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, write to the Free Software Foundation, Inc.,
  16  * 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
  17  *
  18  * The full GNU General Public License is included in this distribution in the
  19  * file called LICENSE.
  20  *
  21  */
  22
  23 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  24
  25 #include <linux/skbuff.h>
  26 #include <linux/netdevice.h>
  27 #include <linux/etherdevice.h>
  28 #include <linux/pkt_sched.h>
  29 #include <linux/spinlock.h>
  30 #include <linux/slab.h>
  31 #include <linux/timer.h>
  32 #include <linux/ip.h>
  33 #include <linux/ipv6.h>
  34 #include <linux/if_arp.h>
  35 #include <linux/if_ether.h>
  36 #include <linux/if_bonding.h>
  37 #include <linux/if_vlan.h>
  38 #include <linux/in.h>
  39 #include <net/ipx.h>
  40 #include <net/arp.h>
  41 #include <net/ipv6.h>
  42 #include <asm/byteorder.h>
  43 #include "bonding.h"
  44 #include "bond_alb.h"
  45
  46
  47
  48 #ifndef __long_aligned
  49 #define __long_aligned __attribute__((aligned((sizeof(long)))))
  50 #endif
  51 static const u8 mac_bcast[ETH_ALEN] __long_aligned = {
  52         0xff, 0xff, 0xff, 0xff, 0xff, 0xff
  53 };
  54 static const u8 mac_v6_allmcast[ETH_ALEN] __long_aligned = {
  55         0x33, 0x33, 0x00, 0x00, 0x00, 0x01
  56 };
  57 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
  58
  59 #pragma pack(1)
  60 struct learning_pkt {
  61         u8 mac_dst[ETH_ALEN];
  62         u8 mac_src[ETH_ALEN];
  63         __be16 type;
  64         u8 padding[ETH_ZLEN - ETH_HLEN];
  65 };
  66
  67 struct arp_pkt {
  68         __be16  hw_addr_space;
  69         __be16  prot_addr_space;
  70         u8      hw_addr_len;
  71         u8      prot_addr_len;
  72         __be16  op_code;
  73         u8      mac_src[ETH_ALEN];      /* sender hardware address */
  74         __be32  ip_src;                 /* sender IP address */
  75         u8      mac_dst[ETH_ALEN];      /* target hardware address */
  76         __be32  ip_dst;                 /* target IP address */
  77 };
  78 #pragma pack()
  79
  80 static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb)
  81 {
  82         return (struct arp_pkt *)skb_network_header(skb);
  83 }
  84
  85 /* Forward declaration */
  86 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[]);
  87
  88 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
  89 {
  90         int i;
  91         u8 hash = 0;
  92
  93         for (i = 0; i < hash_size; i++) {
  94                 hash ^= hash_start[i];
  95         }
  96
  97         return hash;
  98 }
  99
 100 /*********************** tlb specific functions ***************************/
 101
 102 static inline void _lock_tx_hashtbl_bh(struct bonding *bond)
 103 {
 104         spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 105 }
 106
 107 static inline void _unlock_tx_hashtbl_bh(struct bonding *bond)
 108 {
 109         spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 110 }
 111
 112 static inline void _lock_tx_hashtbl(struct bonding *bond)
 113 {
 114         spin_lock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 115 }
 116
 117 static inline void _unlock_tx_hashtbl(struct bonding *bond)
 118 {
 119         spin_unlock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 120 }
 121
 122 /* Caller must hold tx_hashtbl lock */
 123 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
 124 {
 125         if (save_load) {
 126                 entry->load_history = 1 + entry->tx_bytes /
 127                                       BOND_TLB_REBALANCE_INTERVAL;
 128                 entry->tx_bytes = 0;
 129         }
 130
 131         entry->tx_slave = NULL;
 132         entry->next = TLB_NULL_INDEX;
 133         entry->prev = TLB_NULL_INDEX;
 134 }
 135
 136 static inline void tlb_init_slave(struct slave *slave)
 137 {
 138         SLAVE_TLB_INFO(slave).load = 0;
 139         SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
 140 }
 141
 142 /* Caller must hold bond lock for read, BH disabled */
 143 static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
 144                          int save_load)
 145 {
 146         struct tlb_client_info *tx_hash_table;
 147         u32 index;
 148
 149         /* clear slave from tx_hashtbl */
 150         tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
 151
 152         /* skip this if we've already freed the tx hash table */
 153         if (tx_hash_table) {
 154                 index = SLAVE_TLB_INFO(slave).head;
 155                 while (index != TLB_NULL_INDEX) {
 156                         u32 next_index = tx_hash_table[index].next;
 157                         tlb_init_table_entry(&tx_hash_table[index], save_load);
 158                         index = next_index;
 159                 }
 160         }
 161
 162         tlb_init_slave(slave);
 163 }
 164
 165 /* Caller must hold bond lock for read */
 166 static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
 167                          int save_load)
 168 {
 169         _lock_tx_hashtbl_bh(bond);
 170         __tlb_clear_slave(bond, slave, save_load);
 171         _unlock_tx_hashtbl_bh(bond);
 172 }
 173
 174 /* Must be called before starting the monitor timer */
 175 static int tlb_initialize(struct bonding *bond)
 176 {
 177         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 178         int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 179         struct tlb_client_info *new_hashtbl;
 180         int i;
 181
 182         new_hashtbl = kzalloc(size, GFP_KERNEL);
 183         if (!new_hashtbl) {
 184                 pr_err("%s: Error: Failed to allocate TLB hash table\n",
 185                        bond->dev->name);
 186                 return -1;
 187         }
 188         _lock_tx_hashtbl_bh(bond);
 189
 190         bond_info->tx_hashtbl = new_hashtbl;
 191
 192         for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
 193                 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
 194         }
 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         long long max_gap;
 225         int i;
 226
 227         least_loaded = NULL;
 228         max_gap = LLONG_MIN;
 229
 230         /* Find the slave with the largest gap */
 231         bond_for_each_slave(bond, slave, i) {
 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
 270                         slave_info->head = hash_index;
 271                         slave_info->load +=
 272                                 hash_table[hash_index].load_history;
 273                 }
 274         }
 275
 276         if (assigned_slave) {
 277                 hash_table[hash_index].tx_bytes += skb_len;
 278         }
 279
 280         return assigned_slave;
 281 }
 282
 283 /* Caller must hold bond lock for read */
 284 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
 285                                         u32 skb_len)
 286 {
 287         struct slave *tx_slave;
 288         /*
 289          * We don't need to disable softirq here, becase
 290          * tlb_choose_channel() is only called by bond_alb_xmit()
 291          * which already has softirq disabled.
 292          */
 293         _lock_tx_hashtbl(bond);
 294         tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
 295         _unlock_tx_hashtbl(bond);
 296         return tx_slave;
 297 }
 298
 299 /*********************** rlb specific functions ***************************/
 300 static inline void _lock_rx_hashtbl_bh(struct bonding *bond)
 301 {
 302         spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 303 }
 304
 305 static inline void _unlock_rx_hashtbl_bh(struct bonding *bond)
 306 {
 307         spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 308 }
 309
 310 static inline void _lock_rx_hashtbl(struct bonding *bond)
 311 {
 312         spin_lock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 313 }
 314
 315 static inline void _unlock_rx_hashtbl(struct bonding *bond)
 316 {
 317         spin_unlock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 318 }
 319
 320 /* when an ARP REPLY is received from a client update its info
 321  * in the rx_hashtbl
 322  */
 323 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 324 {
 325         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 326         struct rlb_client_info *client_info;
 327         u32 hash_index;
 328
 329         _lock_rx_hashtbl_bh(bond);
 330
 331         hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
 332         client_info = &(bond_info->rx_hashtbl[hash_index]);
 333
 334         if ((client_info->assigned) &&
 335             (client_info->ip_src == arp->ip_dst) &&
 336             (client_info->ip_dst == arp->ip_src) &&
 337             (compare_ether_addr_64bits(client_info->mac_dst, arp->mac_src))) {
 338                 /* update the clients MAC address */
 339                 memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
 340                 client_info->ntt = 1;
 341                 bond_info->rx_ntt = 1;
 342         }
 343
 344         _unlock_rx_hashtbl_bh(bond);
 345 }
 346
 347 static void rlb_arp_recv(struct sk_buff *skb, struct bonding *bond,
 348                          struct slave *slave)
 349 {
 350         struct arp_pkt *arp;
 351
 352         if (skb->protocol != cpu_to_be16(ETH_P_ARP))
 353                 return;
 354
 355         arp = (struct arp_pkt *) skb->data;
 356         if (!arp) {
 357                 pr_debug("Packet has no ARP data\n");
 358                 return;
 359         }
 360
 361         if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
 362                 return;
 363
 364         if (skb->len < sizeof(struct arp_pkt)) {
 365                 pr_debug("Packet is too small to be an ARP\n");
 366                 return;
 367         }
 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 }
 375
 376 /* Caller must hold bond lock for read */
 377 static struct slave *rlb_next_rx_slave(struct bonding *bond)
 378 {
 379         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 380         struct slave *rx_slave, *slave, *start_at;
 381         int i = 0;
 382
 383         if (bond_info->next_rx_slave) {
 384                 start_at = bond_info->next_rx_slave;
 385         } else {
 386                 start_at = bond->first_slave;
 387         }
 388
 389         rx_slave = NULL;
 390
 391         bond_for_each_slave_from(bond, slave, i, start_at) {
 392                 if (SLAVE_IS_OK(slave)) {
 393                         if (!rx_slave) {
 394                                 rx_slave = slave;
 395                         } else if (slave->speed > rx_slave->speed) {
 396                                 rx_slave = slave;
 397                         }
 398                 }
 399         }
 400
 401         if (rx_slave) {
 402                 bond_info->next_rx_slave = rx_slave->next;
 403         }
 404
 405         return rx_slave;
 406 }
 407
 408 /* teach the switch the mac of a disabled slave
 409  * on the primary for fault tolerance
 410  *
 411  * Caller must hold bond->curr_slave_lock for write or bond lock for write
 412  */
 413 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
 414 {
 415         if (!bond->curr_active_slave) {
 416                 return;
 417         }
 418
 419         if (!bond->alb_info.primary_is_promisc) {
 420                 if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1))
 421                         bond->alb_info.primary_is_promisc = 1;
 422                 else
 423                         bond->alb_info.primary_is_promisc = 0;
 424         }
 425
 426         bond->alb_info.rlb_promisc_timeout_counter = 0;
 427
 428         alb_send_learning_packets(bond->curr_active_slave, addr);
 429 }
 430
 431 /* slave being removed should not be active at this point
 432  *
 433  * Caller must hold bond lock for read
 434  */
 435 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 436 {
 437         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 438         struct rlb_client_info *rx_hash_table;
 439         u32 index, next_index;
 440
 441         /* clear slave from rx_hashtbl */
 442         _lock_rx_hashtbl_bh(bond);
 443
 444         rx_hash_table = bond_info->rx_hashtbl;
 445         index = bond_info->rx_hashtbl_head;
 446         for (; index != RLB_NULL_INDEX; index = next_index) {
 447                 next_index = rx_hash_table[index].next;
 448                 if (rx_hash_table[index].slave == slave) {
 449                         struct slave *assigned_slave = rlb_next_rx_slave(bond);
 450
 451                         if (assigned_slave) {
 452                                 rx_hash_table[index].slave = assigned_slave;
 453                                 if (compare_ether_addr_64bits(rx_hash_table[index].mac_dst,
 454                                                               mac_bcast)) {
 455                                         bond_info->rx_hashtbl[index].ntt = 1;
 456                                         bond_info->rx_ntt = 1;
 457                                         /* A slave has been removed from the
 458                                          * table because it is either disabled
 459                                          * or being released. We must retry the
 460                                          * update to avoid clients from not
 461                                          * being updated & disconnecting when
 462                                          * there is stress
 463                                          */
 464                                         bond_info->rlb_update_retry_counter =
 465                                                 RLB_UPDATE_RETRY;
 466                                 }
 467                         } else {  /* there is no active slave */
 468                                 rx_hash_table[index].slave = NULL;
 469                         }
 470                 }
 471         }
 472
 473         _unlock_rx_hashtbl_bh(bond);
 474
 475         write_lock_bh(&bond->curr_slave_lock);
 476
 477         if (slave != bond->curr_active_slave) {
 478                 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 479         }
 480
 481         write_unlock_bh(&bond->curr_slave_lock);
 482 }
 483
 484 static void rlb_update_client(struct rlb_client_info *client_info)
 485 {
 486         int i;
 487
 488         if (!client_info->slave) {
 489                 return;
 490         }
 491
 492         for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 493                 struct sk_buff *skb;
 494
 495                 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 496                                  client_info->ip_dst,
 497                                  client_info->slave->dev,
 498                                  client_info->ip_src,
 499                                  client_info->mac_dst,
 500                                  client_info->slave->dev->dev_addr,
 501                                  client_info->mac_dst);
 502                 if (!skb) {
 503                         pr_err("%s: Error: failed to create an ARP packet\n",
 504                                client_info->slave->dev->master->name);
 505                         continue;
 506                 }
 507
 508                 skb->dev = client_info->slave->dev;
 509
 510                 if (client_info->tag) {
 511                         skb = vlan_put_tag(skb, client_info->vlan_id);
 512                         if (!skb) {
 513                                 pr_err("%s: Error: failed to insert VLAN tag\n",
 514                                        client_info->slave->dev->master->name);
 515                                 continue;
 516                         }
 517                 }
 518
 519                 arp_xmit(skb);
 520         }
 521 }
 522
 523 /* sends ARP REPLIES that update the clients that need updating */
 524 static void rlb_update_rx_clients(struct bonding *bond)
 525 {
 526         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 527         struct rlb_client_info *client_info;
 528         u32 hash_index;
 529
 530         _lock_rx_hashtbl_bh(bond);
 531
 532         hash_index = bond_info->rx_hashtbl_head;
 533         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 534                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 535                 if (client_info->ntt) {
 536                         rlb_update_client(client_info);
 537                         if (bond_info->rlb_update_retry_counter == 0) {
 538                                 client_info->ntt = 0;
 539                         }
 540                 }
 541         }
 542
 543         /* do not update the entries again until this counter is zero so that
 544          * not to confuse the clients.
 545          */
 546         bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 547
 548         _unlock_rx_hashtbl_bh(bond);
 549 }
 550
 551 /* The slave was assigned a new mac address - update the clients */
 552 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 553 {
 554         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 555         struct rlb_client_info *client_info;
 556         int ntt = 0;
 557         u32 hash_index;
 558
 559         _lock_rx_hashtbl_bh(bond);
 560
 561         hash_index = bond_info->rx_hashtbl_head;
 562         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 563                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 564
 565                 if ((client_info->slave == slave) &&
 566                     compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
 567                         client_info->ntt = 1;
 568                         ntt = 1;
 569                 }
 570         }
 571
 572         // update the team's flag only after the whole iteration
 573         if (ntt) {
 574                 bond_info->rx_ntt = 1;
 575                 //fasten the change
 576                 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 577         }
 578
 579         _unlock_rx_hashtbl_bh(bond);
 580 }
 581
 582 /* mark all clients using src_ip to be updated */
 583 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
 584 {
 585         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 586         struct rlb_client_info *client_info;
 587         u32 hash_index;
 588
 589         _lock_rx_hashtbl(bond);
 590
 591         hash_index = bond_info->rx_hashtbl_head;
 592         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 593                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 594
 595                 if (!client_info->slave) {
 596                         pr_err("%s: Error: found a client with no channel in the client's hash table\n",
 597                                bond->dev->name);
 598                         continue;
 599                 }
 600                 /*update all clients using this src_ip, that are not assigned
 601                  * to the team's address (curr_active_slave) and have a known
 602                  * unicast mac address.
 603                  */
 604                 if ((client_info->ip_src == src_ip) &&
 605                     compare_ether_addr_64bits(client_info->slave->dev->dev_addr,
 606                            bond->dev->dev_addr) &&
 607                     compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
 608                         client_info->ntt = 1;
 609                         bond_info->rx_ntt = 1;
 610                 }
 611         }
 612
 613         _unlock_rx_hashtbl(bond);
 614 }
 615
 616 /* Caller must hold both bond and ptr locks for read */
 617 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
 618 {
 619         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 620         struct arp_pkt *arp = arp_pkt(skb);
 621         struct slave *assigned_slave;
 622         struct rlb_client_info *client_info;
 623         u32 hash_index = 0;
 624
 625         _lock_rx_hashtbl(bond);
 626
 627         hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
 628         client_info = &(bond_info->rx_hashtbl[hash_index]);
 629
 630         if (client_info->assigned) {
 631                 if ((client_info->ip_src == arp->ip_src) &&
 632                     (client_info->ip_dst == arp->ip_dst)) {
 633                         /* the entry is already assigned to this client */
 634                         if (compare_ether_addr_64bits(arp->mac_dst, mac_bcast)) {
 635                                 /* update mac address from arp */
 636                                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 637                         }
 638
 639                         assigned_slave = client_info->slave;
 640                         if (assigned_slave) {
 641                                 _unlock_rx_hashtbl(bond);
 642                                 return assigned_slave;
 643                         }
 644                 } else {
 645                         /* the entry is already assigned to some other client,
 646                          * move the old client to primary (curr_active_slave) so
 647                          * that the new client can be assigned to this entry.
 648                          */
 649                         if (bond->curr_active_slave &&
 650                             client_info->slave != bond->curr_active_slave) {
 651                                 client_info->slave = bond->curr_active_slave;
 652                                 rlb_update_client(client_info);
 653                         }
 654                 }
 655         }
 656         /* assign a new slave */
 657         assigned_slave = rlb_next_rx_slave(bond);
 658
 659         if (assigned_slave) {
 660                 client_info->ip_src = arp->ip_src;
 661                 client_info->ip_dst = arp->ip_dst;
 662                 /* arp->mac_dst is broadcast for arp reqeusts.
 663                  * will be updated with clients actual unicast mac address
 664                  * upon receiving an arp reply.
 665                  */
 666                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 667                 client_info->slave = assigned_slave;
 668
 669                 if (compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
 670                         client_info->ntt = 1;
 671                         bond->alb_info.rx_ntt = 1;
 672                 } else {
 673                         client_info->ntt = 0;
 674                 }
 675
 676                 if (bond_vlan_used(bond)) {
 677                         if (!vlan_get_tag(skb, &client_info->vlan_id))
 678                                 client_info->tag = 1;
 679                 }
 680
 681                 if (!client_info->assigned) {
 682                         u32 prev_tbl_head = bond_info->rx_hashtbl_head;
 683                         bond_info->rx_hashtbl_head = hash_index;
 684                         client_info->next = prev_tbl_head;
 685                         if (prev_tbl_head != RLB_NULL_INDEX) {
 686                                 bond_info->rx_hashtbl[prev_tbl_head].prev =
 687                                         hash_index;
 688                         }
 689                         client_info->assigned = 1;
 690                 }
 691         }
 692
 693         _unlock_rx_hashtbl(bond);
 694
 695         return assigned_slave;
 696 }
 697
 698 /* chooses (and returns) transmit channel for arp reply
 699  * does not choose channel for other arp types since they are
 700  * sent on the curr_active_slave
 701  */
 702 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 703 {
 704         struct arp_pkt *arp = arp_pkt(skb);
 705         struct slave *tx_slave = NULL;
 706
 707         if (arp->op_code == htons(ARPOP_REPLY)) {
 708                 /* the arp must be sent on the selected
 709                 * rx channel
 710                 */
 711                 tx_slave = rlb_choose_channel(skb, bond);
 712                 if (tx_slave) {
 713                         memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
 714                 }
 715                 pr_debug("Server sent ARP Reply packet\n");
 716         } else if (arp->op_code == htons(ARPOP_REQUEST)) {
 717                 /* Create an entry in the rx_hashtbl for this client as a
 718                  * place holder.
 719                  * When the arp reply is received the entry will be updated
 720                  * with the correct unicast address of the client.
 721                  */
 722                 rlb_choose_channel(skb, bond);
 723
 724                 /* The ARP reply packets must be delayed so that
 725                  * they can cancel out the influence of the ARP request.
 726                  */
 727                 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 728
 729                 /* arp requests are broadcast and are sent on the primary
 730                  * the arp request will collapse all clients on the subnet to
 731                  * the primary slave. We must register these clients to be
 732                  * updated with their assigned mac.
 733                  */
 734                 rlb_req_update_subnet_clients(bond, arp->ip_src);
 735                 pr_debug("Server sent ARP Request packet\n");
 736         }
 737
 738         return tx_slave;
 739 }
 740
 741 /* Caller must hold bond lock for read */
 742 static void rlb_rebalance(struct bonding *bond)
 743 {
 744         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 745         struct slave *assigned_slave;
 746         struct rlb_client_info *client_info;
 747         int ntt;
 748         u32 hash_index;
 749
 750         _lock_rx_hashtbl_bh(bond);
 751
 752         ntt = 0;
 753         hash_index = bond_info->rx_hashtbl_head;
 754         for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
 755                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 756                 assigned_slave = rlb_next_rx_slave(bond);
 757                 if (assigned_slave && (client_info->slave != assigned_slave)) {
 758                         client_info->slave = assigned_slave;
 759                         client_info->ntt = 1;
 760                         ntt = 1;
 761                 }
 762         }
 763
 764         /* update the team's flag only after the whole iteration */
 765         if (ntt) {
 766                 bond_info->rx_ntt = 1;
 767         }
 768         _unlock_rx_hashtbl_bh(bond);
 769 }
 770
 771 /* Caller must hold rx_hashtbl lock */
 772 static void rlb_init_table_entry(struct rlb_client_info *entry)
 773 {
 774         memset(entry, 0, sizeof(struct rlb_client_info));
 775         entry->next = RLB_NULL_INDEX;
 776         entry->prev = RLB_NULL_INDEX;
 777 }
 778
 779 static int rlb_initialize(struct bonding *bond)
 780 {
 781         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 782         struct rlb_client_info  *new_hashtbl;
 783         int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 784         int i;
 785
 786         new_hashtbl = kmalloc(size, GFP_KERNEL);
 787         if (!new_hashtbl) {
 788                 pr_err("%s: Error: Failed to allocate RLB hash table\n",
 789                        bond->dev->name);
 790                 return -1;
 791         }
 792         _lock_rx_hashtbl_bh(bond);
 793
 794         bond_info->rx_hashtbl = new_hashtbl;
 795
 796         bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
 797
 798         for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
 799                 rlb_init_table_entry(bond_info->rx_hashtbl + i);
 800         }
 801
 802         _unlock_rx_hashtbl_bh(bond);
 803
 804         /* register to receive ARPs */
 805         bond->recv_probe = rlb_arp_recv;
 806
 807         return 0;
 808 }
 809
 810 static void rlb_deinitialize(struct bonding *bond)
 811 {
 812         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 813
 814         _lock_rx_hashtbl_bh(bond);
 815
 816         kfree(bond_info->rx_hashtbl);
 817         bond_info->rx_hashtbl = NULL;
 818         bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
 819
 820         _unlock_rx_hashtbl_bh(bond);
 821 }
 822
 823 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 824 {
 825         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 826         u32 curr_index;
 827
 828         _lock_rx_hashtbl_bh(bond);
 829
 830         curr_index = bond_info->rx_hashtbl_head;
 831         while (curr_index != RLB_NULL_INDEX) {
 832                 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 833                 u32 next_index = bond_info->rx_hashtbl[curr_index].next;
 834                 u32 prev_index = bond_info->rx_hashtbl[curr_index].prev;
 835
 836                 if (curr->tag && (curr->vlan_id == vlan_id)) {
 837                         if (curr_index == bond_info->rx_hashtbl_head) {
 838                                 bond_info->rx_hashtbl_head = next_index;
 839                         }
 840                         if (prev_index != RLB_NULL_INDEX) {
 841                                 bond_info->rx_hashtbl[prev_index].next = next_index;
 842                         }
 843                         if (next_index != RLB_NULL_INDEX) {
 844                                 bond_info->rx_hashtbl[next_index].prev = prev_index;
 845                         }
 846
 847                         rlb_init_table_entry(curr);
 848                 }
 849
 850                 curr_index = next_index;
 851         }
 852
 853         _unlock_rx_hashtbl_bh(bond);
 854 }
 855
 856 /*********************** tlb/rlb shared functions *********************/
 857
 858 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
 859 {
 860         struct bonding *bond = bond_get_bond_by_slave(slave);
 861         struct learning_pkt pkt;
 862         int size = sizeof(struct learning_pkt);
 863         int i;
 864
 865         memset(&pkt, 0, size);
 866         memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
 867         memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
 868         pkt.type = cpu_to_be16(ETH_P_LOOP);
 869
 870         for (i = 0; i < MAX_LP_BURST; i++) {
 871                 struct sk_buff *skb;
 872                 char *data;
 873
 874                 skb = dev_alloc_skb(size);
 875                 if (!skb) {
 876                         return;
 877                 }
 878
 879                 data = skb_put(skb, size);
 880                 memcpy(data, &pkt, size);
 881
 882                 skb_reset_mac_header(skb);
 883                 skb->network_header = skb->mac_header + ETH_HLEN;
 884                 skb->protocol = pkt.type;
 885                 skb->priority = TC_PRIO_CONTROL;
 886                 skb->dev = slave->dev;
 887
 888                 if (bond_vlan_used(bond)) {
 889                         struct vlan_entry *vlan;
 890
 891                         vlan = bond_next_vlan(bond,
 892                                               bond->alb_info.current_alb_vlan);
 893
 894                         bond->alb_info.current_alb_vlan = vlan;
 895                         if (!vlan) {
 896                                 kfree_skb(skb);
 897                                 continue;
 898                         }
 899
 900                         skb = vlan_put_tag(skb, vlan->vlan_id);
 901                         if (!skb) {
 902                                 pr_err("%s: Error: failed to insert VLAN tag\n",
 903                                        bond->dev->name);
 904                                 continue;
 905                         }
 906                 }
 907
 908                 dev_queue_xmit(skb);
 909         }
 910 }
 911
 912 /* hw is a boolean parameter that determines whether we should try and
 913  * set the hw address of the device as well as the hw address of the
 914  * net_device
 915  */
 916 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[], int hw)
 917 {
 918         struct net_device *dev = slave->dev;
 919         struct sockaddr s_addr;
 920
 921         if (!hw) {
 922                 memcpy(dev->dev_addr, addr, dev->addr_len);
 923                 return 0;
 924         }
 925
 926         /* for rlb each slave must have a unique hw mac addresses so that */
 927         /* each slave will receive packets destined to a different mac */
 928         memcpy(s_addr.sa_data, addr, dev->addr_len);
 929         s_addr.sa_family = dev->type;
 930         if (dev_set_mac_address(dev, &s_addr)) {
 931                 pr_err("%s: Error: dev_set_mac_address of dev %s failed!\n"
 932                        "ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
 933                        dev->master->name, dev->name);
 934                 return -EOPNOTSUPP;
 935         }
 936         return 0;
 937 }
 938
 939 /*
 940  * Swap MAC addresses between two slaves.
 941  *
 942  * Called with RTNL held, and no other locks.
 943  *
 944  */
 945
 946 static void alb_swap_mac_addr(struct bonding *bond, struct slave *slave1, struct slave *slave2)
 947 {
 948         u8 tmp_mac_addr[ETH_ALEN];
 949
 950         memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
 951         alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr, bond->alb_info.rlb_enabled);
 952         alb_set_slave_mac_addr(slave2, tmp_mac_addr, bond->alb_info.rlb_enabled);
 953
 954 }
 955
 956 /*
 957  * Send learning packets after MAC address swap.
 958  *
 959  * Called with RTNL and no other locks
 960  */
 961 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
 962                                 struct slave *slave2)
 963 {
 964         int slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
 965         struct slave *disabled_slave = NULL;
 966
 967         ASSERT_RTNL();
 968
 969         /* fasten the change in the switch */
 970         if (SLAVE_IS_OK(slave1)) {
 971                 alb_send_learning_packets(slave1, slave1->dev->dev_addr);
 972                 if (bond->alb_info.rlb_enabled) {
 973                         /* inform the clients that the mac address
 974                          * has changed
 975                          */
 976                         rlb_req_update_slave_clients(bond, slave1);
 977                 }
 978         } else {
 979                 disabled_slave = slave1;
 980         }
 981
 982         if (SLAVE_IS_OK(slave2)) {
 983                 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
 984                 if (bond->alb_info.rlb_enabled) {
 985                         /* inform the clients that the mac address
 986                          * has changed
 987                          */
 988                         rlb_req_update_slave_clients(bond, slave2);
 989                 }
 990         } else {
 991                 disabled_slave = slave2;
 992         }
 993
 994         if (bond->alb_info.rlb_enabled && slaves_state_differ) {
 995                 /* A disabled slave was assigned an active mac addr */
 996                 rlb_teach_disabled_mac_on_primary(bond,
 997                                                   disabled_slave->dev->dev_addr);
 998         }
 999 }
1000
1001 /**
1002  * alb_change_hw_addr_on_detach
1003  * @bond: bonding we're working on
1004  * @slave: the slave that was just detached
1005  *
1006  * We assume that @slave was already detached from the slave list.
1007  *
1008  * If @slave's permanent hw address is different both from its current
1009  * address and from @bond's address, then somewhere in the bond there's
1010  * a slave that has @slave's permanet address as its current address.
1011  * We'll make sure that that slave no longer uses @slave's permanent address.
1012  *
1013  * Caller must hold RTNL and no other locks
1014  */
1015 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1016 {
1017         int perm_curr_diff;
1018         int perm_bond_diff;
1019
1020         perm_curr_diff = compare_ether_addr_64bits(slave->perm_hwaddr,
1021                                                    slave->dev->dev_addr);
1022         perm_bond_diff = compare_ether_addr_64bits(slave->perm_hwaddr,
1023                                                    bond->dev->dev_addr);
1024
1025         if (perm_curr_diff && perm_bond_diff) {
1026                 struct slave *tmp_slave;
1027                 int i, found = 0;
1028
1029                 bond_for_each_slave(bond, tmp_slave, i) {
1030                         if (!compare_ether_addr_64bits(slave->perm_hwaddr,
1031                                                        tmp_slave->dev->dev_addr)) {
1032                                 found = 1;
1033                                 break;
1034                         }
1035                 }
1036
1037                 if (found) {
1038                         /* locking: needs RTNL and nothing else */
1039                         alb_swap_mac_addr(bond, slave, tmp_slave);
1040                         alb_fasten_mac_swap(bond, slave, tmp_slave);
1041                 }
1042         }
1043 }
1044
1045 /**
1046  * alb_handle_addr_collision_on_attach
1047  * @bond: bonding we're working on
1048  * @slave: the slave that was just attached
1049  *
1050  * checks uniqueness of slave's mac address and handles the case the
1051  * new slave uses the bonds mac address.
1052  *
1053  * If the permanent hw address of @slave is @bond's hw address, we need to
1054  * find a different hw address to give @slave, that isn't in use by any other
1055  * slave in the bond. This address must be, of course, one of the permanent
1056  * addresses of the other slaves.
1057  *
1058  * We go over the slave list, and for each slave there we compare its
1059  * permanent hw address with the current address of all the other slaves.
1060  * If no match was found, then we've found a slave with a permanent address
1061  * that isn't used by any other slave in the bond, so we can assign it to
1062  * @slave.
1063  *
1064  * assumption: this function is called before @slave is attached to the
1065  *             bond slave list.
1066  *
1067  * caller must hold the bond lock for write since the mac addresses are compared
1068  * and may be swapped.
1069  */
1070 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1071 {
1072         struct slave *tmp_slave1, *tmp_slave2, *free_mac_slave;
1073         struct slave *has_bond_addr = bond->curr_active_slave;
1074         int i, j, found = 0;
1075
1076         if (bond->slave_cnt == 0) {
1077                 /* this is the first slave */
1078                 return 0;
1079         }
1080
1081         /* if slave's mac address differs from bond's mac address
1082          * check uniqueness of slave's mac address against the other
1083          * slaves in the bond.
1084          */
1085         if (compare_ether_addr_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1086                 bond_for_each_slave(bond, tmp_slave1, i) {
1087                         if (!compare_ether_addr_64bits(tmp_slave1->dev->dev_addr,
1088                                                        slave->dev->dev_addr)) {
1089                                 found = 1;
1090                                 break;
1091                         }
1092                 }
1093
1094                 if (!found)
1095                         return 0;
1096
1097                 /* Try setting slave mac to bond address and fall-through
1098                    to code handling that situation below... */
1099                 alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1100                                        bond->alb_info.rlb_enabled);
1101         }
1102
1103         /* The slave's address is equal to the address of the bond.
1104          * Search for a spare address in the bond for this slave.
1105          */
1106         free_mac_slave = NULL;
1107
1108         bond_for_each_slave(bond, tmp_slave1, i) {
1109                 found = 0;
1110                 bond_for_each_slave(bond, tmp_slave2, j) {
1111                         if (!compare_ether_addr_64bits(tmp_slave1->perm_hwaddr,
1112                                                        tmp_slave2->dev->dev_addr)) {
1113                                 found = 1;
1114                                 break;
1115                         }
1116                 }
1117
1118                 if (!found) {
1119                         /* no slave has tmp_slave1's perm addr
1120                          * as its curr addr
1121                          */
1122                         free_mac_slave = tmp_slave1;
1123                         break;
1124                 }
1125
1126                 if (!has_bond_addr) {
1127                         if (!compare_ether_addr_64bits(tmp_slave1->dev->dev_addr,
1128                                                        bond->dev->dev_addr)) {
1129
1130                                 has_bond_addr = tmp_slave1;
1131                         }
1132                 }
1133         }
1134
1135         if (free_mac_slave) {
1136                 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1137                                        bond->alb_info.rlb_enabled);
1138
1139                 pr_warning("%s: Warning: the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1140                            bond->dev->name, slave->dev->name,
1141                            free_mac_slave->dev->name);
1142
1143         } else if (has_bond_addr) {
1144                 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",
1145                        bond->dev->name, slave->dev->name);
1146                 return -EFAULT;
1147         }
1148
1149         return 0;
1150 }
1151
1152 /**
1153  * alb_set_mac_address
1154  * @bond:
1155  * @addr:
1156  *
1157  * In TLB mode all slaves are configured to the bond's hw address, but set
1158  * their dev_addr field to different addresses (based on their permanent hw
1159  * addresses).
1160  *
1161  * For each slave, this function sets the interface to the new address and then
1162  * changes its dev_addr field to its previous value.
1163  *
1164  * Unwinding assumes bond's mac address has not yet changed.
1165  */
1166 static int alb_set_mac_address(struct bonding *bond, void *addr)
1167 {
1168         struct sockaddr sa;
1169         struct slave *slave, *stop_at;
1170         char tmp_addr[ETH_ALEN];
1171         int res;
1172         int i;
1173
1174         if (bond->alb_info.rlb_enabled) {
1175                 return 0;
1176         }
1177
1178         bond_for_each_slave(bond, slave, i) {
1179                 /* save net_device's current hw address */
1180                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1181
1182                 res = dev_set_mac_address(slave->dev, addr);
1183
1184                 /* restore net_device's hw address */
1185                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1186
1187                 if (res)
1188                         goto unwind;
1189         }
1190
1191         return 0;
1192
1193 unwind:
1194         memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1195         sa.sa_family = bond->dev->type;
1196
1197         /* unwind from head to the slave that failed */
1198         stop_at = slave;
1199         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
1200                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1201                 dev_set_mac_address(slave->dev, &sa);
1202                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1203         }
1204
1205         return res;
1206 }
1207
1208 /************************ exported alb funcions ************************/
1209
1210 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1211 {
1212         int res;
1213
1214         res = tlb_initialize(bond);
1215         if (res) {
1216                 return res;
1217         }
1218
1219         if (rlb_enabled) {
1220                 bond->alb_info.rlb_enabled = 1;
1221                 /* initialize rlb */
1222                 res = rlb_initialize(bond);
1223                 if (res) {
1224                         tlb_deinitialize(bond);
1225                         return res;
1226                 }
1227         } else {
1228                 bond->alb_info.rlb_enabled = 0;
1229         }
1230
1231         return 0;
1232 }
1233
1234 void bond_alb_deinitialize(struct bonding *bond)
1235 {
1236         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1237
1238         tlb_deinitialize(bond);
1239
1240         if (bond_info->rlb_enabled) {
1241                 rlb_deinitialize(bond);
1242         }
1243 }
1244
1245 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1246 {
1247         struct bonding *bond = netdev_priv(bond_dev);
1248         struct ethhdr *eth_data;
1249         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1250         struct slave *tx_slave = NULL;
1251         static const __be32 ip_bcast = htonl(0xffffffff);
1252         int hash_size = 0;
1253         int do_tx_balance = 1;
1254         u32 hash_index = 0;
1255         const u8 *hash_start = NULL;
1256         int res = 1;
1257         struct ipv6hdr *ip6hdr;
1258
1259         skb_reset_mac_header(skb);
1260         eth_data = eth_hdr(skb);
1261
1262         /* make sure that the curr_active_slave do not change during tx
1263          */
1264         read_lock(&bond->curr_slave_lock);
1265
1266         switch (ntohs(skb->protocol)) {
1267         case ETH_P_IP: {
1268                 const struct iphdr *iph = ip_hdr(skb);
1269
1270                 if (!compare_ether_addr_64bits(eth_data->h_dest, mac_bcast) ||
1271                     (iph->daddr == ip_bcast) ||
1272                     (iph->protocol == IPPROTO_IGMP)) {
1273                         do_tx_balance = 0;
1274                         break;
1275                 }
1276                 hash_start = (char *)&(iph->daddr);
1277                 hash_size = sizeof(iph->daddr);
1278         }
1279                 break;
1280         case ETH_P_IPV6:
1281                 /* IPv6 doesn't really use broadcast mac address, but leave
1282                  * that here just in case.
1283                  */
1284                 if (!compare_ether_addr_64bits(eth_data->h_dest, mac_bcast)) {
1285                         do_tx_balance = 0;
1286                         break;
1287                 }
1288
1289                 /* IPv6 uses all-nodes multicast as an equivalent to
1290                  * broadcasts in IPv4.
1291                  */
1292                 if (!compare_ether_addr_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1293                         do_tx_balance = 0;
1294                         break;
1295                 }
1296
1297                 /* Additianally, DAD probes should not be tx-balanced as that
1298                  * will lead to false positives for duplicate addresses and
1299                  * prevent address configuration from working.
1300                  */
1301                 ip6hdr = ipv6_hdr(skb);
1302                 if (ipv6_addr_any(&ip6hdr->saddr)) {
1303                         do_tx_balance = 0;
1304                         break;
1305                 }
1306
1307                 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1308                 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1309                 break;
1310         case ETH_P_IPX:
1311                 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1312                         /* something is wrong with this packet */
1313                         do_tx_balance = 0;
1314                         break;
1315                 }
1316
1317                 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1318                         /* The only protocol worth balancing in
1319                          * this family since it has an "ARP" like
1320                          * mechanism
1321                          */
1322                         do_tx_balance = 0;
1323                         break;
1324                 }
1325
1326                 hash_start = (char*)eth_data->h_dest;
1327                 hash_size = ETH_ALEN;
1328                 break;
1329         case ETH_P_ARP:
1330                 do_tx_balance = 0;
1331                 if (bond_info->rlb_enabled) {
1332                         tx_slave = rlb_arp_xmit(skb, bond);
1333                 }
1334                 break;
1335         default:
1336                 do_tx_balance = 0;
1337                 break;
1338         }
1339
1340         if (do_tx_balance) {
1341                 hash_index = _simple_hash(hash_start, hash_size);
1342                 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1343         }
1344
1345         if (!tx_slave) {
1346                 /* unbalanced or unassigned, send through primary */
1347                 tx_slave = bond->curr_active_slave;
1348                 bond_info->unbalanced_load += skb->len;
1349         }
1350
1351         if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1352                 if (tx_slave != bond->curr_active_slave) {
1353                         memcpy(eth_data->h_source,
1354                                tx_slave->dev->dev_addr,
1355                                ETH_ALEN);
1356                 }
1357
1358                 res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1359         } else {
1360                 if (tx_slave) {
1361                         _lock_tx_hashtbl(bond);
1362                         __tlb_clear_slave(bond, tx_slave, 0);
1363                         _unlock_tx_hashtbl(bond);
1364                 }
1365         }
1366
1367         if (res) {
1368                 /* no suitable interface, frame not sent */
1369                 dev_kfree_skb(skb);
1370         }
1371         read_unlock(&bond->curr_slave_lock);
1372
1373         return NETDEV_TX_OK;
1374 }
1375
1376 void bond_alb_monitor(struct work_struct *work)
1377 {
1378         struct bonding *bond = container_of(work, struct bonding,
1379                                             alb_work.work);
1380         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1381         struct slave *slave;
1382         int i;
1383
1384         read_lock(&bond->lock);
1385
1386         if (bond->slave_cnt == 0) {
1387                 bond_info->tx_rebalance_counter = 0;
1388                 bond_info->lp_counter = 0;
1389                 goto re_arm;
1390         }
1391
1392         bond_info->tx_rebalance_counter++;
1393         bond_info->lp_counter++;
1394
1395         /* send learning packets */
1396         if (bond_info->lp_counter >= BOND_ALB_LP_TICKS) {
1397                 /* change of curr_active_slave involves swapping of mac addresses.
1398                  * in order to avoid this swapping from happening while
1399                  * sending the learning packets, the curr_slave_lock must be held for
1400                  * read.
1401                  */
1402                 read_lock(&bond->curr_slave_lock);
1403
1404                 bond_for_each_slave(bond, slave, i) {
1405                         alb_send_learning_packets(slave, slave->dev->dev_addr);
1406                 }
1407
1408                 read_unlock(&bond->curr_slave_lock);
1409
1410                 bond_info->lp_counter = 0;
1411         }
1412
1413         /* rebalance tx traffic */
1414         if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1415
1416                 read_lock(&bond->curr_slave_lock);
1417
1418                 bond_for_each_slave(bond, slave, i) {
1419                         tlb_clear_slave(bond, slave, 1);
1420                         if (slave == bond->curr_active_slave) {
1421                                 SLAVE_TLB_INFO(slave).load =
1422                                         bond_info->unbalanced_load /
1423                                                 BOND_TLB_REBALANCE_INTERVAL;
1424                                 bond_info->unbalanced_load = 0;
1425                         }
1426                 }
1427
1428                 read_unlock(&bond->curr_slave_lock);
1429
1430                 bond_info->tx_rebalance_counter = 0;
1431         }
1432
1433         /* handle rlb stuff */
1434         if (bond_info->rlb_enabled) {
1435                 if (bond_info->primary_is_promisc &&
1436                     (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1437
1438                         /*
1439                          * dev_set_promiscuity requires rtnl and
1440                          * nothing else.  Avoid race with bond_close.
1441                          */
1442                         read_unlock(&bond->lock);
1443                         if (!rtnl_trylock()) {
1444                                 read_lock(&bond->lock);
1445                                 goto re_arm;
1446                         }
1447
1448                         bond_info->rlb_promisc_timeout_counter = 0;
1449
1450                         /* If the primary was set to promiscuous mode
1451                          * because a slave was disabled then
1452                          * it can now leave promiscuous mode.
1453                          */
1454                         dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1455                         bond_info->primary_is_promisc = 0;
1456
1457                         rtnl_unlock();
1458                         read_lock(&bond->lock);
1459                 }
1460
1461                 if (bond_info->rlb_rebalance) {
1462                         bond_info->rlb_rebalance = 0;
1463                         rlb_rebalance(bond);
1464                 }
1465
1466                 /* check if clients need updating */
1467                 if (bond_info->rx_ntt) {
1468                         if (bond_info->rlb_update_delay_counter) {
1469                                 --bond_info->rlb_update_delay_counter;
1470                         } else {
1471                                 rlb_update_rx_clients(bond);
1472                                 if (bond_info->rlb_update_retry_counter) {
1473                                         --bond_info->rlb_update_retry_counter;
1474                                 } else {
1475                                         bond_info->rx_ntt = 0;
1476                                 }
1477                         }
1478                 }
1479         }
1480
1481 re_arm:
1482         queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1483
1484         read_unlock(&bond->lock);
1485 }
1486
1487 /* assumption: called before the slave is attached to the bond
1488  * and not locked by the bond lock
1489  */
1490 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1491 {
1492         int res;
1493
1494         res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1495                                      bond->alb_info.rlb_enabled);
1496         if (res) {
1497                 return res;
1498         }
1499
1500         /* caller must hold the bond lock for write since the mac addresses
1501          * are compared and may be swapped.
1502          */
1503         read_lock(&bond->lock);
1504
1505         res = alb_handle_addr_collision_on_attach(bond, slave);
1506
1507         read_unlock(&bond->lock);
1508
1509         if (res) {
1510                 return res;
1511         }
1512
1513         tlb_init_slave(slave);
1514
1515         /* order a rebalance ASAP */
1516         bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1517
1518         if (bond->alb_info.rlb_enabled) {
1519                 bond->alb_info.rlb_rebalance = 1;
1520         }
1521
1522         return 0;
1523 }
1524
1525 /*
1526  * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1527  * if necessary.
1528  *
1529  * Caller must hold RTNL and no other locks
1530  */
1531 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1532 {
1533         if (bond->slave_cnt > 1) {
1534                 alb_change_hw_addr_on_detach(bond, slave);
1535         }
1536
1537         tlb_clear_slave(bond, slave, 0);
1538
1539         if (bond->alb_info.rlb_enabled) {
1540                 bond->alb_info.next_rx_slave = NULL;
1541                 rlb_clear_slave(bond, slave);
1542         }
1543 }
1544
1545 /* Caller must hold bond lock for read */
1546 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1547 {
1548         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1549
1550         if (link == BOND_LINK_DOWN) {
1551                 tlb_clear_slave(bond, slave, 0);
1552                 if (bond->alb_info.rlb_enabled) {
1553                         rlb_clear_slave(bond, slave);
1554                 }
1555         } else if (link == BOND_LINK_UP) {
1556                 /* order a rebalance ASAP */
1557                 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1558                 if (bond->alb_info.rlb_enabled) {
1559                         bond->alb_info.rlb_rebalance = 1;
1560                         /* If the updelay module parameter is smaller than the
1561                          * forwarding delay of the switch the rebalance will
1562                          * not work because the rebalance arp replies will
1563                          * not be forwarded to the clients..
1564                          */
1565                 }
1566         }
1567 }
1568
1569 /**
1570  * bond_alb_handle_active_change - assign new curr_active_slave
1571  * @bond: our bonding struct
1572  * @new_slave: new slave to assign
1573  *
1574  * Set the bond->curr_active_slave to @new_slave and handle
1575  * mac address swapping and promiscuity changes as needed.
1576  *
1577  * If new_slave is NULL, caller must hold curr_slave_lock or
1578  * bond->lock for write.
1579  *
1580  * If new_slave is not NULL, caller must hold RTNL, bond->lock for
1581  * read and curr_slave_lock for write.  Processing here may sleep, so
1582  * no other locks may be held.
1583  */
1584 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1585         __releases(&bond->curr_slave_lock)
1586         __releases(&bond->lock)
1587         __acquires(&bond->lock)
1588         __acquires(&bond->curr_slave_lock)
1589 {
1590         struct slave *swap_slave;
1591         int i;
1592
1593         if (bond->curr_active_slave == new_slave) {
1594                 return;
1595         }
1596
1597         if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1598                 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1599                 bond->alb_info.primary_is_promisc = 0;
1600                 bond->alb_info.rlb_promisc_timeout_counter = 0;
1601         }
1602
1603         swap_slave = bond->curr_active_slave;
1604         bond->curr_active_slave = new_slave;
1605
1606         if (!new_slave || (bond->slave_cnt == 0)) {
1607                 return;
1608         }
1609
1610         /* set the new curr_active_slave to the bonds mac address
1611          * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1612          */
1613         if (!swap_slave) {
1614                 struct slave *tmp_slave;
1615                 /* find slave that is holding the bond's mac address */
1616                 bond_for_each_slave(bond, tmp_slave, i) {
1617                         if (!compare_ether_addr_64bits(tmp_slave->dev->dev_addr,
1618                                                        bond->dev->dev_addr)) {
1619                                 swap_slave = tmp_slave;
1620                                 break;
1621                         }
1622                 }
1623         }
1624
1625         /*
1626          * Arrange for swap_slave and new_slave to temporarily be
1627          * ignored so we can mess with their MAC addresses without
1628          * fear of interference from transmit activity.
1629          */
1630         if (swap_slave) {
1631                 tlb_clear_slave(bond, swap_slave, 1);
1632         }
1633         tlb_clear_slave(bond, new_slave, 1);
1634
1635         write_unlock_bh(&bond->curr_slave_lock);
1636         read_unlock(&bond->lock);
1637
1638         ASSERT_RTNL();
1639
1640         /* curr_active_slave must be set before calling alb_swap_mac_addr */
1641         if (swap_slave) {
1642                 /* swap mac address */
1643                 alb_swap_mac_addr(bond, swap_slave, new_slave);
1644         } else {
1645                 /* set the new_slave to the bond mac address */
1646                 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1647                                        bond->alb_info.rlb_enabled);
1648         }
1649
1650         if (swap_slave) {
1651                 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1652                 read_lock(&bond->lock);
1653         } else {
1654                 read_lock(&bond->lock);
1655                 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1656         }
1657
1658         write_lock_bh(&bond->curr_slave_lock);
1659 }
1660
1661 /*
1662  * Called with RTNL
1663  */
1664 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1665         __acquires(&bond->lock)
1666         __releases(&bond->lock)
1667 {
1668         struct bonding *bond = netdev_priv(bond_dev);
1669         struct sockaddr *sa = addr;
1670         struct slave *slave, *swap_slave;
1671         int res;
1672         int i;
1673
1674         if (!is_valid_ether_addr(sa->sa_data)) {
1675                 return -EADDRNOTAVAIL;
1676         }
1677
1678         res = alb_set_mac_address(bond, addr);
1679         if (res) {
1680                 return res;
1681         }
1682
1683         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1684
1685         /* If there is no curr_active_slave there is nothing else to do.
1686          * Otherwise we'll need to pass the new address to it and handle
1687          * duplications.
1688          */
1689         if (!bond->curr_active_slave) {
1690                 return 0;
1691         }
1692
1693         swap_slave = NULL;
1694
1695         bond_for_each_slave(bond, slave, i) {
1696                 if (!compare_ether_addr_64bits(slave->dev->dev_addr,
1697                                                bond_dev->dev_addr)) {
1698                         swap_slave = slave;
1699                         break;
1700                 }
1701         }
1702
1703         if (swap_slave) {
1704                 alb_swap_mac_addr(bond, swap_slave, bond->curr_active_slave);
1705                 alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave);
1706         } else {
1707                 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr,
1708                                        bond->alb_info.rlb_enabled);
1709
1710                 read_lock(&bond->lock);
1711                 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1712                 if (bond->alb_info.rlb_enabled) {
1713                         /* inform clients mac address has changed */
1714                         rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1715                 }
1716                 read_unlock(&bond->lock);
1717         }
1718
1719         return 0;
1720 }
1721
1722 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1723 {
1724         if (bond->alb_info.current_alb_vlan &&
1725             (bond->alb_info.current_alb_vlan->vlan_id == vlan_id)) {
1726                 bond->alb_info.current_alb_vlan = NULL;
1727         }
1728
1729         if (bond->alb_info.rlb_enabled) {
1730                 rlb_clear_vlan(bond, vlan_id);
1731         }
1732 }
1733