drivers/net/bonding/bond_alb.c

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