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