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 <asm/byteorder.h>
  23 #include <net/bonding.h>
  24 #include <net/bond_alb.h>
  25
  26 static const u8 mac_v6_allmcast[ETH_ALEN + 2] __long_aligned = {
  27         0x33, 0x33, 0x00, 0x00, 0x00, 0x01
  28 };
  29 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
  30
  31 #pragma pack(1)
  32 struct learning_pkt {
  33         u8 mac_dst[ETH_ALEN];
  34         u8 mac_src[ETH_ALEN];
  35         __be16 type;
  36         u8 padding[ETH_ZLEN - ETH_HLEN];
  37 };
  38
  39 struct arp_pkt {
  40         __be16  hw_addr_space;
  41         __be16  prot_addr_space;
  42         u8      hw_addr_len;
  43         u8      prot_addr_len;
  44         __be16  op_code;
  45         u8      mac_src[ETH_ALEN];      /* sender hardware address */
  46         __be32  ip_src;                 /* sender IP address */
  47         u8      mac_dst[ETH_ALEN];      /* target hardware address */
  48         __be32  ip_dst;                 /* target IP address */
  49 };
  50 #pragma pack()
  51
  52 /* Forward declaration */
  53 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
  54                                       bool strict_match);
  55 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp);
  56 static void rlb_src_unlink(struct bonding *bond, u32 index);
  57 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash,
  58                          u32 ip_dst_hash);
  59
  60 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
  61 {
  62         int i;
  63         u8 hash = 0;
  64
  65         for (i = 0; i < hash_size; i++)
  66                 hash ^= hash_start[i];
  67
  68         return hash;
  69 }
  70
  71 /*********************** tlb specific functions ***************************/
  72
  73 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
  74 {
  75         if (save_load) {
  76                 entry->load_history = 1 + entry->tx_bytes /
  77                                       BOND_TLB_REBALANCE_INTERVAL;
  78                 entry->tx_bytes = 0;
  79         }
  80
  81         entry->tx_slave = NULL;
  82         entry->next = TLB_NULL_INDEX;
  83         entry->prev = TLB_NULL_INDEX;
  84 }
  85
  86 static inline void tlb_init_slave(struct slave *slave)
  87 {
  88         SLAVE_TLB_INFO(slave).load = 0;
  89         SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
  90 }
  91
  92 static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
  93                          int save_load)
  94 {
  95         struct tlb_client_info *tx_hash_table;
  96         u32 index;
  97
  98         /* clear slave from tx_hashtbl */
  99         tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
 100
 101         /* skip this if we've already freed the tx hash table */
 102         if (tx_hash_table) {
 103                 index = SLAVE_TLB_INFO(slave).head;
 104                 while (index != TLB_NULL_INDEX) {
 105                         u32 next_index = tx_hash_table[index].next;
 106
 107                         tlb_init_table_entry(&tx_hash_table[index], save_load);
 108                         index = next_index;
 109                 }
 110         }
 111
 112         tlb_init_slave(slave);
 113 }
 114
 115 static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
 116                          int save_load)
 117 {
 118         spin_lock_bh(&bond->mode_lock);
 119         __tlb_clear_slave(bond, slave, save_load);
 120         spin_unlock_bh(&bond->mode_lock);
 121 }
 122
 123 /* Must be called before starting the monitor timer */
 124 static int tlb_initialize(struct bonding *bond)
 125 {
 126         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 127         int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 128         struct tlb_client_info *new_hashtbl;
 129         int i;
 130
 131         new_hashtbl = kzalloc(size, GFP_KERNEL);
 132         if (!new_hashtbl)
 133                 return -ENOMEM;
 134
 135         spin_lock_bh(&bond->mode_lock);
 136
 137         bond_info->tx_hashtbl = new_hashtbl;
 138
 139         for (i = 0; i < TLB_HASH_TABLE_SIZE; i++)
 140                 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
 141
 142         spin_unlock_bh(&bond->mode_lock);
 143
 144         return 0;
 145 }
 146
 147 /* Must be called only after all slaves have been released */
 148 static void tlb_deinitialize(struct bonding *bond)
 149 {
 150         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 151
 152         spin_lock_bh(&bond->mode_lock);
 153
 154         kfree(bond_info->tx_hashtbl);
 155         bond_info->tx_hashtbl = NULL;
 156
 157         spin_unlock_bh(&bond->mode_lock);
 158 }
 159
 160 static long long compute_gap(struct slave *slave)
 161 {
 162         return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
 163                (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
 164 }
 165
 166 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
 167 {
 168         struct slave *slave, *least_loaded;
 169         struct list_head *iter;
 170         long long max_gap;
 171
 172         least_loaded = NULL;
 173         max_gap = LLONG_MIN;
 174
 175         /* Find the slave with the largest gap */
 176         bond_for_each_slave_rcu(bond, slave, iter) {
 177                 if (bond_slave_can_tx(slave)) {
 178                         long long gap = compute_gap(slave);
 179
 180                         if (max_gap < gap) {
 181                                 least_loaded = slave;
 182                                 max_gap = gap;
 183                         }
 184                 }
 185         }
 186
 187         return least_loaded;
 188 }
 189
 190 static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
 191                                                 u32 skb_len)
 192 {
 193         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 194         struct tlb_client_info *hash_table;
 195         struct slave *assigned_slave;
 196
 197         hash_table = bond_info->tx_hashtbl;
 198         assigned_slave = hash_table[hash_index].tx_slave;
 199         if (!assigned_slave) {
 200                 assigned_slave = tlb_get_least_loaded_slave(bond);
 201
 202                 if (assigned_slave) {
 203                         struct tlb_slave_info *slave_info =
 204                                 &(SLAVE_TLB_INFO(assigned_slave));
 205                         u32 next_index = slave_info->head;
 206
 207                         hash_table[hash_index].tx_slave = assigned_slave;
 208                         hash_table[hash_index].next = next_index;
 209                         hash_table[hash_index].prev = TLB_NULL_INDEX;
 210
 211                         if (next_index != TLB_NULL_INDEX)
 212                                 hash_table[next_index].prev = hash_index;
 213
 214                         slave_info->head = hash_index;
 215                         slave_info->load +=
 216                                 hash_table[hash_index].load_history;
 217                 }
 218         }
 219
 220         if (assigned_slave)
 221                 hash_table[hash_index].tx_bytes += skb_len;
 222
 223         return assigned_slave;
 224 }
 225
 226 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
 227                                         u32 skb_len)
 228 {
 229         struct slave *tx_slave;
 230
 231         /* We don't need to disable softirq here, because
 232          * tlb_choose_channel() is only called by bond_alb_xmit()
 233          * which already has softirq disabled.
 234          */
 235         spin_lock(&bond->mode_lock);
 236         tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
 237         spin_unlock(&bond->mode_lock);
 238
 239         return tx_slave;
 240 }
 241
 242 /*********************** rlb specific functions ***************************/
 243
 244 /* when an ARP REPLY is received from a client update its info
 245  * in the rx_hashtbl
 246  */
 247 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 248 {
 249         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 250         struct rlb_client_info *client_info;
 251         u32 hash_index;
 252
 253         spin_lock_bh(&bond->mode_lock);
 254
 255         hash_index = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
 256         client_info = &(bond_info->rx_hashtbl[hash_index]);
 257
 258         if ((client_info->assigned) &&
 259             (client_info->ip_src == arp->ip_dst) &&
 260             (client_info->ip_dst == arp->ip_src) &&
 261             (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) {
 262                 /* update the clients MAC address */
 263                 ether_addr_copy(client_info->mac_dst, arp->mac_src);
 264                 client_info->ntt = 1;
 265                 bond_info->rx_ntt = 1;
 266         }
 267
 268         spin_unlock_bh(&bond->mode_lock);
 269 }
 270
 271 static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
 272                         struct slave *slave)
 273 {
 274         struct arp_pkt *arp, _arp;
 275
 276         if (skb->protocol != cpu_to_be16(ETH_P_ARP))
 277                 goto out;
 278
 279         arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp);
 280         if (!arp)
 281                 goto out;
 282
 283         /* We received an ARP from arp->ip_src.
 284          * We might have used this IP address previously (on the bonding host
 285          * itself or on a system that is bridged together with the bond).
 286          * However, if arp->mac_src is different than what is stored in
 287          * rx_hashtbl, some other host is now using the IP and we must prevent
 288          * sending out client updates with this IP address and the old MAC
 289          * address.
 290          * Clean up all hash table entries that have this address as ip_src but
 291          * have a different mac_src.
 292          */
 293         rlb_purge_src_ip(bond, arp);
 294
 295         if (arp->op_code == htons(ARPOP_REPLY)) {
 296                 /* update rx hash table for this ARP */
 297                 rlb_update_entry_from_arp(bond, arp);
 298                 slave_dbg(bond->dev, slave->dev, "Server received an ARP Reply from client\n");
 299         }
 300 out:
 301         return RX_HANDLER_ANOTHER;
 302 }
 303
 304 /* Caller must hold rcu_read_lock() */
 305 static struct slave *__rlb_next_rx_slave(struct bonding *bond)
 306 {
 307         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 308         struct slave *before = NULL, *rx_slave = NULL, *slave;
 309         struct list_head *iter;
 310         bool found = false;
 311
 312         bond_for_each_slave_rcu(bond, slave, iter) {
 313                 if (!bond_slave_can_tx(slave))
 314                         continue;
 315                 if (!found) {
 316                         if (!before || before->speed < slave->speed)
 317                                 before = slave;
 318                 } else {
 319                         if (!rx_slave || rx_slave->speed < slave->speed)
 320                                 rx_slave = slave;
 321                 }
 322                 if (slave == bond_info->rx_slave)
 323                         found = true;
 324         }
 325         /* we didn't find anything after the current or we have something
 326          * better before and up to the current slave
 327          */
 328         if (!rx_slave || (before && rx_slave->speed < before->speed))
 329                 rx_slave = before;
 330
 331         if (rx_slave)
 332                 bond_info->rx_slave = rx_slave;
 333
 334         return rx_slave;
 335 }
 336
 337 /* Caller must hold RTNL, rcu_read_lock is obtained only to silence checkers */
 338 static struct slave *rlb_next_rx_slave(struct bonding *bond)
 339 {
 340         struct slave *rx_slave;
 341
 342         ASSERT_RTNL();
 343
 344         rcu_read_lock();
 345         rx_slave = __rlb_next_rx_slave(bond);
 346         rcu_read_unlock();
 347
 348         return rx_slave;
 349 }
 350
 351 /* teach the switch the mac of a disabled slave
 352  * on the primary for fault tolerance
 353  *
 354  * Caller must hold RTNL
 355  */
 356 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
 357 {
 358         struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 359
 360         if (!curr_active)
 361                 return;
 362
 363         if (!bond->alb_info.primary_is_promisc) {
 364                 if (!dev_set_promiscuity(curr_active->dev, 1))
 365                         bond->alb_info.primary_is_promisc = 1;
 366                 else
 367                         bond->alb_info.primary_is_promisc = 0;
 368         }
 369
 370         bond->alb_info.rlb_promisc_timeout_counter = 0;
 371
 372         alb_send_learning_packets(curr_active, addr, true);
 373 }
 374
 375 /* slave being removed should not be active at this point
 376  *
 377  * Caller must hold rtnl.
 378  */
 379 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 380 {
 381         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 382         struct rlb_client_info *rx_hash_table;
 383         u32 index, next_index;
 384
 385         /* clear slave from rx_hashtbl */
 386         spin_lock_bh(&bond->mode_lock);
 387
 388         rx_hash_table = bond_info->rx_hashtbl;
 389         index = bond_info->rx_hashtbl_used_head;
 390         for (; index != RLB_NULL_INDEX; index = next_index) {
 391                 next_index = rx_hash_table[index].used_next;
 392                 if (rx_hash_table[index].slave == slave) {
 393                         struct slave *assigned_slave = rlb_next_rx_slave(bond);
 394
 395                         if (assigned_slave) {
 396                                 rx_hash_table[index].slave = assigned_slave;
 397                                 if (is_valid_ether_addr(rx_hash_table[index].mac_dst)) {
 398                                         bond_info->rx_hashtbl[index].ntt = 1;
 399                                         bond_info->rx_ntt = 1;
 400                                         /* A slave has been removed from the
 401                                          * table because it is either disabled
 402                                          * or being released. We must retry the
 403                                          * update to avoid clients from not
 404                                          * being updated & disconnecting when
 405                                          * there is stress
 406                                          */
 407                                         bond_info->rlb_update_retry_counter =
 408                                                 RLB_UPDATE_RETRY;
 409                                 }
 410                         } else {  /* there is no active slave */
 411                                 rx_hash_table[index].slave = NULL;
 412                         }
 413                 }
 414         }
 415
 416         spin_unlock_bh(&bond->mode_lock);
 417
 418         if (slave != rtnl_dereference(bond->curr_active_slave))
 419                 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 420 }
 421
 422 static void rlb_update_client(struct rlb_client_info *client_info)
 423 {
 424         int i;
 425
 426         if (!client_info->slave || !is_valid_ether_addr(client_info->mac_dst))
 427                 return;
 428
 429         for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 430                 struct sk_buff *skb;
 431
 432                 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 433                                  client_info->ip_dst,
 434                                  client_info->slave->dev,
 435                                  client_info->ip_src,
 436                                  client_info->mac_dst,
 437                                  client_info->slave->dev->dev_addr,
 438                                  client_info->mac_dst);
 439                 if (!skb) {
 440                         slave_err(client_info->slave->bond->dev,
 441                                   client_info->slave->dev,
 442                                   "failed to create an ARP packet\n");
 443                         continue;
 444                 }
 445
 446                 skb->dev = client_info->slave->dev;
 447
 448                 if (client_info->vlan_id) {
 449                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 450                                                client_info->vlan_id);
 451                 }
 452
 453                 arp_xmit(skb);
 454         }
 455 }
 456
 457 /* sends ARP REPLIES that update the clients that need updating */
 458 static void rlb_update_rx_clients(struct bonding *bond)
 459 {
 460         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 461         struct rlb_client_info *client_info;
 462         u32 hash_index;
 463
 464         spin_lock_bh(&bond->mode_lock);
 465
 466         hash_index = bond_info->rx_hashtbl_used_head;
 467         for (; hash_index != RLB_NULL_INDEX;
 468              hash_index = client_info->used_next) {
 469                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 470                 if (client_info->ntt) {
 471                         rlb_update_client(client_info);
 472                         if (bond_info->rlb_update_retry_counter == 0)
 473                                 client_info->ntt = 0;
 474                 }
 475         }
 476
 477         /* do not update the entries again until this counter is zero so that
 478          * not to confuse the clients.
 479          */
 480         bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 481
 482         spin_unlock_bh(&bond->mode_lock);
 483 }
 484
 485 /* The slave was assigned a new mac address - update the clients */
 486 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 487 {
 488         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 489         struct rlb_client_info *client_info;
 490         int ntt = 0;
 491         u32 hash_index;
 492
 493         spin_lock_bh(&bond->mode_lock);
 494
 495         hash_index = bond_info->rx_hashtbl_used_head;
 496         for (; hash_index != RLB_NULL_INDEX;
 497              hash_index = client_info->used_next) {
 498                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 499
 500                 if ((client_info->slave == slave) &&
 501                     is_valid_ether_addr(client_info->mac_dst)) {
 502                         client_info->ntt = 1;
 503                         ntt = 1;
 504                 }
 505         }
 506
 507         /* update the team's flag only after the whole iteration */
 508         if (ntt) {
 509                 bond_info->rx_ntt = 1;
 510                 /* fasten the change */
 511                 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 512         }
 513
 514         spin_unlock_bh(&bond->mode_lock);
 515 }
 516
 517 /* mark all clients using src_ip to be updated */
 518 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
 519 {
 520         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 521         struct rlb_client_info *client_info;
 522         u32 hash_index;
 523
 524         spin_lock(&bond->mode_lock);
 525
 526         hash_index = bond_info->rx_hashtbl_used_head;
 527         for (; hash_index != RLB_NULL_INDEX;
 528              hash_index = client_info->used_next) {
 529                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 530
 531                 if (!client_info->slave) {
 532                         netdev_err(bond->dev, "found a client with no channel in the client's hash table\n");
 533                         continue;
 534                 }
 535                 /* update all clients using this src_ip, that are not assigned
 536                  * to the team's address (curr_active_slave) and have a known
 537                  * unicast mac address.
 538                  */
 539                 if ((client_info->ip_src == src_ip) &&
 540                     !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
 541                                              bond->dev->dev_addr) &&
 542                     is_valid_ether_addr(client_info->mac_dst)) {
 543                         client_info->ntt = 1;
 544                         bond_info->rx_ntt = 1;
 545                 }
 546         }
 547
 548         spin_unlock(&bond->mode_lock);
 549 }
 550
 551 static struct slave *rlb_choose_channel(struct sk_buff *skb,
 552                                         struct bonding *bond,
 553                                         const struct arp_pkt *arp)
 554 {
 555         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 556         struct slave *assigned_slave, *curr_active_slave;
 557         struct rlb_client_info *client_info;
 558         u32 hash_index = 0;
 559
 560         spin_lock(&bond->mode_lock);
 561
 562         curr_active_slave = rcu_dereference(bond->curr_active_slave);
 563
 564         hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
 565         client_info = &(bond_info->rx_hashtbl[hash_index]);
 566
 567         if (client_info->assigned) {
 568                 if ((client_info->ip_src == arp->ip_src) &&
 569                     (client_info->ip_dst == arp->ip_dst)) {
 570                         /* the entry is already assigned to this client */
 571                         if (!is_broadcast_ether_addr(arp->mac_dst)) {
 572                                 /* update mac address from arp */
 573                                 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
 574                         }
 575                         ether_addr_copy(client_info->mac_src, arp->mac_src);
 576
 577                         assigned_slave = client_info->slave;
 578                         if (assigned_slave) {
 579                                 spin_unlock(&bond->mode_lock);
 580                                 return assigned_slave;
 581                         }
 582                 } else {
 583                         /* the entry is already assigned to some other client,
 584                          * move the old client to primary (curr_active_slave) so
 585                          * that the new client can be assigned to this entry.
 586                          */
 587                         if (curr_active_slave &&
 588                             client_info->slave != curr_active_slave) {
 589                                 client_info->slave = curr_active_slave;
 590                                 rlb_update_client(client_info);
 591                         }
 592                 }
 593         }
 594         /* assign a new slave */
 595         assigned_slave = __rlb_next_rx_slave(bond);
 596
 597         if (assigned_slave) {
 598                 if (!(client_info->assigned &&
 599                       client_info->ip_src == arp->ip_src)) {
 600                         /* ip_src is going to be updated,
 601                          * fix the src hash list
 602                          */
 603                         u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
 604                                                     sizeof(arp->ip_src));
 605                         rlb_src_unlink(bond, hash_index);
 606                         rlb_src_link(bond, hash_src, hash_index);
 607                 }
 608
 609                 client_info->ip_src = arp->ip_src;
 610                 client_info->ip_dst = arp->ip_dst;
 611                 /* arp->mac_dst is broadcast for arp requests.
 612                  * will be updated with clients actual unicast mac address
 613                  * upon receiving an arp reply.
 614                  */
 615                 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
 616                 ether_addr_copy(client_info->mac_src, arp->mac_src);
 617                 client_info->slave = assigned_slave;
 618
 619                 if (is_valid_ether_addr(client_info->mac_dst)) {
 620                         client_info->ntt = 1;
 621                         bond->alb_info.rx_ntt = 1;
 622                 } else {
 623                         client_info->ntt = 0;
 624                 }
 625
 626                 if (vlan_get_tag(skb, &client_info->vlan_id))
 627                         client_info->vlan_id = 0;
 628
 629                 if (!client_info->assigned) {
 630                         u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
 631
 632                         bond_info->rx_hashtbl_used_head = hash_index;
 633                         client_info->used_next = prev_tbl_head;
 634                         if (prev_tbl_head != RLB_NULL_INDEX) {
 635                                 bond_info->rx_hashtbl[prev_tbl_head].used_prev =
 636                                         hash_index;
 637                         }
 638                         client_info->assigned = 1;
 639                 }
 640         }
 641
 642         spin_unlock(&bond->mode_lock);
 643
 644         return assigned_slave;
 645 }
 646
 647 /* chooses (and returns) transmit channel for arp reply
 648  * does not choose channel for other arp types since they are
 649  * sent on the curr_active_slave
 650  */
 651 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 652 {
 653         struct slave *tx_slave = NULL;
 654         struct arp_pkt *arp;
 655
 656         if (!pskb_network_may_pull(skb, sizeof(*arp)))
 657                 return NULL;
 658         arp = (struct arp_pkt *)skb_network_header(skb);
 659
 660         /* Don't modify or load balance ARPs that do not originate locally
 661          * (e.g.,arrive via a bridge).
 662          */
 663         if (!bond_slave_has_mac_rx(bond, arp->mac_src))
 664                 return NULL;
 665
 666         if (arp->op_code == htons(ARPOP_REPLY)) {
 667                 /* the arp must be sent on the selected rx channel */
 668                 tx_slave = rlb_choose_channel(skb, bond, arp);
 669                 if (tx_slave)
 670                         bond_hw_addr_copy(arp->mac_src, tx_slave->dev->dev_addr,
 671                                           tx_slave->dev->addr_len);
 672                 netdev_dbg(bond->dev, "(slave %s): Server sent ARP Reply packet\n",
 673                            tx_slave ? tx_slave->dev->name : "NULL");
 674         } else if (arp->op_code == htons(ARPOP_REQUEST)) {
 675                 /* Create an entry in the rx_hashtbl for this client as a
 676                  * place holder.
 677                  * When the arp reply is received the entry will be updated
 678                  * with the correct unicast address of the client.
 679                  */
 680                 tx_slave = rlb_choose_channel(skb, bond, arp);
 681
 682                 /* The ARP reply packets must be delayed so that
 683                  * they can cancel out the influence of the ARP request.
 684                  */
 685                 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 686
 687                 /* arp requests are broadcast and are sent on the primary
 688                  * the arp request will collapse all clients on the subnet to
 689                  * the primary slave. We must register these clients to be
 690                  * updated with their assigned mac.
 691                  */
 692                 rlb_req_update_subnet_clients(bond, arp->ip_src);
 693                 netdev_dbg(bond->dev, "(slave %s): Server sent ARP Request packet\n",
 694                            tx_slave ? tx_slave->dev->name : "NULL");
 695         }
 696
 697         return tx_slave;
 698 }
 699
 700 static void rlb_rebalance(struct bonding *bond)
 701 {
 702         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 703         struct slave *assigned_slave;
 704         struct rlb_client_info *client_info;
 705         int ntt;
 706         u32 hash_index;
 707
 708         spin_lock_bh(&bond->mode_lock);
 709
 710         ntt = 0;
 711         hash_index = bond_info->rx_hashtbl_used_head;
 712         for (; hash_index != RLB_NULL_INDEX;
 713              hash_index = client_info->used_next) {
 714                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 715                 assigned_slave = __rlb_next_rx_slave(bond);
 716                 if (assigned_slave && (client_info->slave != assigned_slave)) {
 717                         client_info->slave = assigned_slave;
 718                         if (!is_zero_ether_addr(client_info->mac_dst)) {
 719                                 client_info->ntt = 1;
 720                                 ntt = 1;
 721                         }
 722                 }
 723         }
 724
 725         /* update the team's flag only after the whole iteration */
 726         if (ntt)
 727                 bond_info->rx_ntt = 1;
 728         spin_unlock_bh(&bond->mode_lock);
 729 }
 730
 731 /* Caller must hold mode_lock */
 732 static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
 733 {
 734         entry->used_next = RLB_NULL_INDEX;
 735         entry->used_prev = RLB_NULL_INDEX;
 736         entry->assigned = 0;
 737         entry->slave = NULL;
 738         entry->vlan_id = 0;
 739 }
 740 static void rlb_init_table_entry_src(struct rlb_client_info *entry)
 741 {
 742         entry->src_first = RLB_NULL_INDEX;
 743         entry->src_prev = RLB_NULL_INDEX;
 744         entry->src_next = RLB_NULL_INDEX;
 745 }
 746
 747 static void rlb_init_table_entry(struct rlb_client_info *entry)
 748 {
 749         memset(entry, 0, sizeof(struct rlb_client_info));
 750         rlb_init_table_entry_dst(entry);
 751         rlb_init_table_entry_src(entry);
 752 }
 753
 754 static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
 755 {
 756         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 757         u32 next_index = bond_info->rx_hashtbl[index].used_next;
 758         u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
 759
 760         if (index == bond_info->rx_hashtbl_used_head)
 761                 bond_info->rx_hashtbl_used_head = next_index;
 762         if (prev_index != RLB_NULL_INDEX)
 763                 bond_info->rx_hashtbl[prev_index].used_next = next_index;
 764         if (next_index != RLB_NULL_INDEX)
 765                 bond_info->rx_hashtbl[next_index].used_prev = prev_index;
 766 }
 767
 768 /* unlink a rlb hash table entry from the src list */
 769 static void rlb_src_unlink(struct bonding *bond, u32 index)
 770 {
 771         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 772         u32 next_index = bond_info->rx_hashtbl[index].src_next;
 773         u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
 774
 775         bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
 776         bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
 777
 778         if (next_index != RLB_NULL_INDEX)
 779                 bond_info->rx_hashtbl[next_index].src_prev = prev_index;
 780
 781         if (prev_index == RLB_NULL_INDEX)
 782                 return;
 783
 784         /* is prev_index pointing to the head of this list? */
 785         if (bond_info->rx_hashtbl[prev_index].src_first == index)
 786                 bond_info->rx_hashtbl[prev_index].src_first = next_index;
 787         else
 788                 bond_info->rx_hashtbl[prev_index].src_next = next_index;
 789
 790 }
 791
 792 static void rlb_delete_table_entry(struct bonding *bond, u32 index)
 793 {
 794         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 795         struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 796
 797         rlb_delete_table_entry_dst(bond, index);
 798         rlb_init_table_entry_dst(entry);
 799
 800         rlb_src_unlink(bond, index);
 801 }
 802
 803 /* add the rx_hashtbl[ip_dst_hash] entry to the list
 804  * of entries with identical ip_src_hash
 805  */
 806 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
 807 {
 808         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 809         u32 next;
 810
 811         bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
 812         next = bond_info->rx_hashtbl[ip_src_hash].src_first;
 813         bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
 814         if (next != RLB_NULL_INDEX)
 815                 bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
 816         bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
 817 }
 818
 819 /* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does
 820  * not match arp->mac_src
 821  */
 822 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
 823 {
 824         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 825         u32 ip_src_hash = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
 826         u32 index;
 827
 828         spin_lock_bh(&bond->mode_lock);
 829
 830         index = bond_info->rx_hashtbl[ip_src_hash].src_first;
 831         while (index != RLB_NULL_INDEX) {
 832                 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 833                 u32 next_index = entry->src_next;
 834
 835                 if (entry->ip_src == arp->ip_src &&
 836                     !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
 837                         rlb_delete_table_entry(bond, index);
 838                 index = next_index;
 839         }
 840         spin_unlock_bh(&bond->mode_lock);
 841 }
 842
 843 static int rlb_initialize(struct bonding *bond)
 844 {
 845         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 846         struct rlb_client_info  *new_hashtbl;
 847         int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 848         int i;
 849
 850         new_hashtbl = kmalloc(size, GFP_KERNEL);
 851         if (!new_hashtbl)
 852                 return -1;
 853
 854         spin_lock_bh(&bond->mode_lock);
 855
 856         bond_info->rx_hashtbl = new_hashtbl;
 857
 858         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 859
 860         for (i = 0; i < RLB_HASH_TABLE_SIZE; i++)
 861                 rlb_init_table_entry(bond_info->rx_hashtbl + i);
 862
 863         spin_unlock_bh(&bond->mode_lock);
 864
 865         /* register to receive ARPs */
 866         bond->recv_probe = rlb_arp_recv;
 867
 868         return 0;
 869 }
 870
 871 static void rlb_deinitialize(struct bonding *bond)
 872 {
 873         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 874
 875         spin_lock_bh(&bond->mode_lock);
 876
 877         kfree(bond_info->rx_hashtbl);
 878         bond_info->rx_hashtbl = NULL;
 879         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 880
 881         spin_unlock_bh(&bond->mode_lock);
 882 }
 883
 884 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 885 {
 886         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 887         u32 curr_index;
 888
 889         spin_lock_bh(&bond->mode_lock);
 890
 891         curr_index = bond_info->rx_hashtbl_used_head;
 892         while (curr_index != RLB_NULL_INDEX) {
 893                 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 894                 u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
 895
 896                 if (curr->vlan_id == vlan_id)
 897                         rlb_delete_table_entry(bond, curr_index);
 898
 899                 curr_index = next_index;
 900         }
 901
 902         spin_unlock_bh(&bond->mode_lock);
 903 }
 904
 905 /*********************** tlb/rlb shared functions *********************/
 906
 907 static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
 908                             __be16 vlan_proto, u16 vid)
 909 {
 910         struct learning_pkt pkt;
 911         struct sk_buff *skb;
 912         int size = sizeof(struct learning_pkt);
 913
 914         memset(&pkt, 0, size);
 915         ether_addr_copy(pkt.mac_dst, mac_addr);
 916         ether_addr_copy(pkt.mac_src, mac_addr);
 917         pkt.type = cpu_to_be16(ETH_P_LOOPBACK);
 918
 919         skb = dev_alloc_skb(size);
 920         if (!skb)
 921                 return;
 922
 923         skb_put_data(skb, &pkt, size);
 924
 925         skb_reset_mac_header(skb);
 926         skb->network_header = skb->mac_header + ETH_HLEN;
 927         skb->protocol = pkt.type;
 928         skb->priority = TC_PRIO_CONTROL;
 929         skb->dev = slave->dev;
 930
 931         slave_dbg(slave->bond->dev, slave->dev,
 932                   "Send learning packet: mac %pM vlan %d\n", mac_addr, vid);
 933
 934         if (vid)
 935                 __vlan_hwaccel_put_tag(skb, vlan_proto, vid);
 936
 937         dev_queue_xmit(skb);
 938 }
 939
 940 struct alb_walk_data {
 941         struct bonding *bond;
 942         struct slave *slave;
 943         u8 *mac_addr;
 944         bool strict_match;
 945 };
 946
 947 static int alb_upper_dev_walk(struct net_device *upper,
 948                               struct netdev_nested_priv *priv)
 949 {
 950         struct alb_walk_data *data = (struct alb_walk_data *)priv->data;
 951         bool strict_match = data->strict_match;
 952         struct bonding *bond = data->bond;
 953         struct slave *slave = data->slave;
 954         u8 *mac_addr = data->mac_addr;
 955         struct bond_vlan_tag *tags;
 956
 957         if (is_vlan_dev(upper) &&
 958             bond->dev->lower_level == upper->lower_level - 1) {
 959                 if (upper->addr_assign_type == NET_ADDR_STOLEN) {
 960                         alb_send_lp_vid(slave, mac_addr,
 961                                         vlan_dev_vlan_proto(upper),
 962                                         vlan_dev_vlan_id(upper));
 963                 } else {
 964                         alb_send_lp_vid(slave, upper->dev_addr,
 965                                         vlan_dev_vlan_proto(upper),
 966                                         vlan_dev_vlan_id(upper));
 967                 }
 968         }
 969
 970         /* If this is a macvlan device, then only send updates
 971          * when strict_match is turned off.
 972          */
 973         if (netif_is_macvlan(upper) && !strict_match) {
 974                 tags = bond_verify_device_path(bond->dev, upper, 0);
 975                 if (IS_ERR_OR_NULL(tags))
 976                         BUG();
 977                 alb_send_lp_vid(slave, upper->dev_addr,
 978                                 tags[0].vlan_proto, tags[0].vlan_id);
 979                 kfree(tags);
 980         }
 981
 982         return 0;
 983 }
 984
 985 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
 986                                       bool strict_match)
 987 {
 988         struct bonding *bond = bond_get_bond_by_slave(slave);
 989         struct netdev_nested_priv priv;
 990         struct alb_walk_data data = {
 991                 .strict_match = strict_match,
 992                 .mac_addr = mac_addr,
 993                 .slave = slave,
 994                 .bond = bond,
 995         };
 996
 997         priv.data = (void *)&data;
 998         /* send untagged */
 999         alb_send_lp_vid(slave, mac_addr, 0, 0);
1000
1001         /* loop through all devices and see if we need to send a packet
1002          * for that device.
1003          */
1004         rcu_read_lock();
1005         netdev_walk_all_upper_dev_rcu(bond->dev, alb_upper_dev_walk, &priv);
1006         rcu_read_unlock();
1007 }
1008
1009 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[],
1010                                   unsigned int len)
1011 {
1012         struct net_device *dev = slave->dev;
1013         struct sockaddr_storage ss;
1014
1015         if (BOND_MODE(slave->bond) == BOND_MODE_TLB) {
1016                 memcpy(dev->dev_addr, addr, len);
1017                 return 0;
1018         }
1019
1020         /* for rlb each slave must have a unique hw mac addresses so that
1021          * each slave will receive packets destined to a different mac
1022          */
1023         memcpy(ss.__data, addr, len);
1024         ss.ss_family = dev->type;
1025         if (dev_set_mac_address(dev, (struct sockaddr *)&ss, NULL)) {
1026                 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");
1027                 return -EOPNOTSUPP;
1028         }
1029         return 0;
1030 }
1031
1032 /* Swap MAC addresses between two slaves.
1033  *
1034  * Called with RTNL held, and no other locks.
1035  */
1036 static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1037 {
1038         u8 tmp_mac_addr[MAX_ADDR_LEN];
1039
1040         bond_hw_addr_copy(tmp_mac_addr, slave1->dev->dev_addr,
1041                           slave1->dev->addr_len);
1042         alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr,
1043                                slave2->dev->addr_len);
1044         alb_set_slave_mac_addr(slave2, tmp_mac_addr,
1045                                slave1->dev->addr_len);
1046
1047 }
1048
1049 /* Send learning packets after MAC address swap.
1050  *
1051  * Called with RTNL and no other locks
1052  */
1053 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1054                                 struct slave *slave2)
1055 {
1056         int slaves_state_differ = (bond_slave_can_tx(slave1) != bond_slave_can_tx(slave2));
1057         struct slave *disabled_slave = NULL;
1058
1059         ASSERT_RTNL();
1060
1061         /* fasten the change in the switch */
1062         if (bond_slave_can_tx(slave1)) {
1063                 alb_send_learning_packets(slave1, slave1->dev->dev_addr, false);
1064                 if (bond->alb_info.rlb_enabled) {
1065                         /* inform the clients that the mac address
1066                          * has changed
1067                          */
1068                         rlb_req_update_slave_clients(bond, slave1);
1069                 }
1070         } else {
1071                 disabled_slave = slave1;
1072         }
1073
1074         if (bond_slave_can_tx(slave2)) {
1075                 alb_send_learning_packets(slave2, slave2->dev->dev_addr, false);
1076                 if (bond->alb_info.rlb_enabled) {
1077                         /* inform the clients that the mac address
1078                          * has changed
1079                          */
1080                         rlb_req_update_slave_clients(bond, slave2);
1081                 }
1082         } else {
1083                 disabled_slave = slave2;
1084         }
1085
1086         if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1087                 /* A disabled slave was assigned an active mac addr */
1088                 rlb_teach_disabled_mac_on_primary(bond,
1089                                                   disabled_slave->dev->dev_addr);
1090         }
1091 }
1092
1093 /**
1094  * alb_change_hw_addr_on_detach
1095  * @bond: bonding we're working on
1096  * @slave: the slave that was just detached
1097  *
1098  * We assume that @slave was already detached from the slave list.
1099  *
1100  * If @slave's permanent hw address is different both from its current
1101  * address and from @bond's address, then somewhere in the bond there's
1102  * a slave that has @slave's permanet address as its current address.
1103  * We'll make sure that slave no longer uses @slave's permanent address.
1104  *
1105  * Caller must hold RTNL and no other locks
1106  */
1107 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1108 {
1109         int perm_curr_diff;
1110         int perm_bond_diff;
1111         struct slave *found_slave;
1112
1113         perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1114                                                   slave->dev->dev_addr);
1115         perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1116                                                   bond->dev->dev_addr);
1117
1118         if (perm_curr_diff && perm_bond_diff) {
1119                 found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1120
1121                 if (found_slave) {
1122                         alb_swap_mac_addr(slave, found_slave);
1123                         alb_fasten_mac_swap(bond, slave, found_slave);
1124                 }
1125         }
1126 }
1127
1128 /**
1129  * alb_handle_addr_collision_on_attach
1130  * @bond: bonding we're working on
1131  * @slave: the slave that was just attached
1132  *
1133  * checks uniqueness of slave's mac address and handles the case the
1134  * new slave uses the bonds mac address.
1135  *
1136  * If the permanent hw address of @slave is @bond's hw address, we need to
1137  * find a different hw address to give @slave, that isn't in use by any other
1138  * slave in the bond. This address must be, of course, one of the permanent
1139  * addresses of the other slaves.
1140  *
1141  * We go over the slave list, and for each slave there we compare its
1142  * permanent hw address with the current address of all the other slaves.
1143  * If no match was found, then we've found a slave with a permanent address
1144  * that isn't used by any other slave in the bond, so we can assign it to
1145  * @slave.
1146  *
1147  * assumption: this function is called before @slave is attached to the
1148  *             bond slave list.
1149  */
1150 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1151 {
1152         struct slave *has_bond_addr = rcu_access_pointer(bond->curr_active_slave);
1153         struct slave *tmp_slave1, *free_mac_slave = NULL;
1154         struct list_head *iter;
1155
1156         if (!bond_has_slaves(bond)) {
1157                 /* this is the first slave */
1158                 return 0;
1159         }
1160
1161         /* if slave's mac address differs from bond's mac address
1162          * check uniqueness of slave's mac address against the other
1163          * slaves in the bond.
1164          */
1165         if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1166                 if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1167                         return 0;
1168
1169                 /* Try setting slave mac to bond address and fall-through
1170                  * to code handling that situation below...
1171                  */
1172                 alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1173                                        bond->dev->addr_len);
1174         }
1175
1176         /* The slave's address is equal to the address of the bond.
1177          * Search for a spare address in the bond for this slave.
1178          */
1179         bond_for_each_slave(bond, tmp_slave1, iter) {
1180                 if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1181                         /* no slave has tmp_slave1's perm addr
1182                          * as its curr addr
1183                          */
1184                         free_mac_slave = tmp_slave1;
1185                         break;
1186                 }
1187
1188                 if (!has_bond_addr) {
1189                         if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1190                                                     bond->dev->dev_addr)) {
1191
1192                                 has_bond_addr = tmp_slave1;
1193                         }
1194                 }
1195         }
1196
1197         if (free_mac_slave) {
1198                 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1199                                        free_mac_slave->dev->addr_len);
1200
1201                 slave_warn(bond->dev, slave->dev, "the slave hw address is in use by the bond; giving it the hw address of %s\n",
1202                            free_mac_slave->dev->name);
1203
1204         } else if (has_bond_addr) {
1205                 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");
1206                 return -EFAULT;
1207         }
1208
1209         return 0;
1210 }
1211
1212 /**
1213  * alb_set_mac_address
1214  * @bond: bonding we're working on
1215  * @addr: MAC address to set
1216  *
1217  * In TLB mode all slaves are configured to the bond's hw address, but set
1218  * their dev_addr field to different addresses (based on their permanent hw
1219  * addresses).
1220  *
1221  * For each slave, this function sets the interface to the new address and then
1222  * changes its dev_addr field to its previous value.
1223  *
1224  * Unwinding assumes bond's mac address has not yet changed.
1225  */
1226 static int alb_set_mac_address(struct bonding *bond, void *addr)
1227 {
1228         struct slave *slave, *rollback_slave;
1229         struct list_head *iter;
1230         struct sockaddr_storage ss;
1231         char tmp_addr[MAX_ADDR_LEN];
1232         int res;
1233
1234         if (bond->alb_info.rlb_enabled)
1235                 return 0;
1236
1237         bond_for_each_slave(bond, slave, iter) {
1238                 /* save net_device's current hw address */
1239                 bond_hw_addr_copy(tmp_addr, slave->dev->dev_addr,
1240                                   slave->dev->addr_len);
1241
1242                 res = dev_set_mac_address(slave->dev, addr, NULL);
1243
1244                 /* restore net_device's hw address */
1245                 bond_hw_addr_copy(slave->dev->dev_addr, tmp_addr,
1246                                   slave->dev->addr_len);
1247
1248                 if (res)
1249                         goto unwind;
1250         }
1251
1252         return 0;
1253
1254 unwind:
1255         memcpy(ss.__data, bond->dev->dev_addr, bond->dev->addr_len);
1256         ss.ss_family = bond->dev->type;
1257
1258         /* unwind from head to the slave that failed */
1259         bond_for_each_slave(bond, rollback_slave, iter) {
1260                 if (rollback_slave == slave)
1261                         break;
1262                 bond_hw_addr_copy(tmp_addr, rollback_slave->dev->dev_addr,
1263                                   rollback_slave->dev->addr_len);
1264                 dev_set_mac_address(rollback_slave->dev,
1265                                     (struct sockaddr *)&ss, NULL);
1266                 bond_hw_addr_copy(rollback_slave->dev->dev_addr, tmp_addr,
1267                                   rollback_slave->dev->addr_len);
1268         }
1269
1270         return res;
1271 }
1272
1273 /************************ exported alb functions ************************/
1274
1275 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1276 {
1277         int res;
1278
1279         res = tlb_initialize(bond);
1280         if (res)
1281                 return res;
1282
1283         if (rlb_enabled) {
1284                 res = rlb_initialize(bond);
1285                 if (res) {
1286                         tlb_deinitialize(bond);
1287                         return res;
1288                 }
1289                 bond->alb_info.rlb_enabled = 1;
1290         } else {
1291                 bond->alb_info.rlb_enabled = 0;
1292         }
1293
1294         return 0;
1295 }
1296
1297 void bond_alb_deinitialize(struct bonding *bond)
1298 {
1299         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1300
1301         tlb_deinitialize(bond);
1302
1303         if (bond_info->rlb_enabled)
1304                 rlb_deinitialize(bond);
1305 }
1306
1307 static netdev_tx_t bond_do_alb_xmit(struct sk_buff *skb, struct bonding *bond,
1308                                     struct slave *tx_slave)
1309 {
1310         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1311         struct ethhdr *eth_data = eth_hdr(skb);
1312
1313         if (!tx_slave) {
1314                 /* unbalanced or unassigned, send through primary */
1315                 tx_slave = rcu_dereference(bond->curr_active_slave);
1316                 if (bond->params.tlb_dynamic_lb)
1317                         bond_info->unbalanced_load += skb->len;
1318         }
1319
1320         if (tx_slave && bond_slave_can_tx(tx_slave)) {
1321                 if (tx_slave != rcu_access_pointer(bond->curr_active_slave)) {
1322                         ether_addr_copy(eth_data->h_source,
1323                                         tx_slave->dev->dev_addr);
1324                 }
1325
1326                 return bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1327         }
1328
1329         if (tx_slave && bond->params.tlb_dynamic_lb) {
1330                 spin_lock(&bond->mode_lock);
1331                 __tlb_clear_slave(bond, tx_slave, 0);
1332                 spin_unlock(&bond->mode_lock);
1333         }
1334
1335         /* no suitable interface, frame not sent */
1336         return bond_tx_drop(bond->dev, skb);
1337 }
1338
1339 struct slave *bond_xmit_tlb_slave_get(struct bonding *bond,
1340                                       struct sk_buff *skb)
1341 {
1342         struct slave *tx_slave = NULL;
1343         struct ethhdr *eth_data;
1344         u32 hash_index;
1345
1346         skb_reset_mac_header(skb);
1347         eth_data = eth_hdr(skb);
1348
1349         /* Do not TX balance any multicast or broadcast */
1350         if (!is_multicast_ether_addr(eth_data->h_dest)) {
1351                 switch (skb->protocol) {
1352                 case htons(ETH_P_IP):
1353                 case htons(ETH_P_IPV6):
1354                         hash_index = bond_xmit_hash(bond, skb);
1355                         if (bond->params.tlb_dynamic_lb) {
1356                                 tx_slave = tlb_choose_channel(bond,
1357                                                               hash_index & 0xFF,
1358                                                               skb->len);
1359                         } else {
1360                                 struct bond_up_slave *slaves;
1361                                 unsigned int count;
1362
1363                                 slaves = rcu_dereference(bond->usable_slaves);
1364                                 count = slaves ? READ_ONCE(slaves->count) : 0;
1365                                 if (likely(count))
1366                                         tx_slave = slaves->arr[hash_index %
1367                                                                count];
1368                         }
1369                         break;
1370                 }
1371         }
1372         return tx_slave;
1373 }
1374
1375 netdev_tx_t bond_tlb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1376 {
1377         struct bonding *bond = netdev_priv(bond_dev);
1378         struct slave *tx_slave;
1379
1380         tx_slave = bond_xmit_tlb_slave_get(bond, skb);
1381         return bond_do_alb_xmit(skb, bond, tx_slave);
1382 }
1383
1384 struct slave *bond_xmit_alb_slave_get(struct bonding *bond,
1385                                       struct sk_buff *skb)
1386 {
1387         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1388         static const __be32 ip_bcast = htonl(0xffffffff);
1389         struct slave *tx_slave = NULL;
1390         const u8 *hash_start = NULL;
1391         bool do_tx_balance = true;
1392         struct ethhdr *eth_data;
1393         u32 hash_index = 0;
1394         int hash_size = 0;
1395
1396         skb_reset_mac_header(skb);
1397         eth_data = eth_hdr(skb);
1398
1399         switch (ntohs(skb->protocol)) {
1400         case ETH_P_IP: {
1401                 const struct iphdr *iph;
1402
1403                 if (is_broadcast_ether_addr(eth_data->h_dest) ||
1404                     !pskb_network_may_pull(skb, sizeof(*iph))) {
1405                         do_tx_balance = false;
1406                         break;
1407                 }
1408                 iph = ip_hdr(skb);
1409                 if (iph->daddr == ip_bcast || iph->protocol == IPPROTO_IGMP) {
1410                         do_tx_balance = false;
1411                         break;
1412                 }
1413                 hash_start = (char *)&(iph->daddr);
1414                 hash_size = sizeof(iph->daddr);
1415                 break;
1416         }
1417         case ETH_P_IPV6: {
1418                 const struct ipv6hdr *ip6hdr;
1419
1420                 /* IPv6 doesn't really use broadcast mac address, but leave
1421                  * that here just in case.
1422                  */
1423                 if (is_broadcast_ether_addr(eth_data->h_dest)) {
1424                         do_tx_balance = false;
1425                         break;
1426                 }
1427
1428                 /* IPv6 uses all-nodes multicast as an equivalent to
1429                  * broadcasts in IPv4.
1430                  */
1431                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1432                         do_tx_balance = false;
1433                         break;
1434                 }
1435
1436                 if (!pskb_network_may_pull(skb, sizeof(*ip6hdr))) {
1437                         do_tx_balance = false;
1438                         break;
1439                 }
1440                 /* Additionally, DAD probes should not be tx-balanced as that
1441                  * will lead to false positives for duplicate addresses and
1442                  * prevent address configuration from working.
1443                  */
1444                 ip6hdr = ipv6_hdr(skb);
1445                 if (ipv6_addr_any(&ip6hdr->saddr)) {
1446                         do_tx_balance = false;
1447                         break;
1448                 }
1449
1450                 hash_start = (char *)&ip6hdr->daddr;
1451                 hash_size = sizeof(ip6hdr->daddr);
1452                 break;
1453         }
1454         case ETH_P_ARP:
1455                 do_tx_balance = false;
1456                 if (bond_info->rlb_enabled)
1457                         tx_slave = rlb_arp_xmit(skb, bond);
1458                 break;
1459         default:
1460                 do_tx_balance = false;
1461                 break;
1462         }
1463
1464         if (do_tx_balance) {
1465                 if (bond->params.tlb_dynamic_lb) {
1466                         hash_index = _simple_hash(hash_start, hash_size);
1467                         tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1468                 } else {
1469                         /*
1470                          * do_tx_balance means we are free to select the tx_slave
1471                          * So we do exactly what tlb would do for hash selection
1472                          */
1473
1474                         struct bond_up_slave *slaves;
1475                         unsigned int count;
1476
1477                         slaves = rcu_dereference(bond->usable_slaves);
1478                         count = slaves ? READ_ONCE(slaves->count) : 0;
1479                         if (likely(count))
1480                                 tx_slave = slaves->arr[bond_xmit_hash(bond, skb) %
1481                                                        count];
1482                 }
1483         }
1484         return tx_slave;
1485 }
1486
1487 netdev_tx_t bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1488 {
1489         struct bonding *bond = netdev_priv(bond_dev);
1490         struct slave *tx_slave = NULL;
1491
1492         tx_slave = bond_xmit_alb_slave_get(bond, skb);
1493         return bond_do_alb_xmit(skb, bond, tx_slave);
1494 }
1495
1496 void bond_alb_monitor(struct work_struct *work)
1497 {
1498         struct bonding *bond = container_of(work, struct bonding,
1499                                             alb_work.work);
1500         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1501         struct list_head *iter;
1502         struct slave *slave;
1503
1504         if (!bond_has_slaves(bond)) {
1505                 atomic_set(&bond_info->tx_rebalance_counter, 0);
1506                 bond_info->lp_counter = 0;
1507                 goto re_arm;
1508         }
1509
1510         rcu_read_lock();
1511
1512         atomic_inc(&bond_info->tx_rebalance_counter);
1513         bond_info->lp_counter++;
1514
1515         /* send learning packets */
1516         if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1517                 bool strict_match;
1518
1519                 bond_for_each_slave_rcu(bond, slave, iter) {
1520                         /* If updating current_active, use all currently
1521                          * user mac addresses (!strict_match).  Otherwise, only
1522                          * use mac of the slave device.
1523                          * In RLB mode, we always use strict matches.
1524                          */
1525                         strict_match = (slave != rcu_access_pointer(bond->curr_active_slave) ||
1526                                         bond_info->rlb_enabled);
1527                         alb_send_learning_packets(slave, slave->dev->dev_addr,
1528                                                   strict_match);
1529                 }
1530                 bond_info->lp_counter = 0;
1531         }
1532
1533         /* rebalance tx traffic */
1534         if (atomic_read(&bond_info->tx_rebalance_counter) >= BOND_TLB_REBALANCE_TICKS) {
1535                 bond_for_each_slave_rcu(bond, slave, iter) {
1536                         tlb_clear_slave(bond, slave, 1);
1537                         if (slave == rcu_access_pointer(bond->curr_active_slave)) {
1538                                 SLAVE_TLB_INFO(slave).load =
1539                                         bond_info->unbalanced_load /
1540                                                 BOND_TLB_REBALANCE_INTERVAL;
1541                                 bond_info->unbalanced_load = 0;
1542                         }
1543                 }
1544                 atomic_set(&bond_info->tx_rebalance_counter, 0);
1545         }
1546
1547         if (bond_info->rlb_enabled) {
1548                 if (bond_info->primary_is_promisc &&
1549                     (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1550
1551                         /* dev_set_promiscuity requires rtnl and
1552                          * nothing else.  Avoid race with bond_close.
1553                          */
1554                         rcu_read_unlock();
1555                         if (!rtnl_trylock())
1556                                 goto re_arm;
1557
1558                         bond_info->rlb_promisc_timeout_counter = 0;
1559
1560                         /* If the primary was set to promiscuous mode
1561                          * because a slave was disabled then
1562                          * it can now leave promiscuous mode.
1563                          */
1564                         dev_set_promiscuity(rtnl_dereference(bond->curr_active_slave)->dev,
1565                                             -1);
1566                         bond_info->primary_is_promisc = 0;
1567
1568                         rtnl_unlock();
1569                         rcu_read_lock();
1570                 }
1571
1572                 if (bond_info->rlb_rebalance) {
1573                         bond_info->rlb_rebalance = 0;
1574                         rlb_rebalance(bond);
1575                 }
1576
1577                 /* check if clients need updating */
1578                 if (bond_info->rx_ntt) {
1579                         if (bond_info->rlb_update_delay_counter) {
1580                                 --bond_info->rlb_update_delay_counter;
1581                         } else {
1582                                 rlb_update_rx_clients(bond);
1583                                 if (bond_info->rlb_update_retry_counter)
1584                                         --bond_info->rlb_update_retry_counter;
1585                                 else
1586                                         bond_info->rx_ntt = 0;
1587                         }
1588                 }
1589         }
1590         rcu_read_unlock();
1591 re_arm:
1592         queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1593 }
1594
1595 /* assumption: called before the slave is attached to the bond
1596  * and not locked by the bond lock
1597  */
1598 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1599 {
1600         int res;
1601
1602         res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1603                                      slave->dev->addr_len);
1604         if (res)
1605                 return res;
1606
1607         res = alb_handle_addr_collision_on_attach(bond, slave);
1608         if (res)
1609                 return res;
1610
1611         tlb_init_slave(slave);
1612
1613         /* order a rebalance ASAP */
1614         atomic_set(&bond->alb_info.tx_rebalance_counter,
1615                    BOND_TLB_REBALANCE_TICKS);
1616
1617         if (bond->alb_info.rlb_enabled)
1618                 bond->alb_info.rlb_rebalance = 1;
1619
1620         return 0;
1621 }
1622
1623 /* Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1624  * if necessary.
1625  *
1626  * Caller must hold RTNL and no other locks
1627  */
1628 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1629 {
1630         if (bond_has_slaves(bond))
1631                 alb_change_hw_addr_on_detach(bond, slave);
1632
1633         tlb_clear_slave(bond, slave, 0);
1634
1635         if (bond->alb_info.rlb_enabled) {
1636                 bond->alb_info.rx_slave = NULL;
1637                 rlb_clear_slave(bond, slave);
1638         }
1639
1640 }
1641
1642 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1643 {
1644         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1645
1646         if (link == BOND_LINK_DOWN) {
1647                 tlb_clear_slave(bond, slave, 0);
1648                 if (bond->alb_info.rlb_enabled)
1649                         rlb_clear_slave(bond, slave);
1650         } else if (link == BOND_LINK_UP) {
1651                 /* order a rebalance ASAP */
1652                 atomic_set(&bond_info->tx_rebalance_counter,
1653                            BOND_TLB_REBALANCE_TICKS);
1654                 if (bond->alb_info.rlb_enabled) {
1655                         bond->alb_info.rlb_rebalance = 1;
1656                         /* If the updelay module parameter is smaller than the
1657                          * forwarding delay of the switch the rebalance will
1658                          * not work because the rebalance arp replies will
1659                          * not be forwarded to the clients..
1660                          */
1661                 }
1662         }
1663
1664         if (bond_is_nondyn_tlb(bond)) {
1665                 if (bond_update_slave_arr(bond, NULL))
1666                         pr_err("Failed to build slave-array for TLB mode.\n");
1667         }
1668 }
1669
1670 /**
1671  * bond_alb_handle_active_change - assign new curr_active_slave
1672  * @bond: our bonding struct
1673  * @new_slave: new slave to assign
1674  *
1675  * Set the bond->curr_active_slave to @new_slave and handle
1676  * mac address swapping and promiscuity changes as needed.
1677  *
1678  * Caller must hold RTNL
1679  */
1680 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1681 {
1682         struct slave *swap_slave;
1683         struct slave *curr_active;
1684
1685         curr_active = rtnl_dereference(bond->curr_active_slave);
1686         if (curr_active == new_slave)
1687                 return;
1688
1689         if (curr_active && bond->alb_info.primary_is_promisc) {
1690                 dev_set_promiscuity(curr_active->dev, -1);
1691                 bond->alb_info.primary_is_promisc = 0;
1692                 bond->alb_info.rlb_promisc_timeout_counter = 0;
1693         }
1694
1695         swap_slave = curr_active;
1696         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1697
1698         if (!new_slave || !bond_has_slaves(bond))
1699                 return;
1700
1701         /* set the new curr_active_slave to the bonds mac address
1702          * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1703          */
1704         if (!swap_slave)
1705                 swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1706
1707         /* Arrange for swap_slave and new_slave to temporarily be
1708          * ignored so we can mess with their MAC addresses without
1709          * fear of interference from transmit activity.
1710          */
1711         if (swap_slave)
1712                 tlb_clear_slave(bond, swap_slave, 1);
1713         tlb_clear_slave(bond, new_slave, 1);
1714
1715         /* in TLB mode, the slave might flip down/up with the old dev_addr,
1716          * and thus filter bond->dev_addr's packets, so force bond's mac
1717          */
1718         if (BOND_MODE(bond) == BOND_MODE_TLB) {
1719                 struct sockaddr_storage ss;
1720                 u8 tmp_addr[MAX_ADDR_LEN];
1721
1722                 bond_hw_addr_copy(tmp_addr, new_slave->dev->dev_addr,
1723                                   new_slave->dev->addr_len);
1724
1725                 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
1726                                   bond->dev->addr_len);
1727                 ss.ss_family = bond->dev->type;
1728                 /* we don't care if it can't change its mac, best effort */
1729                 dev_set_mac_address(new_slave->dev, (struct sockaddr *)&ss,
1730                                     NULL);
1731
1732                 bond_hw_addr_copy(new_slave->dev->dev_addr, tmp_addr,
1733                                   new_slave->dev->addr_len);
1734         }
1735
1736         /* curr_active_slave must be set before calling alb_swap_mac_addr */
1737         if (swap_slave) {
1738                 /* swap mac address */
1739                 alb_swap_mac_addr(swap_slave, new_slave);
1740                 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1741         } else {
1742                 /* set the new_slave to the bond mac address */
1743                 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1744                                        bond->dev->addr_len);
1745                 alb_send_learning_packets(new_slave, bond->dev->dev_addr,
1746                                           false);
1747         }
1748 }
1749
1750 /* Called with RTNL */
1751 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1752 {
1753         struct bonding *bond = netdev_priv(bond_dev);
1754         struct sockaddr_storage *ss = addr;
1755         struct slave *curr_active;
1756         struct slave *swap_slave;
1757         int res;
1758
1759         if (!is_valid_ether_addr(ss->__data))
1760                 return -EADDRNOTAVAIL;
1761
1762         res = alb_set_mac_address(bond, addr);
1763         if (res)
1764                 return res;
1765
1766         bond_hw_addr_copy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
1767
1768         /* If there is no curr_active_slave there is nothing else to do.
1769          * Otherwise we'll need to pass the new address to it and handle
1770          * duplications.
1771          */
1772         curr_active = rtnl_dereference(bond->curr_active_slave);
1773         if (!curr_active)
1774                 return 0;
1775
1776         swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1777
1778         if (swap_slave) {
1779                 alb_swap_mac_addr(swap_slave, curr_active);
1780                 alb_fasten_mac_swap(bond, swap_slave, curr_active);
1781         } else {
1782                 alb_set_slave_mac_addr(curr_active, bond_dev->dev_addr,
1783                                        bond_dev->addr_len);
1784
1785                 alb_send_learning_packets(curr_active,
1786                                           bond_dev->dev_addr, false);
1787                 if (bond->alb_info.rlb_enabled) {
1788                         /* inform clients mac address has changed */
1789                         rlb_req_update_slave_clients(bond, curr_active);
1790                 }
1791         }
1792
1793         return 0;
1794 }
1795
1796 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1797 {
1798         if (bond->alb_info.rlb_enabled)
1799                 rlb_clear_vlan(bond, vlan_id);
1800 }
1801