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 #include <linux/skbuff.h>
23 #include <linux/netdevice.h>
24 #include <linux/etherdevice.h>
25 #include <linux/pkt_sched.h>
26 #include <linux/spinlock.h>
27 #include <linux/slab.h>
28 #include <linux/timer.h>
30 #include <linux/ipv6.h>
31 #include <linux/if_arp.h>
32 #include <linux/if_ether.h>
33 #include <linux/if_bonding.h>
34 #include <linux/if_vlan.h>
39 #include <asm/byteorder.h>
40 #include <net/bonding.h>
41 #include <net/bond_alb.h>
43 static const u8 mac_v6_allmcast[ETH_ALEN + 2] __long_aligned = {
44 0x33, 0x33, 0x00, 0x00, 0x00, 0x01
46 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
53 u8 padding[ETH_ZLEN - ETH_HLEN];
58 __be16 prot_addr_space;
62 u8 mac_src[ETH_ALEN]; /* sender hardware address */
63 __be32 ip_src; /* sender IP address */
64 u8 mac_dst[ETH_ALEN]; /* target hardware address */
65 __be32 ip_dst; /* target IP address */
69 /* Forward declaration */
70 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
72 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp);
73 static void rlb_src_unlink(struct bonding *bond, u32 index);
74 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash,
77 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
82 for (i = 0; i < hash_size; i++)
83 hash ^= hash_start[i];
88 /*********************** tlb specific functions ***************************/
90 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
93 entry->load_history = 1 + entry->tx_bytes /
94 BOND_TLB_REBALANCE_INTERVAL;
98 entry->tx_slave = NULL;
99 entry->next = TLB_NULL_INDEX;
100 entry->prev = TLB_NULL_INDEX;
103 static inline void tlb_init_slave(struct slave *slave)
105 SLAVE_TLB_INFO(slave).load = 0;
106 SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
109 static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
112 struct tlb_client_info *tx_hash_table;
115 /* clear slave from tx_hashtbl */
116 tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
118 /* skip this if we've already freed the tx hash table */
120 index = SLAVE_TLB_INFO(slave).head;
121 while (index != TLB_NULL_INDEX) {
122 u32 next_index = tx_hash_table[index].next;
123 tlb_init_table_entry(&tx_hash_table[index], save_load);
128 tlb_init_slave(slave);
131 static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
134 spin_lock_bh(&bond->mode_lock);
135 __tlb_clear_slave(bond, slave, save_load);
136 spin_unlock_bh(&bond->mode_lock);
139 /* Must be called before starting the monitor timer */
140 static int tlb_initialize(struct bonding *bond)
142 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
143 int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
144 struct tlb_client_info *new_hashtbl;
147 new_hashtbl = kzalloc(size, GFP_KERNEL);
151 spin_lock_bh(&bond->mode_lock);
153 bond_info->tx_hashtbl = new_hashtbl;
155 for (i = 0; i < TLB_HASH_TABLE_SIZE; i++)
156 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
158 spin_unlock_bh(&bond->mode_lock);
163 /* Must be called only after all slaves have been released */
164 static void tlb_deinitialize(struct bonding *bond)
166 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
168 spin_lock_bh(&bond->mode_lock);
170 kfree(bond_info->tx_hashtbl);
171 bond_info->tx_hashtbl = NULL;
173 spin_unlock_bh(&bond->mode_lock);
176 static long long compute_gap(struct slave *slave)
178 return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
179 (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
182 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
184 struct slave *slave, *least_loaded;
185 struct list_head *iter;
191 /* Find the slave with the largest gap */
192 bond_for_each_slave_rcu(bond, slave, iter) {
193 if (bond_slave_can_tx(slave)) {
194 long long gap = compute_gap(slave);
197 least_loaded = slave;
206 static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
209 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
210 struct tlb_client_info *hash_table;
211 struct slave *assigned_slave;
213 hash_table = bond_info->tx_hashtbl;
214 assigned_slave = hash_table[hash_index].tx_slave;
215 if (!assigned_slave) {
216 assigned_slave = tlb_get_least_loaded_slave(bond);
218 if (assigned_slave) {
219 struct tlb_slave_info *slave_info =
220 &(SLAVE_TLB_INFO(assigned_slave));
221 u32 next_index = slave_info->head;
223 hash_table[hash_index].tx_slave = assigned_slave;
224 hash_table[hash_index].next = next_index;
225 hash_table[hash_index].prev = TLB_NULL_INDEX;
227 if (next_index != TLB_NULL_INDEX)
228 hash_table[next_index].prev = hash_index;
230 slave_info->head = hash_index;
232 hash_table[hash_index].load_history;
237 hash_table[hash_index].tx_bytes += skb_len;
239 return assigned_slave;
242 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
245 struct slave *tx_slave;
247 /* We don't need to disable softirq here, becase
248 * tlb_choose_channel() is only called by bond_alb_xmit()
249 * which already has softirq disabled.
251 spin_lock(&bond->mode_lock);
252 tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
253 spin_unlock(&bond->mode_lock);
258 /*********************** rlb specific functions ***************************/
260 /* when an ARP REPLY is received from a client update its info
263 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
265 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
266 struct rlb_client_info *client_info;
269 spin_lock_bh(&bond->mode_lock);
271 hash_index = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
272 client_info = &(bond_info->rx_hashtbl[hash_index]);
274 if ((client_info->assigned) &&
275 (client_info->ip_src == arp->ip_dst) &&
276 (client_info->ip_dst == arp->ip_src) &&
277 (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) {
278 /* update the clients MAC address */
279 ether_addr_copy(client_info->mac_dst, arp->mac_src);
280 client_info->ntt = 1;
281 bond_info->rx_ntt = 1;
284 spin_unlock_bh(&bond->mode_lock);
287 static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
290 struct arp_pkt *arp, _arp;
292 if (skb->protocol != cpu_to_be16(ETH_P_ARP))
295 arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp);
299 /* We received an ARP from arp->ip_src.
300 * We might have used this IP address previously (on the bonding host
301 * itself or on a system that is bridged together with the bond).
302 * However, if arp->mac_src is different than what is stored in
303 * rx_hashtbl, some other host is now using the IP and we must prevent
304 * sending out client updates with this IP address and the old MAC
306 * Clean up all hash table entries that have this address as ip_src but
307 * have a different mac_src.
309 rlb_purge_src_ip(bond, arp);
311 if (arp->op_code == htons(ARPOP_REPLY)) {
312 /* update rx hash table for this ARP */
313 rlb_update_entry_from_arp(bond, arp);
314 netdev_dbg(bond->dev, "Server received an ARP Reply from client\n");
317 return RX_HANDLER_ANOTHER;
320 /* Caller must hold rcu_read_lock() */
321 static struct slave *__rlb_next_rx_slave(struct bonding *bond)
323 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
324 struct slave *before = NULL, *rx_slave = NULL, *slave;
325 struct list_head *iter;
328 bond_for_each_slave_rcu(bond, slave, iter) {
329 if (!bond_slave_can_tx(slave))
332 if (!before || before->speed < slave->speed)
335 if (!rx_slave || rx_slave->speed < slave->speed)
338 if (slave == bond_info->rx_slave)
341 /* we didn't find anything after the current or we have something
342 * better before and up to the current slave
344 if (!rx_slave || (before && rx_slave->speed < before->speed))
348 bond_info->rx_slave = rx_slave;
353 /* Caller must hold RTNL, rcu_read_lock is obtained only to silence checkers */
354 static struct slave *rlb_next_rx_slave(struct bonding *bond)
356 struct slave *rx_slave;
361 rx_slave = __rlb_next_rx_slave(bond);
367 /* teach the switch the mac of a disabled slave
368 * on the primary for fault tolerance
370 * Caller must hold RTNL
372 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
374 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
379 if (!bond->alb_info.primary_is_promisc) {
380 if (!dev_set_promiscuity(curr_active->dev, 1))
381 bond->alb_info.primary_is_promisc = 1;
383 bond->alb_info.primary_is_promisc = 0;
386 bond->alb_info.rlb_promisc_timeout_counter = 0;
388 alb_send_learning_packets(curr_active, addr, true);
391 /* slave being removed should not be active at this point
393 * Caller must hold rtnl.
395 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
397 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
398 struct rlb_client_info *rx_hash_table;
399 u32 index, next_index;
401 /* clear slave from rx_hashtbl */
402 spin_lock_bh(&bond->mode_lock);
404 rx_hash_table = bond_info->rx_hashtbl;
405 index = bond_info->rx_hashtbl_used_head;
406 for (; index != RLB_NULL_INDEX; index = next_index) {
407 next_index = rx_hash_table[index].used_next;
408 if (rx_hash_table[index].slave == slave) {
409 struct slave *assigned_slave = rlb_next_rx_slave(bond);
411 if (assigned_slave) {
412 rx_hash_table[index].slave = assigned_slave;
413 if (is_valid_ether_addr(rx_hash_table[index].mac_dst)) {
414 bond_info->rx_hashtbl[index].ntt = 1;
415 bond_info->rx_ntt = 1;
416 /* A slave has been removed from the
417 * table because it is either disabled
418 * or being released. We must retry the
419 * update to avoid clients from not
420 * being updated & disconnecting when
423 bond_info->rlb_update_retry_counter =
426 } else { /* there is no active slave */
427 rx_hash_table[index].slave = NULL;
432 spin_unlock_bh(&bond->mode_lock);
434 if (slave != rtnl_dereference(bond->curr_active_slave))
435 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
438 static void rlb_update_client(struct rlb_client_info *client_info)
442 if (!client_info->slave || !is_valid_ether_addr(client_info->mac_dst))
445 for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
448 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
450 client_info->slave->dev,
452 client_info->mac_dst,
453 client_info->slave->dev->dev_addr,
454 client_info->mac_dst);
456 netdev_err(client_info->slave->bond->dev,
457 "failed to create an ARP packet\n");
461 skb->dev = client_info->slave->dev;
463 if (client_info->vlan_id) {
464 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
465 client_info->vlan_id);
472 /* sends ARP REPLIES that update the clients that need updating */
473 static void rlb_update_rx_clients(struct bonding *bond)
475 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
476 struct rlb_client_info *client_info;
479 spin_lock_bh(&bond->mode_lock);
481 hash_index = bond_info->rx_hashtbl_used_head;
482 for (; hash_index != RLB_NULL_INDEX;
483 hash_index = client_info->used_next) {
484 client_info = &(bond_info->rx_hashtbl[hash_index]);
485 if (client_info->ntt) {
486 rlb_update_client(client_info);
487 if (bond_info->rlb_update_retry_counter == 0)
488 client_info->ntt = 0;
492 /* do not update the entries again until this counter is zero so that
493 * not to confuse the clients.
495 bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
497 spin_unlock_bh(&bond->mode_lock);
500 /* The slave was assigned a new mac address - update the clients */
501 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
503 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
504 struct rlb_client_info *client_info;
508 spin_lock_bh(&bond->mode_lock);
510 hash_index = bond_info->rx_hashtbl_used_head;
511 for (; hash_index != RLB_NULL_INDEX;
512 hash_index = client_info->used_next) {
513 client_info = &(bond_info->rx_hashtbl[hash_index]);
515 if ((client_info->slave == slave) &&
516 is_valid_ether_addr(client_info->mac_dst)) {
517 client_info->ntt = 1;
522 /* update the team's flag only after the whole iteration */
524 bond_info->rx_ntt = 1;
525 /* fasten the change */
526 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
529 spin_unlock_bh(&bond->mode_lock);
532 /* mark all clients using src_ip to be updated */
533 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
535 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
536 struct rlb_client_info *client_info;
539 spin_lock(&bond->mode_lock);
541 hash_index = bond_info->rx_hashtbl_used_head;
542 for (; hash_index != RLB_NULL_INDEX;
543 hash_index = client_info->used_next) {
544 client_info = &(bond_info->rx_hashtbl[hash_index]);
546 if (!client_info->slave) {
547 netdev_err(bond->dev, "found a client with no channel in the client's hash table\n");
550 /* update all clients using this src_ip, that are not assigned
551 * to the team's address (curr_active_slave) and have a known
552 * unicast mac address.
554 if ((client_info->ip_src == src_ip) &&
555 !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
556 bond->dev->dev_addr) &&
557 is_valid_ether_addr(client_info->mac_dst)) {
558 client_info->ntt = 1;
559 bond_info->rx_ntt = 1;
563 spin_unlock(&bond->mode_lock);
566 static struct slave *rlb_choose_channel(struct sk_buff *skb,
567 struct bonding *bond,
568 const struct arp_pkt *arp)
570 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
571 struct slave *assigned_slave, *curr_active_slave;
572 struct rlb_client_info *client_info;
575 spin_lock(&bond->mode_lock);
577 curr_active_slave = rcu_dereference(bond->curr_active_slave);
579 hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
580 client_info = &(bond_info->rx_hashtbl[hash_index]);
582 if (client_info->assigned) {
583 if ((client_info->ip_src == arp->ip_src) &&
584 (client_info->ip_dst == arp->ip_dst)) {
585 /* the entry is already assigned to this client */
586 if (!is_broadcast_ether_addr(arp->mac_dst)) {
587 /* update mac address from arp */
588 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
590 ether_addr_copy(client_info->mac_src, arp->mac_src);
592 assigned_slave = client_info->slave;
593 if (assigned_slave) {
594 spin_unlock(&bond->mode_lock);
595 return assigned_slave;
598 /* the entry is already assigned to some other client,
599 * move the old client to primary (curr_active_slave) so
600 * that the new client can be assigned to this entry.
602 if (curr_active_slave &&
603 client_info->slave != curr_active_slave) {
604 client_info->slave = curr_active_slave;
605 rlb_update_client(client_info);
609 /* assign a new slave */
610 assigned_slave = __rlb_next_rx_slave(bond);
612 if (assigned_slave) {
613 if (!(client_info->assigned &&
614 client_info->ip_src == arp->ip_src)) {
615 /* ip_src is going to be updated,
616 * fix the src hash list
618 u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
619 sizeof(arp->ip_src));
620 rlb_src_unlink(bond, hash_index);
621 rlb_src_link(bond, hash_src, hash_index);
624 client_info->ip_src = arp->ip_src;
625 client_info->ip_dst = arp->ip_dst;
626 /* arp->mac_dst is broadcast for arp reqeusts.
627 * will be updated with clients actual unicast mac address
628 * upon receiving an arp reply.
630 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
631 ether_addr_copy(client_info->mac_src, arp->mac_src);
632 client_info->slave = assigned_slave;
634 if (is_valid_ether_addr(client_info->mac_dst)) {
635 client_info->ntt = 1;
636 bond->alb_info.rx_ntt = 1;
638 client_info->ntt = 0;
641 if (vlan_get_tag(skb, &client_info->vlan_id))
642 client_info->vlan_id = 0;
644 if (!client_info->assigned) {
645 u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
646 bond_info->rx_hashtbl_used_head = hash_index;
647 client_info->used_next = prev_tbl_head;
648 if (prev_tbl_head != RLB_NULL_INDEX) {
649 bond_info->rx_hashtbl[prev_tbl_head].used_prev =
652 client_info->assigned = 1;
656 spin_unlock(&bond->mode_lock);
658 return assigned_slave;
661 /* chooses (and returns) transmit channel for arp reply
662 * does not choose channel for other arp types since they are
663 * sent on the curr_active_slave
665 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
667 struct slave *tx_slave = NULL;
670 if (!pskb_network_may_pull(skb, sizeof(*arp)))
672 arp = (struct arp_pkt *)skb_network_header(skb);
674 /* Don't modify or load balance ARPs that do not originate
675 * from the bond itself or a VLAN directly above the bond.
677 if (!bond_slave_has_mac_rcu(bond, arp->mac_src))
680 if (arp->op_code == htons(ARPOP_REPLY)) {
681 /* the arp must be sent on the selected rx channel */
682 tx_slave = rlb_choose_channel(skb, bond, arp);
684 bond_hw_addr_copy(arp->mac_src, tx_slave->dev->dev_addr,
685 tx_slave->dev->addr_len);
686 netdev_dbg(bond->dev, "Server sent ARP Reply packet\n");
687 } else if (arp->op_code == htons(ARPOP_REQUEST)) {
688 /* Create an entry in the rx_hashtbl for this client as a
690 * When the arp reply is received the entry will be updated
691 * with the correct unicast address of the client.
693 rlb_choose_channel(skb, bond, arp);
695 /* The ARP reply packets must be delayed so that
696 * they can cancel out the influence of the ARP request.
698 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
700 /* arp requests are broadcast and are sent on the primary
701 * the arp request will collapse all clients on the subnet to
702 * the primary slave. We must register these clients to be
703 * updated with their assigned mac.
705 rlb_req_update_subnet_clients(bond, arp->ip_src);
706 netdev_dbg(bond->dev, "Server sent ARP Request packet\n");
712 static void rlb_rebalance(struct bonding *bond)
714 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
715 struct slave *assigned_slave;
716 struct rlb_client_info *client_info;
720 spin_lock_bh(&bond->mode_lock);
723 hash_index = bond_info->rx_hashtbl_used_head;
724 for (; hash_index != RLB_NULL_INDEX;
725 hash_index = client_info->used_next) {
726 client_info = &(bond_info->rx_hashtbl[hash_index]);
727 assigned_slave = __rlb_next_rx_slave(bond);
728 if (assigned_slave && (client_info->slave != assigned_slave)) {
729 client_info->slave = assigned_slave;
730 if (!is_zero_ether_addr(client_info->mac_dst)) {
731 client_info->ntt = 1;
737 /* update the team's flag only after the whole iteration */
739 bond_info->rx_ntt = 1;
740 spin_unlock_bh(&bond->mode_lock);
743 /* Caller must hold mode_lock */
744 static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
746 entry->used_next = RLB_NULL_INDEX;
747 entry->used_prev = RLB_NULL_INDEX;
752 static void rlb_init_table_entry_src(struct rlb_client_info *entry)
754 entry->src_first = RLB_NULL_INDEX;
755 entry->src_prev = RLB_NULL_INDEX;
756 entry->src_next = RLB_NULL_INDEX;
759 static void rlb_init_table_entry(struct rlb_client_info *entry)
761 memset(entry, 0, sizeof(struct rlb_client_info));
762 rlb_init_table_entry_dst(entry);
763 rlb_init_table_entry_src(entry);
766 static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
768 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
769 u32 next_index = bond_info->rx_hashtbl[index].used_next;
770 u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
772 if (index == bond_info->rx_hashtbl_used_head)
773 bond_info->rx_hashtbl_used_head = next_index;
774 if (prev_index != RLB_NULL_INDEX)
775 bond_info->rx_hashtbl[prev_index].used_next = next_index;
776 if (next_index != RLB_NULL_INDEX)
777 bond_info->rx_hashtbl[next_index].used_prev = prev_index;
780 /* unlink a rlb hash table entry from the src list */
781 static void rlb_src_unlink(struct bonding *bond, u32 index)
783 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
784 u32 next_index = bond_info->rx_hashtbl[index].src_next;
785 u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
787 bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
788 bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
790 if (next_index != RLB_NULL_INDEX)
791 bond_info->rx_hashtbl[next_index].src_prev = prev_index;
793 if (prev_index == RLB_NULL_INDEX)
796 /* is prev_index pointing to the head of this list? */
797 if (bond_info->rx_hashtbl[prev_index].src_first == index)
798 bond_info->rx_hashtbl[prev_index].src_first = next_index;
800 bond_info->rx_hashtbl[prev_index].src_next = next_index;
804 static void rlb_delete_table_entry(struct bonding *bond, u32 index)
806 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
807 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
809 rlb_delete_table_entry_dst(bond, index);
810 rlb_init_table_entry_dst(entry);
812 rlb_src_unlink(bond, index);
815 /* add the rx_hashtbl[ip_dst_hash] entry to the list
816 * of entries with identical ip_src_hash
818 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
820 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
823 bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
824 next = bond_info->rx_hashtbl[ip_src_hash].src_first;
825 bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
826 if (next != RLB_NULL_INDEX)
827 bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
828 bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
831 /* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does
832 * not match arp->mac_src
834 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
836 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
837 u32 ip_src_hash = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
840 spin_lock_bh(&bond->mode_lock);
842 index = bond_info->rx_hashtbl[ip_src_hash].src_first;
843 while (index != RLB_NULL_INDEX) {
844 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
845 u32 next_index = entry->src_next;
846 if (entry->ip_src == arp->ip_src &&
847 !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
848 rlb_delete_table_entry(bond, index);
851 spin_unlock_bh(&bond->mode_lock);
854 static int rlb_initialize(struct bonding *bond)
856 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
857 struct rlb_client_info *new_hashtbl;
858 int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
861 new_hashtbl = kmalloc(size, GFP_KERNEL);
865 spin_lock_bh(&bond->mode_lock);
867 bond_info->rx_hashtbl = new_hashtbl;
869 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
871 for (i = 0; i < RLB_HASH_TABLE_SIZE; i++)
872 rlb_init_table_entry(bond_info->rx_hashtbl + i);
874 spin_unlock_bh(&bond->mode_lock);
876 /* register to receive ARPs */
877 bond->recv_probe = rlb_arp_recv;
882 static void rlb_deinitialize(struct bonding *bond)
884 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
886 spin_lock_bh(&bond->mode_lock);
888 kfree(bond_info->rx_hashtbl);
889 bond_info->rx_hashtbl = NULL;
890 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
892 spin_unlock_bh(&bond->mode_lock);
895 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
897 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
900 spin_lock_bh(&bond->mode_lock);
902 curr_index = bond_info->rx_hashtbl_used_head;
903 while (curr_index != RLB_NULL_INDEX) {
904 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
905 u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
907 if (curr->vlan_id == vlan_id)
908 rlb_delete_table_entry(bond, curr_index);
910 curr_index = next_index;
913 spin_unlock_bh(&bond->mode_lock);
916 /*********************** tlb/rlb shared functions *********************/
918 static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
919 __be16 vlan_proto, u16 vid)
921 struct learning_pkt pkt;
923 int size = sizeof(struct learning_pkt);
925 memset(&pkt, 0, size);
926 ether_addr_copy(pkt.mac_dst, mac_addr);
927 ether_addr_copy(pkt.mac_src, mac_addr);
928 pkt.type = cpu_to_be16(ETH_P_LOOPBACK);
930 skb = dev_alloc_skb(size);
934 skb_put_data(skb, &pkt, size);
936 skb_reset_mac_header(skb);
937 skb->network_header = skb->mac_header + ETH_HLEN;
938 skb->protocol = pkt.type;
939 skb->priority = TC_PRIO_CONTROL;
940 skb->dev = slave->dev;
942 netdev_dbg(slave->bond->dev,
943 "Send learning packet: dev %s mac %pM vlan %d\n",
944 slave->dev->name, mac_addr, vid);
947 __vlan_hwaccel_put_tag(skb, vlan_proto, vid);
952 struct alb_walk_data {
953 struct bonding *bond;
959 static int alb_upper_dev_walk(struct net_device *upper, void *_data)
961 struct alb_walk_data *data = _data;
962 bool strict_match = data->strict_match;
963 struct bonding *bond = data->bond;
964 struct slave *slave = data->slave;
965 u8 *mac_addr = data->mac_addr;
966 struct bond_vlan_tag *tags;
968 if (is_vlan_dev(upper) &&
969 bond->nest_level == vlan_get_encap_level(upper) - 1) {
970 if (upper->addr_assign_type == NET_ADDR_STOLEN) {
971 alb_send_lp_vid(slave, mac_addr,
972 vlan_dev_vlan_proto(upper),
973 vlan_dev_vlan_id(upper));
975 alb_send_lp_vid(slave, upper->dev_addr,
976 vlan_dev_vlan_proto(upper),
977 vlan_dev_vlan_id(upper));
981 /* If this is a macvlan device, then only send updates
982 * when strict_match is turned off.
984 if (netif_is_macvlan(upper) && !strict_match) {
985 tags = bond_verify_device_path(bond->dev, upper, 0);
986 if (IS_ERR_OR_NULL(tags))
989 alb_send_lp_vid(slave, upper->dev_addr,
990 tags[0].vlan_proto, tags[0].vlan_id);
997 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
1000 struct bonding *bond = bond_get_bond_by_slave(slave);
1001 struct alb_walk_data data = {
1002 .strict_match = strict_match,
1003 .mac_addr = mac_addr,
1009 alb_send_lp_vid(slave, mac_addr, 0, 0);
1011 /* loop through all devices and see if we need to send a packet
1015 netdev_walk_all_upper_dev_rcu(bond->dev, alb_upper_dev_walk, &data);
1019 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[],
1022 struct net_device *dev = slave->dev;
1023 struct sockaddr_storage ss;
1025 if (BOND_MODE(slave->bond) == BOND_MODE_TLB) {
1026 memcpy(dev->dev_addr, addr, len);
1030 /* for rlb each slave must have a unique hw mac addresses so that
1031 * each slave will receive packets destined to a different mac
1033 memcpy(ss.__data, addr, len);
1034 ss.ss_family = dev->type;
1035 if (dev_set_mac_address(dev, (struct sockaddr *)&ss)) {
1036 netdev_err(slave->bond->dev, "dev_set_mac_address of dev %s failed! ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
1043 /* Swap MAC addresses between two slaves.
1045 * Called with RTNL held, and no other locks.
1047 static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1049 u8 tmp_mac_addr[MAX_ADDR_LEN];
1051 bond_hw_addr_copy(tmp_mac_addr, slave1->dev->dev_addr,
1052 slave1->dev->addr_len);
1053 alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr,
1054 slave2->dev->addr_len);
1055 alb_set_slave_mac_addr(slave2, tmp_mac_addr,
1056 slave1->dev->addr_len);
1060 /* Send learning packets after MAC address swap.
1062 * Called with RTNL and no other locks
1064 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1065 struct slave *slave2)
1067 int slaves_state_differ = (bond_slave_can_tx(slave1) != bond_slave_can_tx(slave2));
1068 struct slave *disabled_slave = NULL;
1072 /* fasten the change in the switch */
1073 if (bond_slave_can_tx(slave1)) {
1074 alb_send_learning_packets(slave1, slave1->dev->dev_addr, false);
1075 if (bond->alb_info.rlb_enabled) {
1076 /* inform the clients that the mac address
1079 rlb_req_update_slave_clients(bond, slave1);
1082 disabled_slave = slave1;
1085 if (bond_slave_can_tx(slave2)) {
1086 alb_send_learning_packets(slave2, slave2->dev->dev_addr, false);
1087 if (bond->alb_info.rlb_enabled) {
1088 /* inform the clients that the mac address
1091 rlb_req_update_slave_clients(bond, slave2);
1094 disabled_slave = slave2;
1097 if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1098 /* A disabled slave was assigned an active mac addr */
1099 rlb_teach_disabled_mac_on_primary(bond,
1100 disabled_slave->dev->dev_addr);
1105 * alb_change_hw_addr_on_detach
1106 * @bond: bonding we're working on
1107 * @slave: the slave that was just detached
1109 * We assume that @slave was already detached from the slave list.
1111 * If @slave's permanent hw address is different both from its current
1112 * address and from @bond's address, then somewhere in the bond there's
1113 * a slave that has @slave's permanet address as its current address.
1114 * We'll make sure that that slave no longer uses @slave's permanent address.
1116 * Caller must hold RTNL and no other locks
1118 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1122 struct slave *found_slave;
1124 perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1125 slave->dev->dev_addr);
1126 perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1127 bond->dev->dev_addr);
1129 if (perm_curr_diff && perm_bond_diff) {
1130 found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1133 alb_swap_mac_addr(slave, found_slave);
1134 alb_fasten_mac_swap(bond, slave, found_slave);
1140 * alb_handle_addr_collision_on_attach
1141 * @bond: bonding we're working on
1142 * @slave: the slave that was just attached
1144 * checks uniqueness of slave's mac address and handles the case the
1145 * new slave uses the bonds mac address.
1147 * If the permanent hw address of @slave is @bond's hw address, we need to
1148 * find a different hw address to give @slave, that isn't in use by any other
1149 * slave in the bond. This address must be, of course, one of the permanent
1150 * addresses of the other slaves.
1152 * We go over the slave list, and for each slave there we compare its
1153 * permanent hw address with the current address of all the other slaves.
1154 * If no match was found, then we've found a slave with a permanent address
1155 * that isn't used by any other slave in the bond, so we can assign it to
1158 * assumption: this function is called before @slave is attached to the
1161 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1163 struct slave *has_bond_addr = rcu_access_pointer(bond->curr_active_slave);
1164 struct slave *tmp_slave1, *free_mac_slave = NULL;
1165 struct list_head *iter;
1167 if (!bond_has_slaves(bond)) {
1168 /* this is the first slave */
1172 /* if slave's mac address differs from bond's mac address
1173 * check uniqueness of slave's mac address against the other
1174 * slaves in the bond.
1176 if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1177 if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1180 /* Try setting slave mac to bond address and fall-through
1181 * to code handling that situation below...
1183 alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1184 bond->dev->addr_len);
1187 /* The slave's address is equal to the address of the bond.
1188 * Search for a spare address in the bond for this slave.
1190 bond_for_each_slave(bond, tmp_slave1, iter) {
1191 if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1192 /* no slave has tmp_slave1's perm addr
1195 free_mac_slave = tmp_slave1;
1199 if (!has_bond_addr) {
1200 if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1201 bond->dev->dev_addr)) {
1203 has_bond_addr = tmp_slave1;
1208 if (free_mac_slave) {
1209 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1210 free_mac_slave->dev->addr_len);
1212 netdev_warn(bond->dev, "the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1213 slave->dev->name, free_mac_slave->dev->name);
1215 } else if (has_bond_addr) {
1216 netdev_err(bond->dev, "the hw address of slave %s is in use by the bond; couldn't find a slave with a free hw address to give it (this should not have happened)\n",
1225 * alb_set_mac_address
1229 * In TLB mode all slaves are configured to the bond's hw address, but set
1230 * their dev_addr field to different addresses (based on their permanent hw
1233 * For each slave, this function sets the interface to the new address and then
1234 * changes its dev_addr field to its previous value.
1236 * Unwinding assumes bond's mac address has not yet changed.
1238 static int alb_set_mac_address(struct bonding *bond, void *addr)
1240 struct slave *slave, *rollback_slave;
1241 struct list_head *iter;
1242 struct sockaddr_storage ss;
1243 char tmp_addr[MAX_ADDR_LEN];
1246 if (bond->alb_info.rlb_enabled)
1249 bond_for_each_slave(bond, slave, iter) {
1250 /* save net_device's current hw address */
1251 bond_hw_addr_copy(tmp_addr, slave->dev->dev_addr,
1252 slave->dev->addr_len);
1254 res = dev_set_mac_address(slave->dev, addr);
1256 /* restore net_device's hw address */
1257 bond_hw_addr_copy(slave->dev->dev_addr, tmp_addr,
1258 slave->dev->addr_len);
1267 memcpy(ss.__data, bond->dev->dev_addr, bond->dev->addr_len);
1268 ss.ss_family = bond->dev->type;
1270 /* unwind from head to the slave that failed */
1271 bond_for_each_slave(bond, rollback_slave, iter) {
1272 if (rollback_slave == slave)
1274 bond_hw_addr_copy(tmp_addr, rollback_slave->dev->dev_addr,
1275 rollback_slave->dev->addr_len);
1276 dev_set_mac_address(rollback_slave->dev,
1277 (struct sockaddr *)&ss);
1278 bond_hw_addr_copy(rollback_slave->dev->dev_addr, tmp_addr,
1279 rollback_slave->dev->addr_len);
1285 /************************ exported alb funcions ************************/
1287 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1291 res = tlb_initialize(bond);
1296 res = rlb_initialize(bond);
1298 tlb_deinitialize(bond);
1301 bond->alb_info.rlb_enabled = 1;
1303 bond->alb_info.rlb_enabled = 0;
1309 void bond_alb_deinitialize(struct bonding *bond)
1311 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1313 tlb_deinitialize(bond);
1315 if (bond_info->rlb_enabled)
1316 rlb_deinitialize(bond);
1319 static netdev_tx_t bond_do_alb_xmit(struct sk_buff *skb, struct bonding *bond,
1320 struct slave *tx_slave)
1322 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1323 struct ethhdr *eth_data = eth_hdr(skb);
1326 /* unbalanced or unassigned, send through primary */
1327 tx_slave = rcu_dereference(bond->curr_active_slave);
1328 if (bond->params.tlb_dynamic_lb)
1329 bond_info->unbalanced_load += skb->len;
1332 if (tx_slave && bond_slave_can_tx(tx_slave)) {
1333 if (tx_slave != rcu_access_pointer(bond->curr_active_slave)) {
1334 ether_addr_copy(eth_data->h_source,
1335 tx_slave->dev->dev_addr);
1338 bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1342 if (tx_slave && bond->params.tlb_dynamic_lb) {
1343 spin_lock(&bond->mode_lock);
1344 __tlb_clear_slave(bond, tx_slave, 0);
1345 spin_unlock(&bond->mode_lock);
1348 /* no suitable interface, frame not sent */
1349 bond_tx_drop(bond->dev, skb);
1351 return NETDEV_TX_OK;
1354 netdev_tx_t bond_tlb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1356 struct bonding *bond = netdev_priv(bond_dev);
1357 struct ethhdr *eth_data;
1358 struct slave *tx_slave = NULL;
1361 skb_reset_mac_header(skb);
1362 eth_data = eth_hdr(skb);
1364 /* Do not TX balance any multicast or broadcast */
1365 if (!is_multicast_ether_addr(eth_data->h_dest)) {
1366 switch (skb->protocol) {
1367 case htons(ETH_P_IP):
1368 case htons(ETH_P_IPX):
1369 /* In case of IPX, it will falback to L2 hash */
1370 case htons(ETH_P_IPV6):
1371 hash_index = bond_xmit_hash(bond, skb);
1372 if (bond->params.tlb_dynamic_lb) {
1373 tx_slave = tlb_choose_channel(bond,
1377 struct bond_up_slave *slaves;
1380 slaves = rcu_dereference(bond->slave_arr);
1381 count = slaves ? READ_ONCE(slaves->count) : 0;
1383 tx_slave = slaves->arr[hash_index %
1389 return bond_do_alb_xmit(skb, bond, tx_slave);
1392 netdev_tx_t bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1394 struct bonding *bond = netdev_priv(bond_dev);
1395 struct ethhdr *eth_data;
1396 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1397 struct slave *tx_slave = NULL;
1398 static const __be32 ip_bcast = htonl(0xffffffff);
1400 bool do_tx_balance = true;
1402 const u8 *hash_start = NULL;
1404 skb_reset_mac_header(skb);
1405 eth_data = eth_hdr(skb);
1407 switch (ntohs(skb->protocol)) {
1409 const struct iphdr *iph;
1411 if (is_broadcast_ether_addr(eth_data->h_dest) ||
1412 !pskb_network_may_pull(skb, sizeof(*iph))) {
1413 do_tx_balance = false;
1417 if (iph->daddr == ip_bcast || iph->protocol == IPPROTO_IGMP) {
1418 do_tx_balance = false;
1421 hash_start = (char *)&(iph->daddr);
1422 hash_size = sizeof(iph->daddr);
1426 const struct ipv6hdr *ip6hdr;
1428 /* IPv6 doesn't really use broadcast mac address, but leave
1429 * that here just in case.
1431 if (is_broadcast_ether_addr(eth_data->h_dest)) {
1432 do_tx_balance = false;
1436 /* IPv6 uses all-nodes multicast as an equivalent to
1437 * broadcasts in IPv4.
1439 if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1440 do_tx_balance = false;
1444 if (!pskb_network_may_pull(skb, sizeof(*ip6hdr))) {
1445 do_tx_balance = false;
1448 /* Additionally, DAD probes should not be tx-balanced as that
1449 * will lead to false positives for duplicate addresses and
1450 * prevent address configuration from working.
1452 ip6hdr = ipv6_hdr(skb);
1453 if (ipv6_addr_any(&ip6hdr->saddr)) {
1454 do_tx_balance = false;
1458 hash_start = (char *)&ip6hdr->daddr;
1459 hash_size = sizeof(ip6hdr->daddr);
1463 const struct ipxhdr *ipxhdr;
1465 if (pskb_network_may_pull(skb, sizeof(*ipxhdr))) {
1466 do_tx_balance = false;
1469 ipxhdr = (struct ipxhdr *)skb_network_header(skb);
1471 if (ipxhdr->ipx_checksum != IPX_NO_CHECKSUM) {
1472 /* something is wrong with this packet */
1473 do_tx_balance = false;
1477 if (ipxhdr->ipx_type != IPX_TYPE_NCP) {
1478 /* The only protocol worth balancing in
1479 * this family since it has an "ARP" like
1482 do_tx_balance = false;
1486 eth_data = eth_hdr(skb);
1487 hash_start = (char *)eth_data->h_dest;
1488 hash_size = ETH_ALEN;
1492 do_tx_balance = false;
1493 if (bond_info->rlb_enabled)
1494 tx_slave = rlb_arp_xmit(skb, bond);
1497 do_tx_balance = false;
1501 if (do_tx_balance) {
1502 if (bond->params.tlb_dynamic_lb) {
1503 hash_index = _simple_hash(hash_start, hash_size);
1504 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1507 * do_tx_balance means we are free to select the tx_slave
1508 * So we do exactly what tlb would do for hash selection
1511 struct bond_up_slave *slaves;
1514 slaves = rcu_dereference(bond->slave_arr);
1515 count = slaves ? READ_ONCE(slaves->count) : 0;
1517 tx_slave = slaves->arr[bond_xmit_hash(bond, skb) %
1522 return bond_do_alb_xmit(skb, bond, tx_slave);
1525 void bond_alb_monitor(struct work_struct *work)
1527 struct bonding *bond = container_of(work, struct bonding,
1529 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1530 struct list_head *iter;
1531 struct slave *slave;
1533 if (!bond_has_slaves(bond)) {
1534 atomic_set(&bond_info->tx_rebalance_counter, 0);
1535 bond_info->lp_counter = 0;
1541 atomic_inc(&bond_info->tx_rebalance_counter);
1542 bond_info->lp_counter++;
1544 /* send learning packets */
1545 if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1548 bond_for_each_slave_rcu(bond, slave, iter) {
1549 /* If updating current_active, use all currently
1550 * user mac addreses (!strict_match). Otherwise, only
1551 * use mac of the slave device.
1552 * In RLB mode, we always use strict matches.
1554 strict_match = (slave != rcu_access_pointer(bond->curr_active_slave) ||
1555 bond_info->rlb_enabled);
1556 alb_send_learning_packets(slave, slave->dev->dev_addr,
1559 bond_info->lp_counter = 0;
1562 /* rebalance tx traffic */
1563 if (atomic_read(&bond_info->tx_rebalance_counter) >= BOND_TLB_REBALANCE_TICKS) {
1564 bond_for_each_slave_rcu(bond, slave, iter) {
1565 tlb_clear_slave(bond, slave, 1);
1566 if (slave == rcu_access_pointer(bond->curr_active_slave)) {
1567 SLAVE_TLB_INFO(slave).load =
1568 bond_info->unbalanced_load /
1569 BOND_TLB_REBALANCE_INTERVAL;
1570 bond_info->unbalanced_load = 0;
1573 atomic_set(&bond_info->tx_rebalance_counter, 0);
1576 if (bond_info->rlb_enabled) {
1577 if (bond_info->primary_is_promisc &&
1578 (++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(rtnl_dereference(bond->curr_active_slave)->dev,
1595 bond_info->primary_is_promisc = 0;
1601 if (bond_info->rlb_rebalance) {
1602 bond_info->rlb_rebalance = 0;
1603 rlb_rebalance(bond);
1606 /* check if clients need updating */
1607 if (bond_info->rx_ntt) {
1608 if (bond_info->rlb_update_delay_counter) {
1609 --bond_info->rlb_update_delay_counter;
1611 rlb_update_rx_clients(bond);
1612 if (bond_info->rlb_update_retry_counter)
1613 --bond_info->rlb_update_retry_counter;
1615 bond_info->rx_ntt = 0;
1621 queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1624 /* assumption: called before the slave is attached to the bond
1625 * and not locked by the bond lock
1627 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1631 res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1632 slave->dev->addr_len);
1636 res = alb_handle_addr_collision_on_attach(bond, slave);
1640 tlb_init_slave(slave);
1642 /* order a rebalance ASAP */
1643 atomic_set(&bond->alb_info.tx_rebalance_counter,
1644 BOND_TLB_REBALANCE_TICKS);
1646 if (bond->alb_info.rlb_enabled)
1647 bond->alb_info.rlb_rebalance = 1;
1652 /* Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1655 * Caller must hold RTNL and no other locks
1657 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1659 if (bond_has_slaves(bond))
1660 alb_change_hw_addr_on_detach(bond, slave);
1662 tlb_clear_slave(bond, slave, 0);
1664 if (bond->alb_info.rlb_enabled) {
1665 bond->alb_info.rx_slave = NULL;
1666 rlb_clear_slave(bond, slave);
1671 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1673 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1675 if (link == BOND_LINK_DOWN) {
1676 tlb_clear_slave(bond, slave, 0);
1677 if (bond->alb_info.rlb_enabled)
1678 rlb_clear_slave(bond, slave);
1679 } else if (link == BOND_LINK_UP) {
1680 /* order a rebalance ASAP */
1681 atomic_set(&bond_info->tx_rebalance_counter,
1682 BOND_TLB_REBALANCE_TICKS);
1683 if (bond->alb_info.rlb_enabled) {
1684 bond->alb_info.rlb_rebalance = 1;
1685 /* If the updelay module parameter is smaller than the
1686 * forwarding delay of the switch the rebalance will
1687 * not work because the rebalance arp replies will
1688 * not be forwarded to the clients..
1693 if (bond_is_nondyn_tlb(bond)) {
1694 if (bond_update_slave_arr(bond, NULL))
1695 pr_err("Failed to build slave-array for TLB mode.\n");
1700 * bond_alb_handle_active_change - assign new curr_active_slave
1701 * @bond: our bonding struct
1702 * @new_slave: new slave to assign
1704 * Set the bond->curr_active_slave to @new_slave and handle
1705 * mac address swapping and promiscuity changes as needed.
1707 * Caller must hold RTNL
1709 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1711 struct slave *swap_slave;
1712 struct slave *curr_active;
1714 curr_active = rtnl_dereference(bond->curr_active_slave);
1715 if (curr_active == new_slave)
1718 if (curr_active && bond->alb_info.primary_is_promisc) {
1719 dev_set_promiscuity(curr_active->dev, -1);
1720 bond->alb_info.primary_is_promisc = 0;
1721 bond->alb_info.rlb_promisc_timeout_counter = 0;
1724 swap_slave = curr_active;
1725 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1727 if (!new_slave || !bond_has_slaves(bond))
1730 /* set the new curr_active_slave to the bonds mac address
1731 * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1734 swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1736 /* Arrange for swap_slave and new_slave to temporarily be
1737 * ignored so we can mess with their MAC addresses without
1738 * fear of interference from transmit activity.
1741 tlb_clear_slave(bond, swap_slave, 1);
1742 tlb_clear_slave(bond, new_slave, 1);
1744 /* in TLB mode, the slave might flip down/up with the old dev_addr,
1745 * and thus filter bond->dev_addr's packets, so force bond's mac
1747 if (BOND_MODE(bond) == BOND_MODE_TLB) {
1748 struct sockaddr_storage ss;
1749 u8 tmp_addr[MAX_ADDR_LEN];
1751 bond_hw_addr_copy(tmp_addr, new_slave->dev->dev_addr,
1752 new_slave->dev->addr_len);
1754 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
1755 bond->dev->addr_len);
1756 ss.ss_family = bond->dev->type;
1757 /* we don't care if it can't change its mac, best effort */
1758 dev_set_mac_address(new_slave->dev, (struct sockaddr *)&ss);
1760 bond_hw_addr_copy(new_slave->dev->dev_addr, tmp_addr,
1761 new_slave->dev->addr_len);
1764 /* curr_active_slave must be set before calling alb_swap_mac_addr */
1766 /* swap mac address */
1767 alb_swap_mac_addr(swap_slave, new_slave);
1768 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1770 /* set the new_slave to the bond mac address */
1771 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1772 bond->dev->addr_len);
1773 alb_send_learning_packets(new_slave, bond->dev->dev_addr,
1778 /* Called with RTNL */
1779 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1781 struct bonding *bond = netdev_priv(bond_dev);
1782 struct sockaddr_storage *ss = addr;
1783 struct slave *curr_active;
1784 struct slave *swap_slave;
1787 if (!is_valid_ether_addr(ss->__data))
1788 return -EADDRNOTAVAIL;
1790 res = alb_set_mac_address(bond, addr);
1794 bond_hw_addr_copy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
1796 /* If there is no curr_active_slave there is nothing else to do.
1797 * Otherwise we'll need to pass the new address to it and handle
1800 curr_active = rtnl_dereference(bond->curr_active_slave);
1804 swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1807 alb_swap_mac_addr(swap_slave, curr_active);
1808 alb_fasten_mac_swap(bond, swap_slave, curr_active);
1810 alb_set_slave_mac_addr(curr_active, bond_dev->dev_addr,
1811 bond_dev->addr_len);
1813 alb_send_learning_packets(curr_active,
1814 bond_dev->dev_addr, false);
1815 if (bond->alb_info.rlb_enabled) {
1816 /* inform clients mac address has changed */
1817 rlb_req_update_slave_clients(bond, curr_active);
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);