                          */
1580                         rcu_read_unlock();
1581                         if (!rtnl_trylock())
1582                                 goto re_arm;
1583
1584                         bond_info->rlb_promisc_timeout_counter = 0;
1585
1586                         /* If the primary was set to promiscuous mode
1587                          * because a slave was disabled then
1588                          * it can now leave promiscuous mode.
1589                          */
1590                         dev_set_promiscuity(rtnl_dereference(bond->curr_active_slave)->dev,
1591                                             -1);
1592                         bond_info->primary_is_promisc = 0;
1593
1594                         rtnl_unlock();
1595                         rcu_read_lock();
1596                 }
1597
1598                 if (bond_info->rlb_rebalance) {
1599                         bond_info->rlb_rebalance = 0;
1600                         rlb_rebalance(bond);
1601                 }
1602
1603                 /* check if clients need updating */
1604                 if (bond_info->rx_ntt) {
1605                         if (bond_info->rlb_update_delay_counter) {
1606                                 --bond_info->rlb_update_delay_counter;
1607                         } else {
1608                                 rlb_update_rx_clients(bond);
1609                                 if (bond_info->rlb_update_retry_counter)
1610                                         --bond_info->rlb_update_retry_counter;
1611                                 else
1612                                         bond_info->rx_ntt = 0;
1613                         }
1614                 }
1615         }
1616         rcu_read_unlock();
1617 re_arm:
1618         queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1619 }
1620
1621 /* assumption: called before the slave is attached to the bond
1622  * and not locked by the bond lock
1623  */
1624 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1625 {
1626         int res;
1627
1628         res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1629                                      slave->dev->addr_len);
1630         if (res)
1631                 return res;
1632
1633         res = alb_handle_addr_collision_on_attach(bond, slave);
1634         if (res)
1635                 return res;
1636
1637         tlb_init_slave(slave);
1638
1639         /* order a rebalance ASAP */
1640         atomic_set(&bond->alb_info.tx_rebalance_counter,
1641                    BOND_TLB_REBALANCE_TICKS);
1642
1643         if (bond->alb_info.rlb_enabled)
1644                 bond->alb_info.rlb_rebalance = 1;
1645
1646         return 0;
1647 }
1648
1649 /* Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1650  * if necessary.
1651  *
1652  * Caller must hold RTNL and no other locks
1653  */
1654 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1655 {
1656         if (bond_has_slaves(bond))
1657                 alb_change_hw_addr_on_detach(bond, slave);
1658
1659         tlb_clear_slave(bond, slave, 0);
1660
1661         if (bond->alb_info.rlb_enabled) {
1662                 bond->alb_info.rx_slave = NULL;
1663                 rlb_clear_slave(bond, slave);
1664         }
1665
1666 }
1667
1668 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1669 {
1670         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1671
1672         if (link == BOND_LINK_DOWN) {
1673                 tlb_clear_slave(bond, slave, 0);
1674                 if (bond->alb_info.rlb_enabled)
1675                         rlb_clear_slave(bond, slave);
1676         } else if (link == BOND_LINK_UP) {
1677                 /* order a rebalance ASAP */
1678                 atomic_set(&bond_info->tx_rebalance_counter,
1679                            BOND_TLB_REBALANCE_TICKS);
1680                 if (bond->alb_info.rlb_enabled) {
1681                         bond->alb_info.rlb_rebalance = 1;
1682                         /* If the updelay module parameter is smaller than the
1683                          * forwarding delay of the switch the rebalance will
1684                          * not work because the rebalance arp replies will
1685                          * not be forwarded to the clients..
1686                          */
1687                 }
1688         }
1689
1690         if (bond_is_nondyn_tlb(bond)) {
1691                 if (bond_update_slave_arr(bond, NULL))
1692                         pr_err("Failed to build slave-array for TLB mode.\n");
1693         }
1694 }
1695
1696 /**
1697  * bond_alb_handle_active_change - assign new curr_active_slave
1698  * @bond: our bonding struct
1699  * @new_slave: new slave to assign
1700  *
1701  * Set the bond->curr_active_slave to @new_slave and handle
1702  * mac address swapping and promiscuity changes as needed.
1703  *
1704  * Caller must hold RTNL
1705  */
1706 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1707 {
1708         struct slave *swap_slave;
1709         struct slave *curr_active;
1710
1711         curr_active = rtnl_dereference(bond->curr_active_slave);
1712         if (curr_active == new_slave)
1713                 return;
1714
1715         if (curr_active && bond->alb_info.primary_is_promisc) {
1716                 dev_set_promiscuity(curr_active->dev, -1);
1717                 bond->alb_info.primary_is_promisc = 0;
1718                 bond->alb_info.rlb_promisc_timeout_counter = 0;
1719         }
1720
1721         swap_slave = curr_active;
1722         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1723
1724         if (!new_slave || !bond_has_slaves(bond))
1725                 return;
1726
1727         /* set the new curr_active_slave to the bonds mac address
1728          * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1729          */
1730         if (!swap_slave)
1731                 swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1732
1733         /* Arrange for swap_slave and new_slave to temporarily be
1734          * ignored so we can mess with their MAC addresses without
1735          * fear of interference from transmit activity.
1736          */
1737         if (swap_slave)
1738                 tlb_clear_slave(bond, swap_slave, 1);
1739         tlb_clear_slave(bond, new_slave, 1);
1740
1741         /* in TLB mode, the slave might flip down/up with the old dev_addr,
1742          * and thus filter bond->dev_addr's packets, so force bond's mac
1743          */
1744         if (BOND_MODE(bond) == BOND_MODE_TLB) {
1745                 struct sockaddr_storage ss;
1746                 u8 tmp_addr[MAX_ADDR_LEN];
1747
1748                 bond_hw_addr_copy(tmp_addr, new_slave->dev->dev_addr,
1749                                   new_slave->dev->addr_len);
1750
1751                 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
1752                                   bond->dev->addr_len);
1753                 ss.ss_family = bond->dev->type;
1754                 /* we don't care if it can't change its mac, best effort */
1755                 dev_set_mac_address(new_slave->dev, (struct sockaddr *)&ss,
1756                                     NULL);
1757
1758                 dev_addr_set(new_slave->dev, tmp_addr);
1759         }
1760
1761         /* curr_active_slave must be set before calling alb_swap_mac_addr */
1762         if (swap_slave) {
1763                 /* swap mac address */
1764                 alb_swap_mac_addr(swap_slave, new_slave);
1765                 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1766         } else {
1767                 /* set the new_slave to the bond mac address */
1768                 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1769                                        bond->dev->addr_len);
1770                 alb_send_learning_packets(new_slave, bond->dev->dev_addr,
1771                                           false);
1772         }
1773 }
1774
1775 /* Called with RTNL */
1776 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1777 {
1778         struct bonding *bond = netdev_priv(bond_dev);
1779         struct sockaddr_storage *ss = addr;
1780         struct slave *curr_active;
1781         struct slave *swap_slave;
1782         int res;
1783
1784         if (!is_valid_ether_addr(ss->__data))
1785                 return -EADDRNOTAVAIL;
1786
1787         res = alb_set_mac_address(bond, addr);
1788         if (res)
1789                 return res;
1790
1791         dev_addr_set(bond_dev, ss->__data);
1792
1793         /* If there is no curr_active_slave there is nothing else to do.
1794          * Otherwise we'll need to pass the new address to it and handle
1795          * duplications.
1796          */
1797         curr_active = rtnl_dereference(bond->curr_active_slave);
1798         if (!curr_active)
1799                 return 0;
1800
1801         swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1802
1803         if (swap_slave) {
1804                 alb_swap_mac_addr(swap_slave, curr_active);
1805                 alb_fasten_mac_swap(bond, swap_slave, curr_active);
1806         } else {
1807                 alb_set_slave_mac_addr(curr_active, bond_dev->dev_addr,
1808                                        bond_dev->addr_len);
1809
1810                 alb_send_learning_packets(curr_active,
1811                                           bond_dev->dev_addr, false);
1812                 if (bond->alb_info.rlb_enabled) {
1813                         /* inform clients mac address has changed */
1814                         rlb_req_update_slave_clients(bond, curr_active);
1815                 }
1816         }
1817
1818         return 0;
1819 }
1820
1821 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1822 {
1823         if (bond->alb_info.rlb_enabled)
1824                 rlb_clear_vlan(bond, vlan_id);
1825 }
1826