linux/drivers/net/bonding/bond_alb.c
<<
>>
Prefs
   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
  45static const u8 mac_bcast[ETH_ALEN + 2] __long_aligned = {
  46        0xff, 0xff, 0xff, 0xff, 0xff, 0xff
  47};
  48static const u8 mac_v6_allmcast[ETH_ALEN + 2] __long_aligned = {
  49        0x33, 0x33, 0x00, 0x00, 0x00, 0x01
  50};
  51static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
  52
  53#pragma pack(1)
  54struct 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
  61struct 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
  74static 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 */
  80static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
  81                                      bool strict_match);
  82static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp);
  83static void rlb_src_unlink(struct bonding *bond, u32 index);
  84static void rlb_src_link(struct bonding *bond, u32 ip_src_hash,
  85                         u32 ip_dst_hash);
  86
  87static 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
 100static 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
 113static 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
 119static 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
 141static 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 */
 150static 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 */
 174static 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
 186static 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
 192static 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
 216static 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
 252static 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 */
 273static 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
 297static 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        }
 326out:
 327        return RX_HANDLER_ANOTHER;
 328}
 329
 330/* Caller must hold rcu_read_lock() */
 331static 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 */
 364static 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 */
 382static 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 */
 405static 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
 449static 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 */
 484static 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 */
 512static 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 */
 544static 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
 577static 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 */
 675static 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                        ether_addr_copy(arp->mac_src, tx_slave->dev->dev_addr);
 691                netdev_dbg(bond->dev, "Server sent ARP Reply packet\n");
 692        } else if (arp->op_code == htons(ARPOP_REQUEST)) {
 693                /* Create an entry in the rx_hashtbl for this client as a
 694                 * place holder.
 695                 * When the arp reply is received the entry will be updated
 696                 * with the correct unicast address of the client.
 697                 */
 698                rlb_choose_channel(skb, bond);
 699
 700                /* The ARP reply packets must be delayed so that
 701                 * they can cancel out the influence of the ARP request.
 702                 */
 703                bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 704
 705                /* arp requests are broadcast and are sent on the primary
 706                 * the arp request will collapse all clients on the subnet to
 707                 * the primary slave. We must register these clients to be
 708                 * updated with their assigned mac.
 709                 */
 710                rlb_req_update_subnet_clients(bond, arp->ip_src);
 711                netdev_dbg(bond->dev, "Server sent ARP Request packet\n");
 712        }
 713
 714        return tx_slave;
 715}
 716
 717static void rlb_rebalance(struct bonding *bond)
 718{
 719        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 720        struct slave *assigned_slave;
 721        struct rlb_client_info *client_info;
 722        int ntt;
 723        u32 hash_index;
 724
 725        spin_lock_bh(&bond->mode_lock);
 726
 727        ntt = 0;
 728        hash_index = bond_info->rx_hashtbl_used_head;
 729        for (; hash_index != RLB_NULL_INDEX;
 730             hash_index = client_info->used_next) {
 731                client_info = &(bond_info->rx_hashtbl[hash_index]);
 732                assigned_slave = __rlb_next_rx_slave(bond);
 733                if (assigned_slave && (client_info->slave != assigned_slave)) {
 734                        client_info->slave = assigned_slave;
 735                        client_info->ntt = 1;
 736                        ntt = 1;
 737                }
 738        }
 739
 740        /* update the team's flag only after the whole iteration */
 741        if (ntt)
 742                bond_info->rx_ntt = 1;
 743        spin_unlock_bh(&bond->mode_lock);
 744}
 745
 746/* Caller must hold mode_lock */
 747static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
 748{
 749        entry->used_next = RLB_NULL_INDEX;
 750        entry->used_prev = RLB_NULL_INDEX;
 751        entry->assigned = 0;
 752        entry->slave = NULL;
 753        entry->vlan_id = 0;
 754}
 755static void rlb_init_table_entry_src(struct rlb_client_info *entry)
 756{
 757        entry->src_first = RLB_NULL_INDEX;
 758        entry->src_prev = RLB_NULL_INDEX;
 759        entry->src_next = RLB_NULL_INDEX;
 760}
 761
 762static void rlb_init_table_entry(struct rlb_client_info *entry)
 763{
 764        memset(entry, 0, sizeof(struct rlb_client_info));
 765        rlb_init_table_entry_dst(entry);
 766        rlb_init_table_entry_src(entry);
 767}
 768
 769static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
 770{
 771        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 772        u32 next_index = bond_info->rx_hashtbl[index].used_next;
 773        u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
 774
 775        if (index == bond_info->rx_hashtbl_used_head)
 776                bond_info->rx_hashtbl_used_head = next_index;
 777        if (prev_index != RLB_NULL_INDEX)
 778                bond_info->rx_hashtbl[prev_index].used_next = next_index;
 779        if (next_index != RLB_NULL_INDEX)
 780                bond_info->rx_hashtbl[next_index].used_prev = prev_index;
 781}
 782
 783/* unlink a rlb hash table entry from the src list */
 784static void rlb_src_unlink(struct bonding *bond, u32 index)
 785{
 786        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 787        u32 next_index = bond_info->rx_hashtbl[index].src_next;
 788        u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
 789
 790        bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
 791        bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
 792
 793        if (next_index != RLB_NULL_INDEX)
 794                bond_info->rx_hashtbl[next_index].src_prev = prev_index;
 795
 796        if (prev_index == RLB_NULL_INDEX)
 797                return;
 798
 799        /* is prev_index pointing to the head of this list? */
 800        if (bond_info->rx_hashtbl[prev_index].src_first == index)
 801                bond_info->rx_hashtbl[prev_index].src_first = next_index;
 802        else
 803                bond_info->rx_hashtbl[prev_index].src_next = next_index;
 804
 805}
 806
 807static void rlb_delete_table_entry(struct bonding *bond, u32 index)
 808{
 809        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 810        struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 811
 812        rlb_delete_table_entry_dst(bond, index);
 813        rlb_init_table_entry_dst(entry);
 814
 815        rlb_src_unlink(bond, index);
 816}
 817
 818/* add the rx_hashtbl[ip_dst_hash] entry to the list
 819 * of entries with identical ip_src_hash
 820 */
 821static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
 822{
 823        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 824        u32 next;
 825
 826        bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
 827        next = bond_info->rx_hashtbl[ip_src_hash].src_first;
 828        bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
 829        if (next != RLB_NULL_INDEX)
 830                bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
 831        bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
 832}
 833
 834/* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does
 835 * not match arp->mac_src
 836 */
 837static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
 838{
 839        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 840        u32 ip_src_hash = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
 841        u32 index;
 842
 843        spin_lock_bh(&bond->mode_lock);
 844
 845        index = bond_info->rx_hashtbl[ip_src_hash].src_first;
 846        while (index != RLB_NULL_INDEX) {
 847                struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 848                u32 next_index = entry->src_next;
 849                if (entry->ip_src == arp->ip_src &&
 850                    !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
 851                                rlb_delete_table_entry(bond, index);
 852                index = next_index;
 853        }
 854        spin_unlock_bh(&bond->mode_lock);
 855}
 856
 857static int rlb_initialize(struct bonding *bond)
 858{
 859        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 860        struct rlb_client_info  *new_hashtbl;
 861        int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 862        int i;
 863
 864        new_hashtbl = kmalloc(size, GFP_KERNEL);
 865        if (!new_hashtbl)
 866                return -1;
 867
 868        spin_lock_bh(&bond->mode_lock);
 869
 870        bond_info->rx_hashtbl = new_hashtbl;
 871
 872        bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 873
 874        for (i = 0; i < RLB_HASH_TABLE_SIZE; i++)
 875                rlb_init_table_entry(bond_info->rx_hashtbl + i);
 876
 877        spin_unlock_bh(&bond->mode_lock);
 878
 879        /* register to receive ARPs */
 880        bond->recv_probe = rlb_arp_recv;
 881
 882        return 0;
 883}
 884
 885static void rlb_deinitialize(struct bonding *bond)
 886{
 887        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 888
 889        spin_lock_bh(&bond->mode_lock);
 890
 891        kfree(bond_info->rx_hashtbl);
 892        bond_info->rx_hashtbl = NULL;
 893        bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 894
 895        spin_unlock_bh(&bond->mode_lock);
 896}
 897
 898static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 899{
 900        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 901        u32 curr_index;
 902
 903        spin_lock_bh(&bond->mode_lock);
 904
 905        curr_index = bond_info->rx_hashtbl_used_head;
 906        while (curr_index != RLB_NULL_INDEX) {
 907                struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 908                u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
 909
 910                if (curr->vlan_id == vlan_id)
 911                        rlb_delete_table_entry(bond, curr_index);
 912
 913                curr_index = next_index;
 914        }
 915
 916        spin_unlock_bh(&bond->mode_lock);
 917}
 918
 919/*********************** tlb/rlb shared functions *********************/
 920
 921static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
 922                            __be16 vlan_proto, u16 vid)
 923{
 924        struct learning_pkt pkt;
 925        struct sk_buff *skb;
 926        int size = sizeof(struct learning_pkt);
 927        char *data;
 928
 929        memset(&pkt, 0, size);
 930        ether_addr_copy(pkt.mac_dst, mac_addr);
 931        ether_addr_copy(pkt.mac_src, mac_addr);
 932        pkt.type = cpu_to_be16(ETH_P_LOOPBACK);
 933
 934        skb = dev_alloc_skb(size);
 935        if (!skb)
 936                return;
 937
 938        data = skb_put(skb, size);
 939        memcpy(data, &pkt, size);
 940
 941        skb_reset_mac_header(skb);
 942        skb->network_header = skb->mac_header + ETH_HLEN;
 943        skb->protocol = pkt.type;
 944        skb->priority = TC_PRIO_CONTROL;
 945        skb->dev = slave->dev;
 946
 947        if (vid)
 948                __vlan_hwaccel_put_tag(skb, vlan_proto, vid);
 949
 950        dev_queue_xmit(skb);
 951}
 952
 953static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
 954                                      bool strict_match)
 955{
 956        struct bonding *bond = bond_get_bond_by_slave(slave);
 957        struct net_device *upper;
 958        struct list_head *iter;
 959        struct bond_vlan_tag *tags;
 960
 961        /* send untagged */
 962        alb_send_lp_vid(slave, mac_addr, 0, 0);
 963
 964        /* loop through all devices and see if we need to send a packet
 965         * for that device.
 966         */
 967        rcu_read_lock();
 968        netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
 969                if (is_vlan_dev(upper) && vlan_get_encap_level(upper) == 0) {
 970                        if (strict_match &&
 971                            ether_addr_equal_64bits(mac_addr,
 972                                                    upper->dev_addr)) {
 973                                alb_send_lp_vid(slave, mac_addr,
 974                                                vlan_dev_vlan_proto(upper),
 975                                                vlan_dev_vlan_id(upper));
 976                        } else if (!strict_match) {
 977                                alb_send_lp_vid(slave, upper->dev_addr,
 978                                                vlan_dev_vlan_proto(upper),
 979                                                vlan_dev_vlan_id(upper));
 980                        }
 981                }
 982
 983                /* If this is a macvlan device, then only send updates
 984                 * when strict_match is turned off.
 985                 */
 986                if (netif_is_macvlan(upper) && !strict_match) {
 987                        tags = bond_verify_device_path(bond->dev, upper, 0);
 988                        if (IS_ERR_OR_NULL(tags))
 989                                BUG();
 990                        alb_send_lp_vid(slave, upper->dev_addr,
 991                                        tags[0].vlan_proto, tags[0].vlan_id);
 992                        kfree(tags);
 993                }
 994        }
 995        rcu_read_unlock();
 996}
 997
 998static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[])
 999{
1000        struct net_device *dev = slave->dev;
1001        struct sockaddr s_addr;
1002
1003        if (BOND_MODE(slave->bond) == BOND_MODE_TLB) {
1004                memcpy(dev->dev_addr, addr, dev->addr_len);
1005                return 0;
1006        }
1007
1008        /* for rlb each slave must have a unique hw mac addresses so that
1009         * each slave will receive packets destined to a different mac
1010         */
1011        memcpy(s_addr.sa_data, addr, dev->addr_len);
1012        s_addr.sa_family = dev->type;
1013        if (dev_set_mac_address(dev, &s_addr)) {
1014                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",
1015                           dev->name);
1016                return -EOPNOTSUPP;
1017        }
1018        return 0;
1019}
1020
1021/* Swap MAC addresses between two slaves.
1022 *
1023 * Called with RTNL held, and no other locks.
1024 */
1025static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1026{
1027        u8 tmp_mac_addr[ETH_ALEN];
1028
1029        ether_addr_copy(tmp_mac_addr, slave1->dev->dev_addr);
1030        alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr);
1031        alb_set_slave_mac_addr(slave2, tmp_mac_addr);
1032
1033}
1034
1035/* Send learning packets after MAC address swap.
1036 *
1037 * Called with RTNL and no other locks
1038 */
1039static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1040                                struct slave *slave2)
1041{
1042        int slaves_state_differ = (bond_slave_can_tx(slave1) != bond_slave_can_tx(slave2));
1043        struct slave *disabled_slave = NULL;
1044
1045        ASSERT_RTNL();
1046
1047        /* fasten the change in the switch */
1048        if (bond_slave_can_tx(slave1)) {
1049                alb_send_learning_packets(slave1, slave1->dev->dev_addr, false);
1050                if (bond->alb_info.rlb_enabled) {
1051                        /* inform the clients that the mac address
1052                         * has changed
1053                         */
1054                        rlb_req_update_slave_clients(bond, slave1);
1055                }
1056        } else {
1057                disabled_slave = slave1;
1058        }
1059
1060        if (bond_slave_can_tx(slave2)) {
1061                alb_send_learning_packets(slave2, slave2->dev->dev_addr, false);
1062                if (bond->alb_info.rlb_enabled) {
1063                        /* inform the clients that the mac address
1064                         * has changed
1065                         */
1066                        rlb_req_update_slave_clients(bond, slave2);
1067                }
1068        } else {
1069                disabled_slave = slave2;
1070        }
1071
1072        if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1073                /* A disabled slave was assigned an active mac addr */
1074                rlb_teach_disabled_mac_on_primary(bond,
1075                                                  disabled_slave->dev->dev_addr);
1076        }
1077}
1078
1079/**
1080 * alb_change_hw_addr_on_detach
1081 * @bond: bonding we're working on
1082 * @slave: the slave that was just detached
1083 *
1084 * We assume that @slave was already detached from the slave list.
1085 *
1086 * If @slave's permanent hw address is different both from its current
1087 * address and from @bond's address, then somewhere in the bond there's
1088 * a slave that has @slave's permanet address as its current address.
1089 * We'll make sure that that slave no longer uses @slave's permanent address.
1090 *
1091 * Caller must hold RTNL and no other locks
1092 */
1093static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1094{
1095        int perm_curr_diff;
1096        int perm_bond_diff;
1097        struct slave *found_slave;
1098
1099        perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1100                                                  slave->dev->dev_addr);
1101        perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1102                                                  bond->dev->dev_addr);
1103
1104        if (perm_curr_diff && perm_bond_diff) {
1105                found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1106
1107                if (found_slave) {
1108                        alb_swap_mac_addr(slave, found_slave);
1109                        alb_fasten_mac_swap(bond, slave, found_slave);
1110                }
1111        }
1112}
1113
1114/**
1115 * alb_handle_addr_collision_on_attach
1116 * @bond: bonding we're working on
1117 * @slave: the slave that was just attached
1118 *
1119 * checks uniqueness of slave's mac address and handles the case the
1120 * new slave uses the bonds mac address.
1121 *
1122 * If the permanent hw address of @slave is @bond's hw address, we need to
1123 * find a different hw address to give @slave, that isn't in use by any other
1124 * slave in the bond. This address must be, of course, one of the permanent
1125 * addresses of the other slaves.
1126 *
1127 * We go over the slave list, and for each slave there we compare its
1128 * permanent hw address with the current address of all the other slaves.
1129 * If no match was found, then we've found a slave with a permanent address
1130 * that isn't used by any other slave in the bond, so we can assign it to
1131 * @slave.
1132 *
1133 * assumption: this function is called before @slave is attached to the
1134 *             bond slave list.
1135 */
1136static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1137{
1138        struct slave *has_bond_addr = rcu_access_pointer(bond->curr_active_slave);
1139        struct slave *tmp_slave1, *free_mac_slave = NULL;
1140        struct list_head *iter;
1141
1142        if (!bond_has_slaves(bond)) {
1143                /* this is the first slave */
1144                return 0;
1145        }
1146
1147        /* if slave's mac address differs from bond's mac address
1148         * check uniqueness of slave's mac address against the other
1149         * slaves in the bond.
1150         */
1151        if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1152                if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1153                        return 0;
1154
1155                /* Try setting slave mac to bond address and fall-through
1156                 * to code handling that situation below...
1157                 */
1158                alb_set_slave_mac_addr(slave, bond->dev->dev_addr);
1159        }
1160
1161        /* The slave's address is equal to the address of the bond.
1162         * Search for a spare address in the bond for this slave.
1163         */
1164        bond_for_each_slave(bond, tmp_slave1, iter) {
1165                if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1166                        /* no slave has tmp_slave1's perm addr
1167                         * as its curr addr
1168                         */
1169                        free_mac_slave = tmp_slave1;
1170                        break;
1171                }
1172
1173                if (!has_bond_addr) {
1174                        if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1175                                                    bond->dev->dev_addr)) {
1176
1177                                has_bond_addr = tmp_slave1;
1178                        }
1179                }
1180        }
1181
1182        if (free_mac_slave) {
1183                alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr);
1184
1185                netdev_warn(bond->dev, "the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1186                            slave->dev->name, free_mac_slave->dev->name);
1187
1188        } else if (has_bond_addr) {
1189                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",
1190                           slave->dev->name);
1191                return -EFAULT;
1192        }
1193
1194        return 0;
1195}
1196
1197/**
1198 * alb_set_mac_address
1199 * @bond:
1200 * @addr:
1201 *
1202 * In TLB mode all slaves are configured to the bond's hw address, but set
1203 * their dev_addr field to different addresses (based on their permanent hw
1204 * addresses).
1205 *
1206 * For each slave, this function sets the interface to the new address and then
1207 * changes its dev_addr field to its previous value.
1208 *
1209 * Unwinding assumes bond's mac address has not yet changed.
1210 */
1211static int alb_set_mac_address(struct bonding *bond, void *addr)
1212{
1213        struct slave *slave, *rollback_slave;
1214        struct list_head *iter;
1215        struct sockaddr sa;
1216        char tmp_addr[ETH_ALEN];
1217        int res;
1218
1219        if (bond->alb_info.rlb_enabled)
1220                return 0;
1221
1222        bond_for_each_slave(bond, slave, iter) {
1223                /* save net_device's current hw address */
1224                ether_addr_copy(tmp_addr, slave->dev->dev_addr);
1225
1226                res = dev_set_mac_address(slave->dev, addr);
1227
1228                /* restore net_device's hw address */
1229                ether_addr_copy(slave->dev->dev_addr, tmp_addr);
1230
1231                if (res)
1232                        goto unwind;
1233        }
1234
1235        return 0;
1236
1237unwind:
1238        memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1239        sa.sa_family = bond->dev->type;
1240
1241        /* unwind from head to the slave that failed */
1242        bond_for_each_slave(bond, rollback_slave, iter) {
1243                if (rollback_slave == slave)
1244                        break;
1245                ether_addr_copy(tmp_addr, rollback_slave->dev->dev_addr);
1246                dev_set_mac_address(rollback_slave->dev, &sa);
1247                ether_addr_copy(rollback_slave->dev->dev_addr, tmp_addr);
1248        }
1249
1250        return res;
1251}
1252
1253/************************ exported alb funcions ************************/
1254
1255int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1256{
1257        int res;
1258
1259        res = tlb_initialize(bond);
1260        if (res)
1261                return res;
1262
1263        if (rlb_enabled) {
1264                bond->alb_info.rlb_enabled = 1;
1265                res = rlb_initialize(bond);
1266                if (res) {
1267                        tlb_deinitialize(bond);
1268                        return res;
1269                }
1270        } else {
1271                bond->alb_info.rlb_enabled = 0;
1272        }
1273
1274        return 0;
1275}
1276
1277void bond_alb_deinitialize(struct bonding *bond)
1278{
1279        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1280
1281        tlb_deinitialize(bond);
1282
1283        if (bond_info->rlb_enabled)
1284                rlb_deinitialize(bond);
1285}
1286
1287static int bond_do_alb_xmit(struct sk_buff *skb, struct bonding *bond,
1288                            struct slave *tx_slave)
1289{
1290        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1291        struct ethhdr *eth_data = eth_hdr(skb);
1292
1293        if (!tx_slave) {
1294                /* unbalanced or unassigned, send through primary */
1295                tx_slave = rcu_dereference(bond->curr_active_slave);
1296                if (bond->params.tlb_dynamic_lb)
1297                        bond_info->unbalanced_load += skb->len;
1298        }
1299
1300        if (tx_slave && bond_slave_can_tx(tx_slave)) {
1301                if (tx_slave != rcu_access_pointer(bond->curr_active_slave)) {
1302                        ether_addr_copy(eth_data->h_source,
1303                                        tx_slave->dev->dev_addr);
1304                }
1305
1306                bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1307                goto out;
1308        }
1309
1310        if (tx_slave && bond->params.tlb_dynamic_lb) {
1311                spin_lock(&bond->mode_lock);
1312                __tlb_clear_slave(bond, tx_slave, 0);
1313                spin_unlock(&bond->mode_lock);
1314        }
1315
1316        /* no suitable interface, frame not sent */
1317        bond_tx_drop(bond->dev, skb);
1318out:
1319        return NETDEV_TX_OK;
1320}
1321
1322int bond_tlb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1323{
1324        struct bonding *bond = netdev_priv(bond_dev);
1325        struct ethhdr *eth_data;
1326        struct slave *tx_slave = NULL;
1327        u32 hash_index;
1328
1329        skb_reset_mac_header(skb);
1330        eth_data = eth_hdr(skb);
1331
1332        /* Do not TX balance any multicast or broadcast */
1333        if (!is_multicast_ether_addr(eth_data->h_dest)) {
1334                switch (skb->protocol) {
1335                case htons(ETH_P_IP):
1336                case htons(ETH_P_IPX):
1337                    /* In case of IPX, it will falback to L2 hash */
1338                case htons(ETH_P_IPV6):
1339                        hash_index = bond_xmit_hash(bond, skb);
1340                        if (bond->params.tlb_dynamic_lb) {
1341                                tx_slave = tlb_choose_channel(bond,
1342                                                              hash_index & 0xFF,
1343                                                              skb->len);
1344                        } else {
1345                                struct bond_up_slave *slaves;
1346                                unsigned int count;
1347
1348                                slaves = rcu_dereference(bond->slave_arr);
1349                                count = slaves ? ACCESS_ONCE(slaves->count) : 0;
1350                                if (likely(count))
1351                                        tx_slave = slaves->arr[hash_index %
1352                                                               count];
1353                        }
1354                        break;
1355                }
1356        }
1357        return bond_do_alb_xmit(skb, bond, tx_slave);
1358}
1359
1360int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1361{
1362        struct bonding *bond = netdev_priv(bond_dev);
1363        struct ethhdr *eth_data;
1364        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1365        struct slave *tx_slave = NULL;
1366        static const __be32 ip_bcast = htonl(0xffffffff);
1367        int hash_size = 0;
1368        bool do_tx_balance = true;
1369        u32 hash_index = 0;
1370        const u8 *hash_start = NULL;
1371        struct ipv6hdr *ip6hdr;
1372
1373        skb_reset_mac_header(skb);
1374        eth_data = eth_hdr(skb);
1375
1376        switch (ntohs(skb->protocol)) {
1377        case ETH_P_IP: {
1378                const struct iphdr *iph = ip_hdr(skb);
1379
1380                if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast) ||
1381                    (iph->daddr == ip_bcast) ||
1382                    (iph->protocol == IPPROTO_IGMP)) {
1383                        do_tx_balance = false;
1384                        break;
1385                }
1386                hash_start = (char *)&(iph->daddr);
1387                hash_size = sizeof(iph->daddr);
1388        }
1389                break;
1390        case ETH_P_IPV6:
1391                /* IPv6 doesn't really use broadcast mac address, but leave
1392                 * that here just in case.
1393                 */
1394                if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast)) {
1395                        do_tx_balance = false;
1396                        break;
1397                }
1398
1399                /* IPv6 uses all-nodes multicast as an equivalent to
1400                 * broadcasts in IPv4.
1401                 */
1402                if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1403                        do_tx_balance = false;
1404                        break;
1405                }
1406
1407                /* Additianally, DAD probes should not be tx-balanced as that
1408                 * will lead to false positives for duplicate addresses and
1409                 * prevent address configuration from working.
1410                 */
1411                ip6hdr = ipv6_hdr(skb);
1412                if (ipv6_addr_any(&ip6hdr->saddr)) {
1413                        do_tx_balance = false;
1414                        break;
1415                }
1416
1417                hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1418                hash_size = sizeof(ipv6_hdr(skb)->daddr);
1419                break;
1420        case ETH_P_IPX:
1421                if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1422                        /* something is wrong with this packet */
1423                        do_tx_balance = false;
1424                        break;
1425                }
1426
1427                if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1428                        /* The only protocol worth balancing in
1429                         * this family since it has an "ARP" like
1430                         * mechanism
1431                         */
1432                        do_tx_balance = false;
1433                        break;
1434                }
1435
1436                hash_start = (char *)eth_data->h_dest;
1437                hash_size = ETH_ALEN;
1438                break;
1439        case ETH_P_ARP:
1440                do_tx_balance = false;
1441                if (bond_info->rlb_enabled)
1442                        tx_slave = rlb_arp_xmit(skb, bond);
1443                break;
1444        default:
1445                do_tx_balance = false;
1446                break;
1447        }
1448
1449        if (do_tx_balance) {
1450                hash_index = _simple_hash(hash_start, hash_size);
1451                tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1452        }
1453
1454        return bond_do_alb_xmit(skb, bond, tx_slave);
1455}
1456
1457void bond_alb_monitor(struct work_struct *work)
1458{
1459        struct bonding *bond = container_of(work, struct bonding,
1460                                            alb_work.work);
1461        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1462        struct list_head *iter;
1463        struct slave *slave;
1464
1465        if (!bond_has_slaves(bond)) {
1466                bond_info->tx_rebalance_counter = 0;
1467                bond_info->lp_counter = 0;
1468                goto re_arm;
1469        }
1470
1471        rcu_read_lock();
1472
1473        bond_info->tx_rebalance_counter++;
1474        bond_info->lp_counter++;
1475
1476        /* send learning packets */
1477        if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1478                bool strict_match;
1479
1480                bond_for_each_slave_rcu(bond, slave, iter) {
1481                        /* If updating current_active, use all currently
1482                         * user mac addreses (!strict_match).  Otherwise, only
1483                         * use mac of the slave device.
1484                         * In RLB mode, we always use strict matches.
1485                         */
1486                        strict_match = (slave != rcu_access_pointer(bond->curr_active_slave) ||
1487                                        bond_info->rlb_enabled);
1488                        alb_send_learning_packets(slave, slave->dev->dev_addr,
1489                                                  strict_match);
1490                }
1491                bond_info->lp_counter = 0;
1492        }
1493
1494        /* rebalance tx traffic */
1495        if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1496                bond_for_each_slave_rcu(bond, slave, iter) {
1497                        tlb_clear_slave(bond, slave, 1);
1498                        if (slave == rcu_access_pointer(bond->curr_active_slave)) {
1499                                SLAVE_TLB_INFO(slave).load =
1500                                        bond_info->unbalanced_load /
1501                                                BOND_TLB_REBALANCE_INTERVAL;
1502                                bond_info->unbalanced_load = 0;
1503                        }
1504                }
1505                bond_info->tx_rebalance_counter = 0;
1506        }
1507
1508        if (bond_info->rlb_enabled) {
1509                if (bond_info->primary_is_promisc &&
1510                    (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1511
1512                        /* dev_set_promiscuity requires rtnl and
1513                         * nothing else.  Avoid race with bond_close.
1514                         */
1515                        rcu_read_unlock();
1516                        if (!rtnl_trylock())
1517                                goto re_arm;
1518
1519                        bond_info->rlb_promisc_timeout_counter = 0;
1520
1521                        /* If the primary was set to promiscuous mode
1522                         * because a slave was disabled then
1523                         * it can now leave promiscuous mode.
1524                         */
1525                        dev_set_promiscuity(rtnl_dereference(bond->curr_active_slave)->dev,
1526                                            -1);
1527                        bond_info->primary_is_promisc = 0;
1528
1529                        rtnl_unlock();
1530                        rcu_read_lock();
1531                }
1532
1533                if (bond_info->rlb_rebalance) {
1534                        bond_info->rlb_rebalance = 0;
1535                        rlb_rebalance(bond);
1536                }
1537
1538                /* check if clients need updating */
1539                if (bond_info->rx_ntt) {
1540                        if (bond_info->rlb_update_delay_counter) {
1541                                --bond_info->rlb_update_delay_counter;
1542                        } else {
1543                                rlb_update_rx_clients(bond);
1544                                if (bond_info->rlb_update_retry_counter)
1545                                        --bond_info->rlb_update_retry_counter;
1546                                else
1547                                        bond_info->rx_ntt = 0;
1548                        }
1549                }
1550        }
1551        rcu_read_unlock();
1552re_arm:
1553        queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1554}
1555
1556/* assumption: called before the slave is attached to the bond
1557 * and not locked by the bond lock
1558 */
1559int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1560{
1561        int res;
1562
1563        res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr);
1564        if (res)
1565                return res;
1566
1567        res = alb_handle_addr_collision_on_attach(bond, slave);
1568        if (res)
1569                return res;
1570
1571        tlb_init_slave(slave);
1572
1573        /* order a rebalance ASAP */
1574        bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1575
1576        if (bond->alb_info.rlb_enabled)
1577                bond->alb_info.rlb_rebalance = 1;
1578
1579        return 0;
1580}
1581
1582/* Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1583 * if necessary.
1584 *
1585 * Caller must hold RTNL and no other locks
1586 */
1587void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1588{
1589        if (bond_has_slaves(bond))
1590                alb_change_hw_addr_on_detach(bond, slave);
1591
1592        tlb_clear_slave(bond, slave, 0);
1593
1594        if (bond->alb_info.rlb_enabled) {
1595                bond->alb_info.rx_slave = NULL;
1596                rlb_clear_slave(bond, slave);
1597        }
1598
1599}
1600
1601void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1602{
1603        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1604
1605        if (link == BOND_LINK_DOWN) {
1606                tlb_clear_slave(bond, slave, 0);
1607                if (bond->alb_info.rlb_enabled)
1608                        rlb_clear_slave(bond, slave);
1609        } else if (link == BOND_LINK_UP) {
1610                /* order a rebalance ASAP */
1611                bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1612                if (bond->alb_info.rlb_enabled) {
1613                        bond->alb_info.rlb_rebalance = 1;
1614                        /* If the updelay module parameter is smaller than the
1615                         * forwarding delay of the switch the rebalance will
1616                         * not work because the rebalance arp replies will
1617                         * not be forwarded to the clients..
1618                         */
1619                }
1620        }
1621
1622        if (bond_is_nondyn_tlb(bond)) {
1623                if (bond_update_slave_arr(bond, NULL))
1624                        pr_err("Failed to build slave-array for TLB mode.\n");
1625        }
1626}
1627
1628/**
1629 * bond_alb_handle_active_change - assign new curr_active_slave
1630 * @bond: our bonding struct
1631 * @new_slave: new slave to assign
1632 *
1633 * Set the bond->curr_active_slave to @new_slave and handle
1634 * mac address swapping and promiscuity changes as needed.
1635 *
1636 * Caller must hold RTNL
1637 */
1638void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1639{
1640        struct slave *swap_slave;
1641        struct slave *curr_active;
1642
1643        curr_active = rtnl_dereference(bond->curr_active_slave);
1644        if (curr_active == new_slave)
1645                return;
1646
1647        if (curr_active && bond->alb_info.primary_is_promisc) {
1648                dev_set_promiscuity(curr_active->dev, -1);
1649                bond->alb_info.primary_is_promisc = 0;
1650                bond->alb_info.rlb_promisc_timeout_counter = 0;
1651        }
1652
1653        swap_slave = curr_active;
1654        rcu_assign_pointer(bond->curr_active_slave, new_slave);
1655
1656        if (!new_slave || !bond_has_slaves(bond))
1657                return;
1658
1659        /* set the new curr_active_slave to the bonds mac address
1660         * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1661         */
1662        if (!swap_slave)
1663                swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1664
1665        /* Arrange for swap_slave and new_slave to temporarily be
1666         * ignored so we can mess with their MAC addresses without
1667         * fear of interference from transmit activity.
1668         */
1669        if (swap_slave)
1670                tlb_clear_slave(bond, swap_slave, 1);
1671        tlb_clear_slave(bond, new_slave, 1);
1672
1673        /* in TLB mode, the slave might flip down/up with the old dev_addr,
1674         * and thus filter bond->dev_addr's packets, so force bond's mac
1675         */
1676        if (BOND_MODE(bond) == BOND_MODE_TLB) {
1677                struct sockaddr sa;
1678                u8 tmp_addr[ETH_ALEN];
1679
1680                ether_addr_copy(tmp_addr, new_slave->dev->dev_addr);
1681
1682                memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1683                sa.sa_family = bond->dev->type;
1684                /* we don't care if it can't change its mac, best effort */
1685                dev_set_mac_address(new_slave->dev, &sa);
1686
1687                ether_addr_copy(new_slave->dev->dev_addr, tmp_addr);
1688        }
1689
1690        /* curr_active_slave must be set before calling alb_swap_mac_addr */
1691        if (swap_slave) {
1692                /* swap mac address */
1693                alb_swap_mac_addr(swap_slave, new_slave);
1694                alb_fasten_mac_swap(bond, swap_slave, new_slave);
1695        } else {
1696                /* set the new_slave to the bond mac address */
1697                alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr);
1698                alb_send_learning_packets(new_slave, bond->dev->dev_addr,
1699                                          false);
1700        }
1701}
1702
1703/* Called with RTNL */
1704int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1705{
1706        struct bonding *bond = netdev_priv(bond_dev);
1707        struct sockaddr *sa = addr;
1708        struct slave *curr_active;
1709        struct slave *swap_slave;
1710        int res;
1711
1712        if (!is_valid_ether_addr(sa->sa_data))
1713                return -EADDRNOTAVAIL;
1714
1715        res = alb_set_mac_address(bond, addr);
1716        if (res)
1717                return res;
1718
1719        memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1720
1721        /* If there is no curr_active_slave there is nothing else to do.
1722         * Otherwise we'll need to pass the new address to it and handle
1723         * duplications.
1724         */
1725        curr_active = rtnl_dereference(bond->curr_active_slave);
1726        if (!curr_active)
1727                return 0;
1728
1729        swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1730
1731        if (swap_slave) {
1732                alb_swap_mac_addr(swap_slave, curr_active);
1733                alb_fasten_mac_swap(bond, swap_slave, curr_active);
1734        } else {
1735                alb_set_slave_mac_addr(curr_active, bond_dev->dev_addr);
1736
1737                alb_send_learning_packets(curr_active,
1738                                          bond_dev->dev_addr, false);
1739                if (bond->alb_info.rlb_enabled) {
1740                        /* inform clients mac address has changed */
1741                        rlb_req_update_slave_clients(bond, curr_active);
1742                }
1743        }
1744
1745        return 0;
1746}
1747
1748void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1749{
1750        if (bond->alb_info.rlb_enabled)
1751                rlb_clear_vlan(bond, vlan_id);
1752}
1753
1754