2 * Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved.
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.
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
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/>.
17 * The full GNU General Public License is included in this distribution in the
18 * file called LICENSE.
22 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
24 #include <linux/skbuff.h>
25 #include <linux/netdevice.h>
26 #include <linux/etherdevice.h>
27 #include <linux/pkt_sched.h>
28 #include <linux/spinlock.h>
29 #include <linux/slab.h>
30 #include <linux/timer.h>
32 #include <linux/ipv6.h>
33 #include <linux/if_arp.h>
34 #include <linux/if_ether.h>
35 #include <linux/if_bonding.h>
36 #include <linux/if_vlan.h>
41 #include <asm/byteorder.h>
47 #ifndef __long_aligned
48 #define __long_aligned __attribute__((aligned((sizeof(long)))))
50 static const u8 mac_bcast[ETH_ALEN] __long_aligned = {
51 0xff, 0xff, 0xff, 0xff, 0xff, 0xff
53 static const u8 mac_v6_allmcast[ETH_ALEN] __long_aligned = {
54 0x33, 0x33, 0x00, 0x00, 0x00, 0x01
56 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
63 u8 padding[ETH_ZLEN - ETH_HLEN];
68 __be16 prot_addr_space;
72 u8 mac_src[ETH_ALEN]; /* sender hardware address */
73 __be32 ip_src; /* sender IP address */
74 u8 mac_dst[ETH_ALEN]; /* target hardware address */
75 __be32 ip_dst; /* target IP address */
79 static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb)
81 return (struct arp_pkt *)skb_network_header(skb);
84 /* Forward declaration */
85 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[]);
86 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp);
87 static void rlb_src_unlink(struct bonding *bond, u32 index);
88 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash,
91 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
96 for (i = 0; i < hash_size; i++)
97 hash ^= hash_start[i];
102 /*********************** tlb specific functions ***************************/
104 static inline void _lock_tx_hashtbl_bh(struct bonding *bond)
106 spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
109 static inline void _unlock_tx_hashtbl_bh(struct bonding *bond)
111 spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
114 static inline void _lock_tx_hashtbl(struct bonding *bond)
116 spin_lock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
119 static inline void _unlock_tx_hashtbl(struct bonding *bond)
121 spin_unlock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
124 /* Caller must hold tx_hashtbl lock */
125 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
128 entry->load_history = 1 + entry->tx_bytes /
129 BOND_TLB_REBALANCE_INTERVAL;
133 entry->tx_slave = NULL;
134 entry->next = TLB_NULL_INDEX;
135 entry->prev = TLB_NULL_INDEX;
138 static inline void tlb_init_slave(struct slave *slave)
140 SLAVE_TLB_INFO(slave).load = 0;
141 SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
144 /* Caller must hold bond lock for read, BH disabled */
145 static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
148 struct tlb_client_info *tx_hash_table;
151 /* clear slave from tx_hashtbl */
152 tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
154 /* skip this if we've already freed the tx hash table */
156 index = SLAVE_TLB_INFO(slave).head;
157 while (index != TLB_NULL_INDEX) {
158 u32 next_index = tx_hash_table[index].next;
159 tlb_init_table_entry(&tx_hash_table[index], save_load);
164 tlb_init_slave(slave);
167 /* Caller must hold bond lock for read */
168 static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
171 _lock_tx_hashtbl_bh(bond);
172 __tlb_clear_slave(bond, slave, save_load);
173 _unlock_tx_hashtbl_bh(bond);
176 /* Must be called before starting the monitor timer */
177 static int tlb_initialize(struct bonding *bond)
179 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
180 int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
181 struct tlb_client_info *new_hashtbl;
184 new_hashtbl = kzalloc(size, GFP_KERNEL);
188 _lock_tx_hashtbl_bh(bond);
190 bond_info->tx_hashtbl = new_hashtbl;
192 for (i = 0; i < TLB_HASH_TABLE_SIZE; i++)
193 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
195 _unlock_tx_hashtbl_bh(bond);
200 /* Must be called only after all slaves have been released */
201 static void tlb_deinitialize(struct bonding *bond)
203 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
205 _lock_tx_hashtbl_bh(bond);
207 kfree(bond_info->tx_hashtbl);
208 bond_info->tx_hashtbl = NULL;
210 _unlock_tx_hashtbl_bh(bond);
213 static long long compute_gap(struct slave *slave)
215 return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
216 (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
219 /* Caller must hold bond lock for read */
220 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
222 struct slave *slave, *least_loaded;
223 struct list_head *iter;
229 /* Find the slave with the largest gap */
230 bond_for_each_slave_rcu(bond, slave, iter) {
231 if (SLAVE_IS_OK(slave)) {
232 long long gap = compute_gap(slave);
235 least_loaded = slave;
244 static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
247 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
248 struct tlb_client_info *hash_table;
249 struct slave *assigned_slave;
251 hash_table = bond_info->tx_hashtbl;
252 assigned_slave = hash_table[hash_index].tx_slave;
253 if (!assigned_slave) {
254 assigned_slave = tlb_get_least_loaded_slave(bond);
256 if (assigned_slave) {
257 struct tlb_slave_info *slave_info =
258 &(SLAVE_TLB_INFO(assigned_slave));
259 u32 next_index = slave_info->head;
261 hash_table[hash_index].tx_slave = assigned_slave;
262 hash_table[hash_index].next = next_index;
263 hash_table[hash_index].prev = TLB_NULL_INDEX;
265 if (next_index != TLB_NULL_INDEX)
266 hash_table[next_index].prev = hash_index;
268 slave_info->head = hash_index;
270 hash_table[hash_index].load_history;
275 hash_table[hash_index].tx_bytes += skb_len;
277 return assigned_slave;
280 /* Caller must hold bond lock for read */
281 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
284 struct slave *tx_slave;
286 * We don't need to disable softirq here, becase
287 * tlb_choose_channel() is only called by bond_alb_xmit()
288 * which already has softirq disabled.
290 _lock_tx_hashtbl(bond);
291 tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
292 _unlock_tx_hashtbl(bond);
296 /*********************** rlb specific functions ***************************/
297 static inline void _lock_rx_hashtbl_bh(struct bonding *bond)
299 spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
302 static inline void _unlock_rx_hashtbl_bh(struct bonding *bond)
304 spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
307 static inline void _lock_rx_hashtbl(struct bonding *bond)
309 spin_lock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
312 static inline void _unlock_rx_hashtbl(struct bonding *bond)
314 spin_unlock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
317 /* when an ARP REPLY is received from a client update its info
320 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
322 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
323 struct rlb_client_info *client_info;
326 _lock_rx_hashtbl_bh(bond);
328 hash_index = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
329 client_info = &(bond_info->rx_hashtbl[hash_index]);
331 if ((client_info->assigned) &&
332 (client_info->ip_src == arp->ip_dst) &&
333 (client_info->ip_dst == arp->ip_src) &&
334 (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) {
335 /* update the clients MAC address */
336 ether_addr_copy(client_info->mac_dst, arp->mac_src);
337 client_info->ntt = 1;
338 bond_info->rx_ntt = 1;
341 _unlock_rx_hashtbl_bh(bond);
344 static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
347 struct arp_pkt *arp, _arp;
349 if (skb->protocol != cpu_to_be16(ETH_P_ARP))
352 arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp);
356 /* We received an ARP from arp->ip_src.
357 * We might have used this IP address previously (on the bonding host
358 * itself or on a system that is bridged together with the bond).
359 * However, if arp->mac_src is different than what is stored in
360 * rx_hashtbl, some other host is now using the IP and we must prevent
361 * sending out client updates with this IP address and the old MAC
363 * Clean up all hash table entries that have this address as ip_src but
364 * have a different mac_src.
366 rlb_purge_src_ip(bond, arp);
368 if (arp->op_code == htons(ARPOP_REPLY)) {
369 /* update rx hash table for this ARP */
370 rlb_update_entry_from_arp(bond, arp);
371 pr_debug("Server received an ARP Reply from client\n");
374 return RX_HANDLER_ANOTHER;
377 /* Caller must hold bond lock for read */
378 static struct slave *rlb_next_rx_slave(struct bonding *bond)
380 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
381 struct slave *before = NULL, *rx_slave = NULL, *slave;
382 struct list_head *iter;
385 bond_for_each_slave(bond, slave, iter) {
386 if (!SLAVE_IS_OK(slave))
389 if (!before || before->speed < slave->speed)
392 if (!rx_slave || rx_slave->speed < slave->speed)
395 if (slave == bond_info->rx_slave)
398 /* we didn't find anything after the current or we have something
399 * better before and up to the current slave
401 if (!rx_slave || (before && rx_slave->speed < before->speed))
405 bond_info->rx_slave = rx_slave;
410 /* Caller must hold rcu_read_lock() for read */
411 static struct slave *__rlb_next_rx_slave(struct bonding *bond)
413 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
414 struct slave *before = NULL, *rx_slave = NULL, *slave;
415 struct list_head *iter;
418 bond_for_each_slave_rcu(bond, slave, iter) {
419 if (!SLAVE_IS_OK(slave))
422 if (!before || before->speed < slave->speed)
425 if (!rx_slave || rx_slave->speed < slave->speed)
428 if (slave == bond_info->rx_slave)
431 /* we didn't find anything after the current or we have something
432 * better before and up to the current slave
434 if (!rx_slave || (before && rx_slave->speed < before->speed))
438 bond_info->rx_slave = rx_slave;
443 /* teach the switch the mac of a disabled slave
444 * on the primary for fault tolerance
446 * Caller must hold bond->curr_slave_lock for write or bond lock for write
448 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
450 if (!bond->curr_active_slave)
453 if (!bond->alb_info.primary_is_promisc) {
454 if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1))
455 bond->alb_info.primary_is_promisc = 1;
457 bond->alb_info.primary_is_promisc = 0;
460 bond->alb_info.rlb_promisc_timeout_counter = 0;
462 alb_send_learning_packets(bond->curr_active_slave, addr);
465 /* slave being removed should not be active at this point
467 * Caller must hold rtnl.
469 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
471 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
472 struct rlb_client_info *rx_hash_table;
473 u32 index, next_index;
475 /* clear slave from rx_hashtbl */
476 _lock_rx_hashtbl_bh(bond);
478 rx_hash_table = bond_info->rx_hashtbl;
479 index = bond_info->rx_hashtbl_used_head;
480 for (; index != RLB_NULL_INDEX; index = next_index) {
481 next_index = rx_hash_table[index].used_next;
482 if (rx_hash_table[index].slave == slave) {
483 struct slave *assigned_slave = rlb_next_rx_slave(bond);
485 if (assigned_slave) {
486 rx_hash_table[index].slave = assigned_slave;
487 if (!ether_addr_equal_64bits(rx_hash_table[index].mac_dst,
489 bond_info->rx_hashtbl[index].ntt = 1;
490 bond_info->rx_ntt = 1;
491 /* A slave has been removed from the
492 * table because it is either disabled
493 * or being released. We must retry the
494 * update to avoid clients from not
495 * being updated & disconnecting when
498 bond_info->rlb_update_retry_counter =
501 } else { /* there is no active slave */
502 rx_hash_table[index].slave = NULL;
507 _unlock_rx_hashtbl_bh(bond);
509 write_lock_bh(&bond->curr_slave_lock);
511 if (slave != bond->curr_active_slave)
512 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
514 write_unlock_bh(&bond->curr_slave_lock);
517 static void rlb_update_client(struct rlb_client_info *client_info)
521 if (!client_info->slave)
524 for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
527 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
529 client_info->slave->dev,
531 client_info->mac_dst,
532 client_info->slave->dev->dev_addr,
533 client_info->mac_dst);
535 pr_err("%s: Error: failed to create an ARP packet\n",
536 client_info->slave->bond->dev->name);
540 skb->dev = client_info->slave->dev;
542 if (client_info->vlan_id) {
543 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), client_info->vlan_id);
545 pr_err("%s: Error: failed to insert VLAN tag\n",
546 client_info->slave->bond->dev->name);
555 /* sends ARP REPLIES that update the clients that need updating */
556 static void rlb_update_rx_clients(struct bonding *bond)
558 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
559 struct rlb_client_info *client_info;
562 _lock_rx_hashtbl_bh(bond);
564 hash_index = bond_info->rx_hashtbl_used_head;
565 for (; hash_index != RLB_NULL_INDEX;
566 hash_index = client_info->used_next) {
567 client_info = &(bond_info->rx_hashtbl[hash_index]);
568 if (client_info->ntt) {
569 rlb_update_client(client_info);
570 if (bond_info->rlb_update_retry_counter == 0)
571 client_info->ntt = 0;
575 /* do not update the entries again until this counter is zero so that
576 * not to confuse the clients.
578 bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
580 _unlock_rx_hashtbl_bh(bond);
583 /* The slave was assigned a new mac address - update the clients */
584 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
586 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
587 struct rlb_client_info *client_info;
591 _lock_rx_hashtbl_bh(bond);
593 hash_index = bond_info->rx_hashtbl_used_head;
594 for (; hash_index != RLB_NULL_INDEX;
595 hash_index = client_info->used_next) {
596 client_info = &(bond_info->rx_hashtbl[hash_index]);
598 if ((client_info->slave == slave) &&
599 !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
600 client_info->ntt = 1;
605 /* update the team's flag only after the whole iteration */
607 bond_info->rx_ntt = 1;
608 /* fasten the change */
609 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
612 _unlock_rx_hashtbl_bh(bond);
615 /* mark all clients using src_ip to be updated */
616 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
618 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
619 struct rlb_client_info *client_info;
622 _lock_rx_hashtbl(bond);
624 hash_index = bond_info->rx_hashtbl_used_head;
625 for (; hash_index != RLB_NULL_INDEX;
626 hash_index = client_info->used_next) {
627 client_info = &(bond_info->rx_hashtbl[hash_index]);
629 if (!client_info->slave) {
630 pr_err("%s: Error: found a client with no channel in the client's hash table\n",
634 /*update all clients using this src_ip, that are not assigned
635 * to the team's address (curr_active_slave) and have a known
636 * unicast mac address.
638 if ((client_info->ip_src == src_ip) &&
639 !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
640 bond->dev->dev_addr) &&
641 !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
642 client_info->ntt = 1;
643 bond_info->rx_ntt = 1;
647 _unlock_rx_hashtbl(bond);
650 /* Caller must hold both bond and ptr locks for read */
651 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
653 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
654 struct arp_pkt *arp = arp_pkt(skb);
655 struct slave *assigned_slave, *curr_active_slave;
656 struct rlb_client_info *client_info;
659 _lock_rx_hashtbl(bond);
661 curr_active_slave = rcu_dereference(bond->curr_active_slave);
663 hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
664 client_info = &(bond_info->rx_hashtbl[hash_index]);
666 if (client_info->assigned) {
667 if ((client_info->ip_src == arp->ip_src) &&
668 (client_info->ip_dst == arp->ip_dst)) {
669 /* the entry is already assigned to this client */
670 if (!ether_addr_equal_64bits(arp->mac_dst, mac_bcast)) {
671 /* update mac address from arp */
672 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
674 ether_addr_copy(client_info->mac_src, arp->mac_src);
676 assigned_slave = client_info->slave;
677 if (assigned_slave) {
678 _unlock_rx_hashtbl(bond);
679 return assigned_slave;
682 /* the entry is already assigned to some other client,
683 * move the old client to primary (curr_active_slave) so
684 * that the new client can be assigned to this entry.
686 if (bond->curr_active_slave &&
687 client_info->slave != curr_active_slave) {
688 client_info->slave = curr_active_slave;
689 rlb_update_client(client_info);
693 /* assign a new slave */
694 assigned_slave = __rlb_next_rx_slave(bond);
696 if (assigned_slave) {
697 if (!(client_info->assigned &&
698 client_info->ip_src == arp->ip_src)) {
699 /* ip_src is going to be updated,
700 * fix the src hash list
702 u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
703 sizeof(arp->ip_src));
704 rlb_src_unlink(bond, hash_index);
705 rlb_src_link(bond, hash_src, hash_index);
708 client_info->ip_src = arp->ip_src;
709 client_info->ip_dst = arp->ip_dst;
710 /* arp->mac_dst is broadcast for arp reqeusts.
711 * will be updated with clients actual unicast mac address
712 * upon receiving an arp reply.
714 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
715 ether_addr_copy(client_info->mac_src, arp->mac_src);
716 client_info->slave = assigned_slave;
718 if (!ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
719 client_info->ntt = 1;
720 bond->alb_info.rx_ntt = 1;
722 client_info->ntt = 0;
725 if (vlan_get_tag(skb, &client_info->vlan_id))
726 client_info->vlan_id = 0;
728 if (!client_info->assigned) {
729 u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
730 bond_info->rx_hashtbl_used_head = hash_index;
731 client_info->used_next = prev_tbl_head;
732 if (prev_tbl_head != RLB_NULL_INDEX) {
733 bond_info->rx_hashtbl[prev_tbl_head].used_prev =
736 client_info->assigned = 1;
740 _unlock_rx_hashtbl(bond);
742 return assigned_slave;
745 /* chooses (and returns) transmit channel for arp reply
746 * does not choose channel for other arp types since they are
747 * sent on the curr_active_slave
749 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
751 struct arp_pkt *arp = arp_pkt(skb);
752 struct slave *tx_slave = NULL;
754 /* Don't modify or load balance ARPs that do not originate locally
755 * (e.g.,arrive via a bridge).
757 if (!bond_slave_has_mac_rcu(bond, arp->mac_src))
760 if (arp->op_code == htons(ARPOP_REPLY)) {
761 /* the arp must be sent on the selected
764 tx_slave = rlb_choose_channel(skb, bond);
766 ether_addr_copy(arp->mac_src, tx_slave->dev->dev_addr);
767 pr_debug("Server sent ARP Reply packet\n");
768 } else if (arp->op_code == htons(ARPOP_REQUEST)) {
769 /* Create an entry in the rx_hashtbl for this client as a
771 * When the arp reply is received the entry will be updated
772 * with the correct unicast address of the client.
774 rlb_choose_channel(skb, bond);
776 /* The ARP reply packets must be delayed so that
777 * they can cancel out the influence of the ARP request.
779 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
781 /* arp requests are broadcast and are sent on the primary
782 * the arp request will collapse all clients on the subnet to
783 * the primary slave. We must register these clients to be
784 * updated with their assigned mac.
786 rlb_req_update_subnet_clients(bond, arp->ip_src);
787 pr_debug("Server sent ARP Request packet\n");
793 /* Caller must hold bond lock for read */
794 static void rlb_rebalance(struct bonding *bond)
796 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
797 struct slave *assigned_slave;
798 struct rlb_client_info *client_info;
802 _lock_rx_hashtbl_bh(bond);
805 hash_index = bond_info->rx_hashtbl_used_head;
806 for (; hash_index != RLB_NULL_INDEX;
807 hash_index = client_info->used_next) {
808 client_info = &(bond_info->rx_hashtbl[hash_index]);
809 assigned_slave = __rlb_next_rx_slave(bond);
810 if (assigned_slave && (client_info->slave != assigned_slave)) {
811 client_info->slave = assigned_slave;
812 client_info->ntt = 1;
817 /* update the team's flag only after the whole iteration */
819 bond_info->rx_ntt = 1;
820 _unlock_rx_hashtbl_bh(bond);
823 /* Caller must hold rx_hashtbl lock */
824 static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
826 entry->used_next = RLB_NULL_INDEX;
827 entry->used_prev = RLB_NULL_INDEX;
832 static void rlb_init_table_entry_src(struct rlb_client_info *entry)
834 entry->src_first = RLB_NULL_INDEX;
835 entry->src_prev = RLB_NULL_INDEX;
836 entry->src_next = RLB_NULL_INDEX;
839 static void rlb_init_table_entry(struct rlb_client_info *entry)
841 memset(entry, 0, sizeof(struct rlb_client_info));
842 rlb_init_table_entry_dst(entry);
843 rlb_init_table_entry_src(entry);
846 static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
848 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
849 u32 next_index = bond_info->rx_hashtbl[index].used_next;
850 u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
852 if (index == bond_info->rx_hashtbl_used_head)
853 bond_info->rx_hashtbl_used_head = next_index;
854 if (prev_index != RLB_NULL_INDEX)
855 bond_info->rx_hashtbl[prev_index].used_next = next_index;
856 if (next_index != RLB_NULL_INDEX)
857 bond_info->rx_hashtbl[next_index].used_prev = prev_index;
860 /* unlink a rlb hash table entry from the src list */
861 static void rlb_src_unlink(struct bonding *bond, u32 index)
863 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
864 u32 next_index = bond_info->rx_hashtbl[index].src_next;
865 u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
867 bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
868 bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
870 if (next_index != RLB_NULL_INDEX)
871 bond_info->rx_hashtbl[next_index].src_prev = prev_index;
873 if (prev_index == RLB_NULL_INDEX)
876 /* is prev_index pointing to the head of this list? */
877 if (bond_info->rx_hashtbl[prev_index].src_first == index)
878 bond_info->rx_hashtbl[prev_index].src_first = next_index;
880 bond_info->rx_hashtbl[prev_index].src_next = next_index;
884 static void rlb_delete_table_entry(struct bonding *bond, u32 index)
886 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
887 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
889 rlb_delete_table_entry_dst(bond, index);
890 rlb_init_table_entry_dst(entry);
892 rlb_src_unlink(bond, index);
895 /* add the rx_hashtbl[ip_dst_hash] entry to the list
896 * of entries with identical ip_src_hash
898 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
900 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
903 bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
904 next = bond_info->rx_hashtbl[ip_src_hash].src_first;
905 bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
906 if (next != RLB_NULL_INDEX)
907 bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
908 bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
911 /* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does
912 * not match arp->mac_src */
913 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
915 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
916 u32 ip_src_hash = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
919 _lock_rx_hashtbl_bh(bond);
921 index = bond_info->rx_hashtbl[ip_src_hash].src_first;
922 while (index != RLB_NULL_INDEX) {
923 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
924 u32 next_index = entry->src_next;
925 if (entry->ip_src == arp->ip_src &&
926 !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
927 rlb_delete_table_entry(bond, index);
930 _unlock_rx_hashtbl_bh(bond);
933 static int rlb_initialize(struct bonding *bond)
935 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
936 struct rlb_client_info *new_hashtbl;
937 int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
940 new_hashtbl = kmalloc(size, GFP_KERNEL);
944 _lock_rx_hashtbl_bh(bond);
946 bond_info->rx_hashtbl = new_hashtbl;
948 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
950 for (i = 0; i < RLB_HASH_TABLE_SIZE; i++)
951 rlb_init_table_entry(bond_info->rx_hashtbl + i);
953 _unlock_rx_hashtbl_bh(bond);
955 /* register to receive ARPs */
956 bond->recv_probe = rlb_arp_recv;
961 static void rlb_deinitialize(struct bonding *bond)
963 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
965 _lock_rx_hashtbl_bh(bond);
967 kfree(bond_info->rx_hashtbl);
968 bond_info->rx_hashtbl = NULL;
969 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
971 _unlock_rx_hashtbl_bh(bond);
974 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
976 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
979 _lock_rx_hashtbl_bh(bond);
981 curr_index = bond_info->rx_hashtbl_used_head;
982 while (curr_index != RLB_NULL_INDEX) {
983 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
984 u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
986 if (curr->vlan_id == vlan_id)
987 rlb_delete_table_entry(bond, curr_index);
989 curr_index = next_index;
992 _unlock_rx_hashtbl_bh(bond);
995 /*********************** tlb/rlb shared functions *********************/
997 static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
1000 struct learning_pkt pkt;
1001 struct sk_buff *skb;
1002 int size = sizeof(struct learning_pkt);
1005 memset(&pkt, 0, size);
1006 ether_addr_copy(pkt.mac_dst, mac_addr);
1007 ether_addr_copy(pkt.mac_src, mac_addr);
1008 pkt.type = cpu_to_be16(ETH_P_LOOPBACK);
1010 skb = dev_alloc_skb(size);
1014 data = skb_put(skb, size);
1015 memcpy(data, &pkt, size);
1017 skb_reset_mac_header(skb);
1018 skb->network_header = skb->mac_header + ETH_HLEN;
1019 skb->protocol = pkt.type;
1020 skb->priority = TC_PRIO_CONTROL;
1021 skb->dev = slave->dev;
1024 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vid);
1026 pr_err("%s: Error: failed to insert VLAN tag\n",
1027 slave->bond->dev->name);
1032 dev_queue_xmit(skb);
1036 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
1038 struct bonding *bond = bond_get_bond_by_slave(slave);
1039 struct net_device *upper;
1040 struct list_head *iter;
1043 alb_send_lp_vid(slave, mac_addr, 0);
1045 /* loop through vlans and send one packet for each */
1047 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
1048 if (upper->priv_flags & IFF_802_1Q_VLAN)
1049 alb_send_lp_vid(slave, mac_addr,
1050 vlan_dev_vlan_id(upper));
1055 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[])
1057 struct net_device *dev = slave->dev;
1058 struct sockaddr s_addr;
1060 if (BOND_MODE(slave->bond) == BOND_MODE_TLB) {
1061 memcpy(dev->dev_addr, addr, dev->addr_len);
1065 /* for rlb each slave must have a unique hw mac addresses so that */
1066 /* each slave will receive packets destined to a different mac */
1067 memcpy(s_addr.sa_data, addr, dev->addr_len);
1068 s_addr.sa_family = dev->type;
1069 if (dev_set_mac_address(dev, &s_addr)) {
1070 pr_err("%s: Error: dev_set_mac_address of dev %s failed!\n"
1071 "ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
1072 slave->bond->dev->name, dev->name);
1079 * Swap MAC addresses between two slaves.
1081 * Called with RTNL held, and no other locks.
1085 static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1087 u8 tmp_mac_addr[ETH_ALEN];
1089 ether_addr_copy(tmp_mac_addr, slave1->dev->dev_addr);
1090 alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr);
1091 alb_set_slave_mac_addr(slave2, tmp_mac_addr);
1096 * Send learning packets after MAC address swap.
1098 * Called with RTNL and no other locks
1100 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1101 struct slave *slave2)
1103 int slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
1104 struct slave *disabled_slave = NULL;
1108 /* fasten the change in the switch */
1109 if (SLAVE_IS_OK(slave1)) {
1110 alb_send_learning_packets(slave1, slave1->dev->dev_addr);
1111 if (bond->alb_info.rlb_enabled) {
1112 /* inform the clients that the mac address
1115 rlb_req_update_slave_clients(bond, slave1);
1118 disabled_slave = slave1;
1121 if (SLAVE_IS_OK(slave2)) {
1122 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
1123 if (bond->alb_info.rlb_enabled) {
1124 /* inform the clients that the mac address
1127 rlb_req_update_slave_clients(bond, slave2);
1130 disabled_slave = slave2;
1133 if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1134 /* A disabled slave was assigned an active mac addr */
1135 rlb_teach_disabled_mac_on_primary(bond,
1136 disabled_slave->dev->dev_addr);
1141 * alb_change_hw_addr_on_detach
1142 * @bond: bonding we're working on
1143 * @slave: the slave that was just detached
1145 * We assume that @slave was already detached from the slave list.
1147 * If @slave's permanent hw address is different both from its current
1148 * address and from @bond's address, then somewhere in the bond there's
1149 * a slave that has @slave's permanet address as its current address.
1150 * We'll make sure that that slave no longer uses @slave's permanent address.
1152 * Caller must hold RTNL and no other locks
1154 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1158 struct slave *found_slave;
1160 perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1161 slave->dev->dev_addr);
1162 perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1163 bond->dev->dev_addr);
1165 if (perm_curr_diff && perm_bond_diff) {
1166 found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1169 /* locking: needs RTNL and nothing else */
1170 alb_swap_mac_addr(slave, found_slave);
1171 alb_fasten_mac_swap(bond, slave, found_slave);
1177 * alb_handle_addr_collision_on_attach
1178 * @bond: bonding we're working on
1179 * @slave: the slave that was just attached
1181 * checks uniqueness of slave's mac address and handles the case the
1182 * new slave uses the bonds mac address.
1184 * If the permanent hw address of @slave is @bond's hw address, we need to
1185 * find a different hw address to give @slave, that isn't in use by any other
1186 * slave in the bond. This address must be, of course, one of the permanent
1187 * addresses of the other slaves.
1189 * We go over the slave list, and for each slave there we compare its
1190 * permanent hw address with the current address of all the other slaves.
1191 * If no match was found, then we've found a slave with a permanent address
1192 * that isn't used by any other slave in the bond, so we can assign it to
1195 * assumption: this function is called before @slave is attached to the
1198 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1200 struct slave *has_bond_addr = bond->curr_active_slave;
1201 struct slave *tmp_slave1, *free_mac_slave = NULL;
1202 struct list_head *iter;
1204 if (!bond_has_slaves(bond)) {
1205 /* this is the first slave */
1209 /* if slave's mac address differs from bond's mac address
1210 * check uniqueness of slave's mac address against the other
1211 * slaves in the bond.
1213 if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1214 if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1217 /* Try setting slave mac to bond address and fall-through
1218 to code handling that situation below... */
1219 alb_set_slave_mac_addr(slave, bond->dev->dev_addr);
1222 /* The slave's address is equal to the address of the bond.
1223 * Search for a spare address in the bond for this slave.
1225 bond_for_each_slave(bond, tmp_slave1, iter) {
1226 if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1227 /* no slave has tmp_slave1's perm addr
1230 free_mac_slave = tmp_slave1;
1234 if (!has_bond_addr) {
1235 if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1236 bond->dev->dev_addr)) {
1238 has_bond_addr = tmp_slave1;
1243 if (free_mac_slave) {
1244 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr);
1246 pr_warn("%s: Warning: the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1247 bond->dev->name, slave->dev->name,
1248 free_mac_slave->dev->name);
1250 } else if (has_bond_addr) {
1251 pr_err("%s: Error: the hw address of slave %s is in use by the bond; couldn't find a slave with a free hw address to give it (this should not have happened)\n",
1252 bond->dev->name, slave->dev->name);
1260 * alb_set_mac_address
1264 * In TLB mode all slaves are configured to the bond's hw address, but set
1265 * their dev_addr field to different addresses (based on their permanent hw
1268 * For each slave, this function sets the interface to the new address and then
1269 * changes its dev_addr field to its previous value.
1271 * Unwinding assumes bond's mac address has not yet changed.
1273 static int alb_set_mac_address(struct bonding *bond, void *addr)
1275 struct slave *slave, *rollback_slave;
1276 struct list_head *iter;
1278 char tmp_addr[ETH_ALEN];
1281 if (bond->alb_info.rlb_enabled)
1284 bond_for_each_slave(bond, slave, iter) {
1285 /* save net_device's current hw address */
1286 ether_addr_copy(tmp_addr, slave->dev->dev_addr);
1288 res = dev_set_mac_address(slave->dev, addr);
1290 /* restore net_device's hw address */
1291 ether_addr_copy(slave->dev->dev_addr, tmp_addr);
1300 memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1301 sa.sa_family = bond->dev->type;
1303 /* unwind from head to the slave that failed */
1304 bond_for_each_slave(bond, rollback_slave, iter) {
1305 if (rollback_slave == slave)
1307 ether_addr_copy(tmp_addr, rollback_slave->dev->dev_addr);
1308 dev_set_mac_address(rollback_slave->dev, &sa);
1309 ether_addr_copy(rollback_slave->dev->dev_addr, tmp_addr);
1315 /************************ exported alb funcions ************************/
1317 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1321 res = tlb_initialize(bond);
1326 bond->alb_info.rlb_enabled = 1;
1327 /* initialize rlb */
1328 res = rlb_initialize(bond);
1330 tlb_deinitialize(bond);
1334 bond->alb_info.rlb_enabled = 0;
1340 void bond_alb_deinitialize(struct bonding *bond)
1342 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1344 tlb_deinitialize(bond);
1346 if (bond_info->rlb_enabled)
1347 rlb_deinitialize(bond);
1350 static int bond_do_alb_xmit(struct sk_buff *skb, struct bonding *bond,
1351 struct slave *tx_slave)
1353 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1354 struct ethhdr *eth_data = eth_hdr(skb);
1357 /* unbalanced or unassigned, send through primary */
1358 tx_slave = rcu_dereference(bond->curr_active_slave);
1359 if (bond->params.tlb_dynamic_lb)
1360 bond_info->unbalanced_load += skb->len;
1363 if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1364 if (tx_slave != rcu_dereference(bond->curr_active_slave)) {
1365 ether_addr_copy(eth_data->h_source,
1366 tx_slave->dev->dev_addr);
1369 bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1373 if (tx_slave && bond->params.tlb_dynamic_lb) {
1374 _lock_tx_hashtbl(bond);
1375 __tlb_clear_slave(bond, tx_slave, 0);
1376 _unlock_tx_hashtbl(bond);
1379 /* no suitable interface, frame not sent */
1380 dev_kfree_skb_any(skb);
1382 return NETDEV_TX_OK;
1385 int bond_tlb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1387 struct bonding *bond = netdev_priv(bond_dev);
1388 struct ethhdr *eth_data;
1389 struct slave *tx_slave = NULL;
1392 skb_reset_mac_header(skb);
1393 eth_data = eth_hdr(skb);
1395 /* Do not TX balance any multicast or broadcast */
1396 if (!is_multicast_ether_addr(eth_data->h_dest)) {
1397 switch (skb->protocol) {
1398 case htons(ETH_P_IP):
1399 case htons(ETH_P_IPX):
1400 /* In case of IPX, it will falback to L2 hash */
1401 case htons(ETH_P_IPV6):
1402 hash_index = bond_xmit_hash(bond, skb);
1403 if (bond->params.tlb_dynamic_lb) {
1404 tx_slave = tlb_choose_channel(bond,
1408 struct list_head *iter;
1409 int idx = hash_index % bond->slave_cnt;
1411 bond_for_each_slave_rcu(bond, tx_slave, iter)
1418 return bond_do_alb_xmit(skb, bond, tx_slave);
1421 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1423 struct bonding *bond = netdev_priv(bond_dev);
1424 struct ethhdr *eth_data;
1425 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1426 struct slave *tx_slave = NULL;
1427 static const __be32 ip_bcast = htonl(0xffffffff);
1429 bool do_tx_balance = true;
1431 const u8 *hash_start = NULL;
1432 struct ipv6hdr *ip6hdr;
1434 skb_reset_mac_header(skb);
1435 eth_data = eth_hdr(skb);
1437 switch (ntohs(skb->protocol)) {
1439 const struct iphdr *iph = ip_hdr(skb);
1441 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast) ||
1442 (iph->daddr == ip_bcast) ||
1443 (iph->protocol == IPPROTO_IGMP)) {
1444 do_tx_balance = false;
1447 hash_start = (char *)&(iph->daddr);
1448 hash_size = sizeof(iph->daddr);
1452 /* IPv6 doesn't really use broadcast mac address, but leave
1453 * that here just in case.
1455 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast)) {
1456 do_tx_balance = false;
1460 /* IPv6 uses all-nodes multicast as an equivalent to
1461 * broadcasts in IPv4.
1463 if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1464 do_tx_balance = false;
1468 /* Additianally, DAD probes should not be tx-balanced as that
1469 * will lead to false positives for duplicate addresses and
1470 * prevent address configuration from working.
1472 ip6hdr = ipv6_hdr(skb);
1473 if (ipv6_addr_any(&ip6hdr->saddr)) {
1474 do_tx_balance = false;
1478 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1479 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1482 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1483 /* something is wrong with this packet */
1484 do_tx_balance = false;
1488 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1489 /* The only protocol worth balancing in
1490 * this family since it has an "ARP" like
1493 do_tx_balance = false;
1497 hash_start = (char *)eth_data->h_dest;
1498 hash_size = ETH_ALEN;
1501 do_tx_balance = false;
1502 if (bond_info->rlb_enabled)
1503 tx_slave = rlb_arp_xmit(skb, bond);
1506 do_tx_balance = false;
1510 if (do_tx_balance) {
1511 hash_index = _simple_hash(hash_start, hash_size);
1512 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1515 return bond_do_alb_xmit(skb, bond, tx_slave);
1518 void bond_alb_monitor(struct work_struct *work)
1520 struct bonding *bond = container_of(work, struct bonding,
1522 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1523 struct list_head *iter;
1524 struct slave *slave;
1526 if (!bond_has_slaves(bond)) {
1527 bond_info->tx_rebalance_counter = 0;
1528 bond_info->lp_counter = 0;
1534 bond_info->tx_rebalance_counter++;
1535 bond_info->lp_counter++;
1537 /* send learning packets */
1538 if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1539 /* change of curr_active_slave involves swapping of mac addresses.
1540 * in order to avoid this swapping from happening while
1541 * sending the learning packets, the curr_slave_lock must be held for
1544 read_lock(&bond->curr_slave_lock);
1546 bond_for_each_slave_rcu(bond, slave, iter)
1547 alb_send_learning_packets(slave, slave->dev->dev_addr);
1549 read_unlock(&bond->curr_slave_lock);
1551 bond_info->lp_counter = 0;
1554 /* rebalance tx traffic */
1555 if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1557 read_lock(&bond->curr_slave_lock);
1559 bond_for_each_slave_rcu(bond, slave, iter) {
1560 tlb_clear_slave(bond, slave, 1);
1561 if (slave == bond->curr_active_slave) {
1562 SLAVE_TLB_INFO(slave).load =
1563 bond_info->unbalanced_load /
1564 BOND_TLB_REBALANCE_INTERVAL;
1565 bond_info->unbalanced_load = 0;
1569 read_unlock(&bond->curr_slave_lock);
1571 bond_info->tx_rebalance_counter = 0;
1574 /* handle rlb stuff */
1575 if (bond_info->rlb_enabled) {
1576 if (bond_info->primary_is_promisc &&
1577 (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1580 * dev_set_promiscuity requires rtnl and
1581 * nothing else. Avoid race with bond_close.
1584 if (!rtnl_trylock())
1587 bond_info->rlb_promisc_timeout_counter = 0;
1589 /* If the primary was set to promiscuous mode
1590 * because a slave was disabled then
1591 * it can now leave promiscuous mode.
1593 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1594 bond_info->primary_is_promisc = 0;
1600 if (bond_info->rlb_rebalance) {
1601 bond_info->rlb_rebalance = 0;
1602 rlb_rebalance(bond);
1605 /* check if clients need updating */
1606 if (bond_info->rx_ntt) {
1607 if (bond_info->rlb_update_delay_counter) {
1608 --bond_info->rlb_update_delay_counter;
1610 rlb_update_rx_clients(bond);
1611 if (bond_info->rlb_update_retry_counter)
1612 --bond_info->rlb_update_retry_counter;
1614 bond_info->rx_ntt = 0;
1620 queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1623 /* assumption: called before the slave is attached to the bond
1624 * and not locked by the bond lock
1626 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1630 res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr);
1634 res = alb_handle_addr_collision_on_attach(bond, slave);
1638 tlb_init_slave(slave);
1640 /* order a rebalance ASAP */
1641 bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1643 if (bond->alb_info.rlb_enabled)
1644 bond->alb_info.rlb_rebalance = 1;
1650 * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1653 * Caller must hold RTNL and no other locks
1655 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1657 if (bond_has_slaves(bond))
1658 alb_change_hw_addr_on_detach(bond, slave);
1660 tlb_clear_slave(bond, slave, 0);
1662 if (bond->alb_info.rlb_enabled) {
1663 bond->alb_info.rx_slave = NULL;
1664 rlb_clear_slave(bond, slave);
1668 /* Caller must hold bond lock for read */
1669 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1671 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1673 if (link == BOND_LINK_DOWN) {
1674 tlb_clear_slave(bond, slave, 0);
1675 if (bond->alb_info.rlb_enabled)
1676 rlb_clear_slave(bond, slave);
1677 } else if (link == BOND_LINK_UP) {
1678 /* order a rebalance ASAP */
1679 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1680 if (bond->alb_info.rlb_enabled) {
1681 bond->alb_info.rlb_rebalance = 1;
1682 /* If the updelay module parameter is smaller than the
1683 * forwarding delay of the switch the rebalance will
1684 * not work because the rebalance arp replies will
1685 * not be forwarded to the clients..
1692 * bond_alb_handle_active_change - assign new curr_active_slave
1693 * @bond: our bonding struct
1694 * @new_slave: new slave to assign
1696 * Set the bond->curr_active_slave to @new_slave and handle
1697 * mac address swapping and promiscuity changes as needed.
1699 * If new_slave is NULL, caller must hold curr_slave_lock or
1700 * bond->lock for write.
1702 * If new_slave is not NULL, caller must hold RTNL, curr_slave_lock
1703 * for write. Processing here may sleep, so no other locks may be held.
1705 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1706 __releases(&bond->curr_slave_lock)
1707 __acquires(&bond->curr_slave_lock)
1709 struct slave *swap_slave;
1711 if (bond->curr_active_slave == new_slave)
1714 if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1715 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1716 bond->alb_info.primary_is_promisc = 0;
1717 bond->alb_info.rlb_promisc_timeout_counter = 0;
1720 swap_slave = bond->curr_active_slave;
1721 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1723 if (!new_slave || !bond_has_slaves(bond))
1726 /* set the new curr_active_slave to the bonds mac address
1727 * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1730 swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1733 * Arrange for swap_slave and new_slave to temporarily be
1734 * ignored so we can mess with their MAC addresses without
1735 * fear of interference from transmit activity.
1738 tlb_clear_slave(bond, swap_slave, 1);
1739 tlb_clear_slave(bond, new_slave, 1);
1741 write_unlock_bh(&bond->curr_slave_lock);
1745 /* in TLB mode, the slave might flip down/up with the old dev_addr,
1746 * and thus filter bond->dev_addr's packets, so force bond's mac
1748 if (BOND_MODE(bond) == BOND_MODE_TLB) {
1750 u8 tmp_addr[ETH_ALEN];
1752 ether_addr_copy(tmp_addr, new_slave->dev->dev_addr);
1754 memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1755 sa.sa_family = bond->dev->type;
1756 /* we don't care if it can't change its mac, best effort */
1757 dev_set_mac_address(new_slave->dev, &sa);
1759 ether_addr_copy(new_slave->dev->dev_addr, tmp_addr);
1762 /* curr_active_slave must be set before calling alb_swap_mac_addr */
1764 /* swap mac address */
1765 alb_swap_mac_addr(swap_slave, new_slave);
1766 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1768 /* set the new_slave to the bond mac address */
1769 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr);
1770 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1773 write_lock_bh(&bond->curr_slave_lock);
1779 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1780 __acquires(&bond->lock)
1781 __releases(&bond->lock)
1783 struct bonding *bond = netdev_priv(bond_dev);
1784 struct sockaddr *sa = addr;
1785 struct slave *swap_slave;
1788 if (!is_valid_ether_addr(sa->sa_data))
1789 return -EADDRNOTAVAIL;
1791 res = alb_set_mac_address(bond, addr);
1795 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1797 /* If there is no curr_active_slave there is nothing else to do.
1798 * Otherwise we'll need to pass the new address to it and handle
1801 if (!bond->curr_active_slave)
1804 swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1807 alb_swap_mac_addr(swap_slave, bond->curr_active_slave);
1808 alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave);
1810 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr);
1812 read_lock(&bond->lock);
1813 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1814 if (bond->alb_info.rlb_enabled) {
1815 /* inform clients mac address has changed */
1816 rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1818 read_unlock(&bond->lock);
1824 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1826 if (bond->alb_info.rlb_enabled)
1827 rlb_clear_vlan(bond, vlan_id);