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 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
 101         return hash;
 102 }
 103
 104 /*********************** tlb specific functions ***************************/
 105
 106 static inline void _lock_tx_hashtbl_bh(struct bonding *bond)
 107 {
 108         spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 109 }
 110
 111 static inline void _unlock_tx_hashtbl_bh(struct bonding *bond)
 112 {
 113         spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 114 }
 115
 116 static inline void _lock_tx_hashtbl(struct bonding *bond)
 117 {
 118         spin_lock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 119 }
 120
 121 static inline void _unlock_tx_hashtbl(struct bonding *bond)
 122 {
 123         spin_unlock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
 124 }
 125
 126 /* Caller must hold tx_hashtbl lock */
 127 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
 128 {
 129         if (save_load) {
 130                 entry->load_history = 1 + entry->tx_bytes /
 131                                       BOND_TLB_REBALANCE_INTERVAL;
 132                 entry->tx_bytes = 0;
 133         }
 134
 135         entry->tx_slave = NULL;
 136         entry->next = TLB_NULL_INDEX;
 137         entry->prev = TLB_NULL_INDEX;
 138 }
 139
 140 static inline void tlb_init_slave(struct slave *slave)
 141 {
 142         SLAVE_TLB_INFO(slave).load = 0;
 143         SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
 144 }
 145
 146 /* Caller must hold bond lock for read, BH disabled */
 147 static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
 148                          int save_load)
 149 {
 150         struct tlb_client_info *tx_hash_table;
 151         u32 index;
 152
 153         /* clear slave from tx_hashtbl */
 154         tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
 155
 156         /* skip this if we've already freed the tx hash table */
 157         if (tx_hash_table) {
 158                 index = SLAVE_TLB_INFO(slave).head;
 159                 while (index != TLB_NULL_INDEX) {
 160                         u32 next_index = tx_hash_table[index].next;
 161                         tlb_init_table_entry(&tx_hash_table[index], save_load);
 162                         index = next_index;
 163                 }
 164         }
 165
 166         tlb_init_slave(slave);
 167 }
 168
 169 /* Caller must hold bond lock for read */
 170 static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
 171                          int save_load)
 172 {
 173         _lock_tx_hashtbl_bh(bond);
 174         __tlb_clear_slave(bond, slave, save_load);
 175         _unlock_tx_hashtbl_bh(bond);
 176 }
 177
 178 /* Must be called before starting the monitor timer */
 179 static int tlb_initialize(struct bonding *bond)
 180 {
 181         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 182         int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 183         struct tlb_client_info *new_hashtbl;
 184         int i;
 185
 186         new_hashtbl = kzalloc(size, GFP_KERNEL);
 187         if (!new_hashtbl)
 188                 return -1;
 189
 190         _lock_tx_hashtbl_bh(bond);
 191
 192         bond_info->tx_hashtbl = new_hashtbl;
 193
 194         for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
 195                 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
 196         }
 197
 198         _unlock_tx_hashtbl_bh(bond);
 199
 200         return 0;
 201 }
 202
 203 /* Must be called only after all slaves have been released */
 204 static void tlb_deinitialize(struct bonding *bond)
 205 {
 206         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 207
 208         _lock_tx_hashtbl_bh(bond);
 209
 210         kfree(bond_info->tx_hashtbl);
 211         bond_info->tx_hashtbl = NULL;
 212
 213         _unlock_tx_hashtbl_bh(bond);
 214 }
 215
 216 static long long compute_gap(struct slave *slave)
 217 {
 218         return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
 219                (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
 220 }
 221
 222 /* Caller must hold bond lock for read */
 223 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
 224 {
 225         struct slave *slave, *least_loaded;
 226         struct list_head *iter;
 227         long long max_gap;
 228
 229         least_loaded = NULL;
 230         max_gap = LLONG_MIN;
 231
 232         /* Find the slave with the largest gap */
 233         bond_for_each_slave(bond, slave, iter) {
 234                 if (SLAVE_IS_OK(slave)) {
 235                         long long gap = compute_gap(slave);
 236
 237                         if (max_gap < gap) {
 238                                 least_loaded = slave;
 239                                 max_gap = gap;
 240                         }
 241                 }
 242         }
 243
 244         return least_loaded;
 245 }
 246
 247 static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
 248                                                 u32 skb_len)
 249 {
 250         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 251         struct tlb_client_info *hash_table;
 252         struct slave *assigned_slave;
 253
 254         hash_table = bond_info->tx_hashtbl;
 255         assigned_slave = hash_table[hash_index].tx_slave;
 256         if (!assigned_slave) {
 257                 assigned_slave = tlb_get_least_loaded_slave(bond);
 258
 259                 if (assigned_slave) {
 260                         struct tlb_slave_info *slave_info =
 261                                 &(SLAVE_TLB_INFO(assigned_slave));
 262                         u32 next_index = slave_info->head;
 263
 264                         hash_table[hash_index].tx_slave = assigned_slave;
 265                         hash_table[hash_index].next = next_index;
 266                         hash_table[hash_index].prev = TLB_NULL_INDEX;
 267
 268                         if (next_index != TLB_NULL_INDEX) {
 269                                 hash_table[next_index].prev = hash_index;
 270                         }
 271
 272                         slave_info->head = hash_index;
 273                         slave_info->load +=
 274                                 hash_table[hash_index].load_history;
 275                 }
 276         }
 277
 278         if (assigned_slave) {
 279                 hash_table[hash_index].tx_bytes += skb_len;
 280         }
 281
 282         return assigned_slave;
 283 }
 284
 285 /* Caller must hold bond lock for read */
 286 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
 287                                         u32 skb_len)
 288 {
 289         struct slave *tx_slave;
 290         /*
 291          * We don't need to disable softirq here, becase
 292          * tlb_choose_channel() is only called by bond_alb_xmit()
 293          * which already has softirq disabled.
 294          */
 295         _lock_tx_hashtbl(bond);
 296         tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
 297         _unlock_tx_hashtbl(bond);
 298         return tx_slave;
 299 }
 300
 301 /*********************** rlb specific functions ***************************/
 302 static inline void _lock_rx_hashtbl_bh(struct bonding *bond)
 303 {
 304         spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 305 }
 306
 307 static inline void _unlock_rx_hashtbl_bh(struct bonding *bond)
 308 {
 309         spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 310 }
 311
 312 static inline void _lock_rx_hashtbl(struct bonding *bond)
 313 {
 314         spin_lock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 315 }
 316
 317 static inline void _unlock_rx_hashtbl(struct bonding *bond)
 318 {
 319         spin_unlock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
 320 }
 321
 322 /* when an ARP REPLY is received from a client update its info
 323  * in the rx_hashtbl
 324  */
 325 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 326 {
 327         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 328         struct rlb_client_info *client_info;
 329         u32 hash_index;
 330
 331         _lock_rx_hashtbl_bh(bond);
 332
 333         hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
 334         client_info = &(bond_info->rx_hashtbl[hash_index]);
 335
 336         if ((client_info->assigned) &&
 337             (client_info->ip_src == arp->ip_dst) &&
 338             (client_info->ip_dst == arp->ip_src) &&
 339             (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) {
 340                 /* update the clients MAC address */
 341                 memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
 342                 client_info->ntt = 1;
 343                 bond_info->rx_ntt = 1;
 344         }
 345
 346         _unlock_rx_hashtbl_bh(bond);
 347 }
 348
 349 static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
 350                         struct slave *slave)
 351 {
 352         struct arp_pkt *arp, _arp;
 353
 354         if (skb->protocol != cpu_to_be16(ETH_P_ARP))
 355                 goto out;
 356
 357         arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp);
 358         if (!arp)
 359                 goto out;
 360
 361         /* We received an ARP from arp->ip_src.
 362          * We might have used this IP address previously (on the bonding host
 363          * itself or on a system that is bridged together with the bond).
 364          * However, if arp->mac_src is different than what is stored in
 365          * rx_hashtbl, some other host is now using the IP and we must prevent
 366          * sending out client updates with this IP address and the old MAC
 367          * address.
 368          * Clean up all hash table entries that have this address as ip_src but
 369          * have a different mac_src.
 370          */
 371         rlb_purge_src_ip(bond, arp);
 372
 373         if (arp->op_code == htons(ARPOP_REPLY)) {
 374                 /* update rx hash table for this ARP */
 375                 rlb_update_entry_from_arp(bond, arp);
 376                 pr_debug("Server received an ARP Reply from client\n");
 377         }
 378 out:
 379         return RX_HANDLER_ANOTHER;
 380 }
 381
 382 /* Caller must hold bond lock for read */
 383 static struct slave *rlb_next_rx_slave(struct bonding *bond)
 384 {
 385         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 386         struct slave *before = NULL, *rx_slave = NULL, *slave;
 387         struct list_head *iter;
 388         bool found = false;
 389
 390         bond_for_each_slave(bond, slave, iter) {
 391                 if (!SLAVE_IS_OK(slave))
 392                         continue;
 393                 if (!found) {
 394                         if (!before || before->speed < slave->speed)
 395                                 before = slave;
 396                 } else {
 397                         if (!rx_slave || rx_slave->speed < slave->speed)
 398                                 rx_slave = slave;
 399                 }
 400                 if (slave == bond_info->rx_slave)
 401                         found = true;
 402         }
 403         /* we didn't find anything after the current or we have something
 404          * better before and up to the current slave
 405          */
 406         if (!rx_slave || (before && rx_slave->speed < before->speed))
 407                 rx_slave = before;
 408
 409         if (rx_slave)
 410                 bond_info->rx_slave = rx_slave;
 411
 412         return rx_slave;
 413 }
 414
 415 /* teach the switch the mac of a disabled slave
 416  * on the primary for fault tolerance
 417  *
 418  * Caller must hold bond->curr_slave_lock for write or bond lock for write
 419  */
 420 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
 421 {
 422         if (!bond->curr_active_slave) {
 423                 return;
 424         }
 425
 426         if (!bond->alb_info.primary_is_promisc) {
 427                 if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1))
 428                         bond->alb_info.primary_is_promisc = 1;
 429                 else
 430                         bond->alb_info.primary_is_promisc = 0;
 431         }
 432
 433         bond->alb_info.rlb_promisc_timeout_counter = 0;
 434
 435         alb_send_learning_packets(bond->curr_active_slave, addr);
 436 }
 437
 438 /* slave being removed should not be active at this point
 439  *
 440  * Caller must hold bond lock for read
 441  */
 442 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 443 {
 444         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 445         struct rlb_client_info *rx_hash_table;
 446         u32 index, next_index;
 447
 448         /* clear slave from rx_hashtbl */
 449         _lock_rx_hashtbl_bh(bond);
 450
 451         rx_hash_table = bond_info->rx_hashtbl;
 452         index = bond_info->rx_hashtbl_used_head;
 453         for (; index != RLB_NULL_INDEX; index = next_index) {
 454                 next_index = rx_hash_table[index].used_next;
 455                 if (rx_hash_table[index].slave == slave) {
 456                         struct slave *assigned_slave = rlb_next_rx_slave(bond);
 457
 458                         if (assigned_slave) {
 459                                 rx_hash_table[index].slave = assigned_slave;
 460                                 if (!ether_addr_equal_64bits(rx_hash_table[index].mac_dst,
 461                                                              mac_bcast)) {
 462                                         bond_info->rx_hashtbl[index].ntt = 1;
 463                                         bond_info->rx_ntt = 1;
 464                                         /* A slave has been removed from the
 465                                          * table because it is either disabled
 466                                          * or being released. We must retry the
 467                                          * update to avoid clients from not
 468                                          * being updated & disconnecting when
 469                                          * there is stress
 470                                          */
 471                                         bond_info->rlb_update_retry_counter =
 472                                                 RLB_UPDATE_RETRY;
 473                                 }
 474                         } else {  /* there is no active slave */
 475                                 rx_hash_table[index].slave = NULL;
 476                         }
 477                 }
 478         }
 479
 480         _unlock_rx_hashtbl_bh(bond);
 481
 482         write_lock_bh(&bond->curr_slave_lock);
 483
 484         if (slave != bond->curr_active_slave) {
 485                 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 486         }
 487
 488         write_unlock_bh(&bond->curr_slave_lock);
 489 }
 490
 491 static void rlb_update_client(struct rlb_client_info *client_info)
 492 {
 493         int i;
 494
 495         if (!client_info->slave) {
 496                 return;
 497         }
 498
 499         for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 500                 struct sk_buff *skb;
 501
 502                 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 503                                  client_info->ip_dst,
 504                                  client_info->slave->dev,
 505                                  client_info->ip_src,
 506                                  client_info->mac_dst,
 507                                  client_info->slave->dev->dev_addr,
 508                                  client_info->mac_dst);
 509                 if (!skb) {
 510                         pr_err("%s: Error: failed to create an ARP packet\n",
 511                                client_info->slave->bond->dev->name);
 512                         continue;
 513                 }
 514
 515                 skb->dev = client_info->slave->dev;
 516
 517                 if (client_info->vlan_id) {
 518                         skb = vlan_put_tag(skb, htons(ETH_P_8021Q), client_info->vlan_id);
 519                         if (!skb) {
 520                                 pr_err("%s: Error: failed to insert VLAN tag\n",
 521                                        client_info->slave->bond->dev->name);
 522                                 continue;
 523                         }
 524                 }
 525
 526                 arp_xmit(skb);
 527         }
 528 }
 529
 530 /* sends ARP REPLIES that update the clients that need updating */
 531 static void rlb_update_rx_clients(struct bonding *bond)
 532 {
 533         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 534         struct rlb_client_info *client_info;
 535         u32 hash_index;
 536
 537         _lock_rx_hashtbl_bh(bond);
 538
 539         hash_index = bond_info->rx_hashtbl_used_head;
 540         for (; hash_index != RLB_NULL_INDEX;
 541              hash_index = client_info->used_next) {
 542                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 543                 if (client_info->ntt) {
 544                         rlb_update_client(client_info);
 545                         if (bond_info->rlb_update_retry_counter == 0) {
 546                                 client_info->ntt = 0;
 547                         }
 548                 }
 549         }
 550
 551         /* do not update the entries again until this counter is zero so that
 552          * not to confuse the clients.
 553          */
 554         bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 555
 556         _unlock_rx_hashtbl_bh(bond);
 557 }
 558
 559 /* The slave was assigned a new mac address - update the clients */
 560 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 561 {
 562         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 563         struct rlb_client_info *client_info;
 564         int ntt = 0;
 565         u32 hash_index;
 566
 567         _lock_rx_hashtbl_bh(bond);
 568
 569         hash_index = bond_info->rx_hashtbl_used_head;
 570         for (; hash_index != RLB_NULL_INDEX;
 571              hash_index = client_info->used_next) {
 572                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 573
 574                 if ((client_info->slave == slave) &&
 575                     !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
 576                         client_info->ntt = 1;
 577                         ntt = 1;
 578                 }
 579         }
 580
 581         // update the team's flag only after the whole iteration
 582         if (ntt) {
 583                 bond_info->rx_ntt = 1;
 584                 //fasten the change
 585                 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 586         }
 587
 588         _unlock_rx_hashtbl_bh(bond);
 589 }
 590
 591 /* mark all clients using src_ip to be updated */
 592 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
 593 {
 594         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 595         struct rlb_client_info *client_info;
 596         u32 hash_index;
 597
 598         _lock_rx_hashtbl(bond);
 599
 600         hash_index = bond_info->rx_hashtbl_used_head;
 601         for (; hash_index != RLB_NULL_INDEX;
 602              hash_index = client_info->used_next) {
 603                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 604
 605                 if (!client_info->slave) {
 606                         pr_err("%s: Error: found a client with no channel in the client's hash table\n",
 607                                bond->dev->name);
 608                         continue;
 609                 }
 610                 /*update all clients using this src_ip, that are not assigned
 611                  * to the team's address (curr_active_slave) and have a known
 612                  * unicast mac address.
 613                  */
 614                 if ((client_info->ip_src == src_ip) &&
 615                     !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
 616                                              bond->dev->dev_addr) &&
 617                     !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
 618                         client_info->ntt = 1;
 619                         bond_info->rx_ntt = 1;
 620                 }
 621         }
 622
 623         _unlock_rx_hashtbl(bond);
 624 }
 625
 626 /* Caller must hold both bond and ptr locks for read */
 627 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
 628 {
 629         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 630         struct arp_pkt *arp = arp_pkt(skb);
 631         struct slave *assigned_slave;
 632         struct rlb_client_info *client_info;
 633         u32 hash_index = 0;
 634
 635         _lock_rx_hashtbl(bond);
 636
 637         hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
 638         client_info = &(bond_info->rx_hashtbl[hash_index]);
 639
 640         if (client_info->assigned) {
 641                 if ((client_info->ip_src == arp->ip_src) &&
 642                     (client_info->ip_dst == arp->ip_dst)) {
 643                         /* the entry is already assigned to this client */
 644                         if (!ether_addr_equal_64bits(arp->mac_dst, mac_bcast)) {
 645                                 /* update mac address from arp */
 646                                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 647                         }
 648                         memcpy(client_info->mac_src, arp->mac_src, ETH_ALEN);
 649
 650                         assigned_slave = client_info->slave;
 651                         if (assigned_slave) {
 652                                 _unlock_rx_hashtbl(bond);
 653                                 return assigned_slave;
 654                         }
 655                 } else {
 656                         /* the entry is already assigned to some other client,
 657                          * move the old client to primary (curr_active_slave) so
 658                          * that the new client can be assigned to this entry.
 659                          */
 660                         if (bond->curr_active_slave &&
 661                             client_info->slave != bond->curr_active_slave) {
 662                                 client_info->slave = bond->curr_active_slave;
 663                                 rlb_update_client(client_info);
 664                         }
 665                 }
 666         }
 667         /* assign a new slave */
 668         assigned_slave = rlb_next_rx_slave(bond);
 669
 670         if (assigned_slave) {
 671                 if (!(client_info->assigned &&
 672                       client_info->ip_src == arp->ip_src)) {
 673                         /* ip_src is going to be updated,
 674                          * fix the src hash list
 675                          */
 676                         u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
 677                                                     sizeof(arp->ip_src));
 678                         rlb_src_unlink(bond, hash_index);
 679                         rlb_src_link(bond, hash_src, hash_index);
 680                 }
 681
 682                 client_info->ip_src = arp->ip_src;
 683                 client_info->ip_dst = arp->ip_dst;
 684                 /* arp->mac_dst is broadcast for arp reqeusts.
 685                  * will be updated with clients actual unicast mac address
 686                  * upon receiving an arp reply.
 687                  */
 688                 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
 689                 memcpy(client_info->mac_src, arp->mac_src, ETH_ALEN);
 690                 client_info->slave = assigned_slave;
 691
 692                 if (!ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
 693                         client_info->ntt = 1;
 694                         bond->alb_info.rx_ntt = 1;
 695                 } else {
 696                         client_info->ntt = 0;
 697                 }
 698
 699                 if (!vlan_get_tag(skb, &client_info->vlan_id))
 700                         client_info->vlan_id = 0;
 701
 702                 if (!client_info->assigned) {
 703                         u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
 704                         bond_info->rx_hashtbl_used_head = hash_index;
 705                         client_info->used_next = prev_tbl_head;
 706                         if (prev_tbl_head != RLB_NULL_INDEX) {
 707                                 bond_info->rx_hashtbl[prev_tbl_head].used_prev =
 708                                         hash_index;
 709                         }
 710                         client_info->assigned = 1;
 711                 }
 712         }
 713
 714         _unlock_rx_hashtbl(bond);
 715
 716         return assigned_slave;
 717 }
 718
 719 /* chooses (and returns) transmit channel for arp reply
 720  * does not choose channel for other arp types since they are
 721  * sent on the curr_active_slave
 722  */
 723 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 724 {
 725         struct arp_pkt *arp = arp_pkt(skb);
 726         struct slave *tx_slave = NULL;
 727
 728         /* Don't modify or load balance ARPs that do not originate locally
 729          * (e.g.,arrive via a bridge).
 730          */
 731         if (!bond_slave_has_mac(bond, arp->mac_src))
 732                 return NULL;
 733
 734         if (arp->op_code == htons(ARPOP_REPLY)) {
 735                 /* the arp must be sent on the selected
 736                 * rx channel
 737                 */
 738                 tx_slave = rlb_choose_channel(skb, bond);
 739                 if (tx_slave) {
 740                         memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
 741                 }
 742                 pr_debug("Server sent ARP Reply packet\n");
 743         } else if (arp->op_code == htons(ARPOP_REQUEST)) {
 744                 /* Create an entry in the rx_hashtbl for this client as a
 745                  * place holder.
 746                  * When the arp reply is received the entry will be updated
 747                  * with the correct unicast address of the client.
 748                  */
 749                 rlb_choose_channel(skb, bond);
 750
 751                 /* The ARP reply packets must be delayed so that
 752                  * they can cancel out the influence of the ARP request.
 753                  */
 754                 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 755
 756                 /* arp requests are broadcast and are sent on the primary
 757                  * the arp request will collapse all clients on the subnet to
 758                  * the primary slave. We must register these clients to be
 759                  * updated with their assigned mac.
 760                  */
 761                 rlb_req_update_subnet_clients(bond, arp->ip_src);
 762                 pr_debug("Server sent ARP Request packet\n");
 763         }
 764
 765         return tx_slave;
 766 }
 767
 768 /* Caller must hold bond lock for read */
 769 static void rlb_rebalance(struct bonding *bond)
 770 {
 771         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 772         struct slave *assigned_slave;
 773         struct rlb_client_info *client_info;
 774         int ntt;
 775         u32 hash_index;
 776
 777         _lock_rx_hashtbl_bh(bond);
 778
 779         ntt = 0;
 780         hash_index = bond_info->rx_hashtbl_used_head;
 781         for (; hash_index != RLB_NULL_INDEX;
 782              hash_index = client_info->used_next) {
 783                 client_info = &(bond_info->rx_hashtbl[hash_index]);
 784                 assigned_slave = rlb_next_rx_slave(bond);
 785                 if (assigned_slave && (client_info->slave != assigned_slave)) {
 786                         client_info->slave = assigned_slave;
 787                         client_info->ntt = 1;
 788                         ntt = 1;
 789                 }
 790         }
 791
 792         /* update the team's flag only after the whole iteration */
 793         if (ntt) {
 794                 bond_info->rx_ntt = 1;
 795         }
 796         _unlock_rx_hashtbl_bh(bond);
 797 }
 798
 799 /* Caller must hold rx_hashtbl lock */
 800 static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
 801 {
 802         entry->used_next = RLB_NULL_INDEX;
 803         entry->used_prev = RLB_NULL_INDEX;
 804         entry->assigned = 0;
 805         entry->slave = NULL;
 806         entry->vlan_id = 0;
 807 }
 808 static void rlb_init_table_entry_src(struct rlb_client_info *entry)
 809 {
 810         entry->src_first = RLB_NULL_INDEX;
 811         entry->src_prev = RLB_NULL_INDEX;
 812         entry->src_next = RLB_NULL_INDEX;
 813 }
 814
 815 static void rlb_init_table_entry(struct rlb_client_info *entry)
 816 {
 817         memset(entry, 0, sizeof(struct rlb_client_info));
 818         rlb_init_table_entry_dst(entry);
 819         rlb_init_table_entry_src(entry);
 820 }
 821
 822 static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
 823 {
 824         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 825         u32 next_index = bond_info->rx_hashtbl[index].used_next;
 826         u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
 827
 828         if (index == bond_info->rx_hashtbl_used_head)
 829                 bond_info->rx_hashtbl_used_head = next_index;
 830         if (prev_index != RLB_NULL_INDEX)
 831                 bond_info->rx_hashtbl[prev_index].used_next = next_index;
 832         if (next_index != RLB_NULL_INDEX)
 833                 bond_info->rx_hashtbl[next_index].used_prev = prev_index;
 834 }
 835
 836 /* unlink a rlb hash table entry from the src list */
 837 static void rlb_src_unlink(struct bonding *bond, u32 index)
 838 {
 839         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 840         u32 next_index = bond_info->rx_hashtbl[index].src_next;
 841         u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
 842
 843         bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
 844         bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
 845
 846         if (next_index != RLB_NULL_INDEX)
 847                 bond_info->rx_hashtbl[next_index].src_prev = prev_index;
 848
 849         if (prev_index == RLB_NULL_INDEX)
 850                 return;
 851
 852         /* is prev_index pointing to the head of this list? */
 853         if (bond_info->rx_hashtbl[prev_index].src_first == index)
 854                 bond_info->rx_hashtbl[prev_index].src_first = next_index;
 855         else
 856                 bond_info->rx_hashtbl[prev_index].src_next = next_index;
 857
 858 }
 859
 860 static void rlb_delete_table_entry(struct bonding *bond, u32 index)
 861 {
 862         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 863         struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 864
 865         rlb_delete_table_entry_dst(bond, index);
 866         rlb_init_table_entry_dst(entry);
 867
 868         rlb_src_unlink(bond, index);
 869 }
 870
 871 /* add the rx_hashtbl[ip_dst_hash] entry to the list
 872  * of entries with identical ip_src_hash
 873  */
 874 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
 875 {
 876         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 877         u32 next;
 878
 879         bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
 880         next = bond_info->rx_hashtbl[ip_src_hash].src_first;
 881         bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
 882         if (next != RLB_NULL_INDEX)
 883                 bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
 884         bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
 885 }
 886
 887 /* deletes all rx_hashtbl entries with  arp->ip_src if their mac_src does
 888  * not match arp->mac_src */
 889 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
 890 {
 891         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 892         u32 ip_src_hash = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
 893         u32 index;
 894
 895         _lock_rx_hashtbl_bh(bond);
 896
 897         index = bond_info->rx_hashtbl[ip_src_hash].src_first;
 898         while (index != RLB_NULL_INDEX) {
 899                 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 900                 u32 next_index = entry->src_next;
 901                 if (entry->ip_src == arp->ip_src &&
 902                     !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
 903                                 rlb_delete_table_entry(bond, index);
 904                 index = next_index;
 905         }
 906         _unlock_rx_hashtbl_bh(bond);
 907 }
 908
 909 static int rlb_initialize(struct bonding *bond)
 910 {
 911         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 912         struct rlb_client_info  *new_hashtbl;
 913         int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 914         int i;
 915
 916         new_hashtbl = kmalloc(size, GFP_KERNEL);
 917         if (!new_hashtbl)
 918                 return -1;
 919
 920         _lock_rx_hashtbl_bh(bond);
 921
 922         bond_info->rx_hashtbl = new_hashtbl;
 923
 924         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 925
 926         for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
 927                 rlb_init_table_entry(bond_info->rx_hashtbl + i);
 928         }
 929
 930         _unlock_rx_hashtbl_bh(bond);
 931
 932         /* register to receive ARPs */
 933         bond->recv_probe = rlb_arp_recv;
 934
 935         return 0;
 936 }
 937
 938 static void rlb_deinitialize(struct bonding *bond)
 939 {
 940         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 941
 942         _lock_rx_hashtbl_bh(bond);
 943
 944         kfree(bond_info->rx_hashtbl);
 945         bond_info->rx_hashtbl = NULL;
 946         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 947
 948         _unlock_rx_hashtbl_bh(bond);
 949 }
 950
 951 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 952 {
 953         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 954         u32 curr_index;
 955
 956         _lock_rx_hashtbl_bh(bond);
 957
 958         curr_index = bond_info->rx_hashtbl_used_head;
 959         while (curr_index != RLB_NULL_INDEX) {
 960                 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 961                 u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
 962
 963                 if (curr->vlan_id == vlan_id)
 964                         rlb_delete_table_entry(bond, curr_index);
 965
 966                 curr_index = next_index;
 967         }
 968
 969         _unlock_rx_hashtbl_bh(bond);
 970 }
 971
 972 /*********************** tlb/rlb shared functions *********************/
 973
 974 static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
 975                             u16 vid)
 976 {
 977         struct learning_pkt pkt;
 978         struct sk_buff *skb;
 979         int size = sizeof(struct learning_pkt);
 980         char *data;
 981
 982         memset(&pkt, 0, size);
 983         memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
 984         memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
 985         pkt.type = cpu_to_be16(ETH_P_LOOP);
 986
 987         skb = dev_alloc_skb(size);
 988         if (!skb)
 989                 return;
 990
 991         data = skb_put(skb, size);
 992         memcpy(data, &pkt, size);
 993
 994         skb_reset_mac_header(skb);
 995         skb->network_header = skb->mac_header + ETH_HLEN;
 996         skb->protocol = pkt.type;
 997         skb->priority = TC_PRIO_CONTROL;
 998         skb->dev = slave->dev;
 999
1000         if (vid) {
1001                 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vid);
1002                 if (!skb) {
1003                         pr_err("%s: Error: failed to insert VLAN tag\n",
1004                                slave->bond->dev->name);
1005                         return;
1006                 }
1007         }
1008
1009         dev_queue_xmit(skb);
1010 }
1011
1012
1013 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
1014 {
1015         struct bonding *bond = bond_get_bond_by_slave(slave);
1016         struct net_device *upper;
1017         struct list_head *iter;
1018
1019         /* send untagged */
1020         alb_send_lp_vid(slave, mac_addr, 0);
1021
1022         /* loop through vlans and send one packet for each */
1023         rcu_read_lock();
1024         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
1025                 if (upper->priv_flags & IFF_802_1Q_VLAN)
1026                         alb_send_lp_vid(slave, mac_addr,
1027                                         vlan_dev_vlan_id(upper));
1028         }
1029         rcu_read_unlock();
1030 }
1031
1032 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[])
1033 {
1034         struct net_device *dev = slave->dev;
1035         struct sockaddr s_addr;
1036
1037         if (slave->bond->params.mode == BOND_MODE_TLB) {
1038                 memcpy(dev->dev_addr, addr, dev->addr_len);
1039                 return 0;
1040         }
1041
1042         /* for rlb each slave must have a unique hw mac addresses so that */
1043         /* each slave will receive packets destined to a different mac */
1044         memcpy(s_addr.sa_data, addr, dev->addr_len);
1045         s_addr.sa_family = dev->type;
1046         if (dev_set_mac_address(dev, &s_addr)) {
1047                 pr_err("%s: Error: dev_set_mac_address of dev %s failed!\n"
1048                        "ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
1049                        slave->bond->dev->name, dev->name);
1050                 return -EOPNOTSUPP;
1051         }
1052         return 0;
1053 }
1054
1055 /*
1056  * Swap MAC addresses between two slaves.
1057  *
1058  * Called with RTNL held, and no other locks.
1059  *
1060  */
1061
1062 static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1063 {
1064         u8 tmp_mac_addr[ETH_ALEN];
1065
1066         memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
1067         alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr);
1068         alb_set_slave_mac_addr(slave2, tmp_mac_addr);
1069
1070 }
1071
1072 /*
1073  * Send learning packets after MAC address swap.
1074  *
1075  * Called with RTNL and no other locks
1076  */
1077 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1078                                 struct slave *slave2)
1079 {
1080         int slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
1081         struct slave *disabled_slave = NULL;
1082
1083         ASSERT_RTNL();
1084
1085         /* fasten the change in the switch */
1086         if (SLAVE_IS_OK(slave1)) {
1087                 alb_send_learning_packets(slave1, slave1->dev->dev_addr);
1088                 if (bond->alb_info.rlb_enabled) {
1089                         /* inform the clients that the mac address
1090                          * has changed
1091                          */
1092                         rlb_req_update_slave_clients(bond, slave1);
1093                 }
1094         } else {
1095                 disabled_slave = slave1;
1096         }
1097
1098         if (SLAVE_IS_OK(slave2)) {
1099                 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
1100                 if (bond->alb_info.rlb_enabled) {
1101                         /* inform the clients that the mac address
1102                          * has changed
1103                          */
1104                         rlb_req_update_slave_clients(bond, slave2);
1105                 }
1106         } else {
1107                 disabled_slave = slave2;
1108         }
1109
1110         if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1111                 /* A disabled slave was assigned an active mac addr */
1112                 rlb_teach_disabled_mac_on_primary(bond,
1113                                                   disabled_slave->dev->dev_addr);
1114         }
1115 }
1116
1117 /**
1118  * alb_change_hw_addr_on_detach
1119  * @bond: bonding we're working on
1120  * @slave: the slave that was just detached
1121  *
1122  * We assume that @slave was already detached from the slave list.
1123  *
1124  * If @slave's permanent hw address is different both from its current
1125  * address and from @bond's address, then somewhere in the bond there's
1126  * a slave that has @slave's permanet address as its current address.
1127  * We'll make sure that that slave no longer uses @slave's permanent address.
1128  *
1129  * Caller must hold RTNL and no other locks
1130  */
1131 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1132 {
1133         int perm_curr_diff;
1134         int perm_bond_diff;
1135         struct slave *found_slave;
1136
1137         perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1138                                                   slave->dev->dev_addr);
1139         perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1140                                                   bond->dev->dev_addr);
1141
1142         if (perm_curr_diff && perm_bond_diff) {
1143                 found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1144
1145                 if (found_slave) {
1146                         /* locking: needs RTNL and nothing else */
1147                         alb_swap_mac_addr(slave, found_slave);
1148                         alb_fasten_mac_swap(bond, slave, found_slave);
1149                 }
1150         }
1151 }
1152
1153 /**
1154  * alb_handle_addr_collision_on_attach
1155  * @bond: bonding we're working on
1156  * @slave: the slave that was just attached
1157  *
1158  * checks uniqueness of slave's mac address and handles the case the
1159  * new slave uses the bonds mac address.
1160  *
1161  * If the permanent hw address of @slave is @bond's hw address, we need to
1162  * find a different hw address to give @slave, that isn't in use by any other
1163  * slave in the bond. This address must be, of course, one of the permanent
1164  * addresses of the other slaves.
1165  *
1166  * We go over the slave list, and for each slave there we compare its
1167  * permanent hw address with the current address of all the other slaves.
1168  * If no match was found, then we've found a slave with a permanent address
1169  * that isn't used by any other slave in the bond, so we can assign it to
1170  * @slave.
1171  *
1172  * assumption: this function is called before @slave is attached to the
1173  *             bond slave list.
1174  */
1175 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1176 {
1177         struct slave *has_bond_addr = bond->curr_active_slave;
1178         struct slave *tmp_slave1, *free_mac_slave = NULL;
1179         struct list_head *iter;
1180
1181         if (!bond_has_slaves(bond)) {
1182                 /* this is the first slave */
1183                 return 0;
1184         }
1185
1186         /* if slave's mac address differs from bond's mac address
1187          * check uniqueness of slave's mac address against the other
1188          * slaves in the bond.
1189          */
1190         if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1191                 if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1192                         return 0;
1193
1194                 /* Try setting slave mac to bond address and fall-through
1195                    to code handling that situation below... */
1196                 alb_set_slave_mac_addr(slave, bond->dev->dev_addr);
1197         }
1198
1199         /* The slave's address is equal to the address of the bond.
1200          * Search for a spare address in the bond for this slave.
1201          */
1202         bond_for_each_slave(bond, tmp_slave1, iter) {
1203                 if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1204                         /* no slave has tmp_slave1's perm addr
1205                          * as its curr addr
1206                          */
1207                         free_mac_slave = tmp_slave1;
1208                         break;
1209                 }
1210
1211                 if (!has_bond_addr) {
1212                         if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1213                                                     bond->dev->dev_addr)) {
1214
1215                                 has_bond_addr = tmp_slave1;
1216                         }
1217                 }
1218         }
1219
1220         if (free_mac_slave) {
1221                 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr);
1222
1223                 pr_warning("%s: Warning: the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1224                            bond->dev->name, slave->dev->name,
1225                            free_mac_slave->dev->name);
1226
1227         } else if (has_bond_addr) {
1228                 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",
1229                        bond->dev->name, slave->dev->name);
1230                 return -EFAULT;
1231         }
1232
1233         return 0;
1234 }
1235
1236 /**
1237  * alb_set_mac_address
1238  * @bond:
1239  * @addr:
1240  *
1241  * In TLB mode all slaves are configured to the bond's hw address, but set
1242  * their dev_addr field to different addresses (based on their permanent hw
1243  * addresses).
1244  *
1245  * For each slave, this function sets the interface to the new address and then
1246  * changes its dev_addr field to its previous value.
1247  *
1248  * Unwinding assumes bond's mac address has not yet changed.
1249  */
1250 static int alb_set_mac_address(struct bonding *bond, void *addr)
1251 {
1252         struct slave *slave, *rollback_slave;
1253         struct list_head *iter;
1254         struct sockaddr sa;
1255         char tmp_addr[ETH_ALEN];
1256         int res;
1257
1258         if (bond->alb_info.rlb_enabled)
1259                 return 0;
1260
1261         bond_for_each_slave(bond, slave, iter) {
1262                 /* save net_device's current hw address */
1263                 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1264
1265                 res = dev_set_mac_address(slave->dev, addr);
1266
1267                 /* restore net_device's hw address */
1268                 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1269
1270                 if (res)
1271                         goto unwind;
1272         }
1273
1274         return 0;
1275
1276 unwind:
1277         memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1278         sa.sa_family = bond->dev->type;
1279
1280         /* unwind from head to the slave that failed */
1281         bond_for_each_slave(bond, rollback_slave, iter) {
1282                 if (rollback_slave == slave)
1283                         break;
1284                 memcpy(tmp_addr, rollback_slave->dev->dev_addr, ETH_ALEN);
1285                 dev_set_mac_address(rollback_slave->dev, &sa);
1286                 memcpy(rollback_slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1287         }
1288
1289         return res;
1290 }
1291
1292 /************************ exported alb funcions ************************/
1293
1294 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1295 {
1296         int res;
1297
1298         res = tlb_initialize(bond);
1299         if (res) {
1300                 return res;
1301         }
1302
1303         if (rlb_enabled) {
1304                 bond->alb_info.rlb_enabled = 1;
1305                 /* initialize rlb */
1306                 res = rlb_initialize(bond);
1307                 if (res) {
1308                         tlb_deinitialize(bond);
1309                         return res;
1310                 }
1311         } else {
1312                 bond->alb_info.rlb_enabled = 0;
1313         }
1314
1315         return 0;
1316 }
1317
1318 void bond_alb_deinitialize(struct bonding *bond)
1319 {
1320         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1321
1322         tlb_deinitialize(bond);
1323
1324         if (bond_info->rlb_enabled) {
1325                 rlb_deinitialize(bond);
1326         }
1327 }
1328
1329 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1330 {
1331         struct bonding *bond = netdev_priv(bond_dev);
1332         struct ethhdr *eth_data;
1333         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1334         struct slave *tx_slave = NULL;
1335         static const __be32 ip_bcast = htonl(0xffffffff);
1336         int hash_size = 0;
1337         int do_tx_balance = 1;
1338         u32 hash_index = 0;
1339         const u8 *hash_start = NULL;
1340         int res = 1;
1341         struct ipv6hdr *ip6hdr;
1342
1343         skb_reset_mac_header(skb);
1344         eth_data = eth_hdr(skb);
1345
1346         /* make sure that the curr_active_slave do not change during tx
1347          */
1348         read_lock(&bond->lock);
1349         read_lock(&bond->curr_slave_lock);
1350
1351         switch (ntohs(skb->protocol)) {
1352         case ETH_P_IP: {
1353                 const struct iphdr *iph = ip_hdr(skb);
1354
1355                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast) ||
1356                     (iph->daddr == ip_bcast) ||
1357                     (iph->protocol == IPPROTO_IGMP)) {
1358                         do_tx_balance = 0;
1359                         break;
1360                 }
1361                 hash_start = (char *)&(iph->daddr);
1362                 hash_size = sizeof(iph->daddr);
1363         }
1364                 break;
1365         case ETH_P_IPV6:
1366                 /* IPv6 doesn't really use broadcast mac address, but leave
1367                  * that here just in case.
1368                  */
1369                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast)) {
1370                         do_tx_balance = 0;
1371                         break;
1372                 }
1373
1374                 /* IPv6 uses all-nodes multicast as an equivalent to
1375                  * broadcasts in IPv4.
1376                  */
1377                 if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1378                         do_tx_balance = 0;
1379                         break;
1380                 }
1381
1382                 /* Additianally, DAD probes should not be tx-balanced as that
1383                  * will lead to false positives for duplicate addresses and
1384                  * prevent address configuration from working.
1385                  */
1386                 ip6hdr = ipv6_hdr(skb);
1387                 if (ipv6_addr_any(&ip6hdr->saddr)) {
1388                         do_tx_balance = 0;
1389                         break;
1390                 }
1391
1392                 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1393                 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1394                 break;
1395         case ETH_P_IPX:
1396                 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1397                         /* something is wrong with this packet */
1398                         do_tx_balance = 0;
1399                         break;
1400                 }
1401
1402                 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1403                         /* The only protocol worth balancing in
1404                          * this family since it has an "ARP" like
1405                          * mechanism
1406                          */
1407                         do_tx_balance = 0;
1408                         break;
1409                 }
1410
1411                 hash_start = (char*)eth_data->h_dest;
1412                 hash_size = ETH_ALEN;
1413                 break;
1414         case ETH_P_ARP:
1415                 do_tx_balance = 0;
1416                 if (bond_info->rlb_enabled) {
1417                         tx_slave = rlb_arp_xmit(skb, bond);
1418                 }
1419                 break;
1420         default:
1421                 do_tx_balance = 0;
1422                 break;
1423         }
1424
1425         if (do_tx_balance) {
1426                 hash_index = _simple_hash(hash_start, hash_size);
1427                 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1428         }
1429
1430         if (!tx_slave) {
1431                 /* unbalanced or unassigned, send through primary */
1432                 tx_slave = bond->curr_active_slave;
1433                 bond_info->unbalanced_load += skb->len;
1434         }
1435
1436         if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1437                 if (tx_slave != bond->curr_active_slave) {
1438                         memcpy(eth_data->h_source,
1439                                tx_slave->dev->dev_addr,
1440                                ETH_ALEN);
1441                 }
1442
1443                 res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1444         } else {
1445                 if (tx_slave) {
1446                         _lock_tx_hashtbl(bond);
1447                         __tlb_clear_slave(bond, tx_slave, 0);
1448                         _unlock_tx_hashtbl(bond);
1449                 }
1450         }
1451
1452         read_unlock(&bond->curr_slave_lock);
1453         read_unlock(&bond->lock);
1454         if (res) {
1455                 /* no suitable interface, frame not sent */
1456                 kfree_skb(skb);
1457         }
1458
1459         return NETDEV_TX_OK;
1460 }
1461
1462 void bond_alb_monitor(struct work_struct *work)
1463 {
1464         struct bonding *bond = container_of(work, struct bonding,
1465                                             alb_work.work);
1466         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1467         struct list_head *iter;
1468         struct slave *slave;
1469
1470         read_lock(&bond->lock);
1471
1472         if (!bond_has_slaves(bond)) {
1473                 bond_info->tx_rebalance_counter = 0;
1474                 bond_info->lp_counter = 0;
1475                 goto re_arm;
1476         }
1477
1478         bond_info->tx_rebalance_counter++;
1479         bond_info->lp_counter++;
1480
1481         /* send learning packets */
1482         if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1483                 /* change of curr_active_slave involves swapping of mac addresses.
1484                  * in order to avoid this swapping from happening while
1485                  * sending the learning packets, the curr_slave_lock must be held for
1486                  * read.
1487                  */
1488                 read_lock(&bond->curr_slave_lock);
1489
1490                 bond_for_each_slave(bond, slave, iter)
1491                         alb_send_learning_packets(slave, slave->dev->dev_addr);
1492
1493                 read_unlock(&bond->curr_slave_lock);
1494
1495                 bond_info->lp_counter = 0;
1496         }
1497
1498         /* rebalance tx traffic */
1499         if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1500
1501                 read_lock(&bond->curr_slave_lock);
1502
1503                 bond_for_each_slave(bond, slave, iter) {
1504                         tlb_clear_slave(bond, slave, 1);
1505                         if (slave == bond->curr_active_slave) {
1506                                 SLAVE_TLB_INFO(slave).load =
1507                                         bond_info->unbalanced_load /
1508                                                 BOND_TLB_REBALANCE_INTERVAL;
1509                                 bond_info->unbalanced_load = 0;
1510                         }
1511                 }
1512
1513                 read_unlock(&bond->curr_slave_lock);
1514
1515                 bond_info->tx_rebalance_counter = 0;
1516         }
1517
1518         /* handle rlb stuff */
1519         if (bond_info->rlb_enabled) {
1520                 if (bond_info->primary_is_promisc &&
1521                     (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1522
1523                         /*
1524                          * dev_set_promiscuity requires rtnl and
1525                          * nothing else.  Avoid race with bond_close.
1526                          */
1527                         read_unlock(&bond->lock);
1528                         if (!rtnl_trylock()) {
1529                                 read_lock(&bond->lock);
1530                                 goto re_arm;
1531                         }
1532
1533                         bond_info->rlb_promisc_timeout_counter = 0;
1534
1535                         /* If the primary was set to promiscuous mode
1536                          * because a slave was disabled then
1537                          * it can now leave promiscuous mode.
1538                          */
1539                         dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1540                         bond_info->primary_is_promisc = 0;
1541
1542                         rtnl_unlock();
1543                         read_lock(&bond->lock);
1544                 }
1545
1546                 if (bond_info->rlb_rebalance) {
1547                         bond_info->rlb_rebalance = 0;
1548                         rlb_rebalance(bond);
1549                 }
1550
1551                 /* check if clients need updating */
1552                 if (bond_info->rx_ntt) {
1553                         if (bond_info->rlb_update_delay_counter) {
1554                                 --bond_info->rlb_update_delay_counter;
1555                         } else {
1556                                 rlb_update_rx_clients(bond);
1557                                 if (bond_info->rlb_update_retry_counter) {
1558                                         --bond_info->rlb_update_retry_counter;
1559                                 } else {
1560                                         bond_info->rx_ntt = 0;
1561                                 }
1562                         }
1563                 }
1564         }
1565
1566 re_arm:
1567         queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1568
1569         read_unlock(&bond->lock);
1570 }
1571
1572 /* assumption: called before the slave is attached to the bond
1573  * and not locked by the bond lock
1574  */
1575 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1576 {
1577         int res;
1578
1579         res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr);
1580         if (res) {
1581                 return res;
1582         }
1583
1584         res = alb_handle_addr_collision_on_attach(bond, slave);
1585         if (res) {
1586                 return res;
1587         }
1588
1589         tlb_init_slave(slave);
1590
1591         /* order a rebalance ASAP */
1592         bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1593
1594         if (bond->alb_info.rlb_enabled) {
1595                 bond->alb_info.rlb_rebalance = 1;
1596         }
1597
1598         return 0;
1599 }
1600
1601 /*
1602  * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1603  * if necessary.
1604  *
1605  * Caller must hold RTNL and no other locks
1606  */
1607 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1608 {
1609         if (bond_has_slaves(bond))
1610                 alb_change_hw_addr_on_detach(bond, slave);
1611
1612         tlb_clear_slave(bond, slave, 0);
1613
1614         if (bond->alb_info.rlb_enabled) {
1615                 bond->alb_info.rx_slave = NULL;
1616                 rlb_clear_slave(bond, slave);
1617         }
1618 }
1619
1620 /* Caller must hold bond lock for read */
1621 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1622 {
1623         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1624
1625         if (link == BOND_LINK_DOWN) {
1626                 tlb_clear_slave(bond, slave, 0);
1627                 if (bond->alb_info.rlb_enabled) {
1628                         rlb_clear_slave(bond, slave);
1629                 }
1630         } else if (link == BOND_LINK_UP) {
1631                 /* order a rebalance ASAP */
1632                 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1633                 if (bond->alb_info.rlb_enabled) {
1634                         bond->alb_info.rlb_rebalance = 1;
1635                         /* If the updelay module parameter is smaller than the
1636                          * forwarding delay of the switch the rebalance will
1637                          * not work because the rebalance arp replies will
1638                          * not be forwarded to the clients..
1639                          */
1640                 }
1641         }
1642 }
1643
1644 /**
1645  * bond_alb_handle_active_change - assign new curr_active_slave
1646  * @bond: our bonding struct
1647  * @new_slave: new slave to assign
1648  *
1649  * Set the bond->curr_active_slave to @new_slave and handle
1650  * mac address swapping and promiscuity changes as needed.
1651  *
1652  * If new_slave is NULL, caller must hold curr_slave_lock or
1653  * bond->lock for write.
1654  *
1655  * If new_slave is not NULL, caller must hold RTNL, bond->lock for
1656  * read and curr_slave_lock for write.  Processing here may sleep, so
1657  * no other locks may be held.
1658  */
1659 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1660         __releases(&bond->curr_slave_lock)
1661         __releases(&bond->lock)
1662         __acquires(&bond->lock)
1663         __acquires(&bond->curr_slave_lock)
1664 {
1665         struct slave *swap_slave;
1666
1667         if (bond->curr_active_slave == new_slave)
1668                 return;
1669
1670         if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1671                 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1672                 bond->alb_info.primary_is_promisc = 0;
1673                 bond->alb_info.rlb_promisc_timeout_counter = 0;
1674         }
1675
1676         swap_slave = bond->curr_active_slave;
1677         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1678
1679         if (!new_slave || !bond_has_slaves(bond))
1680                 return;
1681
1682         /* set the new curr_active_slave to the bonds mac address
1683          * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1684          */
1685         if (!swap_slave)
1686                 swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1687
1688         /*
1689          * Arrange for swap_slave and new_slave to temporarily be
1690          * ignored so we can mess with their MAC addresses without
1691          * fear of interference from transmit activity.
1692          */
1693         if (swap_slave)
1694                 tlb_clear_slave(bond, swap_slave, 1);
1695         tlb_clear_slave(bond, new_slave, 1);
1696
1697         write_unlock_bh(&bond->curr_slave_lock);
1698         read_unlock(&bond->lock);
1699
1700         ASSERT_RTNL();
1701
1702         /* curr_active_slave must be set before calling alb_swap_mac_addr */
1703         if (swap_slave) {
1704                 /* swap mac address */
1705                 alb_swap_mac_addr(swap_slave, new_slave);
1706                 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1707                 read_lock(&bond->lock);
1708         } else {
1709                 /* set the new_slave to the bond mac address */
1710                 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr);
1711                 read_lock(&bond->lock);
1712                 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1713         }
1714
1715         write_lock_bh(&bond->curr_slave_lock);
1716 }
1717
1718 /*
1719  * Called with RTNL
1720  */
1721 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1722         __acquires(&bond->lock)
1723         __releases(&bond->lock)
1724 {
1725         struct bonding *bond = netdev_priv(bond_dev);
1726         struct sockaddr *sa = addr;
1727         struct slave *swap_slave;
1728         int res;
1729
1730         if (!is_valid_ether_addr(sa->sa_data)) {
1731                 return -EADDRNOTAVAIL;
1732         }
1733
1734         res = alb_set_mac_address(bond, addr);
1735         if (res) {
1736                 return res;
1737         }
1738
1739         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1740
1741         /* If there is no curr_active_slave there is nothing else to do.
1742          * Otherwise we'll need to pass the new address to it and handle
1743          * duplications.
1744          */
1745         if (!bond->curr_active_slave) {
1746                 return 0;
1747         }
1748
1749         swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1750
1751         if (swap_slave) {
1752                 alb_swap_mac_addr(swap_slave, bond->curr_active_slave);
1753                 alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave);
1754         } else {
1755                 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr);
1756
1757                 read_lock(&bond->lock);
1758                 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1759                 if (bond->alb_info.rlb_enabled) {
1760                         /* inform clients mac address has changed */
1761                         rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1762                 }
1763                 read_unlock(&bond->lock);
1764         }
1765
1766         return 0;
1767 }
1768
1769 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1770 {
1771         if (bond->alb_info.rlb_enabled) {
1772                 rlb_clear_vlan(bond, vlan_id);
1773         }
1774 }
1775