1/*
   2 * Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved.
   3 *
   4 * This program is free software; you can redistribute it and/or modify it
   5 * under the terms of the GNU General Public License as published by the
   6 * Free Software Foundation; either version 2 of the License, or
   7 * (at your option) any later version.
   8 *
   9 * This program is distributed in the hope that it will be useful, but
  10 * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  11 * or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12 * for more details.
  13 *
  14 * You should have received a copy of the GNU General Public License along
  15 * with this program; if not, see <http://www.gnu.org/licenses/>.
  16 *
  17 * The full GNU General Public License is included in this distribution in the
  18 * file called LICENSE.
  19 *
  20 */
  21
  22#include <linux/skbuff.h>
  23#include <linux/netdevice.h>
  24#include <linux/etherdevice.h>
  25#include <linux/pkt_sched.h>
  26#include <linux/spinlock.h>
  27#include <linux/slab.h>
  28#include <linux/timer.h>
  29#include <linux/ip.h>
  30#include <linux/ipv6.h>
  31#include <linux/if_arp.h>
  32#include <linux/if_ether.h>
  33#include <linux/if_bonding.h>
  34#include <linux/if_vlan.h>
  35#include <linux/in.h>
  36#include <net/ipx.h>
  37#include <net/arp.h>
  38#include <net/ipv6.h>
  39#include <asm/byteorder.h>
  40#include <net/bonding.h>
  41#include <net/bond_alb.h>
  42
  43
  44
  45static const u8 mac_bcast[ETH_ALEN + 2] __long_aligned = {
  46	0xff, 0xff, 0xff, 0xff, 0xff, 0xff
  47};
  48static const u8 mac_v6_allmcast[ETH_ALEN + 2] __long_aligned = {
  49	0x33, 0x33, 0x00, 0x00, 0x00, 0x01
  50};
  51static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
  52
  53#pragma pack(1)
  54struct learning_pkt {
  55	u8 mac_dst[ETH_ALEN];
  56	u8 mac_src[ETH_ALEN];
  57	__be16 type;
  58	u8 padding[ETH_ZLEN - ETH_HLEN];
  59};
  60
  61struct arp_pkt {
  62	__be16  hw_addr_space;
  63	__be16  prot_addr_space;
  64	u8      hw_addr_len;
  65	u8      prot_addr_len;
  66	__be16  op_code;
  67	u8      mac_src[ETH_ALEN];	/* sender hardware address */
  68	__be32  ip_src;			/* sender IP address */
  69	u8      mac_dst[ETH_ALEN];	/* target hardware address */
  70	__be32  ip_dst;			/* target IP address */
  71};
  72#pragma pack()
  73
  74static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb)
  75{
  76	return (struct arp_pkt *)skb_network_header(skb);
  77}
  78
  79/* Forward declaration */
  80static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
  81				      bool strict_match);
  82static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp);
  83static void rlb_src_unlink(struct bonding *bond, u32 index);
  84static void rlb_src_link(struct bonding *bond, u32 ip_src_hash,
  85			 u32 ip_dst_hash);
  86
  87static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
  88{
  89	int i;
  90	u8 hash = 0;
  91
  92	for (i = 0; i < hash_size; i++)
  93		hash ^= hash_start[i];
  94
  95	return hash;
  96}
  97
  98/*********************** tlb specific functions ***************************/
  99
 100static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
 101{
 102	if (save_load) {
 103		entry->load_history = 1 + entry->tx_bytes /
 104				      BOND_TLB_REBALANCE_INTERVAL;
 105		entry->tx_bytes = 0;
 106	}
 107
 108	entry->tx_slave = NULL;
 109	entry->next = TLB_NULL_INDEX;
 110	entry->prev = TLB_NULL_INDEX;
 111}
 112
 113static inline void tlb_init_slave(struct slave *slave)
 114{
 115	SLAVE_TLB_INFO(slave).load = 0;
 116	SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
 117}
 118
 119static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
 120			 int save_load)
 121{
 122	struct tlb_client_info *tx_hash_table;
 123	u32 index;
 124
 125	/* clear slave from tx_hashtbl */
 126	tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
 127
 128	/* skip this if we've already freed the tx hash table */
 129	if (tx_hash_table) {
 130		index = SLAVE_TLB_INFO(slave).head;
 131		while (index != TLB_NULL_INDEX) {
 132			u32 next_index = tx_hash_table[index].next;
 133			tlb_init_table_entry(&tx_hash_table[index], save_load);
 134			index = next_index;
 135		}
 136	}
 137
 138	tlb_init_slave(slave);
 139}
 140
 141static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
 142			 int save_load)
 143{
 144	spin_lock_bh(&bond->mode_lock);
 145	__tlb_clear_slave(bond, slave, save_load);
 146	spin_unlock_bh(&bond->mode_lock);
 147}
 148
 149/* Must be called before starting the monitor timer */
 150static int tlb_initialize(struct bonding *bond)
 151{
 152	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 153	int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 154	struct tlb_client_info *new_hashtbl;
 155	int i;
 156
 157	new_hashtbl = kzalloc(size, GFP_KERNEL);
 158	if (!new_hashtbl)
 159		return -ENOMEM;
 160
 161	spin_lock_bh(&bond->mode_lock);
 162
 163	bond_info->tx_hashtbl = new_hashtbl;
 164
 165	for (i = 0; i < TLB_HASH_TABLE_SIZE; i++)
 166		tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
 167
 168	spin_unlock_bh(&bond->mode_lock);
 169
 170	return 0;
 171}
 172
 173/* Must be called only after all slaves have been released */
 174static void tlb_deinitialize(struct bonding *bond)
 175{
 176	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 177
 178	spin_lock_bh(&bond->mode_lock);
 179
 180	kfree(bond_info->tx_hashtbl);
 181	bond_info->tx_hashtbl = NULL;
 182
 183	spin_unlock_bh(&bond->mode_lock);
 184}
 185
 186static long long compute_gap(struct slave *slave)
 187{
 188	return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
 189	       (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
 190}
 191
 192static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
 193{
 194	struct slave *slave, *least_loaded;
 195	struct list_head *iter;
 196	long long max_gap;
 197
 198	least_loaded = NULL;
 199	max_gap = LLONG_MIN;
 200
 201	/* Find the slave with the largest gap */
 202	bond_for_each_slave_rcu(bond, slave, iter) {
 203		if (bond_slave_can_tx(slave)) {
 204			long long gap = compute_gap(slave);
 205
 206			if (max_gap < gap) {
 207				least_loaded = slave;
 208				max_gap = gap;
 209			}
 210		}
 211	}
 212
 213	return least_loaded;
 214}
 215
 216static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
 217						u32 skb_len)
 218{
 219	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 220	struct tlb_client_info *hash_table;
 221	struct slave *assigned_slave;
 222
 223	hash_table = bond_info->tx_hashtbl;
 224	assigned_slave = hash_table[hash_index].tx_slave;
 225	if (!assigned_slave) {
 226		assigned_slave = tlb_get_least_loaded_slave(bond);
 227
 228		if (assigned_slave) {
 229			struct tlb_slave_info *slave_info =
 230				&(SLAVE_TLB_INFO(assigned_slave));
 231			u32 next_index = slave_info->head;
 232
 233			hash_table[hash_index].tx_slave = assigned_slave;
 234			hash_table[hash_index].next = next_index;
 235			hash_table[hash_index].prev = TLB_NULL_INDEX;
 236
 237			if (next_index != TLB_NULL_INDEX)
 238				hash_table[next_index].prev = hash_index;
 239
 240			slave_info->head = hash_index;
 241			slave_info->load +=
 242				hash_table[hash_index].load_history;
 243		}
 244	}
 245
 246	if (assigned_slave)
 247		hash_table[hash_index].tx_bytes += skb_len;
 248
 249	return assigned_slave;
 250}
 251
 252static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
 253					u32 skb_len)
 254{
 255	struct slave *tx_slave;
 256
 257	/* We don't need to disable softirq here, becase
 258	 * tlb_choose_channel() is only called by bond_alb_xmit()
 259	 * which already has softirq disabled.
 260	 */
 261	spin_lock(&bond->mode_lock);
 262	tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
 263	spin_unlock(&bond->mode_lock);
 264
 265	return tx_slave;
 266}
 267
 268/*********************** rlb specific functions ***************************/
 269
 270/* when an ARP REPLY is received from a client update its info
 271 * in the rx_hashtbl
 272 */
 273static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 274{
 275	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 276	struct rlb_client_info *client_info;
 277	u32 hash_index;
 278
 279	spin_lock_bh(&bond->mode_lock);
 280
 281	hash_index = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
 282	client_info = &(bond_info->rx_hashtbl[hash_index]);
 283
 284	if ((client_info->assigned) &&
 285	    (client_info->ip_src == arp->ip_dst) &&
 286	    (client_info->ip_dst == arp->ip_src) &&
 287	    (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) {
 288		/* update the clients MAC address */
 289		ether_addr_copy(client_info->mac_dst, arp->mac_src);
 290		client_info->ntt = 1;
 291		bond_info->rx_ntt = 1;
 292	}
 293
 294	spin_unlock_bh(&bond->mode_lock);
 295}
 296
 297static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
 298			struct slave *slave)
 299{
 300	struct arp_pkt *arp, _arp;
 301
 302	if (skb->protocol != cpu_to_be16(ETH_P_ARP))
 303		goto out;
 304
 305	arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp);
 306	if (!arp)
 307		goto out;
 308
 309	/* We received an ARP from arp->ip_src.
 310	 * We might have used this IP address previously (on the bonding host
 311	 * itself or on a system that is bridged together with the bond).
 312	 * However, if arp->mac_src is different than what is stored in
 313	 * rx_hashtbl, some other host is now using the IP and we must prevent
 314	 * sending out client updates with this IP address and the old MAC
 315	 * address.
 316	 * Clean up all hash table entries that have this address as ip_src but
 317	 * have a different mac_src.
 318	 */
 319	rlb_purge_src_ip(bond, arp);
 320
 321	if (arp->op_code == htons(ARPOP_REPLY)) {
 322		/* update rx hash table for this ARP */
 323		rlb_update_entry_from_arp(bond, arp);
 324		netdev_dbg(bond->dev, "Server received an ARP Reply from client\n");
 325	}
 326out:
 327	return RX_HANDLER_ANOTHER;
 328}
 329
 330/* Caller must hold rcu_read_lock() */
 331static struct slave *__rlb_next_rx_slave(struct bonding *bond)
 332{
 333	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 334	struct slave *before = NULL, *rx_slave = NULL, *slave;
 335	struct list_head *iter;
 336	bool found = false;
 337
 338	bond_for_each_slave_rcu(bond, slave, iter) {
 339		if (!bond_slave_can_tx(slave))
 340			continue;
 341		if (!found) {
 342			if (!before || before->speed < slave->speed)
 343				before = slave;
 344		} else {
 345			if (!rx_slave || rx_slave->speed < slave->speed)
 346				rx_slave = slave;
 347		}
 348		if (slave == bond_info->rx_slave)
 349			found = true;
 350	}
 351	/* we didn't find anything after the current or we have something
 352	 * better before and up to the current slave
 353	 */
 354	if (!rx_slave || (before && rx_slave->speed < before->speed))
 355		rx_slave = before;
 356
 357	if (rx_slave)
 358		bond_info->rx_slave = rx_slave;
 359
 360	return rx_slave;
 361}
 362
 363/* Caller must hold RTNL, rcu_read_lock is obtained only to silence checkers */
 364static struct slave *rlb_next_rx_slave(struct bonding *bond)
 365{
 366	struct slave *rx_slave;
 367
 368	ASSERT_RTNL();
 369
 370	rcu_read_lock();
 371	rx_slave = __rlb_next_rx_slave(bond);
 372	rcu_read_unlock();
 373
 374	return rx_slave;
 375}
 376
 377/* teach the switch the mac of a disabled slave
 378 * on the primary for fault tolerance
 379 *
 380 * Caller must hold RTNL
 381 */
 382static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
 383{
 384	struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 385
 386	if (!curr_active)
 387		return;
 388
 389	if (!bond->alb_info.primary_is_promisc) {
 390		if (!dev_set_promiscuity(curr_active->dev, 1))
 391			bond->alb_info.primary_is_promisc = 1;
 392		else
 393			bond->alb_info.primary_is_promisc = 0;
 394	}
 395
 396	bond->alb_info.rlb_promisc_timeout_counter = 0;
 397
 398	alb_send_learning_packets(curr_active, addr, true);
 399}
 400
 401/* slave being removed should not be active at this point
 402 *
 403 * Caller must hold rtnl.
 404 */
 405static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 406{
 407	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 408	struct rlb_client_info *rx_hash_table;
 409	u32 index, next_index;
 410
 411	/* clear slave from rx_hashtbl */
 412	spin_lock_bh(&bond->mode_lock);
 413
 414	rx_hash_table = bond_info->rx_hashtbl;
 415	index = bond_info->rx_hashtbl_used_head;
 416	for (; index != RLB_NULL_INDEX; index = next_index) {
 417		next_index = rx_hash_table[index].used_next;
 418		if (rx_hash_table[index].slave == slave) {
 419			struct slave *assigned_slave = rlb_next_rx_slave(bond);
 420
 421			if (assigned_slave) {
 422				rx_hash_table[index].slave = assigned_slave;
 423				if (!ether_addr_equal_64bits(rx_hash_table[index].mac_dst,
 424							     mac_bcast)) {
 425					bond_info->rx_hashtbl[index].ntt = 1;
 426					bond_info->rx_ntt = 1;
 427					/* A slave has been removed from the
 428					 * table because it is either disabled
 429					 * or being released. We must retry the
 430					 * update to avoid clients from not
 431					 * being updated & disconnecting when
 432					 * there is stress
 433					 */
 434					bond_info->rlb_update_retry_counter =
 435						RLB_UPDATE_RETRY;
 436				}
 437			} else {  /* there is no active slave */
 438				rx_hash_table[index].slave = NULL;
 439			}
 440		}
 441	}
 442
 443	spin_unlock_bh(&bond->mode_lock);
 444
 445	if (slave != rtnl_dereference(bond->curr_active_slave))
 446		rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 447}
 448
 449static void rlb_update_client(struct rlb_client_info *client_info)
 450{
 451	int i;
 452
 453	if (!client_info->slave || !is_valid_ether_addr(client_info->mac_dst))
 454		return;
 455
 456	for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 457		struct sk_buff *skb;
 458
 459		skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 460				 client_info->ip_dst,
 461				 client_info->slave->dev,
 462				 client_info->ip_src,
 463				 client_info->mac_dst,
 464				 client_info->slave->dev->dev_addr,
 465				 client_info->mac_dst);
 466		if (!skb) {
 467			netdev_err(client_info->slave->bond->dev,
 468				   "failed to create an ARP packet\n");
 469			continue;
 470		}
 471
 472		skb->dev = client_info->slave->dev;
 473
 474		if (client_info->vlan_id) {
 475			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 476					       client_info->vlan_id);
 477		}
 478
 479		arp_xmit(skb);
 480	}
 481}
 482
 483/* sends ARP REPLIES that update the clients that need updating */
 484static void rlb_update_rx_clients(struct bonding *bond)
 485{
 486	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 487	struct rlb_client_info *client_info;
 488	u32 hash_index;
 489
 490	spin_lock_bh(&bond->mode_lock);
 491
 492	hash_index = bond_info->rx_hashtbl_used_head;
 493	for (; hash_index != RLB_NULL_INDEX;
 494	     hash_index = client_info->used_next) {
 495		client_info = &(bond_info->rx_hashtbl[hash_index]);
 496		if (client_info->ntt) {
 497			rlb_update_client(client_info);
 498			if (bond_info->rlb_update_retry_counter == 0)
 499				client_info->ntt = 0;
 500		}
 501	}
 502
 503	/* do not update the entries again until this counter is zero so that
 504	 * not to confuse the clients.
 505	 */
 506	bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 507
 508	spin_unlock_bh(&bond->mode_lock);
 509}
 510
 511/* The slave was assigned a new mac address - update the clients */
 512static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 513{
 514	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 515	struct rlb_client_info *client_info;
 516	int ntt = 0;
 517	u32 hash_index;
 518
 519	spin_lock_bh(&bond->mode_lock);
 520
 521	hash_index = bond_info->rx_hashtbl_used_head;
 522	for (; hash_index != RLB_NULL_INDEX;
 523	     hash_index = client_info->used_next) {
 524		client_info = &(bond_info->rx_hashtbl[hash_index]);
 525
 526		if ((client_info->slave == slave) &&
 527		    !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
 528			client_info->ntt = 1;
 529			ntt = 1;
 530		}
 531	}
 532
 533	/* update the team's flag only after the whole iteration */
 534	if (ntt) {
 535		bond_info->rx_ntt = 1;
 536		/* fasten the change */
 537		bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 538	}
 539
 540	spin_unlock_bh(&bond->mode_lock);
 541}
 542
 543/* mark all clients using src_ip to be updated */
 544static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
 545{
 546	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 547	struct rlb_client_info *client_info;
 548	u32 hash_index;
 549
 550	spin_lock(&bond->mode_lock);
 551
 552	hash_index = bond_info->rx_hashtbl_used_head;
 553	for (; hash_index != RLB_NULL_INDEX;
 554	     hash_index = client_info->used_next) {
 555		client_info = &(bond_info->rx_hashtbl[hash_index]);
 556
 557		if (!client_info->slave) {
 558			netdev_err(bond->dev, "found a client with no channel in the client's hash table\n");
 559			continue;
 560		}
 561		/* update all clients using this src_ip, that are not assigned
 562		 * to the team's address (curr_active_slave) and have a known
 563		 * unicast mac address.
 564		 */
 565		if ((client_info->ip_src == src_ip) &&
 566		    !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
 567					     bond->dev->dev_addr) &&
 568		    !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
 569			client_info->ntt = 1;
 570			bond_info->rx_ntt = 1;
 571		}
 572	}
 573
 574	spin_unlock(&bond->mode_lock);
 575}
 576
 577static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
 578{
 579	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 580	struct arp_pkt *arp = arp_pkt(skb);
 581	struct slave *assigned_slave, *curr_active_slave;
 582	struct rlb_client_info *client_info;
 583	u32 hash_index = 0;
 584
 585	spin_lock(&bond->mode_lock);
 586
 587	curr_active_slave = rcu_dereference(bond->curr_active_slave);
 588
 589	hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
 590	client_info = &(bond_info->rx_hashtbl[hash_index]);
 591
 592	if (client_info->assigned) {
 593		if ((client_info->ip_src == arp->ip_src) &&
 594		    (client_info->ip_dst == arp->ip_dst)) {
 595			/* the entry is already assigned to this client */
 596			if (!ether_addr_equal_64bits(arp->mac_dst, mac_bcast)) {
 597				/* update mac address from arp */
 598				ether_addr_copy(client_info->mac_dst, arp->mac_dst);
 599			}
 600			ether_addr_copy(client_info->mac_src, arp->mac_src);
 601
 602			assigned_slave = client_info->slave;
 603			if (assigned_slave) {
 604				spin_unlock(&bond->mode_lock);
 605				return assigned_slave;
 606			}
 607		} else {
 608			/* the entry is already assigned to some other client,
 609			 * move the old client to primary (curr_active_slave) so
 610			 * that the new client can be assigned to this entry.
 611			 */
 612			if (curr_active_slave &&
 613			    client_info->slave != curr_active_slave) {
 614				client_info->slave = curr_active_slave;
 615				rlb_update_client(client_info);
 616			}
 617		}
 618	}
 619	/* assign a new slave */
 620	assigned_slave = __rlb_next_rx_slave(bond);
 621
 622	if (assigned_slave) {
 623		if (!(client_info->assigned &&
 624		      client_info->ip_src == arp->ip_src)) {
 625			/* ip_src is going to be updated,
 626			 * fix the src hash list
 627			 */
 628			u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
 629						    sizeof(arp->ip_src));
 630			rlb_src_unlink(bond, hash_index);
 631			rlb_src_link(bond, hash_src, hash_index);
 632		}
 633
 634		client_info->ip_src = arp->ip_src;
 635		client_info->ip_dst = arp->ip_dst;
 636		/* arp->mac_dst is broadcast for arp reqeusts.
 637		 * will be updated with clients actual unicast mac address
 638		 * upon receiving an arp reply.
 639		 */
 640		ether_addr_copy(client_info->mac_dst, arp->mac_dst);
 641		ether_addr_copy(client_info->mac_src, arp->mac_src);
 642		client_info->slave = assigned_slave;
 643
 644		if (!ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
 645			client_info->ntt = 1;
 646			bond->alb_info.rx_ntt = 1;
 647		} else {
 648			client_info->ntt = 0;
 649		}
 650
 651		if (vlan_get_tag(skb, &client_info->vlan_id))
 652			client_info->vlan_id = 0;
 653
 654		if (!client_info->assigned) {
 655			u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
 656			bond_info->rx_hashtbl_used_head = hash_index;
 657			client_info->used_next = prev_tbl_head;
 658			if (prev_tbl_head != RLB_NULL_INDEX) {
 659				bond_info->rx_hashtbl[prev_tbl_head].used_prev =
 660					hash_index;
 661			}
 662			client_info->assigned = 1;
 663		}
 664	}
 665
 666	spin_unlock(&bond->mode_lock);
 667
 668	return assigned_slave;
 669}
 670
 671/* chooses (and returns) transmit channel for arp reply
 672 * does not choose channel for other arp types since they are
 673 * sent on the curr_active_slave
 674 */
 675static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 676{
 677	struct arp_pkt *arp = arp_pkt(skb);
 678	struct slave *tx_slave = NULL;
 679
 680	/* Don't modify or load balance ARPs that do not originate locally
 681	 * (e.g.,arrive via a bridge).
 682	 */
 683	if (!bond_slave_has_mac_rx(bond, arp->mac_src))
 684		return NULL;
 685
 686	if (arp->op_code == htons(ARPOP_REPLY)) {
 687		/* the arp must be sent on the selected rx channel */
 688		tx_slave = rlb_choose_channel(skb, bond);
 689		if (tx_slave)
 690			bond_hw_addr_copy(arp->mac_src, tx_slave->dev->dev_addr,
 691					  tx_slave->dev->addr_len);
 692		netdev_dbg(bond->dev, "Server sent ARP Reply packet\n");
 693	} else if (arp->op_code == htons(ARPOP_REQUEST)) {
 694		/* Create an entry in the rx_hashtbl for this client as a
 695		 * place holder.
 696		 * When the arp reply is received the entry will be updated
 697		 * with the correct unicast address of the client.
 698		 */
 699		rlb_choose_channel(skb, bond);
 700
 701		/* The ARP reply packets must be delayed so that
 702		 * they can cancel out the influence of the ARP request.
 703		 */
 704		bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 705
 706		/* arp requests are broadcast and are sent on the primary
 707		 * the arp request will collapse all clients on the subnet to
 708		 * the primary slave. We must register these clients to be
 709		 * updated with their assigned mac.
 710		 */
 711		rlb_req_update_subnet_clients(bond, arp->ip_src);
 712		netdev_dbg(bond->dev, "Server sent ARP Request packet\n");
 713	}
 714
 715	return tx_slave;
 716}
 717
 718static void rlb_rebalance(struct bonding *bond)
 719{
 720	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 721	struct slave *assigned_slave;
 722	struct rlb_client_info *client_info;
 723	int ntt;
 724	u32 hash_index;
 725
 726	spin_lock_bh(&bond->mode_lock);
 727
 728	ntt = 0;
 729	hash_index = bond_info->rx_hashtbl_used_head;
 730	for (; hash_index != RLB_NULL_INDEX;
 731	     hash_index = client_info->used_next) {
 732		client_info = &(bond_info->rx_hashtbl[hash_index]);
 733		assigned_slave = __rlb_next_rx_slave(bond);
 734		if (assigned_slave && (client_info->slave != assigned_slave)) {
 735			client_info->slave = assigned_slave;
 736			client_info->ntt = 1;
 737			ntt = 1;
 738		}
 739	}
 740
 741	/* update the team's flag only after the whole iteration */
 742	if (ntt)
 743		bond_info->rx_ntt = 1;
 744	spin_unlock_bh(&bond->mode_lock);
 745}
 746
 747/* Caller must hold mode_lock */
 748static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
 749{
 750	entry->used_next = RLB_NULL_INDEX;
 751	entry->used_prev = RLB_NULL_INDEX;
 752	entry->assigned = 0;
 753	entry->slave = NULL;
 754	entry->vlan_id = 0;
 755}
 756static void rlb_init_table_entry_src(struct rlb_client_info *entry)
 757{
 758	entry->src_first = RLB_NULL_INDEX;
 759	entry->src_prev = RLB_NULL_INDEX;
 760	entry->src_next = RLB_NULL_INDEX;
 761}
 762
 763static void rlb_init_table_entry(struct rlb_client_info *entry)
 764{
 765	memset(entry, 0, sizeof(struct rlb_client_info));
 766	rlb_init_table_entry_dst(entry);
 767	rlb_init_table_entry_src(entry);
 768}
 769
 770static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
 771{
 772	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 773	u32 next_index = bond_info->rx_hashtbl[index].used_next;
 774	u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
 775
 776	if (index == bond_info->rx_hashtbl_used_head)
 777		bond_info->rx_hashtbl_used_head = next_index;
 778	if (prev_index != RLB_NULL_INDEX)
 779		bond_info->rx_hashtbl[prev_index].used_next = next_index;
 780	if (next_index != RLB_NULL_INDEX)
 781		bond_info->rx_hashtbl[next_index].used_prev = prev_index;
 782}
 783
 784/* unlink a rlb hash table entry from the src list */
 785static void rlb_src_unlink(struct bonding *bond, u32 index)
 786{
 787	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 788	u32 next_index = bond_info->rx_hashtbl[index].src_next;
 789	u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
 790
 791	bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
 792	bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
 793
 794	if (next_index != RLB_NULL_INDEX)
 795		bond_info->rx_hashtbl[next_index].src_prev = prev_index;
 796
 797	if (prev_index == RLB_NULL_INDEX)
 798		return;
 799
 800	/* is prev_index pointing to the head of this list? */
 801	if (bond_info->rx_hashtbl[prev_index].src_first == index)
 802		bond_info->rx_hashtbl[prev_index].src_first = next_index;
 803	else
 804		bond_info->rx_hashtbl[prev_index].src_next = next_index;
 805
 806}
 807
 808static void rlb_delete_table_entry(struct bonding *bond, u32 index)
 809{
 810	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 811	struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 812
 813	rlb_delete_table_entry_dst(bond, index);
 814	rlb_init_table_entry_dst(entry);
 815
 816	rlb_src_unlink(bond, index);
 817}
 818
 819/* add the rx_hashtbl[ip_dst_hash] entry to the list
 820 * of entries with identical ip_src_hash
 821 */
 822static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
 823{
 824	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 825	u32 next;
 826
 827	bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
 828	next = bond_info->rx_hashtbl[ip_src_hash].src_first;
 829	bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
 830	if (next != RLB_NULL_INDEX)
 831		bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
 832	bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
 833}
 834
 835/* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does
 836 * not match arp->mac_src
 837 */
 838static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
 839{
 840	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 841	u32 ip_src_hash = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
 842	u32 index;
 843
 844	spin_lock_bh(&bond->mode_lock);
 845
 846	index = bond_info->rx_hashtbl[ip_src_hash].src_first;
 847	while (index != RLB_NULL_INDEX) {
 848		struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 849		u32 next_index = entry->src_next;
 850		if (entry->ip_src == arp->ip_src &&
 851		    !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
 852				rlb_delete_table_entry(bond, index);
 853		index = next_index;
 854	}
 855	spin_unlock_bh(&bond->mode_lock);
 856}
 857
 858static int rlb_initialize(struct bonding *bond)
 859{
 860	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 861	struct rlb_client_info	*new_hashtbl;
 862	int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 863	int i;
 864
 865	new_hashtbl = kmalloc(size, GFP_KERNEL);
 866	if (!new_hashtbl)
 867		return -1;
 868
 869	spin_lock_bh(&bond->mode_lock);
 870
 871	bond_info->rx_hashtbl = new_hashtbl;
 872
 873	bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 874
 875	for (i = 0; i < RLB_HASH_TABLE_SIZE; i++)
 876		rlb_init_table_entry(bond_info->rx_hashtbl + i);
 877
 878	spin_unlock_bh(&bond->mode_lock);
 879
 880	/* register to receive ARPs */
 881	bond->recv_probe = rlb_arp_recv;
 882
 883	return 0;
 884}
 885
 886static void rlb_deinitialize(struct bonding *bond)
 887{
 888	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 889
 890	spin_lock_bh(&bond->mode_lock);
 891
 892	kfree(bond_info->rx_hashtbl);
 893	bond_info->rx_hashtbl = NULL;
 894	bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 895
 896	spin_unlock_bh(&bond->mode_lock);
 897}
 898
 899static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 900{
 901	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 902	u32 curr_index;
 903
 904	spin_lock_bh(&bond->mode_lock);
 905
 906	curr_index = bond_info->rx_hashtbl_used_head;
 907	while (curr_index != RLB_NULL_INDEX) {
 908		struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 909		u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
 910
 911		if (curr->vlan_id == vlan_id)
 912			rlb_delete_table_entry(bond, curr_index);
 913
 914		curr_index = next_index;
 915	}
 916
 917	spin_unlock_bh(&bond->mode_lock);
 918}
 919
 920/*********************** tlb/rlb shared functions *********************/
 921
 922static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
 923			    __be16 vlan_proto, u16 vid)
 924{
 925	struct learning_pkt pkt;
 926	struct sk_buff *skb;
 927	int size = sizeof(struct learning_pkt);
 
 928
 929	memset(&pkt, 0, size);
 930	ether_addr_copy(pkt.mac_dst, mac_addr);
 931	ether_addr_copy(pkt.mac_src, mac_addr);
 932	pkt.type = cpu_to_be16(ETH_P_LOOPBACK);
 933
 934	skb = dev_alloc_skb(size);
 935	if (!skb)
 936		return;
 937
 938	skb_put_data(skb, &pkt, size);
 
 939
 940	skb_reset_mac_header(skb);
 941	skb->network_header = skb->mac_header + ETH_HLEN;
 942	skb->protocol = pkt.type;
 943	skb->priority = TC_PRIO_CONTROL;
 944	skb->dev = slave->dev;
 945
 946	netdev_dbg(slave->bond->dev,
 947		   "Send learning packet: dev %s mac %pM vlan %d\n",
 948		   slave->dev->name, mac_addr, vid);
 949
 950	if (vid)
 951		__vlan_hwaccel_put_tag(skb, vlan_proto, vid);
 952
 953	dev_queue_xmit(skb);
 954}
 955
 956struct alb_walk_data {
 957	struct bonding *bond;
 958	struct slave *slave;
 959	u8 *mac_addr;
 960	bool strict_match;
 961};
 962
 963static int alb_upper_dev_walk(struct net_device *upper, void *_data)
 964{
 965	struct alb_walk_data *data = _data;
 966	bool strict_match = data->strict_match;
 967	struct bonding *bond = data->bond;
 968	struct slave *slave = data->slave;
 969	u8 *mac_addr = data->mac_addr;
 970	struct bond_vlan_tag *tags;
 971
 972	if (is_vlan_dev(upper) &&
 973	    bond->nest_level == vlan_get_encap_level(upper) - 1) {
 974		if (upper->addr_assign_type == NET_ADDR_STOLEN) {
 
 975			alb_send_lp_vid(slave, mac_addr,
 976					vlan_dev_vlan_proto(upper),
 977					vlan_dev_vlan_id(upper));
 978		} else {
 979			alb_send_lp_vid(slave, upper->dev_addr,
 980					vlan_dev_vlan_proto(upper),
 981					vlan_dev_vlan_id(upper));
 982		}
 983	}
 984
 985	/* If this is a macvlan device, then only send updates
 986	 * when strict_match is turned off.
 987	 */
 988	if (netif_is_macvlan(upper) && !strict_match) {
 989		tags = bond_verify_device_path(bond->dev, upper, 0);
 990		if (IS_ERR_OR_NULL(tags))
 991			BUG();
 992		alb_send_lp_vid(slave, upper->dev_addr,
 993				tags[0].vlan_proto, tags[0].vlan_id);
 994		kfree(tags);
 995	}
 996
 997	return 0;
 998}
 999
1000static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
1001				      bool strict_match)
1002{
1003	struct bonding *bond = bond_get_bond_by_slave(slave);
1004	struct alb_walk_data data = {
1005		.strict_match = strict_match,
1006		.mac_addr = mac_addr,
1007		.slave = slave,
1008		.bond = bond,
1009	};
1010
1011	/* send untagged */
1012	alb_send_lp_vid(slave, mac_addr, 0, 0);
1013
1014	/* loop through all devices and see if we need to send a packet
1015	 * for that device.
1016	 */
1017	rcu_read_lock();
1018	netdev_walk_all_upper_dev_rcu(bond->dev, alb_upper_dev_walk, &data);
1019	rcu_read_unlock();
1020}
1021
1022static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[],
1023				  unsigned int len)
1024{
1025	struct net_device *dev = slave->dev;
1026	struct sockaddr_storage ss;
1027
1028	if (BOND_MODE(slave->bond) == BOND_MODE_TLB) {
1029		memcpy(dev->dev_addr, addr, len);
1030		return 0;
1031	}
1032
1033	/* for rlb each slave must have a unique hw mac addresses so that
1034	 * each slave will receive packets destined to a different mac
1035	 */
1036	memcpy(ss.__data, addr, len);
1037	ss.ss_family = dev->type;
1038	if (dev_set_mac_address(dev, (struct sockaddr *)&ss)) {
1039		netdev_err(slave->bond->dev, "dev_set_mac_address of dev %s failed! ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
1040			   dev->name);
1041		return -EOPNOTSUPP;
1042	}
1043	return 0;
1044}
1045
1046/* Swap MAC addresses between two slaves.
1047 *
1048 * Called with RTNL held, and no other locks.
1049 */
1050static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1051{
1052	u8 tmp_mac_addr[MAX_ADDR_LEN];
1053
1054	bond_hw_addr_copy(tmp_mac_addr, slave1->dev->dev_addr,
1055			  slave1->dev->addr_len);
1056	alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr,
1057			       slave2->dev->addr_len);
1058	alb_set_slave_mac_addr(slave2, tmp_mac_addr,
1059			       slave1->dev->addr_len);
1060
1061}
1062
1063/* Send learning packets after MAC address swap.
1064 *
1065 * Called with RTNL and no other locks
1066 */
1067static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1068				struct slave *slave2)
1069{
1070	int slaves_state_differ = (bond_slave_can_tx(slave1) != bond_slave_can_tx(slave2));
1071	struct slave *disabled_slave = NULL;
1072
1073	ASSERT_RTNL();
1074
1075	/* fasten the change in the switch */
1076	if (bond_slave_can_tx(slave1)) {
1077		alb_send_learning_packets(slave1, slave1->dev->dev_addr, false);
1078		if (bond->alb_info.rlb_enabled) {
1079			/* inform the clients that the mac address
1080			 * has changed
1081			 */
1082			rlb_req_update_slave_clients(bond, slave1);
1083		}
1084	} else {
1085		disabled_slave = slave1;
1086	}
1087
1088	if (bond_slave_can_tx(slave2)) {
1089		alb_send_learning_packets(slave2, slave2->dev->dev_addr, false);
1090		if (bond->alb_info.rlb_enabled) {
1091			/* inform the clients that the mac address
1092			 * has changed
1093			 */
1094			rlb_req_update_slave_clients(bond, slave2);
1095		}
1096	} else {
1097		disabled_slave = slave2;
1098	}
1099
1100	if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1101		/* A disabled slave was assigned an active mac addr */
1102		rlb_teach_disabled_mac_on_primary(bond,
1103						  disabled_slave->dev->dev_addr);
1104	}
1105}
1106
1107/**
1108 * alb_change_hw_addr_on_detach
1109 * @bond: bonding we're working on
1110 * @slave: the slave that was just detached
1111 *
1112 * We assume that @slave was already detached from the slave list.
1113 *
1114 * If @slave's permanent hw address is different both from its current
1115 * address and from @bond's address, then somewhere in the bond there's
1116 * a slave that has @slave's permanet address as its current address.
1117 * We'll make sure that that slave no longer uses @slave's permanent address.
1118 *
1119 * Caller must hold RTNL and no other locks
1120 */
1121static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1122{
1123	int perm_curr_diff;
1124	int perm_bond_diff;
1125	struct slave *found_slave;
1126
1127	perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1128						  slave->dev->dev_addr);
1129	perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1130						  bond->dev->dev_addr);
1131
1132	if (perm_curr_diff && perm_bond_diff) {
1133		found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1134
1135		if (found_slave) {
1136			alb_swap_mac_addr(slave, found_slave);
1137			alb_fasten_mac_swap(bond, slave, found_slave);
1138		}
1139	}
1140}
1141
1142/**
1143 * alb_handle_addr_collision_on_attach
1144 * @bond: bonding we're working on
1145 * @slave: the slave that was just attached
1146 *
1147 * checks uniqueness of slave's mac address and handles the case the
1148 * new slave uses the bonds mac address.
1149 *
1150 * If the permanent hw address of @slave is @bond's hw address, we need to
1151 * find a different hw address to give @slave, that isn't in use by any other
1152 * slave in the bond. This address must be, of course, one of the permanent
1153 * addresses of the other slaves.
1154 *
1155 * We go over the slave list, and for each slave there we compare its
1156 * permanent hw address with the current address of all the other slaves.
1157 * If no match was found, then we've found a slave with a permanent address
1158 * that isn't used by any other slave in the bond, so we can assign it to
1159 * @slave.
1160 *
1161 * assumption: this function is called before @slave is attached to the
1162 *	       bond slave list.
1163 */
1164static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1165{
1166	struct slave *has_bond_addr = rcu_access_pointer(bond->curr_active_slave);
1167	struct slave *tmp_slave1, *free_mac_slave = NULL;
1168	struct list_head *iter;
1169
1170	if (!bond_has_slaves(bond)) {
1171		/* this is the first slave */
1172		return 0;
1173	}
1174
1175	/* if slave's mac address differs from bond's mac address
1176	 * check uniqueness of slave's mac address against the other
1177	 * slaves in the bond.
1178	 */
1179	if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1180		if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1181			return 0;
1182
1183		/* Try setting slave mac to bond address and fall-through
1184		 * to code handling that situation below...
1185		 */
1186		alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1187				       bond->dev->addr_len);
1188	}
1189
1190	/* The slave's address is equal to the address of the bond.
1191	 * Search for a spare address in the bond for this slave.
1192	 */
1193	bond_for_each_slave(bond, tmp_slave1, iter) {
1194		if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1195			/* no slave has tmp_slave1's perm addr
1196			 * as its curr addr
1197			 */
1198			free_mac_slave = tmp_slave1;
1199			break;
1200		}
1201
1202		if (!has_bond_addr) {
1203			if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1204						    bond->dev->dev_addr)) {
1205
1206				has_bond_addr = tmp_slave1;
1207			}
1208		}
1209	}
1210
1211	if (free_mac_slave) {
1212		alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1213				       free_mac_slave->dev->addr_len);
1214
1215		netdev_warn(bond->dev, "the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1216			    slave->dev->name, free_mac_slave->dev->name);
1217
1218	} else if (has_bond_addr) {
1219		netdev_err(bond->dev, "the hw address of slave %s is in use by the bond; couldn't find a slave with a free hw address to give it (this should not have happened)\n",
1220			   slave->dev->name);
1221		return -EFAULT;
1222	}
1223
1224	return 0;
1225}
1226
1227/**
1228 * alb_set_mac_address
1229 * @bond:
1230 * @addr:
1231 *
1232 * In TLB mode all slaves are configured to the bond's hw address, but set
1233 * their dev_addr field to different addresses (based on their permanent hw
1234 * addresses).
1235 *
1236 * For each slave, this function sets the interface to the new address and then
1237 * changes its dev_addr field to its previous value.
1238 *
1239 * Unwinding assumes bond's mac address has not yet changed.
1240 */
1241static int alb_set_mac_address(struct bonding *bond, void *addr)
1242{
1243	struct slave *slave, *rollback_slave;
1244	struct list_head *iter;
1245	struct sockaddr_storage ss;
1246	char tmp_addr[MAX_ADDR_LEN];
1247	int res;
1248
1249	if (bond->alb_info.rlb_enabled)
1250		return 0;
1251
1252	bond_for_each_slave(bond, slave, iter) {
1253		/* save net_device's current hw address */
1254		bond_hw_addr_copy(tmp_addr, slave->dev->dev_addr,
1255				  slave->dev->addr_len);
1256
1257		res = dev_set_mac_address(slave->dev, addr);
1258
1259		/* restore net_device's hw address */
1260		bond_hw_addr_copy(slave->dev->dev_addr, tmp_addr,
1261				  slave->dev->addr_len);
1262
1263		if (res)
1264			goto unwind;
1265	}
1266
1267	return 0;
1268
1269unwind:
1270	memcpy(ss.__data, bond->dev->dev_addr, bond->dev->addr_len);
1271	ss.ss_family = bond->dev->type;
1272
1273	/* unwind from head to the slave that failed */
1274	bond_for_each_slave(bond, rollback_slave, iter) {
1275		if (rollback_slave == slave)
1276			break;
1277		bond_hw_addr_copy(tmp_addr, rollback_slave->dev->dev_addr,
1278				  rollback_slave->dev->addr_len);
1279		dev_set_mac_address(rollback_slave->dev,
1280				    (struct sockaddr *)&ss);
1281		bond_hw_addr_copy(rollback_slave->dev->dev_addr, tmp_addr,
1282				  rollback_slave->dev->addr_len);
1283	}
1284
1285	return res;
1286}
1287
1288/************************ exported alb funcions ************************/
1289
1290int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1291{
1292	int res;
1293
1294	res = tlb_initialize(bond);
1295	if (res)
1296		return res;
1297
1298	if (rlb_enabled) {
1299		bond->alb_info.rlb_enabled = 1;
1300		res = rlb_initialize(bond);
1301		if (res) {
1302			tlb_deinitialize(bond);
1303			return res;
1304		}
1305	} else {
1306		bond->alb_info.rlb_enabled = 0;
1307	}
1308
1309	return 0;
1310}
1311
1312void bond_alb_deinitialize(struct bonding *bond)
1313{
1314	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1315
1316	tlb_deinitialize(bond);
1317
1318	if (bond_info->rlb_enabled)
1319		rlb_deinitialize(bond);
1320}
1321
1322static int bond_do_alb_xmit(struct sk_buff *skb, struct bonding *bond,
1323			    struct slave *tx_slave)
1324{
1325	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1326	struct ethhdr *eth_data = eth_hdr(skb);
1327
1328	if (!tx_slave) {
1329		/* unbalanced or unassigned, send through primary */
1330		tx_slave = rcu_dereference(bond->curr_active_slave);
1331		if (bond->params.tlb_dynamic_lb)
1332			bond_info->unbalanced_load += skb->len;
1333	}
1334
1335	if (tx_slave && bond_slave_can_tx(tx_slave)) {
1336		if (tx_slave != rcu_access_pointer(bond->curr_active_slave)) {
1337			ether_addr_copy(eth_data->h_source,
1338					tx_slave->dev->dev_addr);
1339		}
1340
1341		bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1342		goto out;
1343	}
1344
1345	if (tx_slave && bond->params.tlb_dynamic_lb) {
1346		spin_lock(&bond->mode_lock);
1347		__tlb_clear_slave(bond, tx_slave, 0);
1348		spin_unlock(&bond->mode_lock);
1349	}
1350
1351	/* no suitable interface, frame not sent */
1352	bond_tx_drop(bond->dev, skb);
1353out:
1354	return NETDEV_TX_OK;
1355}
1356
1357int bond_tlb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1358{
1359	struct bonding *bond = netdev_priv(bond_dev);
1360	struct ethhdr *eth_data;
1361	struct slave *tx_slave = NULL;
1362	u32 hash_index;
1363
1364	skb_reset_mac_header(skb);
1365	eth_data = eth_hdr(skb);
1366
1367	/* Do not TX balance any multicast or broadcast */
1368	if (!is_multicast_ether_addr(eth_data->h_dest)) {
1369		switch (skb->protocol) {
1370		case htons(ETH_P_IP):
1371		case htons(ETH_P_IPX):
1372		    /* In case of IPX, it will falback to L2 hash */
1373		case htons(ETH_P_IPV6):
1374			hash_index = bond_xmit_hash(bond, skb);
1375			if (bond->params.tlb_dynamic_lb) {
1376				tx_slave = tlb_choose_channel(bond,
1377							      hash_index & 0xFF,
1378							      skb->len);
1379			} else {
1380				struct bond_up_slave *slaves;
1381				unsigned int count;
1382
1383				slaves = rcu_dereference(bond->slave_arr);
1384				count = slaves ? READ_ONCE(slaves->count) : 0;
1385				if (likely(count))
1386					tx_slave = slaves->arr[hash_index %
1387							       count];
1388			}
1389			break;
1390		}
1391	}
1392	return bond_do_alb_xmit(skb, bond, tx_slave);
1393}
1394
1395int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1396{
1397	struct bonding *bond = netdev_priv(bond_dev);
1398	struct ethhdr *eth_data;
1399	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1400	struct slave *tx_slave = NULL;
1401	static const __be32 ip_bcast = htonl(0xffffffff);
1402	int hash_size = 0;
1403	bool do_tx_balance = true;
1404	u32 hash_index = 0;
1405	const u8 *hash_start = NULL;
1406	struct ipv6hdr *ip6hdr;
1407
1408	skb_reset_mac_header(skb);
1409	eth_data = eth_hdr(skb);
1410
1411	switch (ntohs(skb->protocol)) {
1412	case ETH_P_IP: {
1413		const struct iphdr *iph = ip_hdr(skb);
1414
1415		if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast) ||
1416		    (iph->daddr == ip_bcast) ||
1417		    (iph->protocol == IPPROTO_IGMP)) {
1418			do_tx_balance = false;
1419			break;
1420		}
1421		hash_start = (char *)&(iph->daddr);
1422		hash_size = sizeof(iph->daddr);
1423	}
1424		break;
1425	case ETH_P_IPV6:
1426		/* IPv6 doesn't really use broadcast mac address, but leave
1427		 * that here just in case.
1428		 */
1429		if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast)) {
1430			do_tx_balance = false;
1431			break;
1432		}
1433
1434		/* IPv6 uses all-nodes multicast as an equivalent to
1435		 * broadcasts in IPv4.
1436		 */
1437		if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1438			do_tx_balance = false;
1439			break;
1440		}
1441
1442		/* Additianally, DAD probes should not be tx-balanced as that
1443		 * will lead to false positives for duplicate addresses and
1444		 * prevent address configuration from working.
1445		 */
1446		ip6hdr = ipv6_hdr(skb);
1447		if (ipv6_addr_any(&ip6hdr->saddr)) {
1448			do_tx_balance = false;
1449			break;
1450		}
1451
1452		hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1453		hash_size = sizeof(ipv6_hdr(skb)->daddr);
1454		break;
1455	case ETH_P_IPX:
1456		if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1457			/* something is wrong with this packet */
1458			do_tx_balance = false;
1459			break;
1460		}
1461
1462		if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1463			/* The only protocol worth balancing in
1464			 * this family since it has an "ARP" like
1465			 * mechanism
1466			 */
1467			do_tx_balance = false;
1468			break;
1469		}
1470
1471		hash_start = (char *)eth_data->h_dest;
1472		hash_size = ETH_ALEN;
1473		break;
1474	case ETH_P_ARP:
1475		do_tx_balance = false;
1476		if (bond_info->rlb_enabled)
1477			tx_slave = rlb_arp_xmit(skb, bond);
1478		break;
1479	default:
1480		do_tx_balance = false;
1481		break;
1482	}
1483
1484	if (do_tx_balance) {
1485		hash_index = _simple_hash(hash_start, hash_size);
1486		tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1487	}
1488
1489	return bond_do_alb_xmit(skb, bond, tx_slave);
1490}
1491
1492void bond_alb_monitor(struct work_struct *work)
1493{
1494	struct bonding *bond = container_of(work, struct bonding,
1495					    alb_work.work);
1496	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1497	struct list_head *iter;
1498	struct slave *slave;
1499
1500	if (!bond_has_slaves(bond)) {
1501		bond_info->tx_rebalance_counter = 0;
1502		bond_info->lp_counter = 0;
1503		goto re_arm;
1504	}
1505
1506	rcu_read_lock();
1507
1508	bond_info->tx_rebalance_counter++;
1509	bond_info->lp_counter++;
1510
1511	/* send learning packets */
1512	if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1513		bool strict_match;
1514
1515		bond_for_each_slave_rcu(bond, slave, iter) {
1516			/* If updating current_active, use all currently
1517			 * user mac addreses (!strict_match).  Otherwise, only
1518			 * use mac of the slave device.
1519			 * In RLB mode, we always use strict matches.
1520			 */
1521			strict_match = (slave != rcu_access_pointer(bond->curr_active_slave) ||
1522					bond_info->rlb_enabled);
1523			alb_send_learning_packets(slave, slave->dev->dev_addr,
1524						  strict_match);
1525		}
1526		bond_info->lp_counter = 0;
1527	}
1528
1529	/* rebalance tx traffic */
1530	if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1531		bond_for_each_slave_rcu(bond, slave, iter) {
1532			tlb_clear_slave(bond, slave, 1);
1533			if (slave == rcu_access_pointer(bond->curr_active_slave)) {
1534				SLAVE_TLB_INFO(slave).load =
1535					bond_info->unbalanced_load /
1536						BOND_TLB_REBALANCE_INTERVAL;
1537				bond_info->unbalanced_load = 0;
1538			}
1539		}
1540		bond_info->tx_rebalance_counter = 0;
1541	}
1542
1543	if (bond_info->rlb_enabled) {
1544		if (bond_info->primary_is_promisc &&
1545		    (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1546
1547			/* dev_set_promiscuity requires rtnl and
1548			 * nothing else.  Avoid race with bond_close.
1549			 */
1550			rcu_read_unlock();
1551			if (!rtnl_trylock())
1552				goto re_arm;
1553
1554			bond_info->rlb_promisc_timeout_counter = 0;
1555
1556			/* If the primary was set to promiscuous mode
1557			 * because a slave was disabled then
1558			 * it can now leave promiscuous mode.
1559			 */
1560			dev_set_promiscuity(rtnl_dereference(bond->curr_active_slave)->dev,
1561					    -1);
1562			bond_info->primary_is_promisc = 0;
1563
1564			rtnl_unlock();
1565			rcu_read_lock();
1566		}
1567
1568		if (bond_info->rlb_rebalance) {
1569			bond_info->rlb_rebalance = 0;
1570			rlb_rebalance(bond);
1571		}
1572
1573		/* check if clients need updating */
1574		if (bond_info->rx_ntt) {
1575			if (bond_info->rlb_update_delay_counter) {
1576				--bond_info->rlb_update_delay_counter;
1577			} else {
1578				rlb_update_rx_clients(bond);
1579				if (bond_info->rlb_update_retry_counter)
1580					--bond_info->rlb_update_retry_counter;
1581				else
1582					bond_info->rx_ntt = 0;
1583			}
1584		}
1585	}
1586	rcu_read_unlock();
1587re_arm:
1588	queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1589}
1590
1591/* assumption: called before the slave is attached to the bond
1592 * and not locked by the bond lock
1593 */
1594int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1595{
1596	int res;
1597
1598	res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1599				     slave->dev->addr_len);
1600	if (res)
1601		return res;
1602
1603	res = alb_handle_addr_collision_on_attach(bond, slave);
1604	if (res)
1605		return res;
1606
1607	tlb_init_slave(slave);
1608
1609	/* order a rebalance ASAP */
1610	bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1611
1612	if (bond->alb_info.rlb_enabled)
1613		bond->alb_info.rlb_rebalance = 1;
1614
1615	return 0;
1616}
1617
1618/* Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1619 * if necessary.
1620 *
1621 * Caller must hold RTNL and no other locks
1622 */
1623void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1624{
1625	if (bond_has_slaves(bond))
1626		alb_change_hw_addr_on_detach(bond, slave);
1627
1628	tlb_clear_slave(bond, slave, 0);
1629
1630	if (bond->alb_info.rlb_enabled) {
1631		bond->alb_info.rx_slave = NULL;
1632		rlb_clear_slave(bond, slave);
1633	}
1634
1635}
1636
1637void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1638{
1639	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1640
1641	if (link == BOND_LINK_DOWN) {
1642		tlb_clear_slave(bond, slave, 0);
1643		if (bond->alb_info.rlb_enabled)
1644			rlb_clear_slave(bond, slave);
1645	} else if (link == BOND_LINK_UP) {
1646		/* order a rebalance ASAP */
1647		bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1648		if (bond->alb_info.rlb_enabled) {
1649			bond->alb_info.rlb_rebalance = 1;
1650			/* If the updelay module parameter is smaller than the
1651			 * forwarding delay of the switch the rebalance will
1652			 * not work because the rebalance arp replies will
1653			 * not be forwarded to the clients..
1654			 */
1655		}
1656	}
1657
1658	if (bond_is_nondyn_tlb(bond)) {
1659		if (bond_update_slave_arr(bond, NULL))
1660			pr_err("Failed to build slave-array for TLB mode.\n");
1661	}
1662}
1663
1664/**
1665 * bond_alb_handle_active_change - assign new curr_active_slave
1666 * @bond: our bonding struct
1667 * @new_slave: new slave to assign
1668 *
1669 * Set the bond->curr_active_slave to @new_slave and handle
1670 * mac address swapping and promiscuity changes as needed.
1671 *
1672 * Caller must hold RTNL
1673 */
1674void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1675{
1676	struct slave *swap_slave;
1677	struct slave *curr_active;
1678
1679	curr_active = rtnl_dereference(bond->curr_active_slave);
1680	if (curr_active == new_slave)
1681		return;
1682
1683	if (curr_active && bond->alb_info.primary_is_promisc) {
1684		dev_set_promiscuity(curr_active->dev, -1);
1685		bond->alb_info.primary_is_promisc = 0;
1686		bond->alb_info.rlb_promisc_timeout_counter = 0;
1687	}
1688
1689	swap_slave = curr_active;
1690	rcu_assign_pointer(bond->curr_active_slave, new_slave);
1691
1692	if (!new_slave || !bond_has_slaves(bond))
1693		return;
1694
1695	/* set the new curr_active_slave to the bonds mac address
1696	 * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1697	 */
1698	if (!swap_slave)
1699		swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1700
1701	/* Arrange for swap_slave and new_slave to temporarily be
1702	 * ignored so we can mess with their MAC addresses without
1703	 * fear of interference from transmit activity.
1704	 */
1705	if (swap_slave)
1706		tlb_clear_slave(bond, swap_slave, 1);
1707	tlb_clear_slave(bond, new_slave, 1);
1708
1709	/* in TLB mode, the slave might flip down/up with the old dev_addr,
1710	 * and thus filter bond->dev_addr's packets, so force bond's mac
1711	 */
1712	if (BOND_MODE(bond) == BOND_MODE_TLB) {
1713		struct sockaddr_storage ss;
1714		u8 tmp_addr[MAX_ADDR_LEN];
1715
1716		bond_hw_addr_copy(tmp_addr, new_slave->dev->dev_addr,
1717				  new_slave->dev->addr_len);
1718
1719		bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
1720				  bond->dev->addr_len);
1721		ss.ss_family = bond->dev->type;
1722		/* we don't care if it can't change its mac, best effort */
1723		dev_set_mac_address(new_slave->dev, (struct sockaddr *)&ss);
1724
1725		bond_hw_addr_copy(new_slave->dev->dev_addr, tmp_addr,
1726				  new_slave->dev->addr_len);
1727	}
1728
1729	/* curr_active_slave must be set before calling alb_swap_mac_addr */
1730	if (swap_slave) {
1731		/* swap mac address */
1732		alb_swap_mac_addr(swap_slave, new_slave);
1733		alb_fasten_mac_swap(bond, swap_slave, new_slave);
1734	} else {
1735		/* set the new_slave to the bond mac address */
1736		alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1737				       bond->dev->addr_len);
1738		alb_send_learning_packets(new_slave, bond->dev->dev_addr,
1739					  false);
1740	}
1741}
1742
1743/* Called with RTNL */
1744int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1745{
1746	struct bonding *bond = netdev_priv(bond_dev);
1747	struct sockaddr_storage *ss = addr;
1748	struct slave *curr_active;
1749	struct slave *swap_slave;
1750	int res;
1751
1752	if (!is_valid_ether_addr(ss->__data))
1753		return -EADDRNOTAVAIL;
1754
1755	res = alb_set_mac_address(bond, addr);
1756	if (res)
1757		return res;
1758
1759	bond_hw_addr_copy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
1760
1761	/* If there is no curr_active_slave there is nothing else to do.
1762	 * Otherwise we'll need to pass the new address to it and handle
1763	 * duplications.
1764	 */
1765	curr_active = rtnl_dereference(bond->curr_active_slave);
1766	if (!curr_active)
1767		return 0;
1768
1769	swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1770
1771	if (swap_slave) {
1772		alb_swap_mac_addr(swap_slave, curr_active);
1773		alb_fasten_mac_swap(bond, swap_slave, curr_active);
1774	} else {
1775		alb_set_slave_mac_addr(curr_active, bond_dev->dev_addr,
1776				       bond_dev->addr_len);
1777
1778		alb_send_learning_packets(curr_active,
1779					  bond_dev->dev_addr, false);
1780		if (bond->alb_info.rlb_enabled) {
1781			/* inform clients mac address has changed */
1782			rlb_req_update_slave_clients(bond, curr_active);
1783		}
1784	}
1785
1786	return 0;
1787}
1788
1789void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1790{
1791	if (bond->alb_info.rlb_enabled)
1792		rlb_clear_vlan(bond, vlan_id);
1793}
1794