drivers/net/bonding/bond_alb.c

   1 /*
   2  * Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms of the GNU General Public License as published by the
   6  * Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful, but
  10  * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  11  * or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License along
  15  * with this program; if not, see <http://www.gnu.org/licenses/>.
  16  *
  17  * The full GNU General Public License is included in this distribution in the
  18  * file called LICENSE.
  19  *
  20  */
  21
  22 #include <linux/skbuff.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/etherdevice.h>
  25 #include <linux/pkt_sched.h>
  26 #include <linux/spinlock.h>
  27 #include <linux/slab.h>
  28 #include <linux/timer.h>
  29 #include <linux/ip.h>
  30 #include <linux/ipv6.h>
  31 #include <linux/if_arp.h>
  32 #include <linux/if_ether.h>
  33 #include <linux/if_bonding.h>
  34 #include <linux/if_vlan.h>
  35 #include <linux/in.h>
  36 #include <net/ipx.h>
  37 #include <net/arp.h>
  38 #include <net/ipv6.h>
  39 #include <asm/byteorder.h>
  40 #include <net/bonding.h>
  41 #include <net/bond_alb.h>
  42
  43 static const u8 mac_v6_allmcast[ETH_ALEN + 2] __long_aligned = {
  44         0x33, 0x33, 0x00, 0x00, 0x00, 0x01
  45 };
  46 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
  47
  48 #pragma pack(1)
  49 struct learning_pkt {
  50         u8 mac_dst[ETH_ALEN];
  51         u8 mac_src[ETH_ALEN];
  52         __be16 type;
  53         u8 padding[ETH_ZLEN - ETH_HLEN];
  54 };
  55
  56 struct arp_pkt {
  57         __be16  hw_addr_space;
  58         __be16  prot_addr_space;
  59         u8      hw_addr_len;
  60         u8      prot_addr_len;
  61         __be16  op_code;
  62         u8      mac_src[ETH_ALEN];      /* sender hardware address */
  63         __be32  ip_src;                 /* sender IP address */
  64         u8      mac_dst[ETH_ALEN];      /* target hardware address */
  65         __be32  ip_dst;                 /* target IP address */
  66 };
  67 #pragma pack()
  68
  69 /* Forward declaration */
  70 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
  71                                       bool strict_match);
  72 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp);
  73 static void rlb_src_unlink(struct bonding *bond, u32 index);
  74 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash,
  75                          u32 ip_dst_hash);
  76
  77 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
  78 {
  79         int i;
  80         u8 hash = 0;
  81
  82         for (i = 0; i < hash_size; i++)
  83                 hash ^= hash_start[i];
  84
  85         return hash;
  86 }
  87
  88 /*********************** tlb specific functions ***************************/
  89
  90 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
  91 {
  92         if (save_load) {
  93                 entry->load_history = 1 + entry->tx_bytes /
  94                                       BOND_TLB_REBALANCE_INTERVAL;
  95                 entry->tx_bytes = 0;
  96         }
  97
  98         entry->tx_slave = NULL;
  99         entry->next = TLB_NULL_INDEX;
 100         entry->prev = TLB_NULL_INDEX;
 101 }
 102
 103 static inline void tlb_init_slave(struct slave *slave)
 104 {
 105         SLAVE_TLB_INFO(slave).load = 0;
 106         SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
 107 }
 108
 109 static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
 110                          int save_load)
 111 {
 112         struct tlb_client_info *tx_hash_table;
 113         u32 index;
 114
 115         /* clear slave from tx_hashtbl */
 116         tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
 117
 118         /* skip this if we've already freed the tx hash table */
 119         if (tx_hash_table) {
 120                 index = SLAVE_TLB_INFO(slave).head;
 121                 while (index != TLB_NULL_INDEX) {
 122                         u32 next_index = tx_hash_table[index].next;
 123                         tlb_init_table_entry(&tx_hash_table[index], save_load);
 124                         index = next_index;
 125                 }
 126         }
 127
 128         tlb_init_slave(slave);
 129 }
 130
 131 static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
 132                          int save_load)
 133 {
 134         spin_lock_bh(&bond->mode_lock);
 135         __tlb_clear_slave(bond, slave, save_load);
 136         spin_unlock_bh(&bond->mode_lock);
 137 }
 138
 139 /* Must be called before starting the monitor timer */
 140 static int tlb_initialize(struct bonding *bond)
 141 {
 142         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 143         int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 144         struct tlb_client_info *new_hashtbl;
 145         int i;
 146
 147         new_hashtbl = kzalloc(size, GFP_KERNEL);
 148         if (!new_hashtbl)
 149                 return -ENOMEM;
 150
 151         spin_lock_bh(&bond->mode_lock);
 152
 153         bond_info->tx_hashtbl = new_hashtbl;
 154
 155         for (i = 0; i < TLB_HASH_TABLE_SIZE; i++)
 156                 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
 157
 158         spin_unlock_bh(&bond->mode_lock);
 159
 160         return 0;
 161 }
 162
 163 /* Must be called only after all slaves have been released */
 164 static void tlb_deinitialize(struct bonding *bond)
 165 {
 166         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 167
 168         spin_lock_bh(&bond->mode_lock);
 169
 170         kfree(bond_info->tx_hashtbl);
 171         bond_info->tx_hashtbl = NULL;
 172
 173         spin_unlock_bh(&bond->mode_lock);
 174 }
 175
 176 static long long compute_gap(struct slave *slave)
 177 {
 178         return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
 179                (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
 180 }
 181
 182 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
 183 {
 184         struct slave *slave, *least_loaded;
 185         struct list_head *iter;
 186         long long max_gap;
 187
 188         least_loaded = NULL;
 189         max_gap = LLONG_MIN;
 190
 191         /* Find the slave with the largest gap */
 192         bond_for_each_slave_rcu(bond, slave, iter) {
 193                 if (bond_slave_can_tx(slave)) {
 194                         long long gap = compute_gap(slave);
 195
 196                         if (max_gap < gap) {
 197                                 least_loaded = slave;
 198                                 max_gap = gap;
 199                         }
 200                 }
 201         }
 202
 203         return least_loaded;
 204 }
 205
 206 static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
 207                                                 u32 skb_len)
 208 {
 209         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 210         struct tlb_client_info *hash_table;
 211         struct slave *assigned_slave;
 212
 213         hash_table = bond_info->tx_hashtbl;
 214         assigned_slave = hash_table[hash_index].tx_slave;
 215         if (!assigned_slave) {
 216                 assigned_slave = tlb_get_least_loaded_slave(bond);
 217
 218                 if (assigned_slave) {
 219                         struct tlb_slave_info *slave_info =
 220                                 &(SLAVE_TLB_INFO(assigned_slave));
 221                         u32 next_index = slave_info->head;
 222
 223                         hash_table[hash_index].tx_slave = assigned_slave;
 224                         hash_table[hash_index].next = next_index;
 225                         hash_table[hash_index].prev = TLB_NULL_INDEX;
 226
 227                         if (next_index != TLB_NULL_INDEX)
 228                                 hash_table[next_index].prev = hash_index;
 229
 230                         slave_info->head = hash_index;
 231                         slave_info->load +=
 232                                 hash_table[hash_index].load_history;
 233                 }
 234         }
 235
 236         if (assigned_slave)
 237                 hash_table[hash_index].tx_bytes += skb_len;
 238
 239         return assigned_slave;
 240 }
 241
 242 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
 243                                         u32 skb_len)
 244 {
 245         struct slave *tx_slave;
 246
 247         /* We don't need to disable softirq here, becase
 248          * tlb_choose_channel() is only called by bond_alb_xmit()
 249          * which already has softirq disabled.
 250          */
 251         spin_lock(&bond->mode_lock);
 252         tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
 253         spin_unlock(&bond->mode_lock);
 254
 255         return tx_slave;
 256 }
 257
 258 /*********************** rlb specific functions ***************************/
 259
 260 /* when an ARP REPLY is received from a client update its info
 261  * in the rx_hashtbl
 262  */
 263 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 264 {
 265         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 266         struct rlb_client_info *client_info;
 267         u32 hash_index;
 268
 269         spin_lock_bh(&bond->mode_lock);
 270
 271         hash_index = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
 272         client_info = &(bond_info->rx_hashtbl[hash_index]);
 273
 274         if ((client_info->assigned) &&
 275             (client_info->ip_src == arp->ip_dst) &&
 276             (client_info->ip_dst == arp->ip_src) &&
 277             (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) {
 278                 /* update the clients MAC address */
 279                 ether_addr_copy(client_info->mac_dst, arp->mac_src);
 280                 client_info->ntt = 1;
 281                 bond_info->rx_ntt = 1;
 282         }
 283
 284         spin_unlock_bh(&bond->mode_lock);
 285 }
 286
 287 static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
 288                         struct slave *slave)
 289 {
 290         struct arp_pkt *arp, _arp;
 291
 292         if (skb->protocol != cpu_to_be16(ETH_P_ARP))
 293                 goto out;
 294
 295         arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp);
 296         if (!arp)
 297                 goto out;
 298
 299         /* We received an ARP from arp->ip_src.
 300          * We might have used this IP address previously (on the bonding host
 301          * itself or on a system that is bridged together with the bond).
 302          * However, if arp->mac_src is different than what is stored in
 303          * rx_hashtbl, some other host is now using the IP and we must prevent
 304          * sending out client updates with this IP address and the old MAC
 305          * address.
 306          * Clean up all hash table entries that have this address as ip_src but
 307          * have a different mac_src.
 308          */
 309         rlb_purge_src_ip(bond, arp);
 310
 311         if (arp->op_code == htons(ARPOP_REPLY)) {
 312                 /* update rx hash table for this ARP */
 313                 rlb_update_entry_from_arp(bond, arp);
 314                 netdev_dbg(bond->dev, "Server received an ARP Reply from client\n");
 315         }
 316 out:
 317         return RX_HANDLER_ANOTHER;
 318 }
 319
 320 /* Caller must hold rcu_read_lock() */
 321 static struct slave *__rlb_next_rx_slave(struct bonding *bond)
 322 {
 323         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 324         struct slave *before = NULL, *rx_slave = NULL, *slave;
 325         struct list_head *iter;
 326         bool found = false;
 327
 328         bond_for_each_slave_rcu(bond, slave, iter) {
 329                 if (!bond_slave_can_tx(slave))
 330                         continue;
 331                 if (!found) {
 332                         if (!before || before->speed < slave->speed)
 333                                 before = slave;
 334                 } else {
 335                         if (!rx_slave || rx_slave->speed < slave->speed)
 336                                 rx_slave = slave;
 337                 }
 338                 if (slave == bond_info->rx_slave)
 339                         found = true;
 340         }
 341         /* we didn't find anything after the current or we have something
 342          * better before and up to the current slave
 343          */
 344         if (!rx_slave || (before && rx_slave->speed < before->speed))
 345                 rx_slave = before;
 346
 347         if (rx_slave)
 348                 bond_info->rx_slave = rx_slave;
 349
 350         return rx_slave;
 351 }
 352
 353 /* Caller must hold RTNL, rcu_read_lock is obtained only to silence checkers */
 354 static struct slave *rlb_next_rx_slave(struct bonding *bond)
 355 {
 356         struct slave *rx_slave;
 357
 358         ASSERT_RTNL();
 359
 360         rcu_read_lock();
 361         rx_slave = __rlb_next_rx_slave(bond);
 362         rcu_read_unlock();
 363
 364         return rx_slave;
 365 }
 366
 367 /* teach the switch the mac of a disabled slave
 368  * on the primary for fault tolerance
 369  *
 370  * Caller must hold RTNL
 371  */
 372 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
 373 {
 374         struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 375
 376         if (!curr_active)
 377                 return;
 378
 379         if (!bond->alb_info.primary_is_promisc) {
 380                 if (!dev_set_promiscuity(curr_active->dev, 1))
 381                         bond->alb_info.primary_is_promisc = 1;
 382                 else
 383                         bond->alb_info.primary_is_promisc = 0;
 384         }
 385
 386         bond->alb_info.rlb_promisc_timeout_counter = 0;
 387
 388         alb_send_learning_packets(curr_active, addr, true);
 389 }
 390
 391 /* slave being removed should not be active at this point
 392  *
 393  * Caller must hold rtnl.
 394  */
 395 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 396 {
 397         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 398         struct rlb_client_info *rx_hash_table;
 399         u32 index, next_index;
 400
 401         /* clear slave from rx_hashtbl */
 402         spin_lock_bh(&bond->mode_lock);
 403
 404         rx_hash_table = bond_info->rx_hashtbl;
 405         index = bond_info->rx_hashtbl_used_head;
 406         for (; index != RLB_NULL_INDEX; index = next_index) {
 407                 next_index = rx_hash_table[index].used_next;
 408                 if (rx_hash_table[index].slave == slave) {
 409                         struct slave *assigned_slave = rlb_next_rx_slave(bond);
 410
 411                         if (assigned_slave) {
 412                                 rx_hash_table[index].slave = assigned_slave;
 413                                 if (is_valid_ether_addr(rx_hash_table[index].mac_dst)) {
 414                                         bond_info->rx_hashtbl[index].ntt = 1;
 415                                         bond_info->rx_ntt = 1;
 416                                         /* A slave has been removed from the
 417                                          * table because it is either disabled
 418                                          * or being released. We must retry the
 419                                          * update to avoid clients from not
 420                                          * being updated & disconnecting when
 421                                          * there is stress
 422                                          */
 423                                         bond_info->rlb_update_retry_counter =
 424                                                 RLB_UPDATE_RETRY;
 425                                 }
 426                         } else {  /* there is no active slave */
 427                                 rx_hash_table[index].slave = NULL;
 428                         }
 429                 }
 430         }
 431
 432         spin_unlock_bh(&bond->mode_lock);
 433
 434         if (slave != rtnl_dereference(bond->curr_active_slave))
 435                 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 436 }
 437
 438 static void rlb_update_client(struct rlb_client_info *client_info)
 439 {
 440         int i;
 441
 442         if (!client_info->slave || !is_valid_ether_addr(client_info->mac_dst))
 443                 return;
 444
 445         for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 446                 struct sk_buff *skb;
 447
 448                 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 449                                  client_info->ip_dst,
 450                                  client_info->slave->dev,
 451                                  client_info->ip_src,
 452                                  client_info->mac_dst,
 453                                  client_info->slave->dev->dev_addr,
 454                                  client_info->mac_dst);
 455                 if (!skb) {
 456                         netdev_err(client_info->slave->bond->dev,
 457                                    "failed to create an ARP packet\n");
 458                         continue;
 459                 }
 460
 461                 skb->dev = client_info->slave->dev;
 462
 463                 if (client_info->vlan_id) {
 464                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 465                                                client_info->vlan_id);
 466                 }
 467
 468                 arp_xmit(skb);
 469         }
 470 }
 471
 472 /* sends ARP REPLIES that update the clients that need updating */
 473 static void rlb_update_rx_clients(struct bonding *bond)
 474 {
 475         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 476         struct rlb_client_info *client_info;
 477         u32 hash_index;
 478
 479         spin_lock_bh(&bond->mode_lock);
 480
 481         hash_index = bond_info->rx_hashtbl_used_head;
 482         for (; hash_index != RLB_NULL_INDEX;
 483              hash_index = client_info->used_next) {
 484                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 485                 if (client_info->ntt) {
 486                         rlb_update_client(client_info);
 487                         if (bond_info->rlb_update_retry_counter == 0)
 488                                 client_info->ntt = 0;
 489                 }
 490         }
 491
 492         /* do not update the entries again until this counter is zero so that
 493          * not to confuse the clients.
 494          */
 495         bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 496
 497         spin_unlock_bh(&bond->mode_lock);
 498 }
 499
 500 /* The slave was assigned a new mac address - update the clients */
 501 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 502 {
 503         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 504         struct rlb_client_info *client_info;
 505         int ntt = 0;
 506         u32 hash_index;
 507
 508         spin_lock_bh(&bond->mode_lock);
 509
 510         hash_index = bond_info->rx_hashtbl_used_head;
 511         for (; hash_index != RLB_NULL_INDEX;
 512              hash_index = client_info->used_next) {
 513                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 514
 515                 if ((client_info->slave == slave) &&
 516                     is_valid_ether_addr(client_info->mac_dst)) {
 517                         client_info->ntt = 1;
 518                         ntt = 1;
 519                 }
 520         }
 521
 522         /* update the team's flag only after the whole iteration */
 523         if (ntt) {
 524                 bond_info->rx_ntt = 1;
 525                 /* fasten the change */
 526                 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 527         }
 528
 529         spin_unlock_bh(&bond->mode_lock);
 530 }
 531
 532 /* mark all clients using src_ip to be updated */
 533 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
 534 {
 535         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 536         struct rlb_client_info *client_info;
 537         u32 hash_index;
 538
 539         spin_lock(&bond->mode_lock);
 540
 541         hash_index = bond_info->rx_hashtbl_used_head;
 542         for (; hash_index != RLB_NULL_INDEX;
 543              hash_index = client_info->used_next) {
 544                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 545
 546                 if (!client_info->slave) {
 547                         netdev_err(bond->dev, "found a client with no channel in the client's hash table\n");
 548                         continue;
 549                 }
 550                 /* update all clients using this src_ip, that are not assigned
 551                  * to the team's address (curr_active_slave) and have a known
 552                  * unicast mac address.
 553                  */
 554                 if ((client_info->ip_src == src_ip) &&
 555                     !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
 556                                              bond->dev->dev_addr) &&
 557                     is_valid_ether_addr(client_info->mac_dst)) {
 558                         client_info->ntt = 1;
 559                         bond_info->rx_ntt = 1;
 560                 }
 561         }
 562
 563         spin_unlock(&bond->mode_lock);
 564 }
 565
 566 static struct slave *rlb_choose_channel(struct sk_buff *skb,
 567                                         struct bonding *bond,
 568                                         const struct arp_pkt *arp)
 569 {
 570         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 571         struct slave *assigned_slave, *curr_active_slave;
 572         struct rlb_client_info *client_info;
 573         u32 hash_index = 0;
 574
 575         spin_lock(&bond->mode_lock);
 576
 577         curr_active_slave = rcu_dereference(bond->curr_active_slave);
 578
 579         hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
 580         client_info = &(bond_info->rx_hashtbl[hash_index]);
 581
 582         if (client_info->assigned) {
 583                 if ((client_info->ip_src == arp->ip_src) &&
 584                     (client_info->ip_dst == arp->ip_dst)) {
 585                         /* the entry is already assigned to this client */
 586                         if (!is_broadcast_ether_addr(arp->mac_dst)) {
 587                                 /* update mac address from arp */
 588                                 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
 589                         }
 590                         ether_addr_copy(client_info->mac_src, arp->mac_src);
 591
 592                         assigned_slave = client_info->slave;
 593                         if (assigned_slave) {
 594                                 spin_unlock(&bond->mode_lock);
 595                                 return assigned_slave;
 596                         }
 597                 } else {
 598                         /* the entry is already assigned to some other client,
 599                          * move the old client to primary (curr_active_slave) so
 600                          * that the new client can be assigned to this entry.
 601                          */
 602                         if (curr_active_slave &&
 603                             client_info->slave != curr_active_slave) {
 604                                 client_info->slave = curr_active_slave;
 605                                 rlb_update_client(client_info);
 606                         }
 607                 }
 608         }
 609         /* assign a new slave */
 610         assigned_slave = __rlb_next_rx_slave(bond);
 611
 612         if (assigned_slave) {
 613                 if (!(client_info->assigned &&
 614                       client_info->ip_src == arp->ip_src)) {
 615                         /* ip_src is going to be updated,
 616                          * fix the src hash list
 617                          */
 618                         u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
 619                                                     sizeof(arp->ip_src));
 620                         rlb_src_unlink(bond, hash_index);
 621                         rlb_src_link(bond, hash_src, hash_index);
 622                 }
 623
 624                 client_info->ip_src = arp->ip_src;
 625                 client_info->ip_dst = arp->ip_dst;
 626                 /* arp->mac_dst is broadcast for arp reqeusts.
 627                  * will be updated with clients actual unicast mac address
 628                  * upon receiving an arp reply.
 629                  */
 630                 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
 631                 ether_addr_copy(client_info->mac_src, arp->mac_src);
 632                 client_info->slave = assigned_slave;
 633
 634                 if (is_valid_ether_addr(client_info->mac_dst)) {
 635                         client_info->ntt = 1;
 636                         bond->alb_info.rx_ntt = 1;
 637                 } else {
 638                         client_info->ntt = 0;
 639                 }
 640
 641                 if (vlan_get_tag(skb, &client_info->vlan_id))
 642                         client_info->vlan_id = 0;
 643
 644                 if (!client_info->assigned) {
 645                         u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
 646                         bond_info->rx_hashtbl_used_head = hash_index;
 647                         client_info->used_next = prev_tbl_head;
 648                         if (prev_tbl_head != RLB_NULL_INDEX) {
 649                                 bond_info->rx_hashtbl[prev_tbl_head].used_prev =
 650                                         hash_index;
 651                         }
 652                         client_info->assigned = 1;
 653                 }
 654         }
 655
 656         spin_unlock(&bond->mode_lock);
 657
 658         return assigned_slave;
 659 }
 660
 661 /* chooses (and returns) transmit channel for arp reply
 662  * does not choose channel for other arp types since they are
 663  * sent on the curr_active_slave
 664  */
 665 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 666 {
 667         struct slave *tx_slave = NULL;
 668         struct arp_pkt *arp;
 669
 670         if (!pskb_network_may_pull(skb, sizeof(*arp)))
 671                 return NULL;
 672         arp = (struct arp_pkt *)skb_network_header(skb);
 673
 674         /* Don't modify or load balance ARPs that do not originate locally
 675          * (e.g.,arrive via a bridge).
 676          */
 677         if (!bond_slave_has_mac_rx(bond, arp->mac_src))
 678                 return NULL;
 679
 680         if (arp->op_code == htons(ARPOP_REPLY)) {
 681                 /* the arp must be sent on the selected rx channel */
 682                 tx_slave = rlb_choose_channel(skb, bond, arp);
 683                 if (tx_slave)
 684                         bond_hw_addr_copy(arp->mac_src, tx_slave->dev->dev_addr,
 685                                           tx_slave->dev->addr_len);
 686                 netdev_dbg(bond->dev, "Server sent ARP Reply packet\n");
 687         } else if (arp->op_code == htons(ARPOP_REQUEST)) {
 688                 /* Create an entry in the rx_hashtbl for this client as a
 689                  * place holder.
 690                  * When the arp reply is received the entry will be updated
 691                  * with the correct unicast address of the client.
 692                  */
 693                 rlb_choose_channel(skb, bond, arp);
 694
 695                 /* The ARP reply packets must be delayed so that
 696                  * they can cancel out the influence of the ARP request.
 697                  */
 698                 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 699
 700                 /* arp requests are broadcast and are sent on the primary
 701                  * the arp request will collapse all clients on the subnet to
 702                  * the primary slave. We must register these clients to be
 703                  * updated with their assigned mac.
 704                  */
 705                 rlb_req_update_subnet_clients(bond, arp->ip_src);
 706                 netdev_dbg(bond->dev, "Server sent ARP Request packet\n");
 707         }
 708
 709         return tx_slave;
 710 }
 711
 712 static void rlb_rebalance(struct bonding *bond)
 713 {
 714         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 715         struct slave *assigned_slave;
 716         struct rlb_client_info *client_info;
 717         int ntt;
 718         u32 hash_index;
 719
 720         spin_lock_bh(&bond->mode_lock);
 721
 722         ntt = 0;
 723         hash_index = bond_info->rx_hashtbl_used_head;
 724         for (; hash_index != RLB_NULL_INDEX;
 725              hash_index = client_info->used_next) {
 726                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 727                 assigned_slave = __rlb_next_rx_slave(bond);
 728                 if (assigned_slave && (client_info->slave != assigned_slave)) {
 729                         client_info->slave = assigned_slave;
 730                         if (!is_zero_ether_addr(client_info->mac_dst)) {
 731                                 client_info->ntt = 1;
 732                                 ntt = 1;
 733                         }
 734                 }
 735         }
 736
 737         /* update the team's flag only after the whole iteration */
 738         if (ntt)
 739                 bond_info->rx_ntt = 1;
 740         spin_unlock_bh(&bond->mode_lock);
 741 }
 742
 743 /* Caller must hold mode_lock */
 744 static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
 745 {
 746         entry->used_next = RLB_NULL_INDEX;
 747         entry->used_prev = RLB_NULL_INDEX;
 748         entry->assigned = 0;
 749         entry->slave = NULL;
 750         entry->vlan_id = 0;
 751 }
 752 static void rlb_init_table_entry_src(struct rlb_client_info *entry)
 753 {
 754         entry->src_first = RLB_NULL_INDEX;
 755         entry->src_prev = RLB_NULL_INDEX;
 756         entry->src_next = RLB_NULL_INDEX;
 757 }
 758
 759 static void rlb_init_table_entry(struct rlb_client_info *entry)
 760 {
 761         memset(entry, 0, sizeof(struct rlb_client_info));
 762         rlb_init_table_entry_dst(entry);
 763         rlb_init_table_entry_src(entry);
 764 }
 765
 766 static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
 767 {
 768         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 769         u32 next_index = bond_info->rx_hashtbl[index].used_next;
 770         u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
 771
 772         if (index == bond_info->rx_hashtbl_used_head)
 773                 bond_info->rx_hashtbl_used_head = next_index;
 774         if (prev_index != RLB_NULL_INDEX)
 775                 bond_info->rx_hashtbl[prev_index].used_next = next_index;
 776         if (next_index != RLB_NULL_INDEX)
 777                 bond_info->rx_hashtbl[next_index].used_prev = prev_index;
 778 }
 779
 780 /* unlink a rlb hash table entry from the src list */
 781 static void rlb_src_unlink(struct bonding *bond, u32 index)
 782 {
 783         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 784         u32 next_index = bond_info->rx_hashtbl[index].src_next;
 785         u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
 786
 787         bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
 788         bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
 789
 790         if (next_index != RLB_NULL_INDEX)
 791                 bond_info->rx_hashtbl[next_index].src_prev = prev_index;
 792
 793         if (prev_index == RLB_NULL_INDEX)
 794                 return;
 795
 796         /* is prev_index pointing to the head of this list? */
 797         if (bond_info->rx_hashtbl[prev_index].src_first == index)
 798                 bond_info->rx_hashtbl[prev_index].src_first = next_index;
 799         else
 800                 bond_info->rx_hashtbl[prev_index].src_next = next_index;
 801
 802 }
 803
 804 static void rlb_delete_table_entry(struct bonding *bond, u32 index)
 805 {
 806         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 807         struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 808
 809         rlb_delete_table_entry_dst(bond, index);
 810         rlb_init_table_entry_dst(entry);
 811
 812         rlb_src_unlink(bond, index);
 813 }
 814
 815 /* add the rx_hashtbl[ip_dst_hash] entry to the list
 816  * of entries with identical ip_src_hash
 817  */
 818 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
 819 {
 820         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 821         u32 next;
 822
 823         bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
 824         next = bond_info->rx_hashtbl[ip_src_hash].src_first;
 825         bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
 826         if (next != RLB_NULL_INDEX)
 827                 bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
 828         bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
 829 }
 830
 831 /* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does
 832  * not match arp->mac_src
 833  */
 834 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
 835 {
 836         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 837         u32 ip_src_hash = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
 838         u32 index;
 839
 840         spin_lock_bh(&bond->mode_lock);
 841
 842         index = bond_info->rx_hashtbl[ip_src_hash].src_first;
 843         while (index != RLB_NULL_INDEX) {
 844                 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 845                 u32 next_index = entry->src_next;
 846                 if (entry->ip_src == arp->ip_src &&
 847                     !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
 848                                 rlb_delete_table_entry(bond, index);
 849                 index = next_index;
 850         }
 851         spin_unlock_bh(&bond->mode_lock);
 852 }
 853
 854 static int rlb_initialize(struct bonding *bond)
 855 {
 856         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 857         struct rlb_client_info  *new_hashtbl;
 858         int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 859         int i;
 860
 861         new_hashtbl = kmalloc(size, GFP_KERNEL);
 862         if (!new_hashtbl)
 863                 return -1;
 864
 865         spin_lock_bh(&bond->mode_lock);
 866
 867         bond_info->rx_hashtbl = new_hashtbl;
 868
 869         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 870
 871         for (i = 0; i < RLB_HASH_TABLE_SIZE; i++)
 872                 rlb_init_table_entry(bond_info->rx_hashtbl + i);
 873
 874         spin_unlock_bh(&bond->mode_lock);
 875
 876         /* register to receive ARPs */
 877         bond->recv_probe = rlb_arp_recv;
 878
 879         return 0;
 880 }
 881
 882 static void rlb_deinitialize(struct bonding *bond)
 883 {
 884         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 885
 886         spin_lock_bh(&bond->mode_lock);
 887
 888         kfree(bond_info->rx_hashtbl);
 889         bond_info->rx_hashtbl = NULL;
 890         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 891
 892         spin_unlock_bh(&bond->mode_lock);
 893 }
 894
 895 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 896 {
 897         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 898         u32 curr_index;
 899
 900         spin_lock_bh(&bond->mode_lock);
 901
 902         curr_index = bond_info->rx_hashtbl_used_head;
 903         while (curr_index != RLB_NULL_INDEX) {
 904                 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 905                 u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
 906
 907                 if (curr->vlan_id == vlan_id)
 908                         rlb_delete_table_entry(bond, curr_index);
 909
 910                 curr_index = next_index;
 911         }
 912
 913         spin_unlock_bh(&bond->mode_lock);
 914 }
 915
 916 /*********************** tlb/rlb shared functions *********************/
 917
 918 static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
 919                             __be16 vlan_proto, u16 vid)
 920 {
 921         struct learning_pkt pkt;
 922         struct sk_buff *skb;
 923         int size = sizeof(struct learning_pkt);
 924
 925         memset(&pkt, 0, size);
 926         ether_addr_copy(pkt.mac_dst, mac_addr);
 927         ether_addr_copy(pkt.mac_src, mac_addr);
 928         pkt.type = cpu_to_be16(ETH_P_LOOPBACK);
 929
 930         skb = dev_alloc_skb(size);
 931         if (!skb)
 932                 return;
 933
 934         skb_put_data(skb, &pkt, size);
 935
 936         skb_reset_mac_header(skb);
 937         skb->network_header = skb->mac_header + ETH_HLEN;
 938         skb->protocol = pkt.type;
 939         skb->priority = TC_PRIO_CONTROL;
 940         skb->dev = slave->dev;
 941
 942         netdev_dbg(slave->bond->dev,
 943                    "Send learning packet: dev %s mac %pM vlan %d\n",
 944                    slave->dev->name, mac_addr, vid);
 945
 946         if (vid)
 947                 __vlan_hwaccel_put_tag(skb, vlan_proto, vid);
 948
 949         dev_queue_xmit(skb);
 950 }
 951
 952 struct alb_walk_data {
 953         struct bonding *bond;
 954         struct slave *slave;
 955         u8 *mac_addr;
 956         bool strict_match;
 957 };
 958
 959 static int alb_upper_dev_walk(struct net_device *upper, void *_data)
 960 {
 961         struct alb_walk_data *data = _data;
 962         bool strict_match = data->strict_match;
 963         struct bonding *bond = data->bond;
 964         struct slave *slave = data->slave;
 965         u8 *mac_addr = data->mac_addr;
 966         struct bond_vlan_tag *tags;
 967
 968         if (is_vlan_dev(upper) &&
 969             bond->nest_level == vlan_get_encap_level(upper) - 1) {
 970                 if (upper->addr_assign_type == NET_ADDR_STOLEN) {
 971                         alb_send_lp_vid(slave, mac_addr,
 972                                         vlan_dev_vlan_proto(upper),
 973                                         vlan_dev_vlan_id(upper));
 974                 } else {
 975                         alb_send_lp_vid(slave, upper->dev_addr,
 976                                         vlan_dev_vlan_proto(upper),
 977                                         vlan_dev_vlan_id(upper));
 978                 }
 979         }
 980
 981         /* If this is a macvlan device, then only send updates
 982          * when strict_match is turned off.
 983          */
 984         if (netif_is_macvlan(upper) && !strict_match) {
 985                 tags = bond_verify_device_path(bond->dev, upper, 0);
 986                 if (IS_ERR_OR_NULL(tags))
 987                         BUG();
 988                 alb_send_lp_vid(slave, upper->dev_addr,
 989                                 tags[0].vlan_proto, tags[0].vlan_id);
 990                 kfree(tags);
 991         }
 992
 993         return 0;
 994 }
 995
 996 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
 997                                       bool strict_match)
 998 {
 999         struct bonding *bond = bond_get_bond_by_slave(slave);
1000         struct alb_walk_data data = {
1001                 .strict_match = strict_match,
1002                 .mac_addr = mac_addr,
1003                 .slave = slave,
1004                 .bond = bond,
1005         };
1006
1007         /* send untagged */
1008         alb_send_lp_vid(slave, mac_addr, 0, 0);
1009
1010         /* loop through all devices and see if we need to send a packet
1011          * for that device.
1012          */
1013         rcu_read_lock();
1014         netdev_walk_all_upper_dev_rcu(bond->dev, alb_upper_dev_walk, &data);
1015         rcu_read_unlock();
1016 }
1017
1018 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[],
1019                                   unsigned int len)
1020 {
1021         struct net_device *dev = slave->dev;
1022         struct sockaddr_storage ss;
1023
1024         if (BOND_MODE(slave->bond) == BOND_MODE_TLB) {
1025                 memcpy(dev->dev_addr, addr, len);
1026                 return 0;
1027         }
1028
1029         /* for rlb each slave must have a unique hw mac addresses so that
1030          * each slave will receive packets destined to a different mac
1031          */
1032         memcpy(ss.__data, addr, len);
1033         ss.ss_family = dev->type;
1034         if (dev_set_mac_address(dev, (struct sockaddr *)&ss)) {
1035                 netdev_err(slave->bond->dev, "dev_set_mac_address of dev %s failed! ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
1036                            dev->name);
1037                 return -EOPNOTSUPP;
1038         }
1039         return 0;
1040 }
1041
1042 /* Swap MAC addresses between two slaves.
1043  *
1044  * Called with RTNL held, and no other locks.
1045  */
1046 static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1047 {
1048         u8 tmp_mac_addr[MAX_ADDR_LEN];
1049
1050         bond_hw_addr_copy(tmp_mac_addr, slave1->dev->dev_addr,
1051                           slave1->dev->addr_len);
1052         alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr,
1053                                slave2->dev->addr_len);
1054         alb_set_slave_mac_addr(slave2, tmp_mac_addr,
1055                                slave1->dev->addr_len);
1056
1057 }
1058
1059 /* Send learning packets after MAC address swap.
1060  *
1061  * Called with RTNL and no other locks
1062  */
1063 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1064                                 struct slave *slave2)
1065 {
1066         int slaves_state_differ = (bond_slave_can_tx(slave1) != bond_slave_can_tx(slave2));
1067         struct slave *disabled_slave = NULL;
1068
1069         ASSERT_RTNL();
1070
1071         /* fasten the change in the switch */
1072         if (bond_slave_can_tx(slave1)) {
1073                 alb_send_learning_packets(slave1, slave1->dev->dev_addr, false);
1074                 if (bond->alb_info.rlb_enabled) {
1075                         /* inform the clients that the mac address
1076                          * has changed
1077                          */
1078                         rlb_req_update_slave_clients(bond, slave1);
1079                 }
1080         } else {
1081                 disabled_slave = slave1;
1082         }
1083
1084         if (bond_slave_can_tx(slave2)) {
1085                 alb_send_learning_packets(slave2, slave2->dev->dev_addr, false);
1086                 if (bond->alb_info.rlb_enabled) {
1087                         /* inform the clients that the mac address
1088                          * has changed
1089                          */
1090                         rlb_req_update_slave_clients(bond, slave2);
1091                 }
1092         } else {
1093                 disabled_slave = slave2;
1094         }
1095
1096         if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1097                 /* A disabled slave was assigned an active mac addr */
1098                 rlb_teach_disabled_mac_on_primary(bond,
1099                                                   disabled_slave->dev->dev_addr);
1100         }
1101 }
1102
1103 /**
1104  * alb_change_hw_addr_on_detach
1105  * @bond: bonding we're working on
1106  * @slave: the slave that was just detached
1107  *
1108  * We assume that @slave was already detached from the slave list.
1109  *
1110  * If @slave's permanent hw address is different both from its current
1111  * address and from @bond's address, then somewhere in the bond there's
1112  * a slave that has @slave's permanet address as its current address.
1113  * We'll make sure that that slave no longer uses @slave's permanent address.
1114  *
1115  * Caller must hold RTNL and no other locks
1116  */
1117 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1118 {
1119         int perm_curr_diff;
1120         int perm_bond_diff;
1121         struct slave *found_slave;
1122
1123         perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1124                                                   slave->dev->dev_addr);
1125         perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1126                                                   bond->dev->dev_addr);
1127
1128         if (perm_curr_diff && perm_bond_diff) {
1129                 found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1130
1131                 if (found_slave) {
1132                         alb_swap_mac_addr(slave, found_slave);
1133                         alb_fasten_mac_swap(bond, slave, found_slave);
1134                 }
1135         }
1136 }
1137
1138 /**
1139  * alb_handle_addr_collision_on_attach
1140  * @bond: bonding we're working on
1141  * @slave: the slave that was just attached
1142  *
1143  * checks uniqueness of slave's mac address and handles the case the
1144  * new slave uses the bonds mac address.
1145  *
1146  * If the permanent hw address of @slave is @bond's hw address, we need to
1147  * find a different hw address to give @slave, that isn't in use by any other
1148  * slave in the bond. This address must be, of course, one of the permanent
1149  * addresses of the other slaves.
1150  *
1151  * We go over the slave list, and for each slave there we compare its
1152  * permanent hw address with the current address of all the other slaves.
1153  * If no match was found, then we've found a slave with a permanent address
1154  * that isn't used by any other slave in the bond, so we can assign it to
1155  * @slave.
1156  *
1157  * assumption: this function is called before @slave is attached to the
1158  *             bond slave list.
1159  */
1160 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1161 {
1162         struct slave *has_bond_addr = rcu_access_pointer(bond->curr_active_slave);
1163         struct slave *tmp_slave1, *free_mac_slave = NULL;
1164         struct list_head *iter;
1165
1166         if (!bond_has_slaves(bond)) {
1167                 /* this is the first slave */
1168                 return 0;
1169         }
1170
1171         /* if slave's mac address differs from bond's mac address
1172          * check uniqueness of slave's mac address against the other
1173          * slaves in the bond.
1174          */
1175         if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1176                 if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1177                         return 0;
1178
1179                 /* Try setting slave mac to bond address and fall-through
1180                  * to code handling that situation below...
1181                  */
1182                 alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1183                                        bond->dev->addr_len);
1184         }
1185
1186         /* The slave's address is equal to the address of the bond.
1187          * Search for a spare address in the bond for this slave.
1188          */
1189         bond_for_each_slave(bond, tmp_slave1, iter) {
1190                 if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1191                         /* no slave has tmp_slave1's perm addr
1192                          * as its curr addr
1193                          */
1194                         free_mac_slave = tmp_slave1;
1195                         break;
1196                 }
1197
1198                 if (!has_bond_addr) {
1199                         if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1200                                                     bond->dev->dev_addr)) {
1201
1202                                 has_bond_addr = tmp_slave1;
1203                         }
1204                 }
1205         }
1206
1207         if (free_mac_slave) {
1208                 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1209                                        free_mac_slave->dev->addr_len);
1210
1211                 netdev_warn(bond->dev, "the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1212                             slave->dev->name, free_mac_slave->dev->name);
1213
1214         } else if (has_bond_addr) {
1215                 netdev_err(bond->dev, "the hw address of slave %s is in use by the bond; couldn't find a slave with a free hw address to give it (this should not have happened)\n",
1216                            slave->dev->name);
1217                 return -EFAULT;
1218         }
1219
1220         return 0;
1221 }
1222
1223 /**
1224  * alb_set_mac_address
1225  * @bond:
1226  * @addr:
1227  *
1228  * In TLB mode all slaves are configured to the bond's hw address, but set
1229  * their dev_addr field to different addresses (based on their permanent hw
1230  * addresses).
1231  *
1232  * For each slave, this function sets the interface to the new address and then
1233  * changes its dev_addr field to its previous value.
1234  *
1235  * Unwinding assumes bond's mac address has not yet changed.
1236  */
1237 static int alb_set_mac_address(struct bonding *bond, void *addr)
1238 {
1239         struct slave *slave, *rollback_slave;
1240         struct list_head *iter;
1241         struct sockaddr_storage ss;
1242         char tmp_addr[MAX_ADDR_LEN];
1243         int res;
1244
1245         if (bond->alb_info.rlb_enabled)
1246                 return 0;
1247
1248         bond_for_each_slave(bond, slave, iter) {
1249                 /* save net_device's current hw address */
1250                 bond_hw_addr_copy(tmp_addr, slave->dev->dev_addr,
1251                                   slave->dev->addr_len);
1252
1253                 res = dev_set_mac_address(slave->dev, addr);
1254
1255                 /* restore net_device's hw address */
1256                 bond_hw_addr_copy(slave->dev->dev_addr, tmp_addr,
1257                                   slave->dev->addr_len);
1258
1259                 if (res)
1260                         goto unwind;
1261         }
1262
1263         return 0;
1264
1265 unwind:
1266         memcpy(ss.__data, bond->dev->dev_addr, bond->dev->addr_len);
1267         ss.ss_family = bond->dev->type;
1268
1269         /* unwind from head to the slave that failed */
1270         bond_for_each_slave(bond, rollback_slave, iter) {
1271                 if (rollback_slave == slave)
1272                         break;
1273                 bond_hw_addr_copy(tmp_addr, rollback_slave->dev->dev_addr,
1274                                   rollback_slave->dev->addr_len);
1275                 dev_set_mac_address(rollback_slave->dev,
1276                                     (struct sockaddr *)&ss);
1277                 bond_hw_addr_copy(rollback_slave->dev->dev_addr, tmp_addr,
1278                                   rollback_slave->dev->addr_len);
1279         }
1280
1281         return res;
1282 }
1283
1284 /************************ exported alb funcions ************************/
1285
1286 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1287 {
1288         int res;
1289
1290         res = tlb_initialize(bond);
1291         if (res)
1292                 return res;
1293
1294         if (rlb_enabled) {
1295                 bond->alb_info.rlb_enabled = 1;
1296                 res = rlb_initialize(bond);
1297                 if (res) {
1298                         tlb_deinitialize(bond);
1299                         return res;
1300                 }
1301         } else {
1302                 bond->alb_info.rlb_enabled = 0;
1303         }
1304
1305         return 0;
1306 }
1307
1308 void bond_alb_deinitialize(struct bonding *bond)
1309 {
1310         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1311
1312         tlb_deinitialize(bond);
1313
1314         if (bond_info->rlb_enabled)
1315                 rlb_deinitialize(bond);
1316 }
1317
1318 static netdev_tx_t bond_do_alb_xmit(struct sk_buff *skb, struct bonding *bond,
1319                                     struct slave *tx_slave)
1320 {
1321         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1322         struct ethhdr *eth_data = eth_hdr(skb);
1323
1324         if (!tx_slave) {
1325                 /* unbalanced or unassigned, send through primary */
1326                 tx_slave = rcu_dereference(bond->curr_active_slave);
1327                 if (bond->params.tlb_dynamic_lb)
1328                         bond_info->unbalanced_load += skb->len;
1329         }
1330
1331         if (tx_slave && bond_slave_can_tx(tx_slave)) {
1332                 if (tx_slave != rcu_access_pointer(bond->curr_active_slave)) {
1333                         ether_addr_copy(eth_data->h_source,
1334                                         tx_slave->dev->dev_addr);
1335                 }
1336
1337                 bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1338                 goto out;
1339         }
1340
1341         if (tx_slave && bond->params.tlb_dynamic_lb) {
1342                 spin_lock(&bond->mode_lock);
1343                 __tlb_clear_slave(bond, tx_slave, 0);
1344                 spin_unlock(&bond->mode_lock);
1345         }
1346
1347         /* no suitable interface, frame not sent */
1348         bond_tx_drop(bond->dev, skb);
1349 out:
1350         return NETDEV_TX_OK;
1351 }
1352
1353 netdev_tx_t bond_tlb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1354 {
1355         struct bonding *bond = netdev_priv(bond_dev);
1356         struct ethhdr *eth_data;
1357         struct slave *tx_slave = NULL;
1358         u32 hash_index;
1359
1360         skb_reset_mac_header(skb);
1361         eth_data = eth_hdr(skb);
1362
1363         /* Do not TX balance any multicast or broadcast */
1364         if (!is_multicast_ether_addr(eth_data->h_dest)) {
1365                 switch (skb->protocol) {
1366                 case htons(ETH_P_IP):
1367                 case htons(ETH_P_IPX):
1368                     /* In case of IPX, it will falback to L2 hash */
1369                 case htons(ETH_P_IPV6):
1370                         hash_index = bond_xmit_hash(bond, skb);
1371                         if (bond->params.tlb_dynamic_lb) {
1372                                 tx_slave = tlb_choose_channel(bond,
1373                                                               hash_index & 0xFF,
1374                                                               skb->len);
1375                         } else {
1376                                 struct bond_up_slave *slaves;
1377                                 unsigned int count;
1378
1379                                 slaves = rcu_dereference(bond->slave_arr);
1380                                 count = slaves ? READ_ONCE(slaves->count) : 0;
1381                                 if (likely(count))
1382                                         tx_slave = slaves->arr[hash_index %
1383                                                                count];
1384                         }
1385                         break;
1386                 }
1387         }
1388         return bond_do_alb_xmit(skb, bond, tx_slave);
1389 }
1390
1391 netdev_tx_t bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1392 {
1393         struct bonding *bond = netdev_priv(bond_dev);
1394         struct ethhdr *eth_data;
1395         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1396         struct slave *tx_slave = NULL;
1397         static const __be32 ip_bcast = htonl(0xffffffff);
1398         int hash_size = 0;
1399         bool do_tx_balance = true;
1400         u32 hash_index = 0;
1401         const u8 *hash_start = NULL;
1402
1403         skb_reset_mac_header(skb);
1404         eth_data = eth_hdr(skb);
1405
1406         switch (ntohs(skb->protocol)) {
1407         case ETH_P_IP: {
1408                 const struct iphdr *iph;
1409
1410                 if (is_broadcast_ether_addr(eth_data->h_dest) ||
1411                     !pskb_network_may_pull(skb, sizeof(*iph))) {
1412                         do_tx_balance = false;
1413                         break;
1414                 }
1415                 iph = ip_hdr(skb);
1416                 if (iph->daddr == ip_bcast || iph->protocol == IPPROTO_IGMP) {
1417                         do_tx_balance = false;
1418                         break;
1419                 }
1420                 hash_start = (char *)&(iph->daddr);
1421                 hash_size = sizeof(iph->daddr);
1422                 break;
1423         }
1424         case ETH_P_IPV6: {
1425                 const struct ipv6hdr *ip6hdr;
1426
1427                 /* IPv6 doesn't really use broadcast mac address, but leave
1428                  * that here just in case.
1429                  */
1430                 if (is_broadcast_ether_addr(eth_data->h_dest)) {
1431                         do_tx_balance = false;
1432                         break;
1433                 }
1434
1435                 /* IPv6 uses all-nodes multicast as an equivalent to
1436                  * broadcasts in IPv4.
1437                  */
1438                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1439                         do_tx_balance = false;
1440                         break;
1441                 }
1442
1443                 if (!pskb_network_may_pull(skb, sizeof(*ip6hdr))) {
1444                         do_tx_balance = false;
1445                         break;
1446                 }
1447                 /* Additionally, DAD probes should not be tx-balanced as that
1448                  * will lead to false positives for duplicate addresses and
1449                  * prevent address configuration from working.
1450                  */
1451                 ip6hdr = ipv6_hdr(skb);
1452                 if (ipv6_addr_any(&ip6hdr->saddr)) {
1453                         do_tx_balance = false;
1454                         break;
1455                 }
1456
1457                 hash_start = (char *)&ip6hdr->daddr;
1458                 hash_size = sizeof(ip6hdr->daddr);
1459                 break;
1460         }
1461         case ETH_P_IPX: {
1462                 const struct ipxhdr *ipxhdr;
1463
1464                 if (pskb_network_may_pull(skb, sizeof(*ipxhdr))) {
1465                         do_tx_balance = false;
1466                         break;
1467                 }
1468                 ipxhdr = (struct ipxhdr *)skb_network_header(skb);
1469
1470                 if (ipxhdr->ipx_checksum != IPX_NO_CHECKSUM) {
1471                         /* something is wrong with this packet */
1472                         do_tx_balance = false;
1473                         break;
1474                 }
1475
1476                 if (ipxhdr->ipx_type != IPX_TYPE_NCP) {
1477                         /* The only protocol worth balancing in
1478                          * this family since it has an "ARP" like
1479                          * mechanism
1480                          */
1481                         do_tx_balance = false;
1482                         break;
1483                 }
1484
1485                 eth_data = eth_hdr(skb);
1486                 hash_start = (char *)eth_data->h_dest;
1487                 hash_size = ETH_ALEN;
1488                 break;
1489         }
1490         case ETH_P_ARP:
1491                 do_tx_balance = false;
1492                 if (bond_info->rlb_enabled)
1493                         tx_slave = rlb_arp_xmit(skb, bond);
1494                 break;
1495         default:
1496                 do_tx_balance = false;
1497                 break;
1498         }
1499
1500         if (do_tx_balance) {
1501                 if (bond->params.tlb_dynamic_lb) {
1502                         hash_index = _simple_hash(hash_start, hash_size);
1503                         tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1504                 } else {
1505                         /*
1506                          * do_tx_balance means we are free to select the tx_slave
1507                          * So we do exactly what tlb would do for hash selection
1508                          */
1509
1510                         struct bond_up_slave *slaves;
1511                         unsigned int count;
1512
1513                         slaves = rcu_dereference(bond->slave_arr);
1514                         count = slaves ? READ_ONCE(slaves->count) : 0;
1515                         if (likely(count))
1516                                 tx_slave = slaves->arr[bond_xmit_hash(bond, skb) %
1517                                                        count];
1518                 }
1519         }
1520
1521         return bond_do_alb_xmit(skb, bond, tx_slave);
1522 }
1523
1524 void bond_alb_monitor(struct work_struct *work)
1525 {
1526         struct bonding *bond = container_of(work, struct bonding,
1527                                             alb_work.work);
1528         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1529         struct list_head *iter;
1530         struct slave *slave;
1531
1532         if (!bond_has_slaves(bond)) {
1533                 atomic_set(&bond_info->tx_rebalance_counter, 0);
1534                 bond_info->lp_counter = 0;
1535                 goto re_arm;
1536         }
1537
1538         rcu_read_lock();
1539
1540         atomic_inc(&bond_info->tx_rebalance_counter);
1541         bond_info->lp_counter++;
1542
1543         /* send learning packets */
1544         if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1545                 bool strict_match;
1546
1547                 bond_for_each_slave_rcu(bond, slave, iter) {
1548                         /* If updating current_active, use all currently
1549                          * user mac addreses (!strict_match).  Otherwise, only
1550                          * use mac of the slave device.
1551                          * In RLB mode, we always use strict matches.
1552                          */
1553                         strict_match = (slave != rcu_access_pointer(bond->curr_active_slave) ||
1554                                         bond_info->rlb_enabled);
1555                         alb_send_learning_packets(slave, slave->dev->dev_addr,
1556                                                   strict_match);
1557                 }
1558                 bond_info->lp_counter = 0;
1559         }
1560
1561         /* rebalance tx traffic */
1562         if (atomic_read(&bond_info->tx_rebalance_counter) >= BOND_TLB_REBALANCE_TICKS) {
1563                 bond_for_each_slave_rcu(bond, slave, iter) {
1564                         tlb_clear_slave(bond, slave, 1);
1565                         if (slave == rcu_access_pointer(bond->curr_active_slave)) {
1566                                 SLAVE_TLB_INFO(slave).load =
1567                                         bond_info->unbalanced_load /
1568                                                 BOND_TLB_REBALANCE_INTERVAL;
1569                                 bond_info->unbalanced_load = 0;
1570                         }
1571                 }
1572                 atomic_set(&bond_info->tx_rebalance_counter, 0);
1573         }
1574
1575         if (bond_info->rlb_enabled) {
1576                 if (bond_info->primary_is_promisc &&
1577                     (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1578
1579                         /* dev_set_promiscuity requires rtnl and
1580                          * nothing else.  Avoid race with bond_close.
1581                          */
1582                         rcu_read_unlock();
1583                         if (!rtnl_trylock())
1584                                 goto re_arm;
1585
1586                         bond_info->rlb_promisc_timeout_counter = 0;
1587
1588                         /* If the primary was set to promiscuous mode
1589                          * because a slave was disabled then
1590                          * it can now leave promiscuous mode.
1591                          */
1592                         dev_set_promiscuity(rtnl_dereference(bond->curr_active_slave)->dev,
1593                                             -1);
1594                         bond_info->primary_is_promisc = 0;
1595
1596                         rtnl_unlock();
1597                         rcu_read_lock();
1598                 }
1599
1600                 if (bond_info->rlb_rebalance) {
1601                         bond_info->rlb_rebalance = 0;
1602                         rlb_rebalance(bond);
1603                 }
1604
1605                 /* check if clients need updating */
1606                 if (bond_info->rx_ntt) {
1607                         if (bond_info->rlb_update_delay_counter) {
1608                                 --bond_info->rlb_update_delay_counter;
1609                         } else {
1610                                 rlb_update_rx_clients(bond);
1611                                 if (bond_info->rlb_update_retry_counter)
1612                                         --bond_info->rlb_update_retry_counter;
1613                                 else
1614                                         bond_info->rx_ntt = 0;
1615                         }
1616                 }
1617         }
1618         rcu_read_unlock();
1619 re_arm:
1620         queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1621 }
1622
1623 /* assumption: called before the slave is attached to the bond
1624  * and not locked by the bond lock
1625  */
1626 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1627 {
1628         int res;
1629
1630         res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1631                                      slave->dev->addr_len);
1632         if (res)
1633                 return res;
1634
1635         res = alb_handle_addr_collision_on_attach(bond, slave);
1636         if (res)
1637                 return res;
1638
1639         tlb_init_slave(slave);
1640
1641         /* order a rebalance ASAP */
1642         atomic_set(&bond->alb_info.tx_rebalance_counter,
1643                    BOND_TLB_REBALANCE_TICKS);
1644
1645         if (bond->alb_info.rlb_enabled)
1646                 bond->alb_info.rlb_rebalance = 1;
1647
1648         return 0;
1649 }
1650
1651 /* Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1652  * if necessary.
1653  *
1654  * Caller must hold RTNL and no other locks
1655  */
1656 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1657 {
1658         if (bond_has_slaves(bond))
1659                 alb_change_hw_addr_on_detach(bond, slave);
1660
1661         tlb_clear_slave(bond, slave, 0);
1662
1663         if (bond->alb_info.rlb_enabled) {
1664                 bond->alb_info.rx_slave = NULL;
1665                 rlb_clear_slave(bond, slave);
1666         }
1667
1668 }
1669
1670 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1671 {
1672         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1673
1674         if (link == BOND_LINK_DOWN) {
1675                 tlb_clear_slave(bond, slave, 0);
1676                 if (bond->alb_info.rlb_enabled)
1677                         rlb_clear_slave(bond, slave);
1678         } else if (link == BOND_LINK_UP) {
1679                 /* order a rebalance ASAP */
1680                 atomic_set(&bond_info->tx_rebalance_counter,
1681                            BOND_TLB_REBALANCE_TICKS);
1682                 if (bond->alb_info.rlb_enabled) {
1683                         bond->alb_info.rlb_rebalance = 1;
1684                         /* If the updelay module parameter is smaller than the
1685                          * forwarding delay of the switch the rebalance will
1686                          * not work because the rebalance arp replies will
1687                          * not be forwarded to the clients..
1688                          */
1689                 }
1690         }
1691
1692         if (bond_is_nondyn_tlb(bond)) {
1693                 if (bond_update_slave_arr(bond, NULL))
1694                         pr_err("Failed to build slave-array for TLB mode.\n");
1695         }
1696 }
1697
1698 /**
1699  * bond_alb_handle_active_change - assign new curr_active_slave
1700  * @bond: our bonding struct
1701  * @new_slave: new slave to assign
1702  *
1703  * Set the bond->curr_active_slave to @new_slave and handle
1704  * mac address swapping and promiscuity changes as needed.
1705  *
1706  * Caller must hold RTNL
1707  */
1708 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1709 {
1710         struct slave *swap_slave;
1711         struct slave *curr_active;
1712
1713         curr_active = rtnl_dereference(bond->curr_active_slave);
1714         if (curr_active == new_slave)
1715                 return;
1716
1717         if (curr_active && bond->alb_info.primary_is_promisc) {
1718                 dev_set_promiscuity(curr_active->dev, -1);
1719                 bond->alb_info.primary_is_promisc = 0;
1720                 bond->alb_info.rlb_promisc_timeout_counter = 0;
1721         }
1722
1723         swap_slave = curr_active;
1724         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1725
1726         if (!new_slave || !bond_has_slaves(bond))
1727                 return;
1728
1729         /* set the new curr_active_slave to the bonds mac address
1730          * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1731          */
1732         if (!swap_slave)
1733                 swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1734
1735         /* Arrange for swap_slave and new_slave to temporarily be
1736          * ignored so we can mess with their MAC addresses without
1737          * fear of interference from transmit activity.
1738          */
1739         if (swap_slave)
1740                 tlb_clear_slave(bond, swap_slave, 1);
1741         tlb_clear_slave(bond, new_slave, 1);
1742
1743         /* in TLB mode, the slave might flip down/up with the old dev_addr,
1744          * and thus filter bond->dev_addr's packets, so force bond's mac
1745          */
1746         if (BOND_MODE(bond) == BOND_MODE_TLB) {
1747                 struct sockaddr_storage ss;
1748                 u8 tmp_addr[MAX_ADDR_LEN];
1749
1750                 bond_hw_addr_copy(tmp_addr, new_slave->dev->dev_addr,
1751                                   new_slave->dev->addr_len);
1752
1753                 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
1754                                   bond->dev->addr_len);
1755                 ss.ss_family = bond->dev->type;
1756                 /* we don't care if it can't change its mac, best effort */
1757                 dev_set_mac_address(new_slave->dev, (struct sockaddr *)&ss);
1758
1759                 bond_hw_addr_copy(new_slave->dev->dev_addr, tmp_addr,
1760                                   new_slave->dev->addr_len);
1761         }
1762
1763         /* curr_active_slave must be set before calling alb_swap_mac_addr */
1764         if (swap_slave) {
1765                 /* swap mac address */
1766                 alb_swap_mac_addr(swap_slave, new_slave);
1767                 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1768         } else {
1769                 /* set the new_slave to the bond mac address */
1770                 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1771                                        bond->dev->addr_len);
1772                 alb_send_learning_packets(new_slave, bond->dev->dev_addr,
1773                                           false);
1774         }
1775 }
1776
1777 /* Called with RTNL */
1778 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1779 {
1780         struct bonding *bond = netdev_priv(bond_dev);
1781         struct sockaddr_storage *ss = addr;
1782         struct slave *curr_active;
1783         struct slave *swap_slave;
1784         int res;
1785
1786         if (!is_valid_ether_addr(ss->__data))
1787                 return -EADDRNOTAVAIL;
1788
1789         res = alb_set_mac_address(bond, addr);
1790         if (res)
1791                 return res;
1792
1793         bond_hw_addr_copy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
1794
1795         /* If there is no curr_active_slave there is nothing else to do.
1796          * Otherwise we'll need to pass the new address to it and handle
1797          * duplications.
1798          */
1799         curr_active = rtnl_dereference(bond->curr_active_slave);
1800         if (!curr_active)
1801                 return 0;
1802
1803         swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1804
1805         if (swap_slave) {
1806                 alb_swap_mac_addr(swap_slave, curr_active);
1807                 alb_fasten_mac_swap(bond, swap_slave, curr_active);
1808         } else {
1809                 alb_set_slave_mac_addr(curr_active, bond_dev->dev_addr,
1810                                        bond_dev->addr_len);
1811
1812                 alb_send_learning_packets(curr_active,
1813                                           bond_dev->dev_addr, false);
1814                 if (bond->alb_info.rlb_enabled) {
1815                         /* inform clients mac address has changed */
1816                         rlb_req_update_slave_clients(bond, curr_active);
1817                 }
1818         }
1819
1820         return 0;
1821 }
1822
1823 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1824 {
1825         if (bond->alb_info.rlb_enabled)
1826                 rlb_clear_vlan(bond, vlan_id);
1827 }
1828