/drivers/net/bonding/bond_alb.c

   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved.
   4 */
   5
   6#include <linux/skbuff.h>
   7#include <linux/netdevice.h>
   8#include <linux/etherdevice.h>
   9#include <linux/pkt_sched.h>
  10#include <linux/spinlock.h>
  11#include <linux/slab.h>
  12#include <linux/timer.h>
  13#include <linux/ip.h>
  14#include <linux/ipv6.h>
  15#include <linux/if_arp.h>
  16#include <linux/if_ether.h>
  17#include <linux/if_bonding.h>
  18#include <linux/if_vlan.h>
  19#include <linux/in.h>
  20#include <net/ipx.h>
  21#include <net/arp.h>
  22#include <net/ipv6.h>
  23#include <asm/byteorder.h>
  24#include <net/bonding.h>
  25#include <net/bond_alb.h>
  26
  27static const u8 mac_v6_allmcast[ETH_ALEN + 2] __long_aligned = {
  28	0x33, 0x33, 0x00, 0x00, 0x00, 0x01
  29};
  30static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
  31
  32#pragma pack(1)
  33struct learning_pkt {
  34	u8 mac_dst[ETH_ALEN];
  35	u8 mac_src[ETH_ALEN];
  36	__be16 type;
  37	u8 padding[ETH_ZLEN - ETH_HLEN];
  38};
  39
  40struct arp_pkt {
  41	__be16  hw_addr_space;
  42	__be16  prot_addr_space;
  43	u8      hw_addr_len;
  44	u8      prot_addr_len;
  45	__be16  op_code;
  46	u8      mac_src[ETH_ALEN];	/* sender hardware address */
  47	__be32  ip_src;			/* sender IP address */
  48	u8      mac_dst[ETH_ALEN];	/* target hardware address */
  49	__be32  ip_dst;			/* target IP address */
  50};
  51#pragma pack()
  52
  53/* Forward declaration */
  54static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
  55				      bool strict_match);
  56static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp);
  57static void rlb_src_unlink(struct bonding *bond, u32 index);
  58static void rlb_src_link(struct bonding *bond, u32 ip_src_hash,
  59			 u32 ip_dst_hash);
  60
  61static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
  62{
  63	int i;
  64	u8 hash = 0;
  65
  66	for (i = 0; i < hash_size; i++)
  67		hash ^= hash_start[i];
  68
  69	return hash;
  70}
  71
  72/*********************** tlb specific functions ***************************/
  73
  74static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
  75{
  76	if (save_load) {
  77		entry->load_history = 1 + entry->tx_bytes /
  78				      BOND_TLB_REBALANCE_INTERVAL;
  79		entry->tx_bytes = 0;
  80	}
  81
  82	entry->tx_slave = NULL;
  83	entry->next = TLB_NULL_INDEX;
  84	entry->prev = TLB_NULL_INDEX;
  85}
  86
  87static inline void tlb_init_slave(struct slave *slave)
  88{
  89	SLAVE_TLB_INFO(slave).load = 0;
  90	SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
  91}
  92
  93static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
  94			 int save_load)
  95{
  96	struct tlb_client_info *tx_hash_table;
  97	u32 index;
  98
  99	/* clear slave from tx_hashtbl */
 100	tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
 101
 102	/* skip this if we've already freed the tx hash table */
 103	if (tx_hash_table) {
 104		index = SLAVE_TLB_INFO(slave).head;
 105		while (index != TLB_NULL_INDEX) {
 106			u32 next_index = tx_hash_table[index].next;
 107			tlb_init_table_entry(&tx_hash_table[index], save_load);
 108			index = next_index;
 109		}
 110	}
 111
 112	tlb_init_slave(slave);
 113}
 114
 115static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
 116			 int save_load)
 117{
 118	spin_lock_bh(&bond->mode_lock);
 119	__tlb_clear_slave(bond, slave, save_load);
 120	spin_unlock_bh(&bond->mode_lock);
 121}
 122
 123/* Must be called before starting the monitor timer */
 124static int tlb_initialize(struct bonding *bond)
 125{
 126	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 127	int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
 128	struct tlb_client_info *new_hashtbl;
 129	int i;
 130
 131	new_hashtbl = kzalloc(size, GFP_KERNEL);
 132	if (!new_hashtbl)
 133		return -ENOMEM;
 134
 135	spin_lock_bh(&bond->mode_lock);
 136
 137	bond_info->tx_hashtbl = new_hashtbl;
 138
 139	for (i = 0; i < TLB_HASH_TABLE_SIZE; i++)
 140		tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
 141
 142	spin_unlock_bh(&bond->mode_lock);
 143
 144	return 0;
 145}
 146
 147/* Must be called only after all slaves have been released */
 148static void tlb_deinitialize(struct bonding *bond)
 149{
 150	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 151
 152	spin_lock_bh(&bond->mode_lock);
 153
 154	kfree(bond_info->tx_hashtbl);
 155	bond_info->tx_hashtbl = NULL;
 156
 157	spin_unlock_bh(&bond->mode_lock);
 158}
 159
 160static long long compute_gap(struct slave *slave)
 161{
 162	return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
 163	       (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
 164}
 165
 166static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
 167{
 168	struct slave *slave, *least_loaded;
 169	struct list_head *iter;
 170	long long max_gap;
 171
 172	least_loaded = NULL;
 173	max_gap = LLONG_MIN;
 174
 175	/* Find the slave with the largest gap */
 176	bond_for_each_slave_rcu(bond, slave, iter) {
 177		if (bond_slave_can_tx(slave)) {
 178			long long gap = compute_gap(slave);
 179
 180			if (max_gap < gap) {
 181				least_loaded = slave;
 182				max_gap = gap;
 183			}
 184		}
 185	}
 186
 187	return least_loaded;
 188}
 189
 190static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
 191						u32 skb_len)
 192{
 193	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 194	struct tlb_client_info *hash_table;
 195	struct slave *assigned_slave;
 196
 197	hash_table = bond_info->tx_hashtbl;
 198	assigned_slave = hash_table[hash_index].tx_slave;
 199	if (!assigned_slave) {
 200		assigned_slave = tlb_get_least_loaded_slave(bond);
 201
 202		if (assigned_slave) {
 203			struct tlb_slave_info *slave_info =
 204				&(SLAVE_TLB_INFO(assigned_slave));
 205			u32 next_index = slave_info->head;
 206
 207			hash_table[hash_index].tx_slave = assigned_slave;
 208			hash_table[hash_index].next = next_index;
 209			hash_table[hash_index].prev = TLB_NULL_INDEX;
 210
 211			if (next_index != TLB_NULL_INDEX)
 212				hash_table[next_index].prev = hash_index;
 213
 214			slave_info->head = hash_index;
 215			slave_info->load +=
 216				hash_table[hash_index].load_history;
 217		}
 218	}
 219
 220	if (assigned_slave)
 221		hash_table[hash_index].tx_bytes += skb_len;
 222
 223	return assigned_slave;
 224}
 225
 226static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
 227					u32 skb_len)
 228{
 229	struct slave *tx_slave;
 230
 231	/* We don't need to disable softirq here, becase
 232	 * tlb_choose_channel() is only called by bond_alb_xmit()
 233	 * which already has softirq disabled.
 234	 */
 235	spin_lock(&bond->mode_lock);
 236	tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
 237	spin_unlock(&bond->mode_lock);
 238
 239	return tx_slave;
 240}
 241
 242/*********************** rlb specific functions ***************************/
 243
 244/* when an ARP REPLY is received from a client update its info
 245 * in the rx_hashtbl
 246 */
 247static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
 248{
 249	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 250	struct rlb_client_info *client_info;
 251	u32 hash_index;
 252
 253	spin_lock_bh(&bond->mode_lock);
 254
 255	hash_index = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
 256	client_info = &(bond_info->rx_hashtbl[hash_index]);
 257
 258	if ((client_info->assigned) &&
 259	    (client_info->ip_src == arp->ip_dst) &&
 260	    (client_info->ip_dst == arp->ip_src) &&
 261	    (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) {
 262		/* update the clients MAC address */
 263		ether_addr_copy(client_info->mac_dst, arp->mac_src);
 264		client_info->ntt = 1;
 265		bond_info->rx_ntt = 1;
 266	}
 267
 268	spin_unlock_bh(&bond->mode_lock);
 269}
 270
 271static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
 272			struct slave *slave)
 273{
 274	struct arp_pkt *arp, _arp;
 275
 276	if (skb->protocol != cpu_to_be16(ETH_P_ARP))
 277		goto out;
 278
 279	arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp);
 280	if (!arp)
 281		goto out;
 282
 283	/* We received an ARP from arp->ip_src.
 284	 * We might have used this IP address previously (on the bonding host
 285	 * itself or on a system that is bridged together with the bond).
 286	 * However, if arp->mac_src is different than what is stored in
 287	 * rx_hashtbl, some other host is now using the IP and we must prevent
 288	 * sending out client updates with this IP address and the old MAC
 289	 * address.
 290	 * Clean up all hash table entries that have this address as ip_src but
 291	 * have a different mac_src.
 292	 */
 293	rlb_purge_src_ip(bond, arp);
 294
 295	if (arp->op_code == htons(ARPOP_REPLY)) {
 296		/* update rx hash table for this ARP */
 297		rlb_update_entry_from_arp(bond, arp);
 298		slave_dbg(bond->dev, slave->dev, "Server received an ARP Reply from client\n");
 299	}
 300out:
 301	return RX_HANDLER_ANOTHER;
 302}
 303
 304/* Caller must hold rcu_read_lock() */
 305static struct slave *__rlb_next_rx_slave(struct bonding *bond)
 306{
 307	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 308	struct slave *before = NULL, *rx_slave = NULL, *slave;
 309	struct list_head *iter;
 310	bool found = false;
 311
 312	bond_for_each_slave_rcu(bond, slave, iter) {
 313		if (!bond_slave_can_tx(slave))
 314			continue;
 315		if (!found) {
 316			if (!before || before->speed < slave->speed)
 317				before = slave;
 318		} else {
 319			if (!rx_slave || rx_slave->speed < slave->speed)
 320				rx_slave = slave;
 321		}
 322		if (slave == bond_info->rx_slave)
 323			found = true;
 324	}
 325	/* we didn't find anything after the current or we have something
 326	 * better before and up to the current slave
 327	 */
 328	if (!rx_slave || (before && rx_slave->speed < before->speed))
 329		rx_slave = before;
 330
 331	if (rx_slave)
 332		bond_info->rx_slave = rx_slave;
 333
 334	return rx_slave;
 335}
 336
 337/* Caller must hold RTNL, rcu_read_lock is obtained only to silence checkers */
 338static struct slave *rlb_next_rx_slave(struct bonding *bond)
 339{
 340	struct slave *rx_slave;
 341
 342	ASSERT_RTNL();
 343
 344	rcu_read_lock();
 345	rx_slave = __rlb_next_rx_slave(bond);
 346	rcu_read_unlock();
 347
 348	return rx_slave;
 349}
 350
 351/* teach the switch the mac of a disabled slave
 352 * on the primary for fault tolerance
 353 *
 354 * Caller must hold RTNL
 355 */
 356static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
 357{
 358	struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 359
 360	if (!curr_active)
 361		return;
 362
 363	if (!bond->alb_info.primary_is_promisc) {
 364		if (!dev_set_promiscuity(curr_active->dev, 1))
 365			bond->alb_info.primary_is_promisc = 1;
 366		else
 367			bond->alb_info.primary_is_promisc = 0;
 368	}
 369
 370	bond->alb_info.rlb_promisc_timeout_counter = 0;
 371
 372	alb_send_learning_packets(curr_active, addr, true);
 373}
 374
 375/* slave being removed should not be active at this point
 376 *
 377 * Caller must hold rtnl.
 378 */
 379static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
 380{
 381	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 382	struct rlb_client_info *rx_hash_table;
 383	u32 index, next_index;
 384
 385	/* clear slave from rx_hashtbl */
 386	spin_lock_bh(&bond->mode_lock);
 387
 388	rx_hash_table = bond_info->rx_hashtbl;
 389	index = bond_info->rx_hashtbl_used_head;
 390	for (; index != RLB_NULL_INDEX; index = next_index) {
 391		next_index = rx_hash_table[index].used_next;
 392		if (rx_hash_table[index].slave == slave) {
 393			struct slave *assigned_slave = rlb_next_rx_slave(bond);
 394
 395			if (assigned_slave) {
 396				rx_hash_table[index].slave = assigned_slave;
 397				if (is_valid_ether_addr(rx_hash_table[index].mac_dst)) {
 398					bond_info->rx_hashtbl[index].ntt = 1;
 399					bond_info->rx_ntt = 1;
 400					/* A slave has been removed from the
 401					 * table because it is either disabled
 402					 * or being released. We must retry the
 403					 * update to avoid clients from not
 404					 * being updated & disconnecting when
 405					 * there is stress
 406					 */
 407					bond_info->rlb_update_retry_counter =
 408						RLB_UPDATE_RETRY;
 409				}
 410			} else {  /* there is no active slave */
 411				rx_hash_table[index].slave = NULL;
 412			}
 413		}
 414	}
 415
 416	spin_unlock_bh(&bond->mode_lock);
 417
 418	if (slave != rtnl_dereference(bond->curr_active_slave))
 419		rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
 420}
 421
 422static void rlb_update_client(struct rlb_client_info *client_info)
 423{
 424	int i;
 425
 426	if (!client_info->slave || !is_valid_ether_addr(client_info->mac_dst))
 427		return;
 428
 429	for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
 430		struct sk_buff *skb;
 431
 432		skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
 433				 client_info->ip_dst,
 434				 client_info->slave->dev,
 435				 client_info->ip_src,
 436				 client_info->mac_dst,
 437				 client_info->slave->dev->dev_addr,
 438				 client_info->mac_dst);
 439		if (!skb) {
 440			slave_err(client_info->slave->bond->dev,
 441				  client_info->slave->dev,
 442				  "failed to create an ARP packet\n");
 443			continue;
 444		}
 445
 446		skb->dev = client_info->slave->dev;
 447
 448		if (client_info->vlan_id) {
 449			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 450					       client_info->vlan_id);
 451		}
 452
 453		arp_xmit(skb);
 454	}
 455}
 456
 457/* sends ARP REPLIES that update the clients that need updating */
 458static void rlb_update_rx_clients(struct bonding *bond)
 459{
 460	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 461	struct rlb_client_info *client_info;
 462	u32 hash_index;
 463
 464	spin_lock_bh(&bond->mode_lock);
 465
 466	hash_index = bond_info->rx_hashtbl_used_head;
 467	for (; hash_index != RLB_NULL_INDEX;
 468	     hash_index = client_info->used_next) {
 469		client_info = &(bond_info->rx_hashtbl[hash_index]);
 470		if (client_info->ntt) {
 471			rlb_update_client(client_info);
 472			if (bond_info->rlb_update_retry_counter == 0)
 473				client_info->ntt = 0;
 474		}
 475	}
 476
 477	/* do not update the entries again until this counter is zero so that
 478	 * not to confuse the clients.
 479	 */
 480	bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
 481
 482	spin_unlock_bh(&bond->mode_lock);
 483}
 484
 485/* The slave was assigned a new mac address - update the clients */
 486static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
 487{
 488	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 489	struct rlb_client_info *client_info;
 490	int ntt = 0;
 491	u32 hash_index;
 492
 493	spin_lock_bh(&bond->mode_lock);
 494
 495	hash_index = bond_info->rx_hashtbl_used_head;
 496	for (; hash_index != RLB_NULL_INDEX;
 497	     hash_index = client_info->used_next) {
 498		client_info = &(bond_info->rx_hashtbl[hash_index]);
 499
 500		if ((client_info->slave == slave) &&
 501		    is_valid_ether_addr(client_info->mac_dst)) {
 502			client_info->ntt = 1;
 503			ntt = 1;
 504		}
 505	}
 506
 507	/* update the team's flag only after the whole iteration */
 508	if (ntt) {
 509		bond_info->rx_ntt = 1;
 510		/* fasten the change */
 511		bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
 512	}
 513
 514	spin_unlock_bh(&bond->mode_lock);
 515}
 516
 517/* mark all clients using src_ip to be updated */
 518static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
 519{
 520	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 521	struct rlb_client_info *client_info;
 522	u32 hash_index;
 523
 524	spin_lock(&bond->mode_lock);
 525
 526	hash_index = bond_info->rx_hashtbl_used_head;
 527	for (; hash_index != RLB_NULL_INDEX;
 528	     hash_index = client_info->used_next) {
 529		client_info = &(bond_info->rx_hashtbl[hash_index]);
 530
 531		if (!client_info->slave) {
 532			netdev_err(bond->dev, "found a client with no channel in the client's hash table\n");
 533			continue;
 534		}
 535		/* update all clients using this src_ip, that are not assigned
 536		 * to the team's address (curr_active_slave) and have a known
 537		 * unicast mac address.
 538		 */
 539		if ((client_info->ip_src == src_ip) &&
 540		    !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
 541					     bond->dev->dev_addr) &&
 542		    is_valid_ether_addr(client_info->mac_dst)) {
 543			client_info->ntt = 1;
 544			bond_info->rx_ntt = 1;
 545		}
 546	}
 547
 548	spin_unlock(&bond->mode_lock);
 549}
 550
 551static struct slave *rlb_choose_channel(struct sk_buff *skb,
 552					struct bonding *bond,
 553					const struct arp_pkt *arp)
 554{
 555	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 556	struct slave *assigned_slave, *curr_active_slave;
 557	struct rlb_client_info *client_info;
 558	u32 hash_index = 0;
 559
 560	spin_lock(&bond->mode_lock);
 561
 562	curr_active_slave = rcu_dereference(bond->curr_active_slave);
 563
 564	hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
 565	client_info = &(bond_info->rx_hashtbl[hash_index]);
 566
 567	if (client_info->assigned) {
 568		if ((client_info->ip_src == arp->ip_src) &&
 569		    (client_info->ip_dst == arp->ip_dst)) {
 570			/* the entry is already assigned to this client */
 571			if (!is_broadcast_ether_addr(arp->mac_dst)) {
 572				/* update mac address from arp */
 573				ether_addr_copy(client_info->mac_dst, arp->mac_dst);
 574			}
 575			ether_addr_copy(client_info->mac_src, arp->mac_src);
 576
 577			assigned_slave = client_info->slave;
 578			if (assigned_slave) {
 579				spin_unlock(&bond->mode_lock);
 580				return assigned_slave;
 581			}
 582		} else {
 583			/* the entry is already assigned to some other client,
 584			 * move the old client to primary (curr_active_slave) so
 585			 * that the new client can be assigned to this entry.
 586			 */
 587			if (curr_active_slave &&
 588			    client_info->slave != curr_active_slave) {
 589				client_info->slave = curr_active_slave;
 590				rlb_update_client(client_info);
 591			}
 592		}
 593	}
 594	/* assign a new slave */
 595	assigned_slave = __rlb_next_rx_slave(bond);
 596
 597	if (assigned_slave) {
 598		if (!(client_info->assigned &&
 599		      client_info->ip_src == arp->ip_src)) {
 600			/* ip_src is going to be updated,
 601			 * fix the src hash list
 602			 */
 603			u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
 604						    sizeof(arp->ip_src));
 605			rlb_src_unlink(bond, hash_index);
 606			rlb_src_link(bond, hash_src, hash_index);
 607		}
 608
 609		client_info->ip_src = arp->ip_src;
 610		client_info->ip_dst = arp->ip_dst;
 611		/* arp->mac_dst is broadcast for arp reqeusts.
 612		 * will be updated with clients actual unicast mac address
 613		 * upon receiving an arp reply.
 614		 */
 615		ether_addr_copy(client_info->mac_dst, arp->mac_dst);
 616		ether_addr_copy(client_info->mac_src, arp->mac_src);
 617		client_info->slave = assigned_slave;
 618
 619		if (is_valid_ether_addr(client_info->mac_dst)) {
 620			client_info->ntt = 1;
 621			bond->alb_info.rx_ntt = 1;
 622		} else {
 623			client_info->ntt = 0;
 624		}
 625
 626		if (vlan_get_tag(skb, &client_info->vlan_id))
 627			client_info->vlan_id = 0;
 628
 629		if (!client_info->assigned) {
 630			u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
 631			bond_info->rx_hashtbl_used_head = hash_index;
 632			client_info->used_next = prev_tbl_head;
 633			if (prev_tbl_head != RLB_NULL_INDEX) {
 634				bond_info->rx_hashtbl[prev_tbl_head].used_prev =
 635					hash_index;
 636			}
 637			client_info->assigned = 1;
 638		}
 639	}
 640
 641	spin_unlock(&bond->mode_lock);
 642
 643	return assigned_slave;
 644}
 645
 646/* chooses (and returns) transmit channel for arp reply
 647 * does not choose channel for other arp types since they are
 648 * sent on the curr_active_slave
 649 */
 650static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
 651{
 652	struct slave *tx_slave = NULL;
 653	struct arp_pkt *arp;
 654
 655	if (!pskb_network_may_pull(skb, sizeof(*arp)))
 656		return NULL;
 657	arp = (struct arp_pkt *)skb_network_header(skb);
 658
 659	/* Don't modify or load balance ARPs that do not originate locally
 660	 * (e.g.,arrive via a bridge).
 661	 */
 662	if (!bond_slave_has_mac_rx(bond, arp->mac_src))
 663		return NULL;
 664
 665	if (arp->op_code == htons(ARPOP_REPLY)) {
 666		/* the arp must be sent on the selected rx channel */
 667		tx_slave = rlb_choose_channel(skb, bond, arp);
 668		if (tx_slave)
 669			bond_hw_addr_copy(arp->mac_src, tx_slave->dev->dev_addr,
 670					  tx_slave->dev->addr_len);
 671		netdev_dbg(bond->dev, "(slave %s): Server sent ARP Reply packet\n",
 672			   tx_slave ? tx_slave->dev->name : "NULL");
 673	} else if (arp->op_code == htons(ARPOP_REQUEST)) {
 674		/* Create an entry in the rx_hashtbl for this client as a
 675		 * place holder.
 676		 * When the arp reply is received the entry will be updated
 677		 * with the correct unicast address of the client.
 678		 */
 679		tx_slave = rlb_choose_channel(skb, bond, arp);
 680
 681		/* The ARP reply packets must be delayed so that
 682		 * they can cancel out the influence of the ARP request.
 683		 */
 684		bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
 685
 686		/* arp requests are broadcast and are sent on the primary
 687		 * the arp request will collapse all clients on the subnet to
 688		 * the primary slave. We must register these clients to be
 689		 * updated with their assigned mac.
 690		 */
 691		rlb_req_update_subnet_clients(bond, arp->ip_src);
 692		netdev_dbg(bond->dev, "(slave %s): Server sent ARP Request packet\n",
 693			   tx_slave ? tx_slave->dev->name : "NULL");
 694	}
 695
 696	return tx_slave;
 697}
 698
 699static void rlb_rebalance(struct bonding *bond)
 700{
 701	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 702	struct slave *assigned_slave;
 703	struct rlb_client_info *client_info;
 704	int ntt;
 705	u32 hash_index;
 706
 707	spin_lock_bh(&bond->mode_lock);
 708
 709	ntt = 0;
 710	hash_index = bond_info->rx_hashtbl_used_head;
 711	for (; hash_index != RLB_NULL_INDEX;
 712	     hash_index = client_info->used_next) {
 713		client_info = &(bond_info->rx_hashtbl[hash_index]);
 714		assigned_slave = __rlb_next_rx_slave(bond);
 715		if (assigned_slave && (client_info->slave != assigned_slave)) {
 716			client_info->slave = assigned_slave;
 717			if (!is_zero_ether_addr(client_info->mac_dst)) {
 718				client_info->ntt = 1;
 719				ntt = 1;
 720			}
 721		}
 722	}
 723
 724	/* update the team's flag only after the whole iteration */
 725	if (ntt)
 726		bond_info->rx_ntt = 1;
 727	spin_unlock_bh(&bond->mode_lock);
 728}
 729
 730/* Caller must hold mode_lock */
 731static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
 732{
 733	entry->used_next = RLB_NULL_INDEX;
 734	entry->used_prev = RLB_NULL_INDEX;
 735	entry->assigned = 0;
 736	entry->slave = NULL;
 737	entry->vlan_id = 0;
 738}
 739static void rlb_init_table_entry_src(struct rlb_client_info *entry)
 740{
 741	entry->src_first = RLB_NULL_INDEX;
 742	entry->src_prev = RLB_NULL_INDEX;
 743	entry->src_next = RLB_NULL_INDEX;
 744}
 745
 746static void rlb_init_table_entry(struct rlb_client_info *entry)
 747{
 748	memset(entry, 0, sizeof(struct rlb_client_info));
 749	rlb_init_table_entry_dst(entry);
 750	rlb_init_table_entry_src(entry);
 751}
 752
 753static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
 754{
 755	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 756	u32 next_index = bond_info->rx_hashtbl[index].used_next;
 757	u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
 758
 759	if (index == bond_info->rx_hashtbl_used_head)
 760		bond_info->rx_hashtbl_used_head = next_index;
 761	if (prev_index != RLB_NULL_INDEX)
 762		bond_info->rx_hashtbl[prev_index].used_next = next_index;
 763	if (next_index != RLB_NULL_INDEX)
 764		bond_info->rx_hashtbl[next_index].used_prev = prev_index;
 765}
 766
 767/* unlink a rlb hash table entry from the src list */
 768static void rlb_src_unlink(struct bonding *bond, u32 index)
 769{
 770	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 771	u32 next_index = bond_info->rx_hashtbl[index].src_next;
 772	u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
 773
 774	bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
 775	bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
 776
 777	if (next_index != RLB_NULL_INDEX)
 778		bond_info->rx_hashtbl[next_index].src_prev = prev_index;
 779
 780	if (prev_index == RLB_NULL_INDEX)
 781		return;
 782
 783	/* is prev_index pointing to the head of this list? */
 784	if (bond_info->rx_hashtbl[prev_index].src_first == index)
 785		bond_info->rx_hashtbl[prev_index].src_first = next_index;
 786	else
 787		bond_info->rx_hashtbl[prev_index].src_next = next_index;
 788
 789}
 790
 791static void rlb_delete_table_entry(struct bonding *bond, u32 index)
 792{
 793	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 794	struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 795
 796	rlb_delete_table_entry_dst(bond, index);
 797	rlb_init_table_entry_dst(entry);
 798
 799	rlb_src_unlink(bond, index);
 800}
 801
 802/* add the rx_hashtbl[ip_dst_hash] entry to the list
 803 * of entries with identical ip_src_hash
 804 */
 805static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
 806{
 807	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 808	u32 next;
 809
 810	bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
 811	next = bond_info->rx_hashtbl[ip_src_hash].src_first;
 812	bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
 813	if (next != RLB_NULL_INDEX)
 814		bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
 815	bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
 816}
 817
 818/* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does
 819 * not match arp->mac_src
 820 */
 821static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
 822{
 823	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 824	u32 ip_src_hash = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
 825	u32 index;
 826
 827	spin_lock_bh(&bond->mode_lock);
 828
 829	index = bond_info->rx_hashtbl[ip_src_hash].src_first;
 830	while (index != RLB_NULL_INDEX) {
 831		struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
 832		u32 next_index = entry->src_next;
 833		if (entry->ip_src == arp->ip_src &&
 834		    !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
 835				rlb_delete_table_entry(bond, index);
 836		index = next_index;
 837	}
 838	spin_unlock_bh(&bond->mode_lock);
 839}
 840
 841static int rlb_initialize(struct bonding *bond)
 842{
 843	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 844	struct rlb_client_info	*new_hashtbl;
 845	int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
 846	int i;
 847
 848	new_hashtbl = kmalloc(size, GFP_KERNEL);
 849	if (!new_hashtbl)
 850		return -1;
 851
 852	spin_lock_bh(&bond->mode_lock);
 853
 854	bond_info->rx_hashtbl = new_hashtbl;
 855
 856	bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 857
 858	for (i = 0; i < RLB_HASH_TABLE_SIZE; i++)
 859		rlb_init_table_entry(bond_info->rx_hashtbl + i);
 860
 861	spin_unlock_bh(&bond->mode_lock);
 862
 863	/* register to receive ARPs */
 864	bond->recv_probe = rlb_arp_recv;
 865
 866	return 0;
 867}
 868
 869static void rlb_deinitialize(struct bonding *bond)
 870{
 871	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 872
 873	spin_lock_bh(&bond->mode_lock);
 874
 875	kfree(bond_info->rx_hashtbl);
 876	bond_info->rx_hashtbl = NULL;
 877	bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
 878
 879	spin_unlock_bh(&bond->mode_lock);
 880}
 881
 882static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
 883{
 884	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
 885	u32 curr_index;
 886
 887	spin_lock_bh(&bond->mode_lock);
 888
 889	curr_index = bond_info->rx_hashtbl_used_head;
 890	while (curr_index != RLB_NULL_INDEX) {
 891		struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
 892		u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
 893
 894		if (curr->vlan_id == vlan_id)
 895			rlb_delete_table_entry(bond, curr_index);
 896
 897		curr_index = next_index;
 898	}
 899
 900	spin_unlock_bh(&bond->mode_lock);
 901}
 902
 903/*********************** tlb/rlb shared functions *********************/
 904
 905static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
 906			    __be16 vlan_proto, u16 vid)
 907{
 908	struct learning_pkt pkt;
 909	struct sk_buff *skb;
 910	int size = sizeof(struct learning_pkt);
 911
 912	memset(&pkt, 0, size);
 913	ether_addr_copy(pkt.mac_dst, mac_addr);
 914	ether_addr_copy(pkt.mac_src, mac_addr);
 915	pkt.type = cpu_to_be16(ETH_P_LOOPBACK);
 916
 917	skb = dev_alloc_skb(size);
 918	if (!skb)
 919		return;
 920
 921	skb_put_data(skb, &pkt, size);
 922
 923	skb_reset_mac_header(skb);
 924	skb->network_header = skb->mac_header + ETH_HLEN;
 925	skb->protocol = pkt.type;
 926	skb->priority = TC_PRIO_CONTROL;
 927	skb->dev = slave->dev;
 928
 929	slave_dbg(slave->bond->dev, slave->dev,
 930		  "Send learning packet: mac %pM vlan %d\n", mac_addr, vid);
 931
 932	if (vid)
 933		__vlan_hwaccel_put_tag(skb, vlan_proto, vid);
 934
 935	dev_queue_xmit(skb);
 936}
 937
 938struct alb_walk_data {
 939	struct bonding *bond;
 940	struct slave *slave;
 941	u8 *mac_addr;
 942	bool strict_match;
 943};
 944
 945static int alb_upper_dev_walk(struct net_device *upper, void *_data)
 946{
 947	struct alb_walk_data *data = _data;
 948	bool strict_match = data->strict_match;
 949	struct bonding *bond = data->bond;
 950	struct slave *slave = data->slave;
 951	u8 *mac_addr = data->mac_addr;
 952	struct bond_vlan_tag *tags;
 953
 954	if (is_vlan_dev(upper) &&
 955	    bond->dev->lower_level == upper->lower_level - 1) {
 956		if (upper->addr_assign_type == NET_ADDR_STOLEN) {
 957			alb_send_lp_vid(slave, mac_addr,
 958					vlan_dev_vlan_proto(upper),
 959					vlan_dev_vlan_id(upper));
 960		} else {
 961			alb_send_lp_vid(slave, upper->dev_addr,
 962					vlan_dev_vlan_proto(upper),
 963					vlan_dev_vlan_id(upper));
 964		}
 965	}
 966
 967	/* If this is a macvlan device, then only send updates
 968	 * when strict_match is turned off.
 969	 */
 970	if (netif_is_macvlan(upper) && !strict_match) {
 971		tags = bond_verify_device_path(bond->dev, upper, 0);
 972		if (IS_ERR_OR_NULL(tags))
 973			BUG();
 974		alb_send_lp_vid(slave, upper->dev_addr,
 975				tags[0].vlan_proto, tags[0].vlan_id);
 976		kfree(tags);
 977	}
 978
 979	return 0;
 980}
 981
 982static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
 983				      bool strict_match)
 984{
 985	struct bonding *bond = bond_get_bond_by_slave(slave);
 986	struct alb_walk_data data = {
 987		.strict_match = strict_match,
 988		.mac_addr = mac_addr,
 989		.slave = slave,
 990		.bond = bond,
 991	};
 992
 993	/* send untagged */
 994	alb_send_lp_vid(slave, mac_addr, 0, 0);
 995
 996	/* loop through all devices and see if we need to send a packet
 997	 * for that device.
 998	 */
 999	rcu_read_lock();
1000	netdev_walk_all_upper_dev_rcu(bond->dev, alb_upper_dev_walk, &data);
1001	rcu_read_unlock();
1002}
1003
1004static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[],
1005				  unsigned int len)
1006{
1007	struct net_device *dev = slave->dev;
1008	struct sockaddr_storage ss;
1009
1010	if (BOND_MODE(slave->bond) == BOND_MODE_TLB) {
1011		memcpy(dev->dev_addr, addr, len);
1012		return 0;
1013	}
1014
1015	/* for rlb each slave must have a unique hw mac addresses so that
1016	 * each slave will receive packets destined to a different mac
1017	 */
1018	memcpy(ss.__data, addr, len);
1019	ss.ss_family = dev->type;
1020	if (dev_set_mac_address(dev, (struct sockaddr *)&ss, NULL)) {
1021		slave_err(slave->bond->dev, dev, "dev_set_mac_address on slave failed! ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n");
1022		return -EOPNOTSUPP;
1023	}
1024	return 0;
1025}
1026
1027/* Swap MAC addresses between two slaves.
1028 *
1029 * Called with RTNL held, and no other locks.
1030 */
1031static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1032{
1033	u8 tmp_mac_addr[MAX_ADDR_LEN];
1034
1035	bond_hw_addr_copy(tmp_mac_addr, slave1->dev->dev_addr,
1036			  slave1->dev->addr_len);
1037	alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr,
1038			       slave2->dev->addr_len);
1039	alb_set_slave_mac_addr(slave2, tmp_mac_addr,
1040			       slave1->dev->addr_len);
1041
1042}
1043
1044/* Send learning packets after MAC address swap.
1045 *
1046 * Called with RTNL and no other locks
1047 */
1048static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1049				struct slave *slave2)
1050{
1051	int slaves_state_differ = (bond_slave_can_tx(slave1) != bond_slave_can_tx(slave2));
1052	struct slave *disabled_slave = NULL;
1053
1054	ASSERT_RTNL();
1055
1056	/* fasten the change in the switch */
1057	if (bond_slave_can_tx(slave1)) {
1058		alb_send_learning_packets(slave1, slave1->dev->dev_addr, false);
1059		if (bond->alb_info.rlb_enabled) {
1060			/* inform the clients that the mac address
1061			 * has changed
1062			 */
1063			rlb_req_update_slave_clients(bond, slave1);
1064		}
1065	} else {
1066		disabled_slave = slave1;
1067	}
1068
1069	if (bond_slave_can_tx(slave2)) {
1070		alb_send_learning_packets(slave2, slave2->dev->dev_addr, false);
1071		if (bond->alb_info.rlb_enabled) {
1072			/* inform the clients that the mac address
1073			 * has changed
1074			 */
1075			rlb_req_update_slave_clients(bond, slave2);
1076		}
1077	} else {
1078		disabled_slave = slave2;
1079	}
1080
1081	if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1082		/* A disabled slave was assigned an active mac addr */
1083		rlb_teach_disabled_mac_on_primary(bond,
1084						  disabled_slave->dev->dev_addr);
1085	}
1086}
1087
1088/**
1089 * alb_change_hw_addr_on_detach
1090 * @bond: bonding we're working on
1091 * @slave: the slave that was just detached
1092 *
1093 * We assume that @slave was already detached from the slave list.
1094 *
1095 * If @slave's permanent hw address is different both from its current
1096 * address and from @bond's address, then somewhere in the bond there's
1097 * a slave that has @slave's permanet address as its current address.
1098 * We'll make sure that that slave no longer uses @slave's permanent address.
1099 *
1100 * Caller must hold RTNL and no other locks
1101 */
1102static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1103{
1104	int perm_curr_diff;
1105	int perm_bond_diff;
1106	struct slave *found_slave;
1107
1108	perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1109						  slave->dev->dev_addr);
1110	perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1111						  bond->dev->dev_addr);
1112
1113	if (perm_curr_diff && perm_bond_diff) {
1114		found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1115
1116		if (found_slave) {
1117			alb_swap_mac_addr(slave, found_slave);
1118			alb_fasten_mac_swap(bond, slave, found_slave);
1119		}
1120	}
1121}
1122
1123/**
1124 * alb_handle_addr_collision_on_attach
1125 * @bond: bonding we're working on
1126 * @slave: the slave that was just attached
1127 *
1128 * checks uniqueness of slave's mac address and handles the case the
1129 * new slave uses the bonds mac address.
1130 *
1131 * If the permanent hw address of @slave is @bond's hw address, we need to
1132 * find a different hw address to give @slave, that isn't in use by any other
1133 * slave in the bond. This address must be, of course, one of the permanent
1134 * addresses of the other slaves.
1135 *
1136 * We go over the slave list, and for each slave there we compare its
1137 * permanent hw address with the current address of all the other slaves.
1138 * If no match was found, then we've found a slave with a permanent address
1139 * that isn't used by any other slave in the bond, so we can assign it to
1140 * @slave.
1141 *
1142 * assumption: this function is called before @slave is attached to the
1143 *	       bond slave list.
1144 */
1145static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1146{
1147	struct slave *has_bond_addr = rcu_access_pointer(bond->curr_active_slave);
1148	struct slave *tmp_slave1, *free_mac_slave = NULL;
1149	struct list_head *iter;
1150
1151	if (!bond_has_slaves(bond)) {
1152		/* this is the first slave */
1153		return 0;
1154	}
1155
1156	/* if slave's mac address differs from bond's mac address
1157	 * check uniqueness of slave's mac address against the other
1158	 * slaves in the bond.
1159	 */
1160	if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1161		if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1162			return 0;
1163
1164		/* Try setting slave mac to bond address and fall-through
1165		 * to code handling that situation below...
1166		 */
1167		alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1168				       bond->dev->addr_len);
1169	}
1170
1171	/* The slave's address is equal to the address of the bond.
1172	 * Search for a spare address in the bond for this slave.
1173	 */
1174	bond_for_each_slave(bond, tmp_slave1, iter) {
1175		if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1176			/* no slave has tmp_slave1's perm addr
1177			 * as its curr addr
1178			 */
1179			free_mac_slave = tmp_slave1;
1180			break;
1181		}
1182
1183		if (!has_bond_addr) {
1184			if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1185						    bond->dev->dev_addr)) {
1186
1187				has_bond_addr = tmp_slave1;
1188			}
1189		}
1190	}
1191
1192	if (free_mac_slave) {
1193		alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1194				       free_mac_slave->dev->addr_len);
1195
1196		slave_warn(bond->dev, slave->dev, "the slave hw address is in use by the bond; giving it the hw address of %s\n",
1197			   free_mac_slave->dev->name);
1198
1199	} else if (has_bond_addr) {
1200		slave_err(bond->dev, slave->dev, "the slave hw address is in use by the bond; couldn't find a slave with a free hw address to give it (this should not have happened)\n");
1201		return -EFAULT;
1202	}
1203
1204	return 0;
1205}
1206
1207/**
1208 * alb_set_mac_address
1209 * @bond:
1210 * @addr:
1211 *
1212 * In TLB mode all slaves are configured to the bond's hw address, but set
1213 * their dev_addr field to different addresses (based on their permanent hw
1214 * addresses).
1215 *
1216 * For each slave, this function sets the interface to the new address and then
1217 * changes its dev_addr field to its previous value.
1218 *
1219 * Unwinding assumes bond's mac address has not yet changed.
1220 */
1221static int alb_set_mac_address(struct bonding *bond, void *addr)
1222{
1223	struct slave *slave, *rollback_slave;
1224	struct list_head *iter;
1225	struct sockaddr_storage ss;
1226	char tmp_addr[MAX_ADDR_LEN];
1227	int res;
1228
1229	if (bond->alb_info.rlb_enabled)
1230		return 0;
1231
1232	bond_for_each_slave(bond, slave, iter) {
1233		/* save net_device's current hw address */
1234		bond_hw_addr_copy(tmp_addr, slave->dev->dev_addr,
1235				  slave->dev->addr_len);
1236
1237		res = dev_set_mac_address(slave->dev, addr, NULL);
1238
1239		/* restore net_device's hw address */
1240		bond_hw_addr_copy(slave->dev->dev_addr, tmp_addr,
1241				  slave->dev->addr_len);
1242
1243		if (res)
1244			goto unwind;
1245	}
1246
1247	return 0;
1248
1249unwind:
1250	memcpy(ss.__data, bond->dev->dev_addr, bond->dev->addr_len);
1251	ss.ss_family = bond->dev->type;
1252
1253	/* unwind from head to the slave that failed */
1254	bond_for_each_slave(bond, rollback_slave, iter) {
1255		if (rollback_slave == slave)
1256			break;
1257		bond_hw_addr_copy(tmp_addr, rollback_slave->dev->dev_addr,
1258				  rollback_slave->dev->addr_len);
1259		dev_set_mac_address(rollback_slave->dev,
1260				    (struct sockaddr *)&ss, NULL);
1261		bond_hw_addr_copy(rollback_slave->dev->dev_addr, tmp_addr,
1262				  rollback_slave->dev->addr_len);
1263	}
1264
1265	return res;
1266}
1267
1268/************************ exported alb funcions ************************/
1269
1270int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1271{
1272	int res;
1273
1274	res = tlb_initialize(bond);
1275	if (res)
1276		return res;
1277
1278	if (rlb_enabled) {
1279		bond->alb_info.rlb_enabled = 1;
1280		res = rlb_initialize(bond);
1281		if (res) {
1282			tlb_deinitialize(bond);
1283			return res;
1284		}
1285	} else {
1286		bond->alb_info.rlb_enabled = 0;
1287	}
1288
1289	return 0;
1290}
1291
1292void bond_alb_deinitialize(struct bonding *bond)
1293{
1294	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1295
1296	tlb_deinitialize(bond);
1297
1298	if (bond_info->rlb_enabled)
1299		rlb_deinitialize(bond);
1300}
1301
1302static netdev_tx_t bond_do_alb_xmit(struct sk_buff *skb, struct bonding *bond,
1303				    struct slave *tx_slave)
1304{
1305	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1306	struct ethhdr *eth_data = eth_hdr(skb);
1307
1308	if (!tx_slave) {
1309		/* unbalanced or unassigned, send through primary */
1310		tx_slave = rcu_dereference(bond->curr_active_slave);
1311		if (bond->params.tlb_dynamic_lb)
1312			bond_info->unbalanced_load += skb->len;
1313	}
1314
1315	if (tx_slave && bond_slave_can_tx(tx_slave)) {
1316		if (tx_slave != rcu_access_pointer(bond->curr_active_slave)) {
1317			ether_addr_copy(eth_data->h_source,
1318					tx_slave->dev->dev_addr);
1319		}
1320
1321		bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1322		goto out;
1323	}
1324
1325	if (tx_slave && bond->params.tlb_dynamic_lb) {
1326		spin_lock(&bond->mode_lock);
1327		__tlb_clear_slave(bond, tx_slave, 0);
1328		spin_unlock(&bond->mode_lock);
1329	}
1330
1331	/* no suitable interface, frame not sent */
1332	bond_tx_drop(bond->dev, skb);
1333out:
1334	return NETDEV_TX_OK;
1335}
1336
1337netdev_tx_t bond_tlb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1338{
1339	struct bonding *bond = netdev_priv(bond_dev);
1340	struct ethhdr *eth_data;
1341	struct slave *tx_slave = NULL;
1342	u32 hash_index;
1343
1344	skb_reset_mac_header(skb);
1345	eth_data = eth_hdr(skb);
1346
1347	/* Do not TX balance any multicast or broadcast */
1348	if (!is_multicast_ether_addr(eth_data->h_dest)) {
1349		switch (skb->protocol) {
1350		case htons(ETH_P_IP):
1351		case htons(ETH_P_IPX):
1352		    /* In case of IPX, it will falback to L2 hash */
1353		case htons(ETH_P_IPV6):
1354			hash_index = bond_xmit_hash(bond, skb);
1355			if (bond->params.tlb_dynamic_lb) {
1356				tx_slave = tlb_choose_channel(bond,
1357							      hash_index & 0xFF,
1358							      skb->len);
1359			} else {
1360				struct bond_up_slave *slaves;
1361				unsigned int count;
1362
1363				slaves = rcu_dereference(bond->slave_arr);
1364				count = slaves ? READ_ONCE(slaves->count) : 0;
1365				if (likely(count))
1366					tx_slave = slaves->arr[hash_index %
1367							       count];
1368			}
1369			break;
1370		}
1371	}
1372	return bond_do_alb_xmit(skb, bond, tx_slave);
1373}
1374
1375netdev_tx_t bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1376{
1377	struct bonding *bond = netdev_priv(bond_dev);
1378	struct ethhdr *eth_data;
1379	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1380	struct slave *tx_slave = NULL;
1381	static const __be32 ip_bcast = htonl(0xffffffff);
1382	int hash_size = 0;
1383	bool do_tx_balance = true;
1384	u32 hash_index = 0;
1385	const u8 *hash_start = NULL;
1386
1387	skb_reset_mac_header(skb);
1388	eth_data = eth_hdr(skb);
1389
1390	switch (ntohs(skb->protocol)) {
1391	case ETH_P_IP: {
1392		const struct iphdr *iph;
1393
1394		if (is_broadcast_ether_addr(eth_data->h_dest) ||
1395		    !pskb_network_may_pull(skb, sizeof(*iph))) {
1396			do_tx_balance = false;
1397			break;
1398		}
1399		iph = ip_hdr(skb);
1400		if (iph->daddr == ip_bcast || iph->protocol == IPPROTO_IGMP) {
1401			do_tx_balance = false;
1402			break;
1403		}
1404		hash_start = (char *)&(iph->daddr);
1405		hash_size = sizeof(iph->daddr);
1406		break;
1407	}
1408	case ETH_P_IPV6: {
1409		const struct ipv6hdr *ip6hdr;
1410
1411		/* IPv6 doesn't really use broadcast mac address, but leave
1412		 * that here just in case.
1413		 */
1414		if (is_broadcast_ether_addr(eth_data->h_dest)) {
1415			do_tx_balance = false;
1416			break;
1417		}
1418
1419		/* IPv6 uses all-nodes multicast as an equivalent to
1420		 * broadcasts in IPv4.
1421		 */
1422		if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1423			do_tx_balance = false;
1424			break;
1425		}
1426
1427		if (!pskb_network_may_pull(skb, sizeof(*ip6hdr))) {
1428			do_tx_balance = false;
1429			break;
1430		}
1431		/* Additionally, DAD probes should not be tx-balanced as that
1432		 * will lead to false positives for duplicate addresses and
1433		 * prevent address configuration from working.
1434		 */
1435		ip6hdr = ipv6_hdr(skb);
1436		if (ipv6_addr_any(&ip6hdr->saddr)) {
1437			do_tx_balance = false;
1438			break;
1439		}
1440
1441		hash_start = (char *)&ip6hdr->daddr;
1442		hash_size = sizeof(ip6hdr->daddr);
1443		break;
1444	}
1445	case ETH_P_IPX: {
1446		const struct ipxhdr *ipxhdr;
1447
1448		if (pskb_network_may_pull(skb, sizeof(*ipxhdr))) {
1449			do_tx_balance = false;
1450			break;
1451		}
1452		ipxhdr = (struct ipxhdr *)skb_network_header(skb);
1453
1454		if (ipxhdr->ipx_checksum != IPX_NO_CHECKSUM) {
1455			/* something is wrong with this packet */
1456			do_tx_balance = false;
1457			break;
1458		}
1459
1460		if (ipxhdr->ipx_type != IPX_TYPE_NCP) {
1461			/* The only protocol worth balancing in
1462			 * this family since it has an "ARP" like
1463			 * mechanism
1464			 */
1465			do_tx_balance = false;
1466			break;
1467		}
1468
1469		eth_data = eth_hdr(skb);
1470		hash_start = (char *)eth_data->h_dest;
1471		hash_size = ETH_ALEN;
1472		break;
1473	}
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		if (bond->params.tlb_dynamic_lb) {
1486			hash_index = _simple_hash(hash_start, hash_size);
1487			tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1488		} else {
1489			/*
1490			 * do_tx_balance means we are free to select the tx_slave
1491			 * So we do exactly what tlb would do for hash selection
1492			 */
1493
1494			struct bond_up_slave *slaves;
1495			unsigned int count;
1496
1497			slaves = rcu_dereference(bond->slave_arr);
1498			count = slaves ? READ_ONCE(slaves->count) : 0;
1499			if (likely(count))
1500				tx_slave = slaves->arr[bond_xmit_hash(bond, skb) %
1501						       count];
1502		}
1503	}
1504
1505	return bond_do_alb_xmit(skb, bond, tx_slave);
1506}
1507
1508void bond_alb_monitor(struct work_struct *work)
1509{
1510	struct bonding *bond = container_of(work, struct bonding,
1511					    alb_work.work);
1512	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1513	struct list_head *iter;
1514	struct slave *slave;
1515
1516	if (!bond_has_slaves(bond)) {
1517		bond_info->tx_rebalance_counter = 0;
1518		bond_info->lp_counter = 0;
1519		goto re_arm;
1520	}
1521
1522	rcu_read_lock();
1523
1524	bond_info->tx_rebalance_counter++;
1525	bond_info->lp_counter++;
1526
1527	/* send learning packets */
1528	if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1529		bool strict_match;
1530
1531		bond_for_each_slave_rcu(bond, slave, iter) {
1532			/* If updating current_active, use all currently
1533			 * user mac addreses (!strict_match).  Otherwise, only
1534			 * use mac of the slave device.
1535			 * In RLB mode, we always use strict matches.
1536			 */
1537			strict_match = (slave != rcu_access_pointer(bond->curr_active_slave) ||
1538					bond_info->rlb_enabled);
1539			alb_send_learning_packets(slave, slave->dev->dev_addr,
1540						  strict_match);
1541		}
1542		bond_info->lp_counter = 0;
1543	}
1544
1545	/* rebalance tx traffic */
1546	if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1547		bond_for_each_slave_rcu(bond, slave, iter) {
1548			tlb_clear_slave(bond, slave, 1);
1549			if (slave == rcu_access_pointer(bond->curr_active_slave)) {
1550				SLAVE_TLB_INFO(slave).load =
1551					bond_info->unbalanced_load /
1552						BOND_TLB_REBALANCE_INTERVAL;
1553				bond_info->unbalanced_load = 0;
1554			}
1555		}
1556		bond_info->tx_rebalance_counter = 0;
1557	}
1558
1559	if (bond_info->rlb_enabled) {
1560		if (bond_info->primary_is_promisc &&
1561		    (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1562
1563			/* dev_set_promiscuity requires rtnl and
1564			 * nothing else.  Avoid race with bond_close.
1565			 */
1566			rcu_read_unlock();
1567			if (!rtnl_trylock())
1568				goto re_arm;
1569
1570			bond_info->rlb_promisc_timeout_counter = 0;
1571
1572			/* If the primary was set to promiscuous mode
1573			 * because a slave was disabled then
1574			 * it can now leave promiscuous mode.
1575			 */
1576			dev_set_promiscuity(rtnl_dereference(bond->curr_active_slave)->dev,
1577					    -1);
1578			bond_info->primary_is_promisc = 0;
1579
1580			rtnl_unlock();
1581			rcu_read_lock();
1582		}
1583
1584		if (bond_info->rlb_rebalance) {
1585			bond_info->rlb_rebalance = 0;
1586			rlb_rebalance(bond);
1587		}
1588
1589		/* check if clients need updating */
1590		if (bond_info->rx_ntt) {
1591			if (bond_info->rlb_update_delay_counter) {
1592				--bond_info->rlb_update_delay_counter;
1593			} else {
1594				rlb_update_rx_clients(bond);
1595				if (bond_info->rlb_update_retry_counter)
1596					--bond_info->rlb_update_retry_counter;
1597				else
1598					bond_info->rx_ntt = 0;
1599			}
1600		}
1601	}
1602	rcu_read_unlock();
1603re_arm:
1604	queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1605}
1606
1607/* assumption: called before the slave is attached to the bond
1608 * and not locked by the bond lock
1609 */
1610int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1611{
1612	int res;
1613
1614	res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1615				     slave->dev->addr_len);
1616	if (res)
1617		return res;
1618
1619	res = alb_handle_addr_collision_on_attach(bond, slave);
1620	if (res)
1621		return res;
1622
1623	tlb_init_slave(slave);
1624
1625	/* order a rebalance ASAP */
1626	bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1627
1628	if (bond->alb_info.rlb_enabled)
1629		bond->alb_info.rlb_rebalance = 1;
1630
1631	return 0;
1632}
1633
1634/* Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1635 * if necessary.
1636 *
1637 * Caller must hold RTNL and no other locks
1638 */
1639void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1640{
1641	if (bond_has_slaves(bond))
1642		alb_change_hw_addr_on_detach(bond, slave);
1643
1644	tlb_clear_slave(bond, slave, 0);
1645
1646	if (bond->alb_info.rlb_enabled) {
1647		bond->alb_info.rx_slave = NULL;
1648		rlb_clear_slave(bond, slave);
1649	}
1650
1651}
1652
1653void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1654{
1655	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1656
1657	if (link == BOND_LINK_DOWN) {
1658		tlb_clear_slave(bond, slave, 0);
1659		if (bond->alb_info.rlb_enabled)
1660			rlb_clear_slave(bond, slave);
1661	} else if (link == BOND_LINK_UP) {
1662		/* order a rebalance ASAP */
1663		bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1664		if (bond->alb_info.rlb_enabled) {
1665			bond->alb_info.rlb_rebalance = 1;
1666			/* If the updelay module parameter is smaller than the
1667			 * forwarding delay of the switch the rebalance will
1668			 * not work because the rebalance arp replies will
1669			 * not be forwarded to the clients..
1670			 */
1671		}
1672	}
1673
1674	if (bond_is_nondyn_tlb(bond)) {
1675		if (bond_update_slave_arr(bond, NULL))
1676			pr_err("Failed to build slave-array for TLB mode.\n");
1677	}
1678}
1679
1680/**
1681 * bond_alb_handle_active_change - assign new curr_active_slave
1682 * @bond: our bonding struct
1683 * @new_slave: new slave to assign
1684 *
1685 * Set the bond->curr_active_slave to @new_slave and handle
1686 * mac address swapping and promiscuity changes as needed.
1687 *
1688 * Caller must hold RTNL
1689 */
1690void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1691{
1692	struct slave *swap_slave;
1693	struct slave *curr_active;
1694
1695	curr_active = rtnl_dereference(bond->curr_active_slave);
1696	if (curr_active == new_slave)
1697		return;
1698
1699	if (curr_active && bond->alb_info.primary_is_promisc) {
1700		dev_set_promiscuity(curr_active->dev, -1);
1701		bond->alb_info.primary_is_promisc = 0;
1702		bond->alb_info.rlb_promisc_timeout_counter = 0;
1703	}
1704
1705	swap_slave = curr_active;
1706	rcu_assign_pointer(bond->curr_active_slave, new_slave);
1707
1708	if (!new_slave || !bond_has_slaves(bond))
1709		return;
1710
1711	/* set the new curr_active_slave to the bonds mac address
1712	 * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1713	 */
1714	if (!swap_slave)
1715		swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1716
1717	/* Arrange for swap_slave and new_slave to temporarily be
1718	 * ignored so we can mess with their MAC addresses without
1719	 * fear of interference from transmit activity.
1720	 */
1721	if (swap_slave)
1722		tlb_clear_slave(bond, swap_slave, 1);
1723	tlb_clear_slave(bond, new_slave, 1);
1724
1725	/* in TLB mode, the slave might flip down/up with the old dev_addr,
1726	 * and thus filter bond->dev_addr's packets, so force bond's mac
1727	 */
1728	if (BOND_MODE(bond) == BOND_MODE_TLB) {
1729		struct sockaddr_storage ss;
1730		u8 tmp_addr[MAX_ADDR_LEN];
1731
1732		bond_hw_addr_copy(tmp_addr, new_slave->dev->dev_addr,
1733				  new_slave->dev->addr_len);
1734
1735		bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
1736				  bond->dev->addr_len);
1737		ss.ss_family = bond->dev->type;
1738		/* we don't care if it can't change its mac, best effort */
1739		dev_set_mac_address(new_slave->dev, (struct sockaddr *)&ss,
1740				    NULL);
1741
1742		bond_hw_addr_copy(new_slave->dev->dev_addr, tmp_addr,
1743				  new_slave->dev->addr_len);
1744	}
1745
1746	/* curr_active_slave must be set before calling alb_swap_mac_addr */
1747	if (swap_slave) {
1748		/* swap mac address */
1749		alb_swap_mac_addr(swap_slave, new_slave);
1750		alb_fasten_mac_swap(bond, swap_slave, new_slave);
1751	} else {
1752		/* set the new_slave to the bond mac address */
1753		alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1754				       bond->dev->addr_len);
1755		alb_send_learning_packets(new_slave, bond->dev->dev_addr,
1756					  false);
1757	}
1758}
1759
1760/* Called with RTNL */
1761int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1762{
1763	struct bonding *bond = netdev_priv(bond_dev);
1764	struct sockaddr_storage *ss = addr;
1765	struct slave *curr_active;
1766	struct slave *swap_slave;
1767	int res;
1768
1769	if (!is_valid_ether_addr(ss->__data))
1770		return -EADDRNOTAVAIL;
1771
1772	res = alb_set_mac_address(bond, addr);
1773	if (res)
1774		return res;
1775
1776	bond_hw_addr_copy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
1777
1778	/* If there is no curr_active_slave there is nothing else to do.
1779	 * Otherwise we'll need to pass the new address to it and handle
1780	 * duplications.
1781	 */
1782	curr_active = rtnl_dereference(bond->curr_active_slave);
1783	if (!curr_active)
1784		return 0;
1785
1786	swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1787
1788	if (swap_slave) {
1789		alb_swap_mac_addr(swap_slave, curr_active);
1790		alb_fasten_mac_swap(bond, swap_slave, curr_active);
1791	} else {
1792		alb_set_slave_mac_addr(curr_active, bond_dev->dev_addr,
1793				       bond_dev->addr_len);
1794
1795		alb_send_learning_packets(curr_active,
1796					  bond_dev->dev_addr, false);
1797		if (bond->alb_info.rlb_enabled) {
1798			/* inform clients mac address has changed */
1799			rlb_req_update_slave_clients(bond, curr_active);
1800		}
1801	}
1802
1803	return 0;
1804}
1805
1806void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1807{
1808	if (bond->alb_info.rlb_enabled)
1809		rlb_clear_vlan(bond, vlan_id);
1810}
1811