linux/drivers/net/bonding/bond_alb.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved.
   4 */
   5
   6#include <linux/skbuff.h>
   7#include <linux/netdevice.h>
   8#include <linux/etherdevice.h>
   9#include <linux/pkt_sched.h>
  10#include <linux/spinlock.h>
  11#include <linux/slab.h>
  12#include <linux/timer.h>
  13#include <linux/ip.h>
  14#include <linux/ipv6.h>
  15#include <linux/if_arp.h>
  16#include <linux/if_ether.h>
  17#include <linux/if_bonding.h>
  18#include <linux/if_vlan.h>
  19#include <linux/in.h>
  20#include <net/ipx.h>
  21#include <net/arp.h>
  22#include <net/ipv6.h>
  23#include <asm/byteorder.h>
  24#include <net/bonding.h>
  25#include <net/bond_alb.h>
  26
  27static const u8 mac_v6_allmcast[ETH_ALEN + 2] __long_aligned = {
  28        0x33, 0x33, 0x00, 0x00, 0x00, 0x01
  29};
  30static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
  31
  32#pragma pack(1)
  33struct learning_pkt {
  34        u8 mac_dst[ETH_ALEN];
  35        u8 mac_src[ETH_ALEN];
  36        __be16 type;
  37        u8 padding[ETH_ZLEN - ETH_HLEN];
  38};
  39
  40struct arp_pkt {
  41        __be16  hw_addr_space;
  42        __be16  prot_addr_space;
  43        u8      hw_addr_len;
  44        u8      prot_addr_len;
  45        __be16  op_code;
  46        u8      mac_src[ETH_ALEN];      /* sender hardware address */
  47        __be32  ip_src;                 /* sender IP address */
  48        u8      mac_dst[ETH_ALEN];      /* target hardware address */
  49        __be32  ip_dst;                 /* target IP address */
  50};
  51#pragma pack()
  52
  53static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb)
  54{
  55        return (struct arp_pkt *)skb_network_header(skb);
  56}
  57
  58/* Forward declaration */
  59static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
  60                                      bool strict_match);
  61static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp);
  62static void rlb_src_unlink(struct bonding *bond, u32 index);
  63static void rlb_src_link(struct bonding *bond, u32 ip_src_hash,
  64                         u32 ip_dst_hash);
  65
  66static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
  67{
  68        int i;
  69        u8 hash = 0;
  70
  71        for (i = 0; i < hash_size; i++)
  72                hash ^= hash_start[i];
  73
  74        return hash;
  75}
  76
  77/*********************** tlb specific functions ***************************/
  78
  79static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
  80{
  81        if (save_load) {
  82                entry->load_history = 1 + entry->tx_bytes /
  83                                      BOND_TLB_REBALANCE_INTERVAL;
  84                entry->tx_bytes = 0;
  85        }
  86
  87        entry->tx_slave = NULL;
  88        entry->next = TLB_NULL_INDEX;
  89        entry->prev = TLB_NULL_INDEX;
  90}
  91
  92static inline void tlb_init_slave(struct slave *slave)
  93{
  94        SLAVE_TLB_INFO(slave).load = 0;
  95        SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
  96}
  97
  98static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
  99                         int save_load)
 100{
 101        struct tlb_client_info *tx_hash_table;
 102        u32 index;
 103
 104        /* clear slave from tx_hashtbl */
 105        tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
 106
 107        /* skip this if we've already freed the tx hash table */
 108        if (tx_hash_table) {
 109                index = SLAVE_TLB_INFO(slave).head;
 110                while (index != TLB_NULL_INDEX) {
 111                        u32 next_index = tx_hash_table[index].next;
 112                        tlb_init_table_entry(&tx_hash_table[index], save_load);
 113                        index = next_index;
 114                }
 115        }
 116
 117        tlb_init_slave(slave);
 118}
 119
 120static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
 121                         int save_load)
 122{
 123        spin_lock_bh(&bond->mode_lock);
 124        __tlb_clear_slave(bond, slave, save_load);
 125        spin_unlock_bh(&bond->mode_lock);
 126}
 127
 128/* Must be called before starting the monitor timer */
 129static int tlb_initialize(struct bonding *bond)
 130{
 131        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 132        int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 133        struct tlb_client_info *new_hashtbl;
 134        int i;
 135
 136        new_hashtbl = kzalloc(size, GFP_KERNEL);
 137        if (!new_hashtbl)
 138                return -ENOMEM;
 139
 140        spin_lock_bh(&bond->mode_lock);
 141
 142        bond_info->tx_hashtbl = new_hashtbl;
 143
 144        for (i = 0; i < TLB_HASH_TABLE_SIZE; i++)
 145                tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
 146
 147        spin_unlock_bh(&bond->mode_lock);
 148
 149        return 0;
 150}
 151
 152/* Must be called only after all slaves have been released */
 153static void tlb_deinitialize(struct bonding *bond)
 154{
 155        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 156
 157        spin_lock_bh(&bond->mode_lock);
 158
 159        kfree(bond_info->tx_hashtbl);
 160        bond_info->tx_hashtbl = NULL;
 161
 162        spin_unlock_bh(&bond->mode_lock);
 163}
 164
 165static long long compute_gap(struct slave *slave)
 166{
 167        return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
 168               (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
 169}
 170
 171static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
 172{
 173        struct slave *slave, *least_loaded;
 174        struct list_head *iter;
 175        long long max_gap;
 176
 177        least_loaded = NULL;
 178        max_gap = LLONG_MIN;
 179
 180        /* Find the slave with the largest gap */
 181        bond_for_each_slave_rcu(bond, slave, iter) {
 182                if (bond_slave_can_tx(slave)) {
 183                        long long gap = compute_gap(slave);
 184
 185                        if (max_gap < gap) {
 186                                least_loaded = slave;
 187                                max_gap = gap;
 188                        }
 189                }
 190        }
 191
 192        return least_loaded;
 193}
 194
 195static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
 196                                                u32 skb_len)
 197{
 198        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 199        struct tlb_client_info *hash_table;
 200        struct slave *assigned_slave;
 201
 202        hash_table = bond_info->tx_hashtbl;
 203        assigned_slave = hash_table[hash_index].tx_slave;
 204        if (!assigned_slave) {
 205                assigned_slave = tlb_get_least_loaded_slave(bond);
 206
 207                if (assigned_slave) {
 208                        struct tlb_slave_info *slave_info =
 209                                &(SLAVE_TLB_INFO(assigned_slave));
 210                        u32 next_index = slave_info->head;
 211
 212                        hash_table[hash_index].tx_slave = assigned_slave;
 213                        hash_table[hash_index].next = next_index;
 214                        hash_table[hash_index].prev = TLB_NULL_INDEX;
 215
 216                        if (next_index != TLB_NULL_INDEX)
 217                                hash_table[next_index].prev = hash_index;
 218
 219                        slave_info->head = hash_index;
 220                        slave_info->load +=
 221                                hash_table[hash_index].load_history;
 222                }
 223        }
 224
 225        if (assigned_slave)
 226                hash_table[hash_index].tx_bytes += skb_len;
 227
 228        return assigned_slave;
 229}
 230
 231static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
 232                                        u32 skb_len)
 233{
 234        struct slave *tx_slave;
 235
 236        /* We don't need to disable softirq here, becase
 237         * tlb_choose_channel() is only called by bond_alb_xmit()
 238         * which already has softirq disabled.
 239         */
 240        spin_lock(&bond->mode_lock);
 241        tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
 242        spin_unlock(&bond->mode_lock);
 243
 244        return tx_slave;
 245}
 246
 247/*********************** rlb specific functions ***************************/
 248
 249/* when an ARP REPLY is received from a client update its info
 250 * in the rx_hashtbl
 251 */
 252static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 253{
 254        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 255        struct rlb_client_info *client_info;
 256        u32 hash_index;
 257
 258        spin_lock_bh(&bond->mode_lock);
 259
 260        hash_index = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
 261        client_info = &(bond_info->rx_hashtbl[hash_index]);
 262
 263        if ((client_info->assigned) &&
 264            (client_info->ip_src == arp->ip_dst) &&
 265            (client_info->ip_dst == arp->ip_src) &&
 266            (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) {
 267                /* update the clients MAC address */
 268                ether_addr_copy(client_info->mac_dst, arp->mac_src);
 269                client_info->ntt = 1;
 270                bond_info->rx_ntt = 1;
 271        }
 272
 273        spin_unlock_bh(&bond->mode_lock);
 274}
 275
 276static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
 277                        struct slave *slave)
 278{
 279        struct arp_pkt *arp, _arp;
 280
 281        if (skb->protocol != cpu_to_be16(ETH_P_ARP))
 282                goto out;
 283
 284        arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp);
 285        if (!arp)
 286                goto out;
 287
 288        /* We received an ARP from arp->ip_src.
 289         * We might have used this IP address previously (on the bonding host
 290         * itself or on a system that is bridged together with the bond).
 291         * However, if arp->mac_src is different than what is stored in
 292         * rx_hashtbl, some other host is now using the IP and we must prevent
 293         * sending out client updates with this IP address and the old MAC
 294         * address.
 295         * Clean up all hash table entries that have this address as ip_src but
 296         * have a different mac_src.
 297         */
 298        rlb_purge_src_ip(bond, arp);
 299
 300        if (arp->op_code == htons(ARPOP_REPLY)) {
 301                /* update rx hash table for this ARP */
 302                rlb_update_entry_from_arp(bond, arp);
 303                slave_dbg(bond->dev, slave->dev, "Server received an ARP Reply from client\n");
 304        }
 305out:
 306        return RX_HANDLER_ANOTHER;
 307}
 308
 309/* Caller must hold rcu_read_lock() */
 310static struct slave *__rlb_next_rx_slave(struct bonding *bond)
 311{
 312        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 313        struct slave *before = NULL, *rx_slave = NULL, *slave;
 314        struct list_head *iter;
 315        bool found = false;
 316
 317        bond_for_each_slave_rcu(bond, slave, iter) {
 318                if (!bond_slave_can_tx(slave))
 319                        continue;
 320                if (!found) {
 321                        if (!before || before->speed < slave->speed)
 322                                before = slave;
 323                } else {
 324                        if (!rx_slave || rx_slave->speed < slave->speed)
 325                                rx_slave = slave;
 326                }
 327                if (slave == bond_info->rx_slave)
 328                        found = true;
 329        }
 330        /* we didn't find anything after the current or we have something
 331         * better before and up to the current slave
 332         */
 333        if (!rx_slave || (before && rx_slave->speed < before->speed))
 334                rx_slave = before;
 335
 336        if (rx_slave)
 337                bond_info->rx_slave = rx_slave;
 338
 339        return rx_slave;
 340}
 341
 342/* Caller must hold RTNL, rcu_read_lock is obtained only to silence checkers */
 343static struct slave *rlb_next_rx_slave(struct bonding *bond)
 344{
 345        struct slave *rx_slave;
 346
 347        ASSERT_RTNL();
 348
 349        rcu_read_lock();
 350        rx_slave = __rlb_next_rx_slave(bond);
 351        rcu_read_unlock();
 352
 353        return rx_slave;
 354}
 355
 356/* teach the switch the mac of a disabled slave
 357 * on the primary for fault tolerance
 358 *
 359 * Caller must hold RTNL
 360 */
 361static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
 362{
 363        struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 364
 365        if (!curr_active)
 366                return;
 367
 368        if (!bond->alb_info.primary_is_promisc) {
 369                if (!dev_set_promiscuity(curr_active->dev, 1))
 370                        bond->alb_info.primary_is_promisc = 1;
 371                else
 372                        bond->alb_info.primary_is_promisc = 0;
 373        }
 374
 375        bond->alb_info.rlb_promisc_timeout_counter = 0;
 376
 377        alb_send_learning_packets(curr_active, addr, true);
 378}
 379
 380/* slave being removed should not be active at this point
 381 *
 382 * Caller must hold rtnl.
 383 */
 384static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 385{
 386        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 387        struct rlb_client_info *rx_hash_table;
 388        u32 index, next_index;
 389
 390        /* clear slave from rx_hashtbl */
 391        spin_lock_bh(&bond->mode_lock);
 392
 393        rx_hash_table = bond_info->rx_hashtbl;
 394        index = bond_info->rx_hashtbl_used_head;
 395        for (; index != RLB_NULL_INDEX; index = next_index) {
 396                next_index = rx_hash_table[index].used_next;
 397                if (rx_hash_table[index].slave == slave) {
 398                        struct slave *assigned_slave = rlb_next_rx_slave(bond);
 399
 400                        if (assigned_slave) {
 401                                rx_hash_table[index].slave = assigned_slave;
 402                                if (is_valid_ether_addr(rx_hash_table[index].mac_dst)) {
 403                                        bond_info->rx_hashtbl[index].ntt = 1;
 404                                        bond_info->rx_ntt = 1;
 405                                        /* A slave has been removed from the
 406                                         * table because it is either disabled
 407                                         * or being released. We must retry the
 408                                         * update to avoid clients from not
 409                                         * being updated & disconnecting when
 410                                         * there is stress
 411                                         */
 412                                        bond_info->rlb_update_retry_counter =
 413                                                RLB_UPDATE_RETRY;
 414                                }
 415                        } else {  /* there is no active slave */
 416                                rx_hash_table[index].slave = NULL;
 417                        }
 418                }
 419        }
 420
 421        spin_unlock_bh(&bond->mode_lock);
 422
 423        if (slave != rtnl_dereference(bond->curr_active_slave))
 424                rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 425}
 426
 427static void rlb_update_client(struct rlb_client_info *client_info)
 428{
 429        int i;
 430
 431        if (!client_info->slave || !is_valid_ether_addr(client_info->mac_dst))
 432                return;
 433
 434        for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 435                struct sk_buff *skb;
 436
 437                skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 438                                 client_info->ip_dst,
 439                                 client_info->slave->dev,
 440                                 client_info->ip_src,
 441                                 client_info->mac_dst,
 442                                 client_info->slave->dev->dev_addr,
 443                                 client_info->mac_dst);
 444                if (!skb) {
 445                        slave_err(client_info->slave->bond->dev,
 446                                  client_info->slave->dev,
 447                                  "failed to create an ARP packet\n");
 448                        continue;
 449                }
 450
 451                skb->dev = client_info->slave->dev;
 452
 453                if (client_info->vlan_id) {
 454                        __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 455                                               client_info->vlan_id);
 456                }
 457
 458                arp_xmit(skb);
 459        }
 460}
 461
 462/* sends ARP REPLIES that update the clients that need updating */
 463static void rlb_update_rx_clients(struct bonding *bond)
 464{
 465        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 466        struct rlb_client_info *client_info;
 467        u32 hash_index;
 468
 469        spin_lock_bh(&bond->mode_lock);
 470
 471        hash_index = bond_info->rx_hashtbl_used_head;
 472        for (; hash_index != RLB_NULL_INDEX;
 473             hash_index = client_info->used_next) {
 474                client_info = &(bond_info->rx_hashtbl[hash_index]);
 475                if (client_info->ntt) {
 476                        rlb_update_client(client_info);
 477                        if (bond_info->rlb_update_retry_counter == 0)
 478                                client_info->ntt = 0;
 479                }
 480        }
 481
 482        /* do not update the entries again until this counter is zero so that
 483         * not to confuse the clients.
 484         */
 485        bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 486
 487        spin_unlock_bh(&bond->mode_lock);
 488}
 489
 490/* The slave was assigned a new mac address - update the clients */
 491static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 492{
 493        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 494        struct rlb_client_info *client_info;
 495        int ntt = 0;
 496        u32 hash_index;
 497
 498        spin_lock_bh(&bond->mode_lock);
 499
 500        hash_index = bond_info->rx_hashtbl_used_head;
 501        for (; hash_index != RLB_NULL_INDEX;
 502             hash_index = client_info->used_next) {
 503                client_info = &(bond_info->rx_hashtbl[hash_index]);
 504
 505                if ((client_info->slave == slave) &&
 506                    is_valid_ether_addr(client_info->mac_dst)) {
 507                        client_info->ntt = 1;
 508                        ntt = 1;
 509                }
 510        }
 511
 512        /* update the team's flag only after the whole iteration */
 513        if (ntt) {
 514                bond_info->rx_ntt = 1;
 515                /* fasten the change */
 516                bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 517        }
 518
 519        spin_unlock_bh(&bond->mode_lock);
 520}
 521
 522/* mark all clients using src_ip to be updated */
 523static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
 524{
 525        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 526        struct rlb_client_info *client_info;
 527        u32 hash_index;
 528
 529        spin_lock(&bond->mode_lock);
 530
 531        hash_index = bond_info->rx_hashtbl_used_head;
 532        for (; hash_index != RLB_NULL_INDEX;
 533             hash_index = client_info->used_next) {
 534                client_info = &(bond_info->rx_hashtbl[hash_index]);
 535
 536                if (!client_info->slave) {
 537                        netdev_err(bond->dev, "found a client with no channel in the client's hash table\n");
 538                        continue;
 539                }
 540                /* update all clients using this src_ip, that are not assigned
 541                 * to the team's address (curr_active_slave) and have a known
 542                 * unicast mac address.
 543                 */
 544                if ((client_info->ip_src == src_ip) &&
 545                    !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
 546                                             bond->dev->dev_addr) &&
 547                    is_valid_ether_addr(client_info->mac_dst)) {
 548                        client_info->ntt = 1;
 549                        bond_info->rx_ntt = 1;
 550                }
 551        }
 552
 553        spin_unlock(&bond->mode_lock);
 554}
 555
 556static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
 557{
 558        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 559        struct arp_pkt *arp = arp_pkt(skb);
 560        struct slave *assigned_slave, *curr_active_slave;
 561        struct rlb_client_info *client_info;
 562        u32 hash_index = 0;
 563
 564        spin_lock(&bond->mode_lock);
 565
 566        curr_active_slave = rcu_dereference(bond->curr_active_slave);
 567
 568        hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
 569        client_info = &(bond_info->rx_hashtbl[hash_index]);
 570
 571        if (client_info->assigned) {
 572                if ((client_info->ip_src == arp->ip_src) &&
 573                    (client_info->ip_dst == arp->ip_dst)) {
 574                        /* the entry is already assigned to this client */
 575                        if (!is_broadcast_ether_addr(arp->mac_dst)) {
 576                                /* update mac address from arp */
 577                                ether_addr_copy(client_info->mac_dst, arp->mac_dst);
 578                        }
 579                        ether_addr_copy(client_info->mac_src, arp->mac_src);
 580
 581                        assigned_slave = client_info->slave;
 582                        if (assigned_slave) {
 583                                spin_unlock(&bond->mode_lock);
 584                                return assigned_slave;
 585                        }
 586                } else {
 587                        /* the entry is already assigned to some other client,
 588                         * move the old client to primary (curr_active_slave) so
 589                         * that the new client can be assigned to this entry.
 590                         */
 591                        if (curr_active_slave &&
 592                            client_info->slave != curr_active_slave) {
 593                                client_info->slave = curr_active_slave;
 594                                rlb_update_client(client_info);
 595                        }
 596                }
 597        }
 598        /* assign a new slave */
 599        assigned_slave = __rlb_next_rx_slave(bond);
 600
 601        if (assigned_slave) {
 602                if (!(client_info->assigned &&
 603                      client_info->ip_src == arp->ip_src)) {
 604                        /* ip_src is going to be updated,
 605                         * fix the src hash list
 606                         */
 607                        u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
 608                                                    sizeof(arp->ip_src));
 609                        rlb_src_unlink(bond, hash_index);
 610                        rlb_src_link(bond, hash_src, hash_index);
 611                }
 612
 613                client_info->ip_src = arp->ip_src;
 614                client_info->ip_dst = arp->ip_dst;
 615                /* arp->mac_dst is broadcast for arp reqeusts.
 616                 * will be updated with clients actual unicast mac address
 617                 * upon receiving an arp reply.
 618                 */
 619                ether_addr_copy(client_info->mac_dst, arp->mac_dst);
 620                ether_addr_copy(client_info->mac_src, arp->mac_src);
 621                client_info->slave = assigned_slave;
 622
 623                if (is_valid_ether_addr(client_info->mac_dst)) {
 624                        client_info->ntt = 1;
 625                        bond->alb_info.rx_ntt = 1;
 626                } else {
 627                        client_info->ntt = 0;
 628                }
 629
 630                if (vlan_get_tag(skb, &client_info->vlan_id))
 631                        client_info->vlan_id = 0;
 632
 633                if (!client_info->assigned) {
 634                        u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
 635                        bond_info->rx_hashtbl_used_head = hash_index;
 636                        client_info->used_next = prev_tbl_head;
 637                        if (prev_tbl_head != RLB_NULL_INDEX) {
 638                                bond_info->rx_hashtbl[prev_tbl_head].used_prev =
 639                                        hash_index;
 640                        }
 641                        client_info->assigned = 1;
 642                }
 643        }
 644
 645        spin_unlock(&bond->mode_lock);
 646
 647        return assigned_slave;
 648}
 649
 650/* chooses (and returns) transmit channel for arp reply
 651 * does not choose channel for other arp types since they are
 652 * sent on the curr_active_slave
 653 */
 654static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 655{
 656        struct arp_pkt *arp = arp_pkt(skb);
 657        struct slave *tx_slave = NULL;
 658
 659        /* Don't modify or load balance ARPs that do not originate locally
 660         * (e.g.,arrive via a bridge).
 661         */
 662        if (!bond_slave_has_mac_rx(bond, arp->mac_src))
 663                return NULL;
 664
 665        if (arp->op_code == htons(ARPOP_REPLY)) {
 666                /* the arp must be sent on the selected rx channel */
 667                tx_slave = rlb_choose_channel(skb, bond);
 668                if (tx_slave)
 669                        bond_hw_addr_copy(arp->mac_src, tx_slave->dev->dev_addr,
 670                                          tx_slave->dev->addr_len);
 671                netdev_dbg(bond->dev, "(slave %s): Server sent ARP Reply packet\n",
 672                           tx_slave ? tx_slave->dev->name : "NULL");
 673        } else if (arp->op_code == htons(ARPOP_REQUEST)) {
 674                /* Create an entry in the rx_hashtbl for this client as a
 675                 * place holder.
 676                 * When the arp reply is received the entry will be updated
 677                 * with the correct unicast address of the client.
 678                 */
 679                tx_slave = rlb_choose_channel(skb, bond);
 680
 681                /* The ARP reply packets must be delayed so that
 682                 * they can cancel out the influence of the ARP request.
 683                 */
 684                bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 685
 686                /* arp requests are broadcast and are sent on the primary
 687                 * the arp request will collapse all clients on the subnet to
 688                 * the primary slave. We must register these clients to be
 689                 * updated with their assigned mac.
 690                 */
 691                rlb_req_update_subnet_clients(bond, arp->ip_src);
 692                netdev_dbg(bond->dev, "(slave %s): Server sent ARP Request packet\n",
 693                           tx_slave ? tx_slave->dev->name : "NULL");
 694        }
 695
 696        return tx_slave;
 697}
 698
 699static void rlb_rebalance(struct bonding *bond)
 700{
 701        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 702        struct slave *assigned_slave;
 703        struct rlb_client_info *client_info;
 704        int ntt;
 705        u32 hash_index;
 706
 707        spin_lock_bh(&bond->mode_lock);
 708
 709        ntt = 0;
 710        hash_index = bond_info->rx_hashtbl_used_head;
 711        for (; hash_index != RLB_NULL_INDEX;
 712             hash_index = client_info->used_next) {
 713                client_info = &(bond_info->rx_hashtbl[hash_index]);
 714                assigned_slave = __rlb_next_rx_slave(bond);
 715                if (assigned_slave && (client_info->slave != assigned_slave)) {
 716                        client_info->slave = assigned_slave;
 717                        if (!is_zero_ether_addr(client_info->mac_dst)) {
 718                                client_info->ntt = 1;
 719                                ntt = 1;
 720                        }
 721                }
 722        }
 723
 724        /* update the team's flag only after the whole iteration */
 725        if (ntt)
 726                bond_info->rx_ntt = 1;
 727        spin_unlock_bh(&bond->mode_lock);
 728}
 729
 730/* Caller must hold mode_lock */
 731static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
 732{
 733        entry->used_next = RLB_NULL_INDEX;
 734        entry->used_prev = RLB_NULL_INDEX;
 735        entry->assigned = 0;
 736        entry->slave = NULL;
 737        entry->vlan_id = 0;
 738}
 739static void rlb_init_table_entry_src(struct rlb_client_info *entry)
 740{
 741        entry->src_first = RLB_NULL_INDEX;
 742        entry->src_prev = RLB_NULL_INDEX;
 743        entry->src_next = RLB_NULL_INDEX;
 744}
 745
 746static void rlb_init_table_entry(struct rlb_client_info *entry)
 747{
 748        memset(entry, 0, sizeof(struct rlb_client_info));
 749        rlb_init_table_entry_dst(entry);
 750        rlb_init_table_entry_src(entry);
 751}
 752
 753static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
 754{
 755        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 756        u32 next_index = bond_info->rx_hashtbl[index].used_next;
 757        u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
 758
 759        if (index == bond_info->rx_hashtbl_used_head)
 760                bond_info->rx_hashtbl_used_head = next_index;
 761        if (prev_index != RLB_NULL_INDEX)
 762                bond_info->rx_hashtbl[prev_index].used_next = next_index;
 763        if (next_index != RLB_NULL_INDEX)
 764                bond_info->rx_hashtbl[next_index].used_prev = prev_index;
 765}
 766
 767/* unlink a rlb hash table entry from the src list */
 768static void rlb_src_unlink(struct bonding *bond, u32 index)
 769{
 770        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 771        u32 next_index = bond_info->rx_hashtbl[index].src_next;
 772        u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
 773
 774        bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
 775        bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
 776
 777        if (next_index != RLB_NULL_INDEX)
 778                bond_info->rx_hashtbl[next_index].src_prev = prev_index;
 779
 780        if (prev_index == RLB_NULL_INDEX)
 781                return;
 782
 783        /* is prev_index pointing to the head of this list? */
 784        if (bond_info->rx_hashtbl[prev_index].src_first == index)
 785                bond_info->rx_hashtbl[prev_index].src_first = next_index;
 786        else
 787                bond_info->rx_hashtbl[prev_index].src_next = next_index;
 788
 789}
 790
 791static void rlb_delete_table_entry(struct bonding *bond, u32 index)
 792{
 793        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 794        struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 795
 796        rlb_delete_table_entry_dst(bond, index);
 797        rlb_init_table_entry_dst(entry);
 798
 799        rlb_src_unlink(bond, index);
 800}
 801
 802/* add the rx_hashtbl[ip_dst_hash] entry to the list
 803 * of entries with identical ip_src_hash
 804 */
 805static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
 806{
 807        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 808        u32 next;
 809
 810        bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
 811        next = bond_info->rx_hashtbl[ip_src_hash].src_first;
 812        bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
 813        if (next != RLB_NULL_INDEX)
 814                bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
 815        bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
 816}
 817
 818/* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does
 819 * not match arp->mac_src
 820 */
 821static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
 822{
 823        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 824        u32 ip_src_hash = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
 825        u32 index;
 826
 827        spin_lock_bh(&bond->mode_lock);
 828
 829        index = bond_info->rx_hashtbl[ip_src_hash].src_first;
 830        while (index != RLB_NULL_INDEX) {
 831                struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 832                u32 next_index = entry->src_next;
 833                if (entry->ip_src == arp->ip_src &&
 834                    !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
 835                                rlb_delete_table_entry(bond, index);
 836                index = next_index;
 837        }
 838        spin_unlock_bh(&bond->mode_lock);
 839}
 840
 841static int rlb_initialize(struct bonding *bond)
 842{
 843        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 844        struct rlb_client_info  *new_hashtbl;
 845        int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 846        int i;
 847
 848        new_hashtbl = kmalloc(size, GFP_KERNEL);
 849        if (!new_hashtbl)
 850                return -1;
 851
 852        spin_lock_bh(&bond->mode_lock);
 853
 854        bond_info->rx_hashtbl = new_hashtbl;
 855
 856        bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 857
 858        for (i = 0; i < RLB_HASH_TABLE_SIZE; i++)
 859                rlb_init_table_entry(bond_info->rx_hashtbl + i);
 860
 861        spin_unlock_bh(&bond->mode_lock);
 862
 863        /* register to receive ARPs */
 864        bond->recv_probe = rlb_arp_recv;
 865
 866        return 0;
 867}
 868
 869static void rlb_deinitialize(struct bonding *bond)
 870{
 871        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 872
 873        spin_lock_bh(&bond->mode_lock);
 874
 875        kfree(bond_info->rx_hashtbl);
 876        bond_info->rx_hashtbl = NULL;
 877        bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 878
 879        spin_unlock_bh(&bond->mode_lock);
 880}
 881
 882static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 883{
 884        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 885        u32 curr_index;
 886
 887        spin_lock_bh(&bond->mode_lock);
 888
 889        curr_index = bond_info->rx_hashtbl_used_head;
 890        while (curr_index != RLB_NULL_INDEX) {
 891                struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 892                u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
 893
 894                if (curr->vlan_id == vlan_id)
 895                        rlb_delete_table_entry(bond, curr_index);
 896
 897                curr_index = next_index;
 898        }
 899
 900        spin_unlock_bh(&bond->mode_lock);
 901}
 902
 903/*********************** tlb/rlb shared functions *********************/
 904
 905static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
 906                            __be16 vlan_proto, u16 vid)
 907{
 908        struct learning_pkt pkt;
 909        struct sk_buff *skb;
 910        int size = sizeof(struct learning_pkt);
 911
 912        memset(&pkt, 0, size);
 913        ether_addr_copy(pkt.mac_dst, mac_addr);
 914        ether_addr_copy(pkt.mac_src, mac_addr);
 915        pkt.type = cpu_to_be16(ETH_P_LOOPBACK);
 916
 917        skb = dev_alloc_skb(size);
 918        if (!skb)
 919                return;
 920
 921        skb_put_data(skb, &pkt, size);
 922
 923        skb_reset_mac_header(skb);
 924        skb->network_header = skb->mac_header + ETH_HLEN;
 925        skb->protocol = pkt.type;
 926        skb->priority = TC_PRIO_CONTROL;
 927        skb->dev = slave->dev;
 928
 929        slave_dbg(slave->bond->dev, slave->dev,
 930                  "Send learning packet: mac %pM vlan %d\n", mac_addr, vid);
 931
 932        if (vid)
 933                __vlan_hwaccel_put_tag(skb, vlan_proto, vid);
 934
 935        dev_queue_xmit(skb);
 936}
 937
 938struct alb_walk_data {
 939        struct bonding *bond;
 940        struct slave *slave;
 941        u8 *mac_addr;
 942        bool strict_match;
 943};
 944
 945static int alb_upper_dev_walk(struct net_device *upper, void *_data)
 946{
 947        struct alb_walk_data *data = _data;
 948        bool strict_match = data->strict_match;
 949        struct bonding *bond = data->bond;
 950        struct slave *slave = data->slave;
 951        u8 *mac_addr = data->mac_addr;
 952        struct bond_vlan_tag *tags;
 953
 954        if (is_vlan_dev(upper) &&
 955            bond->dev->lower_level == upper->lower_level - 1) {
 956                if (upper->addr_assign_type == NET_ADDR_STOLEN) {
 957                        alb_send_lp_vid(slave, mac_addr,
 958                                        vlan_dev_vlan_proto(upper),
 959                                        vlan_dev_vlan_id(upper));
 960                } else {
 961                        alb_send_lp_vid(slave, upper->dev_addr,
 962                                        vlan_dev_vlan_proto(upper),
 963                                        vlan_dev_vlan_id(upper));
 964                }
 965        }
 966
 967        /* If this is a macvlan device, then only send updates
 968         * when strict_match is turned off.
 969         */
 970        if (netif_is_macvlan(upper) && !strict_match) {
 971                tags = bond_verify_device_path(bond->dev, upper, 0);
 972                if (IS_ERR_OR_NULL(tags))
 973                        BUG();
 974                alb_send_lp_vid(slave, upper->dev_addr,
 975                                tags[0].vlan_proto, tags[0].vlan_id);
 976                kfree(tags);
 977        }
 978
 979        return 0;
 980}
 981
 982static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
 983                                      bool strict_match)
 984{
 985        struct bonding *bond = bond_get_bond_by_slave(slave);
 986        struct alb_walk_data data = {
 987                .strict_match = strict_match,
 988                .mac_addr = mac_addr,
 989                .slave = slave,
 990                .bond = bond,
 991        };
 992
 993        /* send untagged */
 994        alb_send_lp_vid(slave, mac_addr, 0, 0);
 995
 996        /* loop through all devices and see if we need to send a packet
 997         * for that device.
 998         */
 999        rcu_read_lock();
1000        netdev_walk_all_upper_dev_rcu(bond->dev, alb_upper_dev_walk, &data);
1001        rcu_read_unlock();
1002}
1003
1004static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[],
1005                                  unsigned int len)
1006{
1007        struct net_device *dev = slave->dev;
1008        struct sockaddr_storage ss;
1009
1010        if (BOND_MODE(slave->bond) == BOND_MODE_TLB) {
1011                memcpy(dev->dev_addr, addr, len);
1012                return 0;
1013        }
1014
1015        /* for rlb each slave must have a unique hw mac addresses so that
1016         * each slave will receive packets destined to a different mac
1017         */
1018        memcpy(ss.__data, addr, len);
1019        ss.ss_family = dev->type;
1020        if (dev_set_mac_address(dev, (struct sockaddr *)&ss, NULL)) {
1021                slave_err(slave->bond->dev, dev, "dev_set_mac_address on slave failed! ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n");
1022                return -EOPNOTSUPP;
1023        }
1024        return 0;
1025}
1026
1027/* Swap MAC addresses between two slaves.
1028 *
1029 * Called with RTNL held, and no other locks.
1030 */
1031static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1032{
1033        u8 tmp_mac_addr[MAX_ADDR_LEN];
1034
1035        bond_hw_addr_copy(tmp_mac_addr, slave1->dev->dev_addr,
1036                          slave1->dev->addr_len);
1037        alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr,
1038                               slave2->dev->addr_len);
1039        alb_set_slave_mac_addr(slave2, tmp_mac_addr,
1040                               slave1->dev->addr_len);
1041
1042}
1043
1044/* Send learning packets after MAC address swap.
1045 *
1046 * Called with RTNL and no other locks
1047 */
1048static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1049                                struct slave *slave2)
1050{
1051        int slaves_state_differ = (bond_slave_can_tx(slave1) != bond_slave_can_tx(slave2));
1052        struct slave *disabled_slave = NULL;
1053
1054        ASSERT_RTNL();
1055
1056        /* fasten the change in the switch */
1057        if (bond_slave_can_tx(slave1)) {
1058                alb_send_learning_packets(slave1, slave1->dev->dev_addr, false);
1059                if (bond->alb_info.rlb_enabled) {
1060                        /* inform the clients that the mac address
1061                         * has changed
1062                         */
1063                        rlb_req_update_slave_clients(bond, slave1);
1064                }
1065        } else {
1066                disabled_slave = slave1;
1067        }
1068
1069        if (bond_slave_can_tx(slave2)) {
1070                alb_send_learning_packets(slave2, slave2->dev->dev_addr, false);
1071                if (bond->alb_info.rlb_enabled) {
1072                        /* inform the clients that the mac address
1073                         * has changed
1074                         */
1075                        rlb_req_update_slave_clients(bond, slave2);
1076                }
1077        } else {
1078                disabled_slave = slave2;
1079        }
1080
1081        if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1082                /* A disabled slave was assigned an active mac addr */
1083                rlb_teach_disabled_mac_on_primary(bond,
1084                                                  disabled_slave->dev->dev_addr);
1085        }
1086}
1087
1088/**
1089 * alb_change_hw_addr_on_detach
1090 * @bond: bonding we're working on
1091 * @slave: the slave that was just detached
1092 *
1093 * We assume that @slave was already detached from the slave list.
1094 *
1095 * If @slave's permanent hw address is different both from its current
1096 * address and from @bond's address, then somewhere in the bond there's
1097 * a slave that has @slave's permanet address as its current address.
1098 * We'll make sure that that slave no longer uses @slave's permanent address.
1099 *
1100 * Caller must hold RTNL and no other locks
1101 */
1102static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1103{
1104        int perm_curr_diff;
1105        int perm_bond_diff;
1106        struct slave *found_slave;
1107
1108        perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1109                                                  slave->dev->dev_addr);
1110        perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1111                                                  bond->dev->dev_addr);
1112
1113        if (perm_curr_diff && perm_bond_diff) {
1114                found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1115
1116                if (found_slave) {
1117                        alb_swap_mac_addr(slave, found_slave);
1118                        alb_fasten_mac_swap(bond, slave, found_slave);
1119                }
1120        }
1121}
1122
1123/**
1124 * alb_handle_addr_collision_on_attach
1125 * @bond: bonding we're working on
1126 * @slave: the slave that was just attached
1127 *
1128 * checks uniqueness of slave's mac address and handles the case the
1129 * new slave uses the bonds mac address.
1130 *
1131 * If the permanent hw address of @slave is @bond's hw address, we need to
1132 * find a different hw address to give @slave, that isn't in use by any other
1133 * slave in the bond. This address must be, of course, one of the permanent
1134 * addresses of the other slaves.
1135 *
1136 * We go over the slave list, and for each slave there we compare its
1137 * permanent hw address with the current address of all the other slaves.
1138 * If no match was found, then we've found a slave with a permanent address
1139 * that isn't used by any other slave in the bond, so we can assign it to
1140 * @slave.
1141 *
1142 * assumption: this function is called before @slave is attached to the
1143 *             bond slave list.
1144 */
1145static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1146{
1147        struct slave *has_bond_addr = rcu_access_pointer(bond->curr_active_slave);
1148        struct slave *tmp_slave1, *free_mac_slave = NULL;
1149        struct list_head *iter;
1150
1151        if (!bond_has_slaves(bond)) {
1152                /* this is the first slave */
1153                return 0;
1154        }
1155
1156        /* if slave's mac address differs from bond's mac address
1157         * check uniqueness of slave's mac address against the other
1158         * slaves in the bond.
1159         */
1160        if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1161                if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1162                        return 0;
1163
1164                /* Try setting slave mac to bond address and fall-through
1165                 * to code handling that situation below...
1166                 */
1167                alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1168                                       bond->dev->addr_len);
1169        }
1170
1171        /* The slave's address is equal to the address of the bond.
1172         * Search for a spare address in the bond for this slave.
1173         */
1174        bond_for_each_slave(bond, tmp_slave1, iter) {
1175                if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1176                        /* no slave has tmp_slave1's perm addr
1177                         * as its curr addr
1178                         */
1179                        free_mac_slave = tmp_slave1;
1180                        break;
1181                }
1182
1183                if (!has_bond_addr) {
1184                        if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1185                                                    bond->dev->dev_addr)) {
1186
1187                                has_bond_addr = tmp_slave1;
1188                        }
1189                }
1190        }
1191
1192        if (free_mac_slave) {
1193                alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1194                                       free_mac_slave->dev->addr_len);
1195
1196                slave_warn(bond->dev, slave->dev, "the slave hw address is in use by the bond; giving it the hw address of %s\n",
1197                           free_mac_slave->dev->name);
1198
1199        } else if (has_bond_addr) {
1200                slave_err(bond->dev, slave->dev, "the slave hw address is in use by the bond; couldn't find a slave with a free hw address to give it (this should not have happened)\n");
1201                return -EFAULT;
1202        }
1203
1204        return 0;
1205}
1206
1207/**
1208 * alb_set_mac_address
1209 * @bond:
1210 * @addr:
1211 *
1212 * In TLB mode all slaves are configured to the bond's hw address, but set
1213 * their dev_addr field to different addresses (based on their permanent hw
1214 * addresses).
1215 *
1216 * For each slave, this function sets the interface to the new address and then
1217 * changes its dev_addr field to its previous value.
1218 *
1219 * Unwinding assumes bond's mac address has not yet changed.
1220 */
1221static int alb_set_mac_address(struct bonding *bond, void *addr)
1222{
1223        struct slave *slave, *rollback_slave;
1224        struct list_head *iter;
1225        struct sockaddr_storage ss;
1226        char tmp_addr[MAX_ADDR_LEN];
1227        int res;
1228
1229        if (bond->alb_info.rlb_enabled)
1230                return 0;
1231
1232        bond_for_each_slave(bond, slave, iter) {
1233                /* save net_device's current hw address */
1234                bond_hw_addr_copy(tmp_addr, slave->dev->dev_addr,
1235                                  slave->dev->addr_len);
1236
1237                res = dev_set_mac_address(slave->dev, addr, NULL);
1238
1239                /* restore net_device's hw address */
1240                bond_hw_addr_copy(slave->dev->dev_addr, tmp_addr,
1241                                  slave->dev->addr_len);
1242
1243                if (res)
1244                        goto unwind;
1245        }
1246
1247        return 0;
1248
1249unwind:
1250        memcpy(ss.__data, bond->dev->dev_addr, bond->dev->addr_len);
1251        ss.ss_family = bond->dev->type;
1252
1253        /* unwind from head to the slave that failed */
1254        bond_for_each_slave(bond, rollback_slave, iter) {
1255                if (rollback_slave == slave)
1256                        break;
1257                bond_hw_addr_copy(tmp_addr, rollback_slave->dev->dev_addr,
1258                                  rollback_slave->dev->addr_len);
1259                dev_set_mac_address(rollback_slave->dev,
1260                                    (struct sockaddr *)&ss, NULL);
1261                bond_hw_addr_copy(rollback_slave->dev->dev_addr, tmp_addr,
1262                                  rollback_slave->dev->addr_len);
1263        }
1264
1265        return res;
1266}
1267
1268/************************ exported alb funcions ************************/
1269
1270int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1271{
1272        int res;
1273
1274        res = tlb_initialize(bond);
1275        if (res)
1276                return res;
1277
1278        if (rlb_enabled) {
1279                bond->alb_info.rlb_enabled = 1;
1280                res = rlb_initialize(bond);
1281                if (res) {
1282                        tlb_deinitialize(bond);
1283                        return res;
1284                }
1285        } else {
1286                bond->alb_info.rlb_enabled = 0;
1287        }
1288
1289        return 0;
1290}
1291
1292void bond_alb_deinitialize(struct bonding *bond)
1293{
1294        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1295
1296        tlb_deinitialize(bond);
1297
1298        if (bond_info->rlb_enabled)
1299                rlb_deinitialize(bond);
1300}
1301
1302static netdev_tx_t bond_do_alb_xmit(struct sk_buff *skb, struct bonding *bond,
1303                                    struct slave *tx_slave)
1304{
1305        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1306        struct ethhdr *eth_data = eth_hdr(skb);
1307
1308        if (!tx_slave) {
1309                /* unbalanced or unassigned, send through primary */
1310                tx_slave = rcu_dereference(bond->curr_active_slave);
1311                if (bond->params.tlb_dynamic_lb)
1312                        bond_info->unbalanced_load += skb->len;
1313        }
1314
1315        if (tx_slave && bond_slave_can_tx(tx_slave)) {
1316                if (tx_slave != rcu_access_pointer(bond->curr_active_slave)) {
1317                        ether_addr_copy(eth_data->h_source,
1318                                        tx_slave->dev->dev_addr);
1319                }
1320
1321                bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1322                goto out;
1323        }
1324
1325        if (tx_slave && bond->params.tlb_dynamic_lb) {
1326                spin_lock(&bond->mode_lock);
1327                __tlb_clear_slave(bond, tx_slave, 0);
1328                spin_unlock(&bond->mode_lock);
1329        }
1330
1331        /* no suitable interface, frame not sent */
1332        bond_tx_drop(bond->dev, skb);
1333out:
1334        return NETDEV_TX_OK;
1335}
1336
1337netdev_tx_t bond_tlb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1338{
1339        struct bonding *bond = netdev_priv(bond_dev);
1340        struct ethhdr *eth_data;
1341        struct slave *tx_slave = NULL;
1342        u32 hash_index;
1343
1344        skb_reset_mac_header(skb);
1345        eth_data = eth_hdr(skb);
1346
1347        /* Do not TX balance any multicast or broadcast */
1348        if (!is_multicast_ether_addr(eth_data->h_dest)) {
1349                switch (skb->protocol) {
1350                case htons(ETH_P_IP):
1351                case htons(ETH_P_IPX):
1352                    /* In case of IPX, it will falback to L2 hash */
1353                case htons(ETH_P_IPV6):
1354                        hash_index = bond_xmit_hash(bond, skb);
1355                        if (bond->params.tlb_dynamic_lb) {
1356                                tx_slave = tlb_choose_channel(bond,
1357                                                              hash_index & 0xFF,
1358                                                              skb->len);
1359                        } else {
1360                                struct bond_up_slave *slaves;
1361                                unsigned int count;
1362
1363                                slaves = rcu_dereference(bond->slave_arr);
1364                                count = slaves ? READ_ONCE(slaves->count) : 0;
1365                                if (likely(count))
1366                                        tx_slave = slaves->arr[hash_index %
1367                                                               count];
1368                        }
1369                        break;
1370                }
1371        }
1372        return bond_do_alb_xmit(skb, bond, tx_slave);
1373}
1374
1375netdev_tx_t bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1376{
1377        struct bonding *bond = netdev_priv(bond_dev);
1378        struct ethhdr *eth_data;
1379        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1380        struct slave *tx_slave = NULL;
1381        static const __be32 ip_bcast = htonl(0xffffffff);
1382        int hash_size = 0;
1383        bool do_tx_balance = true;
1384        u32 hash_index = 0;
1385        const u8 *hash_start = NULL;
1386        struct ipv6hdr *ip6hdr;
1387
1388        skb_reset_mac_header(skb);
1389        eth_data = eth_hdr(skb);
1390
1391        switch (ntohs(skb->protocol)) {
1392        case ETH_P_IP: {
1393                const struct iphdr *iph = ip_hdr(skb);
1394
1395                if (is_broadcast_ether_addr(eth_data->h_dest) ||
1396                    iph->daddr == ip_bcast ||
1397                    iph->protocol == IPPROTO_IGMP) {
1398                        do_tx_balance = false;
1399                        break;
1400                }
1401                hash_start = (char *)&(iph->daddr);
1402                hash_size = sizeof(iph->daddr);
1403        }
1404                break;
1405        case ETH_P_IPV6:
1406                /* IPv6 doesn't really use broadcast mac address, but leave
1407                 * that here just in case.
1408                 */
1409                if (is_broadcast_ether_addr(eth_data->h_dest)) {
1410                        do_tx_balance = false;
1411                        break;
1412                }
1413
1414                /* IPv6 uses all-nodes multicast as an equivalent to
1415                 * broadcasts in IPv4.
1416                 */
1417                if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1418                        do_tx_balance = false;
1419                        break;
1420                }
1421
1422                /* Additianally, DAD probes should not be tx-balanced as that
1423                 * will lead to false positives for duplicate addresses and
1424                 * prevent address configuration from working.
1425                 */
1426                ip6hdr = ipv6_hdr(skb);
1427                if (ipv6_addr_any(&ip6hdr->saddr)) {
1428                        do_tx_balance = false;
1429                        break;
1430                }
1431
1432                hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1433                hash_size = sizeof(ipv6_hdr(skb)->daddr);
1434                break;
1435        case ETH_P_IPX:
1436                if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1437                        /* something is wrong with this packet */
1438                        do_tx_balance = false;
1439                        break;
1440                }
1441
1442                if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1443                        /* The only protocol worth balancing in
1444                         * this family since it has an "ARP" like
1445                         * mechanism
1446                         */
1447                        do_tx_balance = false;
1448                        break;
1449                }
1450
1451                hash_start = (char *)eth_data->h_dest;
1452                hash_size = ETH_ALEN;
1453                break;
1454        case ETH_P_ARP:
1455                do_tx_balance = false;
1456                if (bond_info->rlb_enabled)
1457                        tx_slave = rlb_arp_xmit(skb, bond);
1458                break;
1459        default:
1460                do_tx_balance = false;
1461                break;
1462        }
1463
1464        if (do_tx_balance) {
1465                if (bond->params.tlb_dynamic_lb) {
1466                        hash_index = _simple_hash(hash_start, hash_size);
1467                        tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1468                } else {
1469                        /*
1470                         * do_tx_balance means we are free to select the tx_slave
1471                         * So we do exactly what tlb would do for hash selection
1472                         */
1473
1474                        struct bond_up_slave *slaves;
1475                        unsigned int count;
1476
1477                        slaves = rcu_dereference(bond->slave_arr);
1478                        count = slaves ? READ_ONCE(slaves->count) : 0;
1479                        if (likely(count))
1480                                tx_slave = slaves->arr[bond_xmit_hash(bond, skb) %
1481                                                       count];
1482                }
1483        }
1484
1485        return bond_do_alb_xmit(skb, bond, tx_slave);
1486}
1487
1488void bond_alb_monitor(struct work_struct *work)
1489{
1490        struct bonding *bond = container_of(work, struct bonding,
1491                                            alb_work.work);
1492        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1493        struct list_head *iter;
1494        struct slave *slave;
1495
1496        if (!bond_has_slaves(bond)) {
1497                bond_info->tx_rebalance_counter = 0;
1498                bond_info->lp_counter = 0;
1499                goto re_arm;
1500        }
1501
1502        rcu_read_lock();
1503
1504        bond_info->tx_rebalance_counter++;
1505        bond_info->lp_counter++;
1506
1507        /* send learning packets */
1508        if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1509                bool strict_match;
1510
1511                bond_for_each_slave_rcu(bond, slave, iter) {
1512                        /* If updating current_active, use all currently
1513                         * user mac addreses (!strict_match).  Otherwise, only
1514                         * use mac of the slave device.
1515                         * In RLB mode, we always use strict matches.
1516                         */
1517                        strict_match = (slave != rcu_access_pointer(bond->curr_active_slave) ||
1518                                        bond_info->rlb_enabled);
1519                        alb_send_learning_packets(slave, slave->dev->dev_addr,
1520                                                  strict_match);
1521                }
1522                bond_info->lp_counter = 0;
1523        }
1524
1525        /* rebalance tx traffic */
1526        if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1527                bond_for_each_slave_rcu(bond, slave, iter) {
1528                        tlb_clear_slave(bond, slave, 1);
1529                        if (slave == rcu_access_pointer(bond->curr_active_slave)) {
1530                                SLAVE_TLB_INFO(slave).load =
1531                                        bond_info->unbalanced_load /
1532                                                BOND_TLB_REBALANCE_INTERVAL;
1533                                bond_info->unbalanced_load = 0;
1534                        }
1535                }
1536                bond_info->tx_rebalance_counter = 0;
1537        }
1538
1539        if (bond_info->rlb_enabled) {
1540                if (bond_info->primary_is_promisc &&
1541                    (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1542
1543                        /* dev_set_promiscuity requires rtnl and
1544                         * nothing else.  Avoid race with bond_close.
1545                         */
1546                        rcu_read_unlock();
1547                        if (!rtnl_trylock())
1548                                goto re_arm;
1549
1550                        bond_info->rlb_promisc_timeout_counter = 0;
1551
1552                        /* If the primary was set to promiscuous mode
1553                         * because a slave was disabled then
1554                         * it can now leave promiscuous mode.
1555                         */
1556                        dev_set_promiscuity(rtnl_dereference(bond->curr_active_slave)->dev,
1557                                            -1);
1558                        bond_info->primary_is_promisc = 0;
1559
1560                        rtnl_unlock();
1561                        rcu_read_lock();
1562                }
1563
1564                if (bond_info->rlb_rebalance) {
1565                        bond_info->rlb_rebalance = 0;
1566                        rlb_rebalance(bond);
1567                }
1568
1569                /* check if clients need updating */
1570                if (bond_info->rx_ntt) {
1571                        if (bond_info->rlb_update_delay_counter) {
1572                                --bond_info->rlb_update_delay_counter;
1573                        } else {
1574                                rlb_update_rx_clients(bond);
1575                                if (bond_info->rlb_update_retry_counter)
1576                                        --bond_info->rlb_update_retry_counter;
1577                                else
1578                                        bond_info->rx_ntt = 0;
1579                        }
1580                }
1581        }
1582        rcu_read_unlock();
1583re_arm:
1584        queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1585}
1586
1587/* assumption: called before the slave is attached to the bond
1588 * and not locked by the bond lock
1589 */
1590int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1591{
1592        int res;
1593
1594        res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1595                                     slave->dev->addr_len);
1596        if (res)
1597                return res;
1598
1599        res = alb_handle_addr_collision_on_attach(bond, slave);
1600        if (res)
1601                return res;
1602
1603        tlb_init_slave(slave);
1604
1605        /* order a rebalance ASAP */
1606        bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1607
1608        if (bond->alb_info.rlb_enabled)
1609                bond->alb_info.rlb_rebalance = 1;
1610
1611        return 0;
1612}
1613
1614/* Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1615 * if necessary.
1616 *
1617 * Caller must hold RTNL and no other locks
1618 */
1619void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1620{
1621        if (bond_has_slaves(bond))
1622                alb_change_hw_addr_on_detach(bond, slave);
1623
1624        tlb_clear_slave(bond, slave, 0);
1625
1626        if (bond->alb_info.rlb_enabled) {
1627                bond->alb_info.rx_slave = NULL;
1628                rlb_clear_slave(bond, slave);
1629        }
1630
1631}
1632
1633void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1634{
1635        struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1636
1637        if (link == BOND_LINK_DOWN) {
1638                tlb_clear_slave(bond, slave, 0);
1639                if (bond->alb_info.rlb_enabled)
1640                        rlb_clear_slave(bond, slave);
1641        } else if (link == BOND_LINK_UP) {
1642                /* order a rebalance ASAP */
1643                bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1644                if (bond->alb_info.rlb_enabled) {
1645                        bond->alb_info.rlb_rebalance = 1;
1646                        /* If the updelay module parameter is smaller than the
1647                         * forwarding delay of the switch the rebalance will
1648                         * not work because the rebalance arp replies will
1649                         * not be forwarded to the clients..
1650                         */
1651                }
1652        }
1653
1654        if (bond_is_nondyn_tlb(bond)) {
1655                if (bond_update_slave_arr(bond, NULL))
1656                        pr_err("Failed to build slave-array for TLB mode.\n");
1657        }
1658}
1659
1660/**
1661 * bond_alb_handle_active_change - assign new curr_active_slave
1662 * @bond: our bonding struct
1663 * @new_slave: new slave to assign
1664 *
1665 * Set the bond->curr_active_slave to @new_slave and handle
1666 * mac address swapping and promiscuity changes as needed.
1667 *
1668 * Caller must hold RTNL
1669 */
1670void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1671{
1672        struct slave *swap_slave;
1673        struct slave *curr_active;
1674
1675        curr_active = rtnl_dereference(bond->curr_active_slave);
1676        if (curr_active == new_slave)
1677                return;
1678
1679        if (curr_active && bond->alb_info.primary_is_promisc) {
1680                dev_set_promiscuity(curr_active->dev, -1);
1681                bond->alb_info.primary_is_promisc = 0;
1682                bond->alb_info.rlb_promisc_timeout_counter = 0;
1683        }
1684
1685        swap_slave = curr_active;
1686        rcu_assign_pointer(bond->curr_active_slave, new_slave);
1687
1688        if (!new_slave || !bond_has_slaves(bond))
1689                return;
1690
1691        /* set the new curr_active_slave to the bonds mac address
1692         * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1693         */
1694        if (!swap_slave)
1695                swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1696
1697        /* Arrange for swap_slave and new_slave to temporarily be
1698         * ignored so we can mess with their MAC addresses without
1699         * fear of interference from transmit activity.
1700         */
1701        if (swap_slave)
1702                tlb_clear_slave(bond, swap_slave, 1);
1703        tlb_clear_slave(bond, new_slave, 1);
1704
1705        /* in TLB mode, the slave might flip down/up with the old dev_addr,
1706         * and thus filter bond->dev_addr's packets, so force bond's mac
1707         */
1708        if (BOND_MODE(bond) == BOND_MODE_TLB) {
1709                struct sockaddr_storage ss;
1710                u8 tmp_addr[MAX_ADDR_LEN];
1711
1712                bond_hw_addr_copy(tmp_addr, new_slave->dev->dev_addr,
1713                                  new_slave->dev->addr_len);
1714
1715                bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
1716                                  bond->dev->addr_len);
1717                ss.ss_family = bond->dev->type;
1718                /* we don't care if it can't change its mac, best effort */
1719                dev_set_mac_address(new_slave->dev, (struct sockaddr *)&ss,
1720                                    NULL);
1721
1722                bond_hw_addr_copy(new_slave->dev->dev_addr, tmp_addr,
1723                                  new_slave->dev->addr_len);
1724        }
1725
1726        /* curr_active_slave must be set before calling alb_swap_mac_addr */
1727        if (swap_slave) {
1728                /* swap mac address */
1729                alb_swap_mac_addr(swap_slave, new_slave);
1730                alb_fasten_mac_swap(bond, swap_slave, new_slave);
1731        } else {
1732                /* set the new_slave to the bond mac address */
1733                alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1734                                       bond->dev->addr_len);
1735                alb_send_learning_packets(new_slave, bond->dev->dev_addr,
1736                                          false);
1737        }
1738}
1739
1740/* Called with RTNL */
1741int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1742{
1743        struct bonding *bond = netdev_priv(bond_dev);
1744        struct sockaddr_storage *ss = addr;
1745        struct slave *curr_active;
1746        struct slave *swap_slave;
1747        int res;
1748
1749        if (!is_valid_ether_addr(ss->__data))
1750                return -EADDRNOTAVAIL;
1751
1752        res = alb_set_mac_address(bond, addr);
1753        if (res)
1754                return res;
1755
1756        bond_hw_addr_copy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
1757
1758        /* If there is no curr_active_slave there is nothing else to do.
1759         * Otherwise we'll need to pass the new address to it and handle
1760         * duplications.
1761         */
1762        curr_active = rtnl_dereference(bond->curr_active_slave);
1763        if (!curr_active)
1764                return 0;
1765
1766        swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1767
1768        if (swap_slave) {
1769                alb_swap_mac_addr(swap_slave, curr_active);
1770                alb_fasten_mac_swap(bond, swap_slave, curr_active);
1771        } else {
1772                alb_set_slave_mac_addr(curr_active, bond_dev->dev_addr,
1773                                       bond_dev->addr_len);
1774
1775                alb_send_learning_packets(curr_active,
1776                                          bond_dev->dev_addr, false);
1777                if (bond->alb_info.rlb_enabled) {
1778                        /* inform clients mac address has changed */
1779                        rlb_req_update_slave_clients(bond, curr_active);
1780                }
1781        }
1782
1783        return 0;
1784}
1785
1786void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1787{
1788        if (bond->alb_info.rlb_enabled)
1789                rlb_clear_vlan(bond, vlan_id);
1790}
1791
1792