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/ipx.h>
  21 #include <net/arp.h>
  22 #include <net/ipv6.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, 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, u8 addr[])
 358 {
 359         struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 360
 361         if (!curr_active)
 362                 return;
 363
 364         if (!bond->alb_info.primary_is_promisc) {
 365                 if (!dev_set_promiscuity(curr_active->dev, 1))
 366                         bond->alb_info.primary_is_promisc = 1;
 367                 else
 368                         bond->alb_info.primary_is_promisc = 0;
 369         }
 370
 371         bond->alb_info.rlb_promisc_timeout_counter = 0;
 372
 373         alb_send_learning_packets(curr_active, addr, true);
 374 }
 375
 376 /* slave being removed should not be active at this point
 377  *
 378  * Caller must hold rtnl.
 379  */
 380 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 381 {
 382         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 383         struct rlb_client_info *rx_hash_table;
 384         u32 index, next_index;
 385
 386         /* clear slave from rx_hashtbl */
 387         spin_lock_bh(&bond->mode_lock);
 388
 389         rx_hash_table = bond_info->rx_hashtbl;
 390         index = bond_info->rx_hashtbl_used_head;
 391         for (; index != RLB_NULL_INDEX; index = next_index) {
 392                 next_index = rx_hash_table[index].used_next;
 393                 if (rx_hash_table[index].slave == slave) {
 394                         struct slave *assigned_slave = rlb_next_rx_slave(bond);
 395
 396                         if (assigned_slave) {
 397                                 rx_hash_table[index].slave = assigned_slave;
 398                                 if (is_valid_ether_addr(rx_hash_table[index].mac_dst)) {
 399                                         bond_info->rx_hashtbl[index].ntt = 1;
 400                                         bond_info->rx_ntt = 1;
 401                                         /* A slave has been removed from the
 402                                          * table because it is either disabled
 403                                          * or being released. We must retry the
 404                                          * update to avoid clients from not
 405                                          * being updated & disconnecting when
 406                                          * there is stress
 407                                          */
 408                                         bond_info->rlb_update_retry_counter =
 409                                                 RLB_UPDATE_RETRY;
 410                                 }
 411                         } else {  /* there is no active slave */
 412                                 rx_hash_table[index].slave = NULL;
 413                         }
 414                 }
 415         }
 416
 417         spin_unlock_bh(&bond->mode_lock);
 418
 419         if (slave != rtnl_dereference(bond->curr_active_slave))
 420                 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 421 }
 422
 423 static void rlb_update_client(struct rlb_client_info *client_info)
 424 {
 425         int i;
 426
 427         if (!client_info->slave || !is_valid_ether_addr(client_info->mac_dst))
 428                 return;
 429
 430         for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 431                 struct sk_buff *skb;
 432
 433                 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 434                                  client_info->ip_dst,
 435                                  client_info->slave->dev,
 436                                  client_info->ip_src,
 437                                  client_info->mac_dst,
 438                                  client_info->slave->dev->dev_addr,
 439                                  client_info->mac_dst);
 440                 if (!skb) {
 441                         slave_err(client_info->slave->bond->dev,
 442                                   client_info->slave->dev,
 443                                   "failed to create an ARP packet\n");
 444                         continue;
 445                 }
 446
 447                 skb->dev = client_info->slave->dev;
 448
 449                 if (client_info->vlan_id) {
 450                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 451                                                client_info->vlan_id);
 452                 }
 453
 454                 arp_xmit(skb);
 455         }
 456 }
 457
 458 /* sends ARP REPLIES that update the clients that need updating */
 459 static void rlb_update_rx_clients(struct bonding *bond)
 460 {
 461         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 462         struct rlb_client_info *client_info;
 463         u32 hash_index;
 464
 465         spin_lock_bh(&bond->mode_lock);
 466
 467         hash_index = bond_info->rx_hashtbl_used_head;
 468         for (; hash_index != RLB_NULL_INDEX;
 469              hash_index = client_info->used_next) {
 470                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 471                 if (client_info->ntt) {
 472                         rlb_update_client(client_info);
 473                         if (bond_info->rlb_update_retry_counter == 0)
 474                                 client_info->ntt = 0;
 475                 }
 476         }
 477
 478         /* do not update the entries again until this counter is zero so that
 479          * not to confuse the clients.
 480          */
 481         bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 482
 483         spin_unlock_bh(&bond->mode_lock);
 484 }
 485
 486 /* The slave was assigned a new mac address - update the clients */
 487 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 488 {
 489         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 490         struct rlb_client_info *client_info;
 491         int ntt = 0;
 492         u32 hash_index;
 493
 494         spin_lock_bh(&bond->mode_lock);
 495
 496         hash_index = bond_info->rx_hashtbl_used_head;
 497         for (; hash_index != RLB_NULL_INDEX;
 498              hash_index = client_info->used_next) {
 499                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 500
 501                 if ((client_info->slave == slave) &&
 502                     is_valid_ether_addr(client_info->mac_dst)) {
 503                         client_info->ntt = 1;
 504                         ntt = 1;
 505                 }
 506         }
 507
 508         /* update the team's flag only after the whole iteration */
 509         if (ntt) {
 510                 bond_info->rx_ntt = 1;
 511                 /* fasten the change */
 512                 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 513         }
 514
 515         spin_unlock_bh(&bond->mode_lock);
 516 }
 517
 518 /* mark all clients using src_ip to be updated */
 519 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
 520 {
 521         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 522         struct rlb_client_info *client_info;
 523         u32 hash_index;
 524
 525         spin_lock(&bond->mode_lock);
 526
 527         hash_index = bond_info->rx_hashtbl_used_head;
 528         for (; hash_index != RLB_NULL_INDEX;
 529              hash_index = client_info->used_next) {
 530                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 531
 532                 if (!client_info->slave) {
 533                         netdev_err(bond->dev, "found a client with no channel in the client's hash table\n");
 534                         continue;
 535                 }
 536                 /* update all clients using this src_ip, that are not assigned
 537                  * to the team's address (curr_active_slave) and have a known
 538                  * unicast mac address.
 539                  */
 540                 if ((client_info->ip_src == src_ip) &&
 541                     !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
 542                                              bond->dev->dev_addr) &&
 543                     is_valid_ether_addr(client_info->mac_dst)) {
 544                         client_info->ntt = 1;
 545                         bond_info->rx_ntt = 1;
 546                 }
 547         }
 548
 549         spin_unlock(&bond->mode_lock);
 550 }
 551
 552 static struct slave *rlb_choose_channel(struct sk_buff *skb,
 553                                         struct bonding *bond,
 554                                         const struct arp_pkt *arp)
 555 {
 556         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 557         struct slave *assigned_slave, *curr_active_slave;
 558         struct rlb_client_info *client_info;
 559         u32 hash_index = 0;
 560
 561         spin_lock(&bond->mode_lock);
 562
 563         curr_active_slave = rcu_dereference(bond->curr_active_slave);
 564
 565         hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
 566         client_info = &(bond_info->rx_hashtbl[hash_index]);
 567
 568         if (client_info->assigned) {
 569                 if ((client_info->ip_src == arp->ip_src) &&
 570                     (client_info->ip_dst == arp->ip_dst)) {
 571                         /* the entry is already assigned to this client */
 572                         if (!is_broadcast_ether_addr(arp->mac_dst)) {
 573                                 /* update mac address from arp */
 574                                 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
 575                         }
 576                         ether_addr_copy(client_info->mac_src, arp->mac_src);
 577
 578                         assigned_slave = client_info->slave;
 579                         if (assigned_slave) {
 580                                 spin_unlock(&bond->mode_lock);
 581                                 return assigned_slave;
 582                         }
 583                 } else {
 584                         /* the entry is already assigned to some other client,
 585                          * move the old client to primary (curr_active_slave) so
 586                          * that the new client can be assigned to this entry.
 587                          */
 588                         if (curr_active_slave &&
 589                             client_info->slave != curr_active_slave) {
 590                                 client_info->slave = curr_active_slave;
 591                                 rlb_update_client(client_info);
 592                         }
 593                 }
 594         }
 595         /* assign a new slave */
 596         assigned_slave = __rlb_next_rx_slave(bond);
 597
 598         if (assigned_slave) {
 599                 if (!(client_info->assigned &&
 600                       client_info->ip_src == arp->ip_src)) {
 601                         /* ip_src is going to be updated,
 602                          * fix the src hash list
 603                          */
 604                         u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
 605                                                     sizeof(arp->ip_src));
 606                         rlb_src_unlink(bond, hash_index);
 607                         rlb_src_link(bond, hash_src, hash_index);
 608                 }
 609
 610                 client_info->ip_src = arp->ip_src;
 611                 client_info->ip_dst = arp->ip_dst;
 612                 /* arp->mac_dst is broadcast for arp requests.
 613                  * will be updated with clients actual unicast mac address
 614                  * upon receiving an arp reply.
 615                  */
 616                 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
 617                 ether_addr_copy(client_info->mac_src, arp->mac_src);
 618                 client_info->slave = assigned_slave;
 619
 620                 if (is_valid_ether_addr(client_info->mac_dst)) {
 621                         client_info->ntt = 1;
 622                         bond->alb_info.rx_ntt = 1;
 623                 } else {
 624                         client_info->ntt = 0;
 625                 }
 626
 627                 if (vlan_get_tag(skb, &client_info->vlan_id))
 628                         client_info->vlan_id = 0;
 629
 630                 if (!client_info->assigned) {
 631                         u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
 632
 633                         bond_info->rx_hashtbl_used_head = hash_index;
 634                         client_info->used_next = prev_tbl_head;
 635                         if (prev_tbl_head != RLB_NULL_INDEX) {
 636                                 bond_info->rx_hashtbl[prev_tbl_head].used_prev =
 637                                         hash_index;
 638                         }
 639                         client_info->assigned = 1;
 640                 }
 641         }
 642
 643         spin_unlock(&bond->mode_lock);
 644
 645         return assigned_slave;
 646 }
 647
 648 /* chooses (and returns) transmit channel for arp reply
 649  * does not choose channel for other arp types since they are
 650  * sent on the curr_active_slave
 651  */
 652 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 653 {
 654         struct slave *tx_slave = NULL;
 655         struct arp_pkt *arp;
 656
 657         if (!pskb_network_may_pull(skb, sizeof(*arp)))
 658                 return NULL;
 659         arp = (struct arp_pkt *)skb_network_header(skb);
 660
 661         /* Don't modify or load balance ARPs that do not originate locally
 662          * (e.g.,arrive via a bridge).
 663          */
 664         if (!bond_slave_has_mac_rx(bond, arp->mac_src))
 665                 return NULL;
 666
 667         if (arp->op_code == htons(ARPOP_REPLY)) {
 668                 /* the arp must be sent on the selected rx channel */
 669                 tx_slave = rlb_choose_channel(skb, bond, arp);
 670                 if (tx_slave)
 671                         bond_hw_addr_copy(arp->mac_src, tx_slave->dev->dev_addr,
 672                                           tx_slave->dev->addr_len);
 673                 netdev_dbg(bond->dev, "(slave %s): Server sent ARP Reply packet\n",
 674                            tx_slave ? tx_slave->dev->name : "NULL");
 675         } else if (arp->op_code == htons(ARPOP_REQUEST)) {
 676                 /* Create an entry in the rx_hashtbl for this client as a
 677                  * place holder.
 678                  * When the arp reply is received the entry will be updated
 679                  * with the correct unicast address of the client.
 680                  */
 681                 tx_slave = rlb_choose_channel(skb, bond, arp);
 682
 683                 /* The ARP reply packets must be delayed so that
 684                  * they can cancel out the influence of the ARP request.
 685                  */
 686                 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 687
 688                 /* arp requests are broadcast and are sent on the primary
 689                  * the arp request will collapse all clients on the subnet to
 690                  * the primary slave. We must register these clients to be
 691                  * updated with their assigned mac.
 692                  */
 693                 rlb_req_update_subnet_clients(bond, arp->ip_src);
 694                 netdev_dbg(bond->dev, "(slave %s): Server sent ARP Request packet\n",
 695                            tx_slave ? tx_slave->dev->name : "NULL");
 696         }
 697
 698         return tx_slave;
 699 }
 700
 701 static void rlb_rebalance(struct bonding *bond)
 702 {
 703         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 704         struct slave *assigned_slave;
 705         struct rlb_client_info *client_info;
 706         int ntt;
 707         u32 hash_index;
 708
 709         spin_lock_bh(&bond->mode_lock);
 710
 711         ntt = 0;
 712         hash_index = bond_info->rx_hashtbl_used_head;
 713         for (; hash_index != RLB_NULL_INDEX;
 714              hash_index = client_info->used_next) {
 715                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 716                 assigned_slave = __rlb_next_rx_slave(bond);
 717                 if (assigned_slave && (client_info->slave != assigned_slave)) {
 718                         client_info->slave = assigned_slave;
 719                         if (!is_zero_ether_addr(client_info->mac_dst)) {
 720                                 client_info->ntt = 1;
 721                                 ntt = 1;
 722                         }
 723                 }
 724         }
 725
 726         /* update the team's flag only after the whole iteration */
 727         if (ntt)
 728                 bond_info->rx_ntt = 1;
 729         spin_unlock_bh(&bond->mode_lock);
 730 }
 731
 732 /* Caller must hold mode_lock */
 733 static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
 734 {
 735         entry->used_next = RLB_NULL_INDEX;
 736         entry->used_prev = RLB_NULL_INDEX;
 737         entry->assigned = 0;
 738         entry->slave = NULL;
 739         entry->vlan_id = 0;
 740 }
 741 static void rlb_init_table_entry_src(struct rlb_client_info *entry)
 742 {
 743         entry->src_first = RLB_NULL_INDEX;
 744         entry->src_prev = RLB_NULL_INDEX;
 745         entry->src_next = RLB_NULL_INDEX;
 746 }
 747
 748 static void rlb_init_table_entry(struct rlb_client_info *entry)
 749 {
 750         memset(entry, 0, sizeof(struct rlb_client_info));
 751         rlb_init_table_entry_dst(entry);
 752         rlb_init_table_entry_src(entry);
 753 }
 754
 755 static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
 756 {
 757         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 758         u32 next_index = bond_info->rx_hashtbl[index].used_next;
 759         u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
 760
 761         if (index == bond_info->rx_hashtbl_used_head)
 762                 bond_info->rx_hashtbl_used_head = next_index;
 763         if (prev_index != RLB_NULL_INDEX)
 764                 bond_info->rx_hashtbl[prev_index].used_next = next_index;
 765         if (next_index != RLB_NULL_INDEX)
 766                 bond_info->rx_hashtbl[next_index].used_prev = prev_index;
 767 }
 768
 769 /* unlink a rlb hash table entry from the src list */
 770 static void rlb_src_unlink(struct bonding *bond, u32 index)
 771 {
 772         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 773         u32 next_index = bond_info->rx_hashtbl[index].src_next;
 774         u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
 775
 776         bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
 777         bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
 778
 779         if (next_index != RLB_NULL_INDEX)
 780                 bond_info->rx_hashtbl[next_index].src_prev = prev_index;
 781
 782         if (prev_index == RLB_NULL_INDEX)
 783                 return;
 784
 785         /* is prev_index pointing to the head of this list? */
 786         if (bond_info->rx_hashtbl[prev_index].src_first == index)
 787                 bond_info->rx_hashtbl[prev_index].src_first = next_index;
 788         else
 789                 bond_info->rx_hashtbl[prev_index].src_next = next_index;
 790
 791 }
 792
 793 static void rlb_delete_table_entry(struct bonding *bond, u32 index)
 794 {
 795         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 796         struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 797
 798         rlb_delete_table_entry_dst(bond, index);
 799         rlb_init_table_entry_dst(entry);
 800
 801         rlb_src_unlink(bond, index);
 802 }
 803
 804 /* add the rx_hashtbl[ip_dst_hash] entry to the list
 805  * of entries with identical ip_src_hash
 806  */
 807 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
 808 {
 809         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 810         u32 next;
 811
 812         bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
 813         next = bond_info->rx_hashtbl[ip_src_hash].src_first;
 814         bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
 815         if (next != RLB_NULL_INDEX)
 816                 bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
 817         bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
 818 }
 819
 820 /* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does
 821  * not match arp->mac_src
 822  */
 823 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
 824 {
 825         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 826         u32 ip_src_hash = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
 827         u32 index;
 828
 829         spin_lock_bh(&bond->mode_lock);
 830
 831         index = bond_info->rx_hashtbl[ip_src_hash].src_first;
 832         while (index != RLB_NULL_INDEX) {
 833                 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 834                 u32 next_index = entry->src_next;
 835
 836                 if (entry->ip_src == arp->ip_src &&
 837                     !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
 838                         rlb_delete_table_entry(bond, index);
 839                 index = next_index;
 840         }
 841         spin_unlock_bh(&bond->mode_lock);
 842 }
 843
 844 static int rlb_initialize(struct bonding *bond)
 845 {
 846         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 847         struct rlb_client_info  *new_hashtbl;
 848         int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 849         int i;
 850
 851         new_hashtbl = kmalloc(size, GFP_KERNEL);
 852         if (!new_hashtbl)
 853                 return -1;
 854
 855         spin_lock_bh(&bond->mode_lock);
 856
 857         bond_info->rx_hashtbl = new_hashtbl;
 858
 859         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 860
 861         for (i = 0; i < RLB_HASH_TABLE_SIZE; i++)
 862                 rlb_init_table_entry(bond_info->rx_hashtbl + i);
 863
 864         spin_unlock_bh(&bond->mode_lock);
 865
 866         /* register to receive ARPs */
 867         bond->recv_probe = rlb_arp_recv;
 868
 869         return 0;
 870 }
 871
 872 static void rlb_deinitialize(struct bonding *bond)
 873 {
 874         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 875
 876         spin_lock_bh(&bond->mode_lock);
 877
 878         kfree(bond_info->rx_hashtbl);
 879         bond_info->rx_hashtbl = NULL;
 880         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 881
 882         spin_unlock_bh(&bond->mode_lock);
 883 }
 884
 885 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 886 {
 887         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 888         u32 curr_index;
 889
 890         spin_lock_bh(&bond->mode_lock);
 891
 892         curr_index = bond_info->rx_hashtbl_used_head;
 893         while (curr_index != RLB_NULL_INDEX) {
 894                 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 895                 u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
 896
 897                 if (curr->vlan_id == vlan_id)
 898                         rlb_delete_table_entry(bond, curr_index);
 899
 900                 curr_index = next_index;
 901         }
 902
 903         spin_unlock_bh(&bond->mode_lock);
 904 }
 905
 906 /*********************** tlb/rlb shared functions *********************/
 907
 908 static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
 909                             __be16 vlan_proto, u16 vid)
 910 {
 911         struct learning_pkt pkt;
 912         struct sk_buff *skb;
 913         int size = sizeof(struct learning_pkt);
 914
 915         memset(&pkt, 0, size);
 916         ether_addr_copy(pkt.mac_dst, mac_addr);
 917         ether_addr_copy(pkt.mac_src, mac_addr);
 918         pkt.type = cpu_to_be16(ETH_P_LOOPBACK);
 919
 920         skb = dev_alloc_skb(size);
 921         if (!skb)
 922                 return;
 923
 924         skb_put_data(skb, &pkt, size);
 925
 926         skb_reset_mac_header(skb);
 927         skb->network_header = skb->mac_header + ETH_HLEN;
 928         skb->protocol = pkt.type;
 929         skb->priority = TC_PRIO_CONTROL;
 930         skb->dev = slave->dev;
 931
 932         slave_dbg(slave->bond->dev, slave->dev,
 933                   "Send learning packet: mac %pM vlan %d\n", mac_addr, vid);
 934
 935         if (vid)
 936                 __vlan_hwaccel_put_tag(skb, vlan_proto, vid);
 937
 938         dev_queue_xmit(skb);
 939 }
 940
 941 struct alb_walk_data {
 942         struct bonding *bond;
 943         struct slave *slave;
 944         u8 *mac_addr;
 945         bool strict_match;
 946 };
 947
 948 static int alb_upper_dev_walk(struct net_device *upper,
 949                               struct netdev_nested_priv *priv)
 950 {
 951         struct alb_walk_data *data = (struct alb_walk_data *)priv->data;
 952         bool strict_match = data->strict_match;
 953         struct bonding *bond = data->bond;
 954         struct slave *slave = data->slave;
 955         u8 *mac_addr = data->mac_addr;
 956         struct bond_vlan_tag *tags;
 957
 958         if (is_vlan_dev(upper) &&
 959             bond->dev->lower_level == upper->lower_level - 1) {
 960                 if (upper->addr_assign_type == NET_ADDR_STOLEN) {
 961                         alb_send_lp_vid(slave, mac_addr,
 962                                         vlan_dev_vlan_proto(upper),
 963                                         vlan_dev_vlan_id(upper));
 964                 } else {
 965                         alb_send_lp_vid(slave, upper->dev_addr,
 966                                         vlan_dev_vlan_proto(upper),
 967                                         vlan_dev_vlan_id(upper));
 968                 }
 969         }
 970
 971         /* If this is a macvlan device, then only send updates
 972          * when strict_match is turned off.
 973          */
 974         if (netif_is_macvlan(upper) && !strict_match) {
 975                 tags = bond_verify_device_path(bond->dev, upper, 0);
 976                 if (IS_ERR_OR_NULL(tags))
 977                         BUG();
 978                 alb_send_lp_vid(slave, upper->dev_addr,
 979                                 tags[0].vlan_proto, tags[0].vlan_id);
 980                 kfree(tags);
 981         }
 982
 983         return 0;
 984 }
 985
 986 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
 987                                       bool strict_match)
 988 {
 989         struct bonding *bond = bond_get_bond_by_slave(slave);
 990         struct netdev_nested_priv priv;
 991         struct alb_walk_data data = {
 992                 .strict_match = strict_match,
 993                 .mac_addr = mac_addr,
 994                 .slave = slave,
 995                 .bond = bond,
 996         };
 997
 998         priv.data = (void *)&data;
 999         /* send untagged */
1000         alb_send_lp_vid(slave, mac_addr, 0, 0);
1001
1002         /* loop through all devices and see if we need to send a packet
1003          * for that device.
1004          */
1005         rcu_read_lock();
1006         netdev_walk_all_upper_dev_rcu(bond->dev, alb_upper_dev_walk, &priv);
1007         rcu_read_unlock();
1008 }
1009
1010 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[],
1011                                   unsigned int len)
1012 {
1013         struct net_device *dev = slave->dev;
1014         struct sockaddr_storage ss;
1015
1016         if (BOND_MODE(slave->bond) == BOND_MODE_TLB) {
1017                 memcpy(dev->dev_addr, addr, len);
1018                 return 0;
1019         }
1020
1021         /* for rlb each slave must have a unique hw mac addresses so that
1022          * each slave will receive packets destined to a different mac
1023          */
1024         memcpy(ss.__data, addr, len);
1025         ss.ss_family = dev->type;
1026         if (dev_set_mac_address(dev, (struct sockaddr *)&ss, NULL)) {
1027                 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");
1028                 return -EOPNOTSUPP;
1029         }
1030         return 0;
1031 }
1032
1033 /* Swap MAC addresses between two slaves.
1034  *
1035  * Called with RTNL held, and no other locks.
1036  */
1037 static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1038 {
1039         u8 tmp_mac_addr[MAX_ADDR_LEN];
1040
1041         bond_hw_addr_copy(tmp_mac_addr, slave1->dev->dev_addr,
1042                           slave1->dev->addr_len);
1043         alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr,
1044                                slave2->dev->addr_len);
1045         alb_set_slave_mac_addr(slave2, tmp_mac_addr,
1046                                slave1->dev->addr_len);
1047
1048 }
1049
1050 /* Send learning packets after MAC address swap.
1051  *
1052  * Called with RTNL and no other locks
1053  */
1054 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1055                                 struct slave *slave2)
1056 {
1057         int slaves_state_differ = (bond_slave_can_tx(slave1) != bond_slave_can_tx(slave2));
1058         struct slave *disabled_slave = NULL;
1059
1060         ASSERT_RTNL();
1061
1062         /* fasten the change in the switch */
1063         if (bond_slave_can_tx(slave1)) {
1064                 alb_send_learning_packets(slave1, slave1->dev->dev_addr, false);
1065                 if (bond->alb_info.rlb_enabled) {
1066                         /* inform the clients that the mac address
1067                          * has changed
1068                          */
1069                         rlb_req_update_slave_clients(bond, slave1);
1070                 }
1071         } else {
1072                 disabled_slave = slave1;
1073         }
1074
1075         if (bond_slave_can_tx(slave2)) {
1076                 alb_send_learning_packets(slave2, slave2->dev->dev_addr, false);
1077                 if (bond->alb_info.rlb_enabled) {
1078                         /* inform the clients that the mac address
1079                          * has changed
1080                          */
1081                         rlb_req_update_slave_clients(bond, slave2);
1082                 }
1083         } else {
1084                 disabled_slave = slave2;
1085         }
1086
1087         if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1088                 /* A disabled slave was assigned an active mac addr */
1089                 rlb_teach_disabled_mac_on_primary(bond,
1090                                                   disabled_slave->dev->dev_addr);
1091         }
1092 }
1093
1094 /**
1095  * alb_change_hw_addr_on_detach
1096  * @bond: bonding we're working on
1097  * @slave: the slave that was just detached
1098  *
1099  * We assume that @slave was already detached from the slave list.
1100  *
1101  * If @slave's permanent hw address is different both from its current
1102  * address and from @bond's address, then somewhere in the bond there's
1103  * a slave that has @slave's permanet address as its current address.
1104  * We'll make sure that slave no longer uses @slave's permanent address.
1105  *
1106  * Caller must hold RTNL and no other locks
1107  */
1108 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1109 {
1110         int perm_curr_diff;
1111         int perm_bond_diff;
1112         struct slave *found_slave;
1113
1114         perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1115                                                   slave->dev->dev_addr);
1116         perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1117                                                   bond->dev->dev_addr);
1118
1119         if (perm_curr_diff && perm_bond_diff) {
1120                 found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1121
1122                 if (found_slave) {
1123                         alb_swap_mac_addr(slave, found_slave);
1124                         alb_fasten_mac_swap(bond, slave, found_slave);
1125                 }
1126         }
1127 }
1128
1129 /**
1130  * alb_handle_addr_collision_on_attach
1131  * @bond: bonding we're working on
1132  * @slave: the slave that was just attached
1133  *
1134  * checks uniqueness of slave's mac address and handles the case the
1135  * new slave uses the bonds mac address.
1136  *
1137  * If the permanent hw address of @slave is @bond's hw address, we need to
1138  * find a different hw address to give @slave, that isn't in use by any other
1139  * slave in the bond. This address must be, of course, one of the permanent
1140  * addresses of the other slaves.
1141  *
1142  * We go over the slave list, and for each slave there we compare its
1143  * permanent hw address with the current address of all the other slaves.
1144  * If no match was found, then we've found a slave with a permanent address
1145  * that isn't used by any other slave in the bond, so we can assign it to
1146  * @slave.
1147  *
1148  * assumption: this function is called before @slave is attached to the
1149  *             bond slave list.
1150  */
1151 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1152 {
1153         struct slave *has_bond_addr = rcu_access_pointer(bond->curr_active_slave);
1154         struct slave *tmp_slave1, *free_mac_slave = NULL;
1155         struct list_head *iter;
1156
1157         if (!bond_has_slaves(bond)) {
1158                 /* this is the first slave */
1159                 return 0;
1160         }
1161
1162         /* if slave's mac address differs from bond's mac address
1163          * check uniqueness of slave's mac address against the other
1164          * slaves in the bond.
1165          */
1166         if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1167                 if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1168                         return 0;
1169
1170                 /* Try setting slave mac to bond address and fall-through
1171                  * to code handling that situation below...
1172                  */
1173                 alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1174                                        bond->dev->addr_len);
1175         }
1176
1177         /* The slave's address is equal to the address of the bond.
1178          * Search for a spare address in the bond for this slave.
1179          */
1180         bond_for_each_slave(bond, tmp_slave1, iter) {
1181                 if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1182                         /* no slave has tmp_slave1's perm addr
1183                          * as its curr addr
1184                          */
1185                         free_mac_slave = tmp_slave1;
1186                         break;
1187                 }
1188
1189                 if (!has_bond_addr) {
1190                         if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1191                                                     bond->dev->dev_addr)) {
1192
1193                                 has_bond_addr = tmp_slave1;
1194                         }
1195                 }
1196         }
1197
1198         if (free_mac_slave) {
1199                 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1200                                        free_mac_slave->dev->addr_len);
1201
1202                 slave_warn(bond->dev, slave->dev, "the slave hw address is in use by the bond; giving it the hw address of %s\n",
1203                            free_mac_slave->dev->name);
1204
1205         } else if (has_bond_addr) {
1206                 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");
1207                 return -EFAULT;
1208         }
1209
1210         return 0;
1211 }
1212
1213 /**
1214  * alb_set_mac_address
1215  * @bond: bonding we're working on
1216  * @addr: MAC address to set
1217  *
1218  * In TLB mode all slaves are configured to the bond's hw address, but set
1219  * their dev_addr field to different addresses (based on their permanent hw
1220  * addresses).
1221  *
1222  * For each slave, this function sets the interface to the new address and then
1223  * changes its dev_addr field to its previous value.
1224  *
1225  * Unwinding assumes bond's mac address has not yet changed.
1226  */
1227 static int alb_set_mac_address(struct bonding *bond, void *addr)
1228 {
1229         struct slave *slave, *rollback_slave;
1230         struct list_head *iter;
1231         struct sockaddr_storage ss;
1232         char tmp_addr[MAX_ADDR_LEN];
1233         int res;
1234
1235         if (bond->alb_info.rlb_enabled)
1236                 return 0;
1237
1238         bond_for_each_slave(bond, slave, iter) {
1239                 /* save net_device's current hw address */
1240                 bond_hw_addr_copy(tmp_addr, slave->dev->dev_addr,
1241                                   slave->dev->addr_len);
1242
1243                 res = dev_set_mac_address(slave->dev, addr, NULL);
1244
1245                 /* restore net_device's hw address */
1246                 bond_hw_addr_copy(slave->dev->dev_addr, tmp_addr,
1247                                   slave->dev->addr_len);
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                 bond_hw_addr_copy(rollback_slave->dev->dev_addr, tmp_addr,
1268                                   rollback_slave->dev->addr_len);
1269         }
1270
1271         return res;
1272 }
1273
1274 /************************ exported alb functions ************************/
1275
1276 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1277 {
1278         int res;
1279
1280         res = tlb_initialize(bond);
1281         if (res)
1282                 return res;
1283
1284         if (rlb_enabled) {
1285                 bond->alb_info.rlb_enabled = 1;
1286                 res = rlb_initialize(bond);
1287                 if (res) {
1288                         tlb_deinitialize(bond);
1289                         return res;
1290                 }
1291         } else {
1292                 bond->alb_info.rlb_enabled = 0;
1293         }
1294
1295         return 0;
1296 }
1297
1298 void bond_alb_deinitialize(struct bonding *bond)
1299 {
1300         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1301
1302         tlb_deinitialize(bond);
1303
1304         if (bond_info->rlb_enabled)
1305                 rlb_deinitialize(bond);
1306 }
1307
1308 static netdev_tx_t bond_do_alb_xmit(struct sk_buff *skb, struct bonding *bond,
1309                                     struct slave *tx_slave)
1310 {
1311         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1312         struct ethhdr *eth_data = eth_hdr(skb);
1313
1314         if (!tx_slave) {
1315                 /* unbalanced or unassigned, send through primary */
1316                 tx_slave = rcu_dereference(bond->curr_active_slave);
1317                 if (bond->params.tlb_dynamic_lb)
1318                         bond_info->unbalanced_load += skb->len;
1319         }
1320
1321         if (tx_slave && bond_slave_can_tx(tx_slave)) {
1322                 if (tx_slave != rcu_access_pointer(bond->curr_active_slave)) {
1323                         ether_addr_copy(eth_data->h_source,
1324                                         tx_slave->dev->dev_addr);
1325                 }
1326
1327                 return bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1328         }
1329
1330         if (tx_slave && bond->params.tlb_dynamic_lb) {
1331                 spin_lock(&bond->mode_lock);
1332                 __tlb_clear_slave(bond, tx_slave, 0);
1333                 spin_unlock(&bond->mode_lock);
1334         }
1335
1336         /* no suitable interface, frame not sent */
1337         return bond_tx_drop(bond->dev, skb);
1338 }
1339
1340 struct slave *bond_xmit_tlb_slave_get(struct bonding *bond,
1341                                       struct sk_buff *skb)
1342 {
1343         struct slave *tx_slave = NULL;
1344         struct ethhdr *eth_data;
1345         u32 hash_index;
1346
1347         skb_reset_mac_header(skb);
1348         eth_data = eth_hdr(skb);
1349
1350         /* Do not TX balance any multicast or broadcast */
1351         if (!is_multicast_ether_addr(eth_data->h_dest)) {
1352                 switch (skb->protocol) {
1353                 case htons(ETH_P_IP):
1354                 case htons(ETH_P_IPX):
1355                     /* In case of IPX, it will falback to L2 hash */
1356                 case htons(ETH_P_IPV6):
1357                         hash_index = bond_xmit_hash(bond, skb);
1358                         if (bond->params.tlb_dynamic_lb) {
1359                                 tx_slave = tlb_choose_channel(bond,
1360                                                               hash_index & 0xFF,
1361                                                               skb->len);
1362                         } else {
1363                                 struct bond_up_slave *slaves;
1364                                 unsigned int count;
1365
1366                                 slaves = rcu_dereference(bond->usable_slaves);
1367                                 count = slaves ? READ_ONCE(slaves->count) : 0;
1368                                 if (likely(count))
1369                                         tx_slave = slaves->arr[hash_index %
1370                                                                count];
1371                         }
1372                         break;
1373                 }
1374         }
1375         return tx_slave;
1376 }
1377
1378 netdev_tx_t bond_tlb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1379 {
1380         struct bonding *bond = netdev_priv(bond_dev);
1381         struct slave *tx_slave;
1382
1383         tx_slave = bond_xmit_tlb_slave_get(bond, skb);
1384         return bond_do_alb_xmit(skb, bond, tx_slave);
1385 }
1386
1387 struct slave *bond_xmit_alb_slave_get(struct bonding *bond,
1388                                       struct sk_buff *skb)
1389 {
1390         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1391         static const __be32 ip_bcast = htonl(0xffffffff);
1392         struct slave *tx_slave = NULL;
1393         const u8 *hash_start = NULL;
1394         bool do_tx_balance = true;
1395         struct ethhdr *eth_data;
1396         u32 hash_index = 0;
1397         int hash_size = 0;
1398
1399         skb_reset_mac_header(skb);
1400         eth_data = eth_hdr(skb);
1401
1402         switch (ntohs(skb->protocol)) {
1403         case ETH_P_IP: {
1404                 const struct iphdr *iph;
1405
1406                 if (is_broadcast_ether_addr(eth_data->h_dest) ||
1407                     !pskb_network_may_pull(skb, sizeof(*iph))) {
1408                         do_tx_balance = false;
1409                         break;
1410                 }
1411                 iph = ip_hdr(skb);
1412                 if (iph->daddr == ip_bcast || iph->protocol == IPPROTO_IGMP) {
1413                         do_tx_balance = false;
1414                         break;
1415                 }
1416                 hash_start = (char *)&(iph->daddr);
1417                 hash_size = sizeof(iph->daddr);
1418                 break;
1419         }
1420         case ETH_P_IPV6: {
1421                 const struct ipv6hdr *ip6hdr;
1422
1423                 /* IPv6 doesn't really use broadcast mac address, but leave
1424                  * that here just in case.
1425                  */
1426                 if (is_broadcast_ether_addr(eth_data->h_dest)) {
1427                         do_tx_balance = false;
1428                         break;
1429                 }
1430
1431                 /* IPv6 uses all-nodes multicast as an equivalent to
1432                  * broadcasts in IPv4.
1433                  */
1434                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1435                         do_tx_balance = false;
1436                         break;
1437                 }
1438
1439                 if (!pskb_network_may_pull(skb, sizeof(*ip6hdr))) {
1440                         do_tx_balance = false;
1441                         break;
1442                 }
1443                 /* Additionally, DAD probes should not be tx-balanced as that
1444                  * will lead to false positives for duplicate addresses and
1445                  * prevent address configuration from working.
1446                  */
1447                 ip6hdr = ipv6_hdr(skb);
1448                 if (ipv6_addr_any(&ip6hdr->saddr)) {
1449                         do_tx_balance = false;
1450                         break;
1451                 }
1452
1453                 hash_start = (char *)&ip6hdr->daddr;
1454                 hash_size = sizeof(ip6hdr->daddr);
1455                 break;
1456         }
1457         case ETH_P_IPX: {
1458                 const struct ipxhdr *ipxhdr;
1459
1460                 if (pskb_network_may_pull(skb, sizeof(*ipxhdr))) {
1461                         do_tx_balance = false;
1462                         break;
1463                 }
1464                 ipxhdr = (struct ipxhdr *)skb_network_header(skb);
1465
1466                 if (ipxhdr->ipx_checksum != IPX_NO_CHECKSUM) {
1467                         /* something is wrong with this packet */
1468                         do_tx_balance = false;
1469                         break;
1470                 }
1471
1472                 if (ipxhdr->ipx_type != IPX_TYPE_NCP) {
1473                         /* The only protocol worth balancing in
1474                          * this family since it has an "ARP" like
1475                          * mechanism
1476                          */
1477                         do_tx_balance = false;
1478                         break;
1479                 }
1480
1481                 eth_data = eth_hdr(skb);
1482                 hash_start = (char *)eth_data->h_dest;
1483                 hash_size = ETH_ALEN;
1484                 break;
1485         }
1486         case ETH_P_ARP:
1487                 do_tx_balance = false;
1488                 if (bond_info->rlb_enabled)
1489                         tx_slave = rlb_arp_xmit(skb, bond);
1490                 break;
1491         default:
1492                 do_tx_balance = false;
1493                 break;
1494         }
1495
1496         if (do_tx_balance) {
1497                 if (bond->params.tlb_dynamic_lb) {
1498                         hash_index = _simple_hash(hash_start, hash_size);
1499                         tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1500                 } else {
1501                         /*
1502                          * do_tx_balance means we are free to select the tx_slave
1503                          * So we do exactly what tlb would do for hash selection
1504                          */
1505
1506                         struct bond_up_slave *slaves;
1507                         unsigned int count;
1508
1509                         slaves = rcu_dereference(bond->usable_slaves);
1510                         count = slaves ? READ_ONCE(slaves->count) : 0;
1511                         if (likely(count))
1512                                 tx_slave = slaves->arr[bond_xmit_hash(bond, skb) %
1513                                                        count];
1514                 }
1515         }
1516         return tx_slave;
1517 }
1518
1519 netdev_tx_t bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1520 {
1521         struct bonding *bond = netdev_priv(bond_dev);
1522         struct slave *tx_slave = NULL;
1523
1524         tx_slave = bond_xmit_alb_slave_get(bond, skb);
1525         return bond_do_alb_xmit(skb, bond, tx_slave);
1526 }
1527
1528 void bond_alb_monitor(struct work_struct *work)
1529 {
1530         struct bonding *bond = container_of(work, struct bonding,
1531                                             alb_work.work);
1532         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1533         struct list_head *iter;
1534         struct slave *slave;
1535
1536         if (!bond_has_slaves(bond)) {
1537                 bond_info->tx_rebalance_counter = 0;
1538                 bond_info->lp_counter = 0;
1539                 goto re_arm;
1540         }
1541
1542         rcu_read_lock();
1543
1544         bond_info->tx_rebalance_counter++;
1545         bond_info->lp_counter++;
1546
1547         /* send learning packets */
1548         if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1549                 bool strict_match;
1550
1551                 bond_for_each_slave_rcu(bond, slave, iter) {
1552                         /* If updating current_active, use all currently
1553                          * user mac addresses (!strict_match).  Otherwise, only
1554                          * use mac of the slave device.
1555                          * In RLB mode, we always use strict matches.
1556                          */
1557                         strict_match = (slave != rcu_access_pointer(bond->curr_active_slave) ||
1558                                         bond_info->rlb_enabled);
1559                         alb_send_learning_packets(slave, slave->dev->dev_addr,
1560                                                   strict_match);
1561                 }
1562                 bond_info->lp_counter = 0;
1563         }
1564
1565         /* rebalance tx traffic */
1566         if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1567                 bond_for_each_slave_rcu(bond, slave, iter) {
1568                         tlb_clear_slave(bond, slave, 1);
1569                         if (slave == rcu_access_pointer(bond->curr_active_slave)) {
1570                                 SLAVE_TLB_INFO(slave).load =
1571                                         bond_info->unbalanced_load /
1572                                                 BOND_TLB_REBALANCE_INTERVAL;
1573                                 bond_info->unbalanced_load = 0;
1574                         }
1575                 }
1576                 bond_info->tx_rebalance_counter = 0;
1577         }
1578
1579         if (bond_info->rlb_enabled) {
1580                 if (bond_info->primary_is_promisc &&
1581                     (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1582
1583                         /* dev_set_promiscuity requires rtnl and
1584                          * nothing else.  Avoid race with bond_close.
1585                          */
1586                         rcu_read_unlock();
1587                         if (!rtnl_trylock())
1588                                 goto re_arm;
1589
1590                         bond_info->rlb_promisc_timeout_counter = 0;
1591
1592                         /* If the primary was set to promiscuous mode
1593                          * because a slave was disabled then
1594                          * it can now leave promiscuous mode.
1595                          */
1596                         dev_set_promiscuity(rtnl_dereference(bond->curr_active_slave)->dev,
1597                                             -1);
1598                         bond_info->primary_is_promisc = 0;
1599
1600                         rtnl_unlock();
1601                         rcu_read_lock();
1602                 }
1603
1604                 if (bond_info->rlb_rebalance) {
1605                         bond_info->rlb_rebalance = 0;
1606                         rlb_rebalance(bond);
1607                 }
1608
1609                 /* check if clients need updating */
1610                 if (bond_info->rx_ntt) {
1611                         if (bond_info->rlb_update_delay_counter) {
1612                                 --bond_info->rlb_update_delay_counter;
1613                         } else {
1614                                 rlb_update_rx_clients(bond);
1615                                 if (bond_info->rlb_update_retry_counter)
1616                                         --bond_info->rlb_update_retry_counter;
1617                                 else
1618                                         bond_info->rx_ntt = 0;
1619                         }
1620                 }
1621         }
1622         rcu_read_unlock();
1623 re_arm:
1624         queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1625 }
1626
1627 /* assumption: called before the slave is attached to the bond
1628  * and not locked by the bond lock
1629  */
1630 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1631 {
1632         int res;
1633
1634         res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1635                                      slave->dev->addr_len);
1636         if (res)
1637                 return res;
1638
1639         res = alb_handle_addr_collision_on_attach(bond, slave);
1640         if (res)
1641                 return res;
1642
1643         tlb_init_slave(slave);
1644
1645         /* order a rebalance ASAP */
1646         bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1647
1648         if (bond->alb_info.rlb_enabled)
1649                 bond->alb_info.rlb_rebalance = 1;
1650
1651         return 0;
1652 }
1653
1654 /* Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1655  * if necessary.
1656  *
1657  * Caller must hold RTNL and no other locks
1658  */
1659 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1660 {
1661         if (bond_has_slaves(bond))
1662                 alb_change_hw_addr_on_detach(bond, slave);
1663
1664         tlb_clear_slave(bond, slave, 0);
1665
1666         if (bond->alb_info.rlb_enabled) {
1667                 bond->alb_info.rx_slave = NULL;
1668                 rlb_clear_slave(bond, slave);
1669         }
1670
1671 }
1672
1673 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1674 {
1675         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1676
1677         if (link == BOND_LINK_DOWN) {
1678                 tlb_clear_slave(bond, slave, 0);
1679                 if (bond->alb_info.rlb_enabled)
1680                         rlb_clear_slave(bond, slave);
1681         } else if (link == BOND_LINK_UP) {
1682                 /* order a rebalance ASAP */
1683                 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1684                 if (bond->alb_info.rlb_enabled) {
1685                         bond->alb_info.rlb_rebalance = 1;
1686                         /* If the updelay module parameter is smaller than the
1687                          * forwarding delay of the switch the rebalance will
1688                          * not work because the rebalance arp replies will
1689                          * not be forwarded to the clients..
1690                          */
1691                 }
1692         }
1693
1694         if (bond_is_nondyn_tlb(bond)) {
1695                 if (bond_update_slave_arr(bond, NULL))
1696                         pr_err("Failed to build slave-array for TLB mode.\n");
1697         }
1698 }
1699
1700 /**
1701  * bond_alb_handle_active_change - assign new curr_active_slave
1702  * @bond: our bonding struct
1703  * @new_slave: new slave to assign
1704  *
1705  * Set the bond->curr_active_slave to @new_slave and handle
1706  * mac address swapping and promiscuity changes as needed.
1707  *
1708  * Caller must hold RTNL
1709  */
1710 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1711 {
1712         struct slave *swap_slave;
1713         struct slave *curr_active;
1714
1715         curr_active = rtnl_dereference(bond->curr_active_slave);
1716         if (curr_active == new_slave)
1717                 return;
1718
1719         if (curr_active && bond->alb_info.primary_is_promisc) {
1720                 dev_set_promiscuity(curr_active->dev, -1);
1721                 bond->alb_info.primary_is_promisc = 0;
1722                 bond->alb_info.rlb_promisc_timeout_counter = 0;
1723         }
1724
1725         swap_slave = curr_active;
1726         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1727
1728         if (!new_slave || !bond_has_slaves(bond))
1729                 return;
1730
1731         /* set the new curr_active_slave to the bonds mac address
1732          * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1733          */
1734         if (!swap_slave)
1735                 swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1736
1737         /* Arrange for swap_slave and new_slave to temporarily be
1738          * ignored so we can mess with their MAC addresses without
1739          * fear of interference from transmit activity.
1740          */
1741         if (swap_slave)
1742                 tlb_clear_slave(bond, swap_slave, 1);
1743         tlb_clear_slave(bond, new_slave, 1);
1744
1745         /* in TLB mode, the slave might flip down/up with the old dev_addr,
1746          * and thus filter bond->dev_addr's packets, so force bond's mac
1747          */
1748         if (BOND_MODE(bond) == BOND_MODE_TLB) {
1749                 struct sockaddr_storage ss;
1750                 u8 tmp_addr[MAX_ADDR_LEN];
1751
1752                 bond_hw_addr_copy(tmp_addr, new_slave->dev->dev_addr,
1753                                   new_slave->dev->addr_len);
1754
1755                 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
1756                                   bond->dev->addr_len);
1757                 ss.ss_family = bond->dev->type;
1758                 /* we don't care if it can't change its mac, best effort */
1759                 dev_set_mac_address(new_slave->dev, (struct sockaddr *)&ss,
1760                                     NULL);
1761
1762                 bond_hw_addr_copy(new_slave->dev->dev_addr, tmp_addr,
1763                                   new_slave->dev->addr_len);
1764         }
1765
1766         /* curr_active_slave must be set before calling alb_swap_mac_addr */
1767         if (swap_slave) {
1768                 /* swap mac address */
1769                 alb_swap_mac_addr(swap_slave, new_slave);
1770                 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1771         } else {
1772                 /* set the new_slave to the bond mac address */
1773                 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1774                                        bond->dev->addr_len);
1775                 alb_send_learning_packets(new_slave, bond->dev->dev_addr,
1776                                           false);
1777         }
1778 }
1779
1780 /* Called with RTNL */
1781 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1782 {
1783         struct bonding *bond = netdev_priv(bond_dev);
1784         struct sockaddr_storage *ss = addr;
1785         struct slave *curr_active;
1786         struct slave *swap_slave;
1787         int res;
1788
1789         if (!is_valid_ether_addr(ss->__data))
1790                 return -EADDRNOTAVAIL;
1791
1792         res = alb_set_mac_address(bond, addr);
1793         if (res)
1794                 return res;
1795
1796         bond_hw_addr_copy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
1797
1798         /* If there is no curr_active_slave there is nothing else to do.
1799          * Otherwise we'll need to pass the new address to it and handle
1800          * duplications.
1801          */
1802         curr_active = rtnl_dereference(bond->curr_active_slave);
1803         if (!curr_active)
1804                 return 0;
1805
1806         swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1807
1808         if (swap_slave) {
1809                 alb_swap_mac_addr(swap_slave, curr_active);
1810                 alb_fasten_mac_swap(bond, swap_slave, curr_active);
1811         } else {
1812                 alb_set_slave_mac_addr(curr_active, bond_dev->dev_addr,
1813                                        bond_dev->addr_len);
1814
1815                 alb_send_learning_packets(curr_active,
1816                                           bond_dev->dev_addr, false);
1817                 if (bond->alb_info.rlb_enabled) {
1818                         /* inform clients mac address has changed */
1819                         rlb_req_update_slave_clients(bond, curr_active);
1820                 }
1821         }
1822
1823         return 0;
1824 }
1825
1826 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1827 {
1828         if (bond->alb_info.rlb_enabled)
1829                 rlb_clear_vlan(bond, vlan_id);
1830 }
1831