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, write to the Free Software Foundation, Inc.,
16 * 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18 * The full GNU General Public License is included in this distribution in the
19 * file called LICENSE.
23 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
25 #include <linux/skbuff.h>
26 #include <linux/netdevice.h>
27 #include <linux/etherdevice.h>
28 #include <linux/pkt_sched.h>
29 #include <linux/spinlock.h>
30 #include <linux/slab.h>
31 #include <linux/timer.h>
33 #include <linux/ipv6.h>
34 #include <linux/if_arp.h>
35 #include <linux/if_ether.h>
36 #include <linux/if_bonding.h>
37 #include <linux/if_vlan.h>
42 #include <asm/byteorder.h>
48 #ifndef __long_aligned
49 #define __long_aligned __attribute__((aligned((sizeof(long)))))
51 static const u8 mac_bcast[ETH_ALEN] __long_aligned = {
52 0xff, 0xff, 0xff, 0xff, 0xff, 0xff
54 static const u8 mac_v6_allmcast[ETH_ALEN] __long_aligned = {
55 0x33, 0x33, 0x00, 0x00, 0x00, 0x01
57 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
64 u8 padding[ETH_ZLEN - ETH_HLEN];
69 __be16 prot_addr_space;
73 u8 mac_src[ETH_ALEN]; /* sender hardware address */
74 __be32 ip_src; /* sender IP address */
75 u8 mac_dst[ETH_ALEN]; /* target hardware address */
76 __be32 ip_dst; /* target IP address */
80 static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb)
82 return (struct arp_pkt *)skb_network_header(skb);
85 /* Forward declaration */
86 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[]);
87 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp);
88 static void rlb_src_unlink(struct bonding *bond, u32 index);
89 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash,
92 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
97 for (i = 0; i < hash_size; i++) {
98 hash ^= hash_start[i];
104 /*********************** tlb specific functions ***************************/
106 static inline void _lock_tx_hashtbl_bh(struct bonding *bond)
108 spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
111 static inline void _unlock_tx_hashtbl_bh(struct bonding *bond)
113 spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
116 static inline void _lock_tx_hashtbl(struct bonding *bond)
118 spin_lock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
121 static inline void _unlock_tx_hashtbl(struct bonding *bond)
123 spin_unlock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
126 /* Caller must hold tx_hashtbl lock */
127 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
130 entry->load_history = 1 + entry->tx_bytes /
131 BOND_TLB_REBALANCE_INTERVAL;
135 entry->tx_slave = NULL;
136 entry->next = TLB_NULL_INDEX;
137 entry->prev = TLB_NULL_INDEX;
140 static inline void tlb_init_slave(struct slave *slave)
142 SLAVE_TLB_INFO(slave).load = 0;
143 SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
146 /* Caller must hold bond lock for read, BH disabled */
147 static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
150 struct tlb_client_info *tx_hash_table;
153 /* clear slave from tx_hashtbl */
154 tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
156 /* skip this if we've already freed the tx hash table */
158 index = SLAVE_TLB_INFO(slave).head;
159 while (index != TLB_NULL_INDEX) {
160 u32 next_index = tx_hash_table[index].next;
161 tlb_init_table_entry(&tx_hash_table[index], save_load);
166 tlb_init_slave(slave);
169 /* Caller must hold bond lock for read */
170 static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
173 _lock_tx_hashtbl_bh(bond);
174 __tlb_clear_slave(bond, slave, save_load);
175 _unlock_tx_hashtbl_bh(bond);
178 /* Must be called before starting the monitor timer */
179 static int tlb_initialize(struct bonding *bond)
181 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
182 int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
183 struct tlb_client_info *new_hashtbl;
186 new_hashtbl = kzalloc(size, GFP_KERNEL);
190 _lock_tx_hashtbl_bh(bond);
192 bond_info->tx_hashtbl = new_hashtbl;
194 for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
195 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
198 _unlock_tx_hashtbl_bh(bond);
203 /* Must be called only after all slaves have been released */
204 static void tlb_deinitialize(struct bonding *bond)
206 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
208 _lock_tx_hashtbl_bh(bond);
210 kfree(bond_info->tx_hashtbl);
211 bond_info->tx_hashtbl = NULL;
213 _unlock_tx_hashtbl_bh(bond);
216 static long long compute_gap(struct slave *slave)
218 return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
219 (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
222 /* Caller must hold bond lock for read */
223 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
225 struct slave *slave, *least_loaded;
226 struct list_head *iter;
232 /* Find the slave with the largest gap */
233 bond_for_each_slave_rcu(bond, slave, iter) {
234 if (SLAVE_IS_OK(slave)) {
235 long long gap = compute_gap(slave);
238 least_loaded = slave;
247 static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
250 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
251 struct tlb_client_info *hash_table;
252 struct slave *assigned_slave;
254 hash_table = bond_info->tx_hashtbl;
255 assigned_slave = hash_table[hash_index].tx_slave;
256 if (!assigned_slave) {
257 assigned_slave = tlb_get_least_loaded_slave(bond);
259 if (assigned_slave) {
260 struct tlb_slave_info *slave_info =
261 &(SLAVE_TLB_INFO(assigned_slave));
262 u32 next_index = slave_info->head;
264 hash_table[hash_index].tx_slave = assigned_slave;
265 hash_table[hash_index].next = next_index;
266 hash_table[hash_index].prev = TLB_NULL_INDEX;
268 if (next_index != TLB_NULL_INDEX) {
269 hash_table[next_index].prev = hash_index;
272 slave_info->head = hash_index;
274 hash_table[hash_index].load_history;
278 if (assigned_slave) {
279 hash_table[hash_index].tx_bytes += skb_len;
282 return assigned_slave;
285 /* Caller must hold bond lock for read */
286 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
289 struct slave *tx_slave;
291 * We don't need to disable softirq here, becase
292 * tlb_choose_channel() is only called by bond_alb_xmit()
293 * which already has softirq disabled.
295 _lock_tx_hashtbl(bond);
296 tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
297 _unlock_tx_hashtbl(bond);
301 /*********************** rlb specific functions ***************************/
302 static inline void _lock_rx_hashtbl_bh(struct bonding *bond)
304 spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
307 static inline void _unlock_rx_hashtbl_bh(struct bonding *bond)
309 spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
312 static inline void _lock_rx_hashtbl(struct bonding *bond)
314 spin_lock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
317 static inline void _unlock_rx_hashtbl(struct bonding *bond)
319 spin_unlock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
322 /* when an ARP REPLY is received from a client update its info
325 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
327 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
328 struct rlb_client_info *client_info;
331 _lock_rx_hashtbl_bh(bond);
333 hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
334 client_info = &(bond_info->rx_hashtbl[hash_index]);
336 if ((client_info->assigned) &&
337 (client_info->ip_src == arp->ip_dst) &&
338 (client_info->ip_dst == arp->ip_src) &&
339 (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) {
340 /* update the clients MAC address */
341 memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
342 client_info->ntt = 1;
343 bond_info->rx_ntt = 1;
346 _unlock_rx_hashtbl_bh(bond);
349 static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
352 struct arp_pkt *arp, _arp;
354 if (skb->protocol != cpu_to_be16(ETH_P_ARP))
357 arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp);
361 /* We received an ARP from arp->ip_src.
362 * We might have used this IP address previously (on the bonding host
363 * itself or on a system that is bridged together with the bond).
364 * However, if arp->mac_src is different than what is stored in
365 * rx_hashtbl, some other host is now using the IP and we must prevent
366 * sending out client updates with this IP address and the old MAC
368 * Clean up all hash table entries that have this address as ip_src but
369 * have a different mac_src.
371 rlb_purge_src_ip(bond, arp);
373 if (arp->op_code == htons(ARPOP_REPLY)) {
374 /* update rx hash table for this ARP */
375 rlb_update_entry_from_arp(bond, arp);
376 pr_debug("Server received an ARP Reply from client\n");
379 return RX_HANDLER_ANOTHER;
382 /* Caller must hold bond lock for read */
383 static struct slave *rlb_next_rx_slave(struct bonding *bond)
385 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
386 struct slave *before = NULL, *rx_slave = NULL, *slave;
387 struct list_head *iter;
390 bond_for_each_slave(bond, slave, iter) {
391 if (!SLAVE_IS_OK(slave))
394 if (!before || before->speed < slave->speed)
397 if (!rx_slave || rx_slave->speed < slave->speed)
400 if (slave == bond_info->rx_slave)
403 /* we didn't find anything after the current or we have something
404 * better before and up to the current slave
406 if (!rx_slave || (before && rx_slave->speed < before->speed))
410 bond_info->rx_slave = rx_slave;
415 /* Caller must hold rcu_read_lock() for read */
416 static struct slave *__rlb_next_rx_slave(struct bonding *bond)
418 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
419 struct slave *before = NULL, *rx_slave = NULL, *slave;
420 struct list_head *iter;
423 bond_for_each_slave_rcu(bond, slave, iter) {
424 if (!SLAVE_IS_OK(slave))
427 if (!before || before->speed < slave->speed)
430 if (!rx_slave || rx_slave->speed < slave->speed)
433 if (slave == bond_info->rx_slave)
436 /* we didn't find anything after the current or we have something
437 * better before and up to the current slave
439 if (!rx_slave || (before && rx_slave->speed < before->speed))
443 bond_info->rx_slave = rx_slave;
448 /* teach the switch the mac of a disabled slave
449 * on the primary for fault tolerance
451 * Caller must hold bond->curr_slave_lock for write or bond lock for write
453 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
455 if (!bond->curr_active_slave) {
459 if (!bond->alb_info.primary_is_promisc) {
460 if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1))
461 bond->alb_info.primary_is_promisc = 1;
463 bond->alb_info.primary_is_promisc = 0;
466 bond->alb_info.rlb_promisc_timeout_counter = 0;
468 alb_send_learning_packets(bond->curr_active_slave, addr);
471 /* slave being removed should not be active at this point
473 * Caller must hold bond lock for read
475 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
477 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
478 struct rlb_client_info *rx_hash_table;
479 u32 index, next_index;
481 /* clear slave from rx_hashtbl */
482 _lock_rx_hashtbl_bh(bond);
484 rx_hash_table = bond_info->rx_hashtbl;
485 index = bond_info->rx_hashtbl_used_head;
486 for (; index != RLB_NULL_INDEX; index = next_index) {
487 next_index = rx_hash_table[index].used_next;
488 if (rx_hash_table[index].slave == slave) {
489 struct slave *assigned_slave = rlb_next_rx_slave(bond);
491 if (assigned_slave) {
492 rx_hash_table[index].slave = assigned_slave;
493 if (!ether_addr_equal_64bits(rx_hash_table[index].mac_dst,
495 bond_info->rx_hashtbl[index].ntt = 1;
496 bond_info->rx_ntt = 1;
497 /* A slave has been removed from the
498 * table because it is either disabled
499 * or being released. We must retry the
500 * update to avoid clients from not
501 * being updated & disconnecting when
504 bond_info->rlb_update_retry_counter =
507 } else { /* there is no active slave */
508 rx_hash_table[index].slave = NULL;
513 _unlock_rx_hashtbl_bh(bond);
515 write_lock_bh(&bond->curr_slave_lock);
517 if (slave != bond->curr_active_slave) {
518 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
521 write_unlock_bh(&bond->curr_slave_lock);
524 static void rlb_update_client(struct rlb_client_info *client_info)
528 if (!client_info->slave) {
532 for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
535 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
537 client_info->slave->dev,
539 client_info->mac_dst,
540 client_info->slave->dev->dev_addr,
541 client_info->mac_dst);
543 pr_err("%s: Error: failed to create an ARP packet\n",
544 client_info->slave->bond->dev->name);
548 skb->dev = client_info->slave->dev;
550 if (client_info->vlan_id) {
551 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), client_info->vlan_id);
553 pr_err("%s: Error: failed to insert VLAN tag\n",
554 client_info->slave->bond->dev->name);
563 /* sends ARP REPLIES that update the clients that need updating */
564 static void rlb_update_rx_clients(struct bonding *bond)
566 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
567 struct rlb_client_info *client_info;
570 _lock_rx_hashtbl_bh(bond);
572 hash_index = bond_info->rx_hashtbl_used_head;
573 for (; hash_index != RLB_NULL_INDEX;
574 hash_index = client_info->used_next) {
575 client_info = &(bond_info->rx_hashtbl[hash_index]);
576 if (client_info->ntt) {
577 rlb_update_client(client_info);
578 if (bond_info->rlb_update_retry_counter == 0) {
579 client_info->ntt = 0;
584 /* do not update the entries again until this counter is zero so that
585 * not to confuse the clients.
587 bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
589 _unlock_rx_hashtbl_bh(bond);
592 /* The slave was assigned a new mac address - update the clients */
593 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
595 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
596 struct rlb_client_info *client_info;
600 _lock_rx_hashtbl_bh(bond);
602 hash_index = bond_info->rx_hashtbl_used_head;
603 for (; hash_index != RLB_NULL_INDEX;
604 hash_index = client_info->used_next) {
605 client_info = &(bond_info->rx_hashtbl[hash_index]);
607 if ((client_info->slave == slave) &&
608 !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
609 client_info->ntt = 1;
614 // update the team's flag only after the whole iteration
616 bond_info->rx_ntt = 1;
618 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
621 _unlock_rx_hashtbl_bh(bond);
624 /* mark all clients using src_ip to be updated */
625 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
627 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
628 struct rlb_client_info *client_info;
631 _lock_rx_hashtbl(bond);
633 hash_index = bond_info->rx_hashtbl_used_head;
634 for (; hash_index != RLB_NULL_INDEX;
635 hash_index = client_info->used_next) {
636 client_info = &(bond_info->rx_hashtbl[hash_index]);
638 if (!client_info->slave) {
639 pr_err("%s: Error: found a client with no channel in the client's hash table\n",
643 /*update all clients using this src_ip, that are not assigned
644 * to the team's address (curr_active_slave) and have a known
645 * unicast mac address.
647 if ((client_info->ip_src == src_ip) &&
648 !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
649 bond->dev->dev_addr) &&
650 !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
651 client_info->ntt = 1;
652 bond_info->rx_ntt = 1;
656 _unlock_rx_hashtbl(bond);
659 /* Caller must hold both bond and ptr locks for read */
660 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
662 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
663 struct arp_pkt *arp = arp_pkt(skb);
664 struct slave *assigned_slave, *curr_active_slave;
665 struct rlb_client_info *client_info;
668 _lock_rx_hashtbl(bond);
670 curr_active_slave = rcu_dereference(bond->curr_active_slave);
672 hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
673 client_info = &(bond_info->rx_hashtbl[hash_index]);
675 if (client_info->assigned) {
676 if ((client_info->ip_src == arp->ip_src) &&
677 (client_info->ip_dst == arp->ip_dst)) {
678 /* the entry is already assigned to this client */
679 if (!ether_addr_equal_64bits(arp->mac_dst, mac_bcast)) {
680 /* update mac address from arp */
681 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
683 memcpy(client_info->mac_src, arp->mac_src, ETH_ALEN);
685 assigned_slave = client_info->slave;
686 if (assigned_slave) {
687 _unlock_rx_hashtbl(bond);
688 return assigned_slave;
691 /* the entry is already assigned to some other client,
692 * move the old client to primary (curr_active_slave) so
693 * that the new client can be assigned to this entry.
695 if (bond->curr_active_slave &&
696 client_info->slave != curr_active_slave) {
697 client_info->slave = curr_active_slave;
698 rlb_update_client(client_info);
702 /* assign a new slave */
703 assigned_slave = __rlb_next_rx_slave(bond);
705 if (assigned_slave) {
706 if (!(client_info->assigned &&
707 client_info->ip_src == arp->ip_src)) {
708 /* ip_src is going to be updated,
709 * fix the src hash list
711 u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
712 sizeof(arp->ip_src));
713 rlb_src_unlink(bond, hash_index);
714 rlb_src_link(bond, hash_src, hash_index);
717 client_info->ip_src = arp->ip_src;
718 client_info->ip_dst = arp->ip_dst;
719 /* arp->mac_dst is broadcast for arp reqeusts.
720 * will be updated with clients actual unicast mac address
721 * upon receiving an arp reply.
723 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
724 memcpy(client_info->mac_src, arp->mac_src, ETH_ALEN);
725 client_info->slave = assigned_slave;
727 if (!ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
728 client_info->ntt = 1;
729 bond->alb_info.rx_ntt = 1;
731 client_info->ntt = 0;
734 if (!vlan_get_tag(skb, &client_info->vlan_id))
735 client_info->vlan_id = 0;
737 if (!client_info->assigned) {
738 u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
739 bond_info->rx_hashtbl_used_head = hash_index;
740 client_info->used_next = prev_tbl_head;
741 if (prev_tbl_head != RLB_NULL_INDEX) {
742 bond_info->rx_hashtbl[prev_tbl_head].used_prev =
745 client_info->assigned = 1;
749 _unlock_rx_hashtbl(bond);
751 return assigned_slave;
754 /* chooses (and returns) transmit channel for arp reply
755 * does not choose channel for other arp types since they are
756 * sent on the curr_active_slave
758 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
760 struct arp_pkt *arp = arp_pkt(skb);
761 struct slave *tx_slave = NULL;
763 /* Don't modify or load balance ARPs that do not originate locally
764 * (e.g.,arrive via a bridge).
766 if (!bond_slave_has_mac_rcu(bond, arp->mac_src))
769 if (arp->op_code == htons(ARPOP_REPLY)) {
770 /* the arp must be sent on the selected
773 tx_slave = rlb_choose_channel(skb, bond);
775 memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
777 pr_debug("Server sent ARP Reply packet\n");
778 } else if (arp->op_code == htons(ARPOP_REQUEST)) {
779 /* Create an entry in the rx_hashtbl for this client as a
781 * When the arp reply is received the entry will be updated
782 * with the correct unicast address of the client.
784 rlb_choose_channel(skb, bond);
786 /* The ARP reply packets must be delayed so that
787 * they can cancel out the influence of the ARP request.
789 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
791 /* arp requests are broadcast and are sent on the primary
792 * the arp request will collapse all clients on the subnet to
793 * the primary slave. We must register these clients to be
794 * updated with their assigned mac.
796 rlb_req_update_subnet_clients(bond, arp->ip_src);
797 pr_debug("Server sent ARP Request packet\n");
803 /* Caller must hold bond lock for read */
804 static void rlb_rebalance(struct bonding *bond)
806 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
807 struct slave *assigned_slave;
808 struct rlb_client_info *client_info;
812 _lock_rx_hashtbl_bh(bond);
815 hash_index = bond_info->rx_hashtbl_used_head;
816 for (; hash_index != RLB_NULL_INDEX;
817 hash_index = client_info->used_next) {
818 client_info = &(bond_info->rx_hashtbl[hash_index]);
819 assigned_slave = rlb_next_rx_slave(bond);
820 if (assigned_slave && (client_info->slave != assigned_slave)) {
821 client_info->slave = assigned_slave;
822 client_info->ntt = 1;
827 /* update the team's flag only after the whole iteration */
829 bond_info->rx_ntt = 1;
831 _unlock_rx_hashtbl_bh(bond);
834 /* Caller must hold rx_hashtbl lock */
835 static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
837 entry->used_next = RLB_NULL_INDEX;
838 entry->used_prev = RLB_NULL_INDEX;
843 static void rlb_init_table_entry_src(struct rlb_client_info *entry)
845 entry->src_first = RLB_NULL_INDEX;
846 entry->src_prev = RLB_NULL_INDEX;
847 entry->src_next = RLB_NULL_INDEX;
850 static void rlb_init_table_entry(struct rlb_client_info *entry)
852 memset(entry, 0, sizeof(struct rlb_client_info));
853 rlb_init_table_entry_dst(entry);
854 rlb_init_table_entry_src(entry);
857 static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
859 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
860 u32 next_index = bond_info->rx_hashtbl[index].used_next;
861 u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
863 if (index == bond_info->rx_hashtbl_used_head)
864 bond_info->rx_hashtbl_used_head = next_index;
865 if (prev_index != RLB_NULL_INDEX)
866 bond_info->rx_hashtbl[prev_index].used_next = next_index;
867 if (next_index != RLB_NULL_INDEX)
868 bond_info->rx_hashtbl[next_index].used_prev = prev_index;
871 /* unlink a rlb hash table entry from the src list */
872 static void rlb_src_unlink(struct bonding *bond, u32 index)
874 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
875 u32 next_index = bond_info->rx_hashtbl[index].src_next;
876 u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
878 bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
879 bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
881 if (next_index != RLB_NULL_INDEX)
882 bond_info->rx_hashtbl[next_index].src_prev = prev_index;
884 if (prev_index == RLB_NULL_INDEX)
887 /* is prev_index pointing to the head of this list? */
888 if (bond_info->rx_hashtbl[prev_index].src_first == index)
889 bond_info->rx_hashtbl[prev_index].src_first = next_index;
891 bond_info->rx_hashtbl[prev_index].src_next = next_index;
895 static void rlb_delete_table_entry(struct bonding *bond, u32 index)
897 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
898 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
900 rlb_delete_table_entry_dst(bond, index);
901 rlb_init_table_entry_dst(entry);
903 rlb_src_unlink(bond, index);
906 /* add the rx_hashtbl[ip_dst_hash] entry to the list
907 * of entries with identical ip_src_hash
909 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
911 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
914 bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
915 next = bond_info->rx_hashtbl[ip_src_hash].src_first;
916 bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
917 if (next != RLB_NULL_INDEX)
918 bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
919 bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
922 /* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does
923 * not match arp->mac_src */
924 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
926 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
927 u32 ip_src_hash = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
930 _lock_rx_hashtbl_bh(bond);
932 index = bond_info->rx_hashtbl[ip_src_hash].src_first;
933 while (index != RLB_NULL_INDEX) {
934 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
935 u32 next_index = entry->src_next;
936 if (entry->ip_src == arp->ip_src &&
937 !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
938 rlb_delete_table_entry(bond, index);
941 _unlock_rx_hashtbl_bh(bond);
944 static int rlb_initialize(struct bonding *bond)
946 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
947 struct rlb_client_info *new_hashtbl;
948 int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
951 new_hashtbl = kmalloc(size, GFP_KERNEL);
955 _lock_rx_hashtbl_bh(bond);
957 bond_info->rx_hashtbl = new_hashtbl;
959 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
961 for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
962 rlb_init_table_entry(bond_info->rx_hashtbl + i);
965 _unlock_rx_hashtbl_bh(bond);
967 /* register to receive ARPs */
968 bond->recv_probe = rlb_arp_recv;
973 static void rlb_deinitialize(struct bonding *bond)
975 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
977 _lock_rx_hashtbl_bh(bond);
979 kfree(bond_info->rx_hashtbl);
980 bond_info->rx_hashtbl = NULL;
981 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
983 _unlock_rx_hashtbl_bh(bond);
986 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
988 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
991 _lock_rx_hashtbl_bh(bond);
993 curr_index = bond_info->rx_hashtbl_used_head;
994 while (curr_index != RLB_NULL_INDEX) {
995 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
996 u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
998 if (curr->vlan_id == vlan_id)
999 rlb_delete_table_entry(bond, curr_index);
1001 curr_index = next_index;
1004 _unlock_rx_hashtbl_bh(bond);
1007 /*********************** tlb/rlb shared functions *********************/
1009 static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
1012 struct learning_pkt pkt;
1013 struct sk_buff *skb;
1014 int size = sizeof(struct learning_pkt);
1017 memset(&pkt, 0, size);
1018 memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
1019 memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
1020 pkt.type = cpu_to_be16(ETH_P_LOOP);
1022 skb = dev_alloc_skb(size);
1026 data = skb_put(skb, size);
1027 memcpy(data, &pkt, size);
1029 skb_reset_mac_header(skb);
1030 skb->network_header = skb->mac_header + ETH_HLEN;
1031 skb->protocol = pkt.type;
1032 skb->priority = TC_PRIO_CONTROL;
1033 skb->dev = slave->dev;
1036 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vid);
1038 pr_err("%s: Error: failed to insert VLAN tag\n",
1039 slave->bond->dev->name);
1044 dev_queue_xmit(skb);
1048 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
1050 struct bonding *bond = bond_get_bond_by_slave(slave);
1051 struct net_device *upper;
1052 struct list_head *iter;
1055 alb_send_lp_vid(slave, mac_addr, 0);
1057 /* loop through vlans and send one packet for each */
1059 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
1060 if (upper->priv_flags & IFF_802_1Q_VLAN)
1061 alb_send_lp_vid(slave, mac_addr,
1062 vlan_dev_vlan_id(upper));
1067 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[])
1069 struct net_device *dev = slave->dev;
1070 struct sockaddr s_addr;
1072 if (slave->bond->params.mode == BOND_MODE_TLB) {
1073 memcpy(dev->dev_addr, addr, dev->addr_len);
1077 /* for rlb each slave must have a unique hw mac addresses so that */
1078 /* each slave will receive packets destined to a different mac */
1079 memcpy(s_addr.sa_data, addr, dev->addr_len);
1080 s_addr.sa_family = dev->type;
1081 if (dev_set_mac_address(dev, &s_addr)) {
1082 pr_err("%s: Error: dev_set_mac_address of dev %s failed!\n"
1083 "ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
1084 slave->bond->dev->name, dev->name);
1091 * Swap MAC addresses between two slaves.
1093 * Called with RTNL held, and no other locks.
1097 static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1099 u8 tmp_mac_addr[ETH_ALEN];
1101 memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
1102 alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr);
1103 alb_set_slave_mac_addr(slave2, tmp_mac_addr);
1108 * Send learning packets after MAC address swap.
1110 * Called with RTNL and no other locks
1112 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1113 struct slave *slave2)
1115 int slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
1116 struct slave *disabled_slave = NULL;
1120 /* fasten the change in the switch */
1121 if (SLAVE_IS_OK(slave1)) {
1122 alb_send_learning_packets(slave1, slave1->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, slave1);
1130 disabled_slave = slave1;
1133 if (SLAVE_IS_OK(slave2)) {
1134 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
1135 if (bond->alb_info.rlb_enabled) {
1136 /* inform the clients that the mac address
1139 rlb_req_update_slave_clients(bond, slave2);
1142 disabled_slave = slave2;
1145 if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1146 /* A disabled slave was assigned an active mac addr */
1147 rlb_teach_disabled_mac_on_primary(bond,
1148 disabled_slave->dev->dev_addr);
1153 * alb_change_hw_addr_on_detach
1154 * @bond: bonding we're working on
1155 * @slave: the slave that was just detached
1157 * We assume that @slave was already detached from the slave list.
1159 * If @slave's permanent hw address is different both from its current
1160 * address and from @bond's address, then somewhere in the bond there's
1161 * a slave that has @slave's permanet address as its current address.
1162 * We'll make sure that that slave no longer uses @slave's permanent address.
1164 * Caller must hold RTNL and no other locks
1166 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1170 struct slave *found_slave;
1172 perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1173 slave->dev->dev_addr);
1174 perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1175 bond->dev->dev_addr);
1177 if (perm_curr_diff && perm_bond_diff) {
1178 found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1181 /* locking: needs RTNL and nothing else */
1182 alb_swap_mac_addr(slave, found_slave);
1183 alb_fasten_mac_swap(bond, slave, found_slave);
1189 * alb_handle_addr_collision_on_attach
1190 * @bond: bonding we're working on
1191 * @slave: the slave that was just attached
1193 * checks uniqueness of slave's mac address and handles the case the
1194 * new slave uses the bonds mac address.
1196 * If the permanent hw address of @slave is @bond's hw address, we need to
1197 * find a different hw address to give @slave, that isn't in use by any other
1198 * slave in the bond. This address must be, of course, one of the permanent
1199 * addresses of the other slaves.
1201 * We go over the slave list, and for each slave there we compare its
1202 * permanent hw address with the current address of all the other slaves.
1203 * If no match was found, then we've found a slave with a permanent address
1204 * that isn't used by any other slave in the bond, so we can assign it to
1207 * assumption: this function is called before @slave is attached to the
1210 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1212 struct slave *has_bond_addr = bond->curr_active_slave;
1213 struct slave *tmp_slave1, *free_mac_slave = NULL;
1214 struct list_head *iter;
1216 if (!bond_has_slaves(bond)) {
1217 /* this is the first slave */
1221 /* if slave's mac address differs from bond's mac address
1222 * check uniqueness of slave's mac address against the other
1223 * slaves in the bond.
1225 if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1226 if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1229 /* Try setting slave mac to bond address and fall-through
1230 to code handling that situation below... */
1231 alb_set_slave_mac_addr(slave, bond->dev->dev_addr);
1234 /* The slave's address is equal to the address of the bond.
1235 * Search for a spare address in the bond for this slave.
1237 bond_for_each_slave(bond, tmp_slave1, iter) {
1238 if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1239 /* no slave has tmp_slave1's perm addr
1242 free_mac_slave = tmp_slave1;
1246 if (!has_bond_addr) {
1247 if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1248 bond->dev->dev_addr)) {
1250 has_bond_addr = tmp_slave1;
1255 if (free_mac_slave) {
1256 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr);
1258 pr_warning("%s: Warning: the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1259 bond->dev->name, slave->dev->name,
1260 free_mac_slave->dev->name);
1262 } else if (has_bond_addr) {
1263 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",
1264 bond->dev->name, slave->dev->name);
1272 * alb_set_mac_address
1276 * In TLB mode all slaves are configured to the bond's hw address, but set
1277 * their dev_addr field to different addresses (based on their permanent hw
1280 * For each slave, this function sets the interface to the new address and then
1281 * changes its dev_addr field to its previous value.
1283 * Unwinding assumes bond's mac address has not yet changed.
1285 static int alb_set_mac_address(struct bonding *bond, void *addr)
1287 struct slave *slave, *rollback_slave;
1288 struct list_head *iter;
1290 char tmp_addr[ETH_ALEN];
1293 if (bond->alb_info.rlb_enabled)
1296 bond_for_each_slave(bond, slave, iter) {
1297 /* save net_device's current hw address */
1298 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1300 res = dev_set_mac_address(slave->dev, addr);
1302 /* restore net_device's hw address */
1303 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1312 memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1313 sa.sa_family = bond->dev->type;
1315 /* unwind from head to the slave that failed */
1316 bond_for_each_slave(bond, rollback_slave, iter) {
1317 if (rollback_slave == slave)
1319 memcpy(tmp_addr, rollback_slave->dev->dev_addr, ETH_ALEN);
1320 dev_set_mac_address(rollback_slave->dev, &sa);
1321 memcpy(rollback_slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1327 /************************ exported alb funcions ************************/
1329 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1333 res = tlb_initialize(bond);
1339 bond->alb_info.rlb_enabled = 1;
1340 /* initialize rlb */
1341 res = rlb_initialize(bond);
1343 tlb_deinitialize(bond);
1347 bond->alb_info.rlb_enabled = 0;
1353 void bond_alb_deinitialize(struct bonding *bond)
1355 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1357 tlb_deinitialize(bond);
1359 if (bond_info->rlb_enabled) {
1360 rlb_deinitialize(bond);
1364 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1366 struct bonding *bond = netdev_priv(bond_dev);
1367 struct ethhdr *eth_data;
1368 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1369 struct slave *tx_slave = NULL;
1370 static const __be32 ip_bcast = htonl(0xffffffff);
1372 int do_tx_balance = 1;
1374 const u8 *hash_start = NULL;
1376 struct ipv6hdr *ip6hdr;
1378 skb_reset_mac_header(skb);
1379 eth_data = eth_hdr(skb);
1381 switch (ntohs(skb->protocol)) {
1383 const struct iphdr *iph = ip_hdr(skb);
1385 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast) ||
1386 (iph->daddr == ip_bcast) ||
1387 (iph->protocol == IPPROTO_IGMP)) {
1391 hash_start = (char *)&(iph->daddr);
1392 hash_size = sizeof(iph->daddr);
1396 /* IPv6 doesn't really use broadcast mac address, but leave
1397 * that here just in case.
1399 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast)) {
1404 /* IPv6 uses all-nodes multicast as an equivalent to
1405 * broadcasts in IPv4.
1407 if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1412 /* Additianally, DAD probes should not be tx-balanced as that
1413 * will lead to false positives for duplicate addresses and
1414 * prevent address configuration from working.
1416 ip6hdr = ipv6_hdr(skb);
1417 if (ipv6_addr_any(&ip6hdr->saddr)) {
1422 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1423 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1426 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1427 /* something is wrong with this packet */
1432 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1433 /* The only protocol worth balancing in
1434 * this family since it has an "ARP" like
1441 hash_start = (char*)eth_data->h_dest;
1442 hash_size = ETH_ALEN;
1446 if (bond_info->rlb_enabled) {
1447 tx_slave = rlb_arp_xmit(skb, bond);
1455 if (do_tx_balance) {
1456 hash_index = _simple_hash(hash_start, hash_size);
1457 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1461 /* unbalanced or unassigned, send through primary */
1462 tx_slave = rcu_dereference(bond->curr_active_slave);
1463 bond_info->unbalanced_load += skb->len;
1466 if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1467 if (tx_slave != rcu_dereference(bond->curr_active_slave)) {
1468 memcpy(eth_data->h_source,
1469 tx_slave->dev->dev_addr,
1473 res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1476 _lock_tx_hashtbl(bond);
1477 __tlb_clear_slave(bond, tx_slave, 0);
1478 _unlock_tx_hashtbl(bond);
1483 /* no suitable interface, frame not sent */
1487 return NETDEV_TX_OK;
1490 void bond_alb_monitor(struct work_struct *work)
1492 struct bonding *bond = container_of(work, struct bonding,
1494 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1495 struct list_head *iter;
1496 struct slave *slave;
1498 if (!rtnl_trylock())
1501 if (!bond_has_slaves(bond)) {
1502 bond_info->tx_rebalance_counter = 0;
1503 bond_info->lp_counter = 0;
1508 bond_info->tx_rebalance_counter++;
1509 bond_info->lp_counter++;
1511 /* send learning packets */
1512 if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1513 /* change of curr_active_slave involves swapping of mac addresses.
1514 * in order to avoid this swapping from happening while
1515 * sending the learning packets, the curr_slave_lock must be held for
1518 read_lock(&bond->curr_slave_lock);
1520 bond_for_each_slave(bond, slave, iter)
1521 alb_send_learning_packets(slave, slave->dev->dev_addr);
1523 read_unlock(&bond->curr_slave_lock);
1525 bond_info->lp_counter = 0;
1528 /* rebalance tx traffic */
1529 if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1531 read_lock(&bond->curr_slave_lock);
1533 bond_for_each_slave(bond, slave, iter) {
1534 tlb_clear_slave(bond, slave, 1);
1535 if (slave == bond->curr_active_slave) {
1536 SLAVE_TLB_INFO(slave).load =
1537 bond_info->unbalanced_load /
1538 BOND_TLB_REBALANCE_INTERVAL;
1539 bond_info->unbalanced_load = 0;
1543 read_unlock(&bond->curr_slave_lock);
1545 bond_info->tx_rebalance_counter = 0;
1548 /* handle rlb stuff */
1549 if (bond_info->rlb_enabled) {
1550 if (bond_info->primary_is_promisc &&
1551 (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1553 bond_info->rlb_promisc_timeout_counter = 0;
1555 /* If the primary was set to promiscuous mode
1556 * because a slave was disabled then
1557 * it can now leave promiscuous mode.
1559 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1560 bond_info->primary_is_promisc = 0;
1563 if (bond_info->rlb_rebalance) {
1564 bond_info->rlb_rebalance = 0;
1565 rlb_rebalance(bond);
1568 /* check if clients need updating */
1569 if (bond_info->rx_ntt) {
1570 if (bond_info->rlb_update_delay_counter) {
1571 --bond_info->rlb_update_delay_counter;
1573 rlb_update_rx_clients(bond);
1574 if (bond_info->rlb_update_retry_counter) {
1575 --bond_info->rlb_update_retry_counter;
1577 bond_info->rx_ntt = 0;
1585 queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1588 /* assumption: called before the slave is attached to the bond
1589 * and not locked by the bond lock
1591 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1595 res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr);
1600 res = alb_handle_addr_collision_on_attach(bond, slave);
1605 tlb_init_slave(slave);
1607 /* order a rebalance ASAP */
1608 bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1610 if (bond->alb_info.rlb_enabled) {
1611 bond->alb_info.rlb_rebalance = 1;
1618 * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1621 * Caller must hold RTNL and no other locks
1623 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1625 if (bond_has_slaves(bond))
1626 alb_change_hw_addr_on_detach(bond, slave);
1628 tlb_clear_slave(bond, slave, 0);
1630 if (bond->alb_info.rlb_enabled) {
1631 bond->alb_info.rx_slave = NULL;
1632 rlb_clear_slave(bond, slave);
1636 /* Caller must hold bond lock for read */
1637 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1639 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1641 if (link == BOND_LINK_DOWN) {
1642 tlb_clear_slave(bond, slave, 0);
1643 if (bond->alb_info.rlb_enabled) {
1644 rlb_clear_slave(bond, slave);
1646 } else if (link == BOND_LINK_UP) {
1647 /* order a rebalance ASAP */
1648 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1649 if (bond->alb_info.rlb_enabled) {
1650 bond->alb_info.rlb_rebalance = 1;
1651 /* If the updelay module parameter is smaller than the
1652 * forwarding delay of the switch the rebalance will
1653 * not work because the rebalance arp replies will
1654 * not be forwarded to the clients..
1661 * bond_alb_handle_active_change - assign new curr_active_slave
1662 * @bond: our bonding struct
1663 * @new_slave: new slave to assign
1665 * Set the bond->curr_active_slave to @new_slave and handle
1666 * mac address swapping and promiscuity changes as needed.
1668 * If new_slave is NULL, caller must hold curr_slave_lock or
1669 * bond->lock for write.
1671 * If new_slave is not NULL, caller must hold RTNL, bond->lock for
1672 * read and curr_slave_lock for write. Processing here may sleep, so
1673 * no other locks may be held.
1675 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1676 __releases(&bond->curr_slave_lock)
1677 __releases(&bond->lock)
1678 __acquires(&bond->lock)
1679 __acquires(&bond->curr_slave_lock)
1681 struct slave *swap_slave;
1683 if (bond->curr_active_slave == new_slave)
1686 if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1687 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1688 bond->alb_info.primary_is_promisc = 0;
1689 bond->alb_info.rlb_promisc_timeout_counter = 0;
1692 swap_slave = bond->curr_active_slave;
1693 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1695 if (!new_slave || !bond_has_slaves(bond))
1698 /* set the new curr_active_slave to the bonds mac address
1699 * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1702 swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1705 * Arrange for swap_slave and new_slave to temporarily be
1706 * ignored so we can mess with their MAC addresses without
1707 * fear of interference from transmit activity.
1710 tlb_clear_slave(bond, swap_slave, 1);
1711 tlb_clear_slave(bond, new_slave, 1);
1713 write_unlock_bh(&bond->curr_slave_lock);
1714 read_unlock(&bond->lock);
1718 /* in TLB mode, the slave might flip down/up with the old dev_addr,
1719 * and thus filter bond->dev_addr's packets, so force bond's mac
1721 if (bond->params.mode == BOND_MODE_TLB) {
1723 u8 tmp_addr[ETH_ALEN];
1725 memcpy(tmp_addr, new_slave->dev->dev_addr, ETH_ALEN);
1727 memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1728 sa.sa_family = bond->dev->type;
1729 /* we don't care if it can't change its mac, best effort */
1730 dev_set_mac_address(new_slave->dev, &sa);
1732 memcpy(new_slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1735 /* curr_active_slave must be set before calling alb_swap_mac_addr */
1737 /* swap mac address */
1738 alb_swap_mac_addr(swap_slave, new_slave);
1739 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1740 read_lock(&bond->lock);
1742 /* set the new_slave to the bond mac address */
1743 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr);
1744 read_lock(&bond->lock);
1745 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1748 write_lock_bh(&bond->curr_slave_lock);
1754 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1755 __acquires(&bond->lock)
1756 __releases(&bond->lock)
1758 struct bonding *bond = netdev_priv(bond_dev);
1759 struct sockaddr *sa = addr;
1760 struct slave *swap_slave;
1763 if (!is_valid_ether_addr(sa->sa_data)) {
1764 return -EADDRNOTAVAIL;
1767 res = alb_set_mac_address(bond, addr);
1772 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1774 /* If there is no curr_active_slave there is nothing else to do.
1775 * Otherwise we'll need to pass the new address to it and handle
1778 if (!bond->curr_active_slave) {
1782 swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1785 alb_swap_mac_addr(swap_slave, bond->curr_active_slave);
1786 alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave);
1788 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr);
1790 read_lock(&bond->lock);
1791 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1792 if (bond->alb_info.rlb_enabled) {
1793 /* inform clients mac address has changed */
1794 rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1796 read_unlock(&bond->lock);
1802 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1804 if (bond->alb_info.rlb_enabled) {
1805 rlb_clear_vlan(bond, vlan_id);