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;
231 /* Find the slave with the largest gap */
232 bond_for_each_slave(bond, slave) {
233 if (SLAVE_IS_OK(slave)) {
234 long long gap = compute_gap(slave);
237 least_loaded = slave;
246 static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
249 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
250 struct tlb_client_info *hash_table;
251 struct slave *assigned_slave;
253 hash_table = bond_info->tx_hashtbl;
254 assigned_slave = hash_table[hash_index].tx_slave;
255 if (!assigned_slave) {
256 assigned_slave = tlb_get_least_loaded_slave(bond);
258 if (assigned_slave) {
259 struct tlb_slave_info *slave_info =
260 &(SLAVE_TLB_INFO(assigned_slave));
261 u32 next_index = slave_info->head;
263 hash_table[hash_index].tx_slave = assigned_slave;
264 hash_table[hash_index].next = next_index;
265 hash_table[hash_index].prev = TLB_NULL_INDEX;
267 if (next_index != TLB_NULL_INDEX) {
268 hash_table[next_index].prev = hash_index;
271 slave_info->head = hash_index;
273 hash_table[hash_index].load_history;
277 if (assigned_slave) {
278 hash_table[hash_index].tx_bytes += skb_len;
281 return assigned_slave;
284 /* Caller must hold bond lock for read */
285 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
288 struct slave *tx_slave;
290 * We don't need to disable softirq here, becase
291 * tlb_choose_channel() is only called by bond_alb_xmit()
292 * which already has softirq disabled.
294 _lock_tx_hashtbl(bond);
295 tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
296 _unlock_tx_hashtbl(bond);
300 /*********************** rlb specific functions ***************************/
301 static inline void _lock_rx_hashtbl_bh(struct bonding *bond)
303 spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
306 static inline void _unlock_rx_hashtbl_bh(struct bonding *bond)
308 spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
311 static inline void _lock_rx_hashtbl(struct bonding *bond)
313 spin_lock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
316 static inline void _unlock_rx_hashtbl(struct bonding *bond)
318 spin_unlock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
321 /* when an ARP REPLY is received from a client update its info
324 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
326 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
327 struct rlb_client_info *client_info;
330 _lock_rx_hashtbl_bh(bond);
332 hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
333 client_info = &(bond_info->rx_hashtbl[hash_index]);
335 if ((client_info->assigned) &&
336 (client_info->ip_src == arp->ip_dst) &&
337 (client_info->ip_dst == arp->ip_src) &&
338 (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) {
339 /* update the clients MAC address */
340 memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
341 client_info->ntt = 1;
342 bond_info->rx_ntt = 1;
345 _unlock_rx_hashtbl_bh(bond);
348 static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
351 struct arp_pkt *arp, _arp;
353 if (skb->protocol != cpu_to_be16(ETH_P_ARP))
356 arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp);
360 /* We received an ARP from arp->ip_src.
361 * We might have used this IP address previously (on the bonding host
362 * itself or on a system that is bridged together with the bond).
363 * However, if arp->mac_src is different than what is stored in
364 * rx_hashtbl, some other host is now using the IP and we must prevent
365 * sending out client updates with this IP address and the old MAC
367 * Clean up all hash table entries that have this address as ip_src but
368 * have a different mac_src.
370 rlb_purge_src_ip(bond, arp);
372 if (arp->op_code == htons(ARPOP_REPLY)) {
373 /* update rx hash table for this ARP */
374 rlb_update_entry_from_arp(bond, arp);
375 pr_debug("Server received an ARP Reply from client\n");
378 return RX_HANDLER_ANOTHER;
381 /* Caller must hold bond lock for read */
382 static struct slave *rlb_next_rx_slave(struct bonding *bond)
384 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
385 struct slave *rx_slave, *slave, *start_at;
388 if (bond_info->next_rx_slave)
389 start_at = bond_info->next_rx_slave;
391 start_at = bond_first_slave(bond);
395 bond_for_each_slave_from(bond, slave, i, start_at) {
396 if (SLAVE_IS_OK(slave)) {
399 } else if (slave->speed > rx_slave->speed) {
406 slave = bond_next_slave(bond, rx_slave);
407 bond_info->next_rx_slave = slave;
413 /* teach the switch the mac of a disabled slave
414 * on the primary for fault tolerance
416 * Caller must hold bond->curr_slave_lock for write or bond lock for write
418 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
420 if (!bond->curr_active_slave) {
424 if (!bond->alb_info.primary_is_promisc) {
425 if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1))
426 bond->alb_info.primary_is_promisc = 1;
428 bond->alb_info.primary_is_promisc = 0;
431 bond->alb_info.rlb_promisc_timeout_counter = 0;
433 alb_send_learning_packets(bond->curr_active_slave, addr);
436 /* slave being removed should not be active at this point
438 * Caller must hold bond lock for read
440 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
442 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
443 struct rlb_client_info *rx_hash_table;
444 u32 index, next_index;
446 /* clear slave from rx_hashtbl */
447 _lock_rx_hashtbl_bh(bond);
449 rx_hash_table = bond_info->rx_hashtbl;
450 index = bond_info->rx_hashtbl_used_head;
451 for (; index != RLB_NULL_INDEX; index = next_index) {
452 next_index = rx_hash_table[index].used_next;
453 if (rx_hash_table[index].slave == slave) {
454 struct slave *assigned_slave = rlb_next_rx_slave(bond);
456 if (assigned_slave) {
457 rx_hash_table[index].slave = assigned_slave;
458 if (!ether_addr_equal_64bits(rx_hash_table[index].mac_dst,
460 bond_info->rx_hashtbl[index].ntt = 1;
461 bond_info->rx_ntt = 1;
462 /* A slave has been removed from the
463 * table because it is either disabled
464 * or being released. We must retry the
465 * update to avoid clients from not
466 * being updated & disconnecting when
469 bond_info->rlb_update_retry_counter =
472 } else { /* there is no active slave */
473 rx_hash_table[index].slave = NULL;
478 _unlock_rx_hashtbl_bh(bond);
480 write_lock_bh(&bond->curr_slave_lock);
482 if (slave != bond->curr_active_slave) {
483 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
486 write_unlock_bh(&bond->curr_slave_lock);
489 static void rlb_update_client(struct rlb_client_info *client_info)
493 if (!client_info->slave) {
497 for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
500 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
502 client_info->slave->dev,
504 client_info->mac_dst,
505 client_info->slave->dev->dev_addr,
506 client_info->mac_dst);
508 pr_err("%s: Error: failed to create an ARP packet\n",
509 client_info->slave->bond->dev->name);
513 skb->dev = client_info->slave->dev;
515 if (client_info->tag) {
516 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), client_info->vlan_id);
518 pr_err("%s: Error: failed to insert VLAN tag\n",
519 client_info->slave->bond->dev->name);
528 /* sends ARP REPLIES that update the clients that need updating */
529 static void rlb_update_rx_clients(struct bonding *bond)
531 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
532 struct rlb_client_info *client_info;
535 _lock_rx_hashtbl_bh(bond);
537 hash_index = bond_info->rx_hashtbl_used_head;
538 for (; hash_index != RLB_NULL_INDEX;
539 hash_index = client_info->used_next) {
540 client_info = &(bond_info->rx_hashtbl[hash_index]);
541 if (client_info->ntt) {
542 rlb_update_client(client_info);
543 if (bond_info->rlb_update_retry_counter == 0) {
544 client_info->ntt = 0;
549 /* do not update the entries again until this counter is zero so that
550 * not to confuse the clients.
552 bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
554 _unlock_rx_hashtbl_bh(bond);
557 /* The slave was assigned a new mac address - update the clients */
558 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
560 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
561 struct rlb_client_info *client_info;
565 _lock_rx_hashtbl_bh(bond);
567 hash_index = bond_info->rx_hashtbl_used_head;
568 for (; hash_index != RLB_NULL_INDEX;
569 hash_index = client_info->used_next) {
570 client_info = &(bond_info->rx_hashtbl[hash_index]);
572 if ((client_info->slave == slave) &&
573 !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
574 client_info->ntt = 1;
579 // update the team's flag only after the whole iteration
581 bond_info->rx_ntt = 1;
583 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
586 _unlock_rx_hashtbl_bh(bond);
589 /* mark all clients using src_ip to be updated */
590 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
592 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
593 struct rlb_client_info *client_info;
596 _lock_rx_hashtbl(bond);
598 hash_index = bond_info->rx_hashtbl_used_head;
599 for (; hash_index != RLB_NULL_INDEX;
600 hash_index = client_info->used_next) {
601 client_info = &(bond_info->rx_hashtbl[hash_index]);
603 if (!client_info->slave) {
604 pr_err("%s: Error: found a client with no channel in the client's hash table\n",
608 /*update all clients using this src_ip, that are not assigned
609 * to the team's address (curr_active_slave) and have a known
610 * unicast mac address.
612 if ((client_info->ip_src == src_ip) &&
613 !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
614 bond->dev->dev_addr) &&
615 !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
616 client_info->ntt = 1;
617 bond_info->rx_ntt = 1;
621 _unlock_rx_hashtbl(bond);
624 /* Caller must hold both bond and ptr locks for read */
625 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
627 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
628 struct arp_pkt *arp = arp_pkt(skb);
629 struct slave *assigned_slave;
630 struct rlb_client_info *client_info;
633 _lock_rx_hashtbl(bond);
635 hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
636 client_info = &(bond_info->rx_hashtbl[hash_index]);
638 if (client_info->assigned) {
639 if ((client_info->ip_src == arp->ip_src) &&
640 (client_info->ip_dst == arp->ip_dst)) {
641 /* the entry is already assigned to this client */
642 if (!ether_addr_equal_64bits(arp->mac_dst, mac_bcast)) {
643 /* update mac address from arp */
644 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
646 memcpy(client_info->mac_src, arp->mac_src, ETH_ALEN);
648 assigned_slave = client_info->slave;
649 if (assigned_slave) {
650 _unlock_rx_hashtbl(bond);
651 return assigned_slave;
654 /* the entry is already assigned to some other client,
655 * move the old client to primary (curr_active_slave) so
656 * that the new client can be assigned to this entry.
658 if (bond->curr_active_slave &&
659 client_info->slave != bond->curr_active_slave) {
660 client_info->slave = bond->curr_active_slave;
661 rlb_update_client(client_info);
665 /* assign a new slave */
666 assigned_slave = rlb_next_rx_slave(bond);
668 if (assigned_slave) {
669 if (!(client_info->assigned &&
670 client_info->ip_src == arp->ip_src)) {
671 /* ip_src is going to be updated,
672 * fix the src hash list
674 u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
675 sizeof(arp->ip_src));
676 rlb_src_unlink(bond, hash_index);
677 rlb_src_link(bond, hash_src, hash_index);
680 client_info->ip_src = arp->ip_src;
681 client_info->ip_dst = arp->ip_dst;
682 /* arp->mac_dst is broadcast for arp reqeusts.
683 * will be updated with clients actual unicast mac address
684 * upon receiving an arp reply.
686 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
687 memcpy(client_info->mac_src, arp->mac_src, ETH_ALEN);
688 client_info->slave = assigned_slave;
690 if (!ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
691 client_info->ntt = 1;
692 bond->alb_info.rx_ntt = 1;
694 client_info->ntt = 0;
697 if (bond_vlan_used(bond)) {
698 if (!vlan_get_tag(skb, &client_info->vlan_id))
699 client_info->tag = 1;
702 if (!client_info->assigned) {
703 u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
704 bond_info->rx_hashtbl_used_head = hash_index;
705 client_info->used_next = prev_tbl_head;
706 if (prev_tbl_head != RLB_NULL_INDEX) {
707 bond_info->rx_hashtbl[prev_tbl_head].used_prev =
710 client_info->assigned = 1;
714 _unlock_rx_hashtbl(bond);
716 return assigned_slave;
719 /* chooses (and returns) transmit channel for arp reply
720 * does not choose channel for other arp types since they are
721 * sent on the curr_active_slave
723 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
725 struct arp_pkt *arp = arp_pkt(skb);
726 struct slave *tx_slave = NULL;
728 /* Don't modify or load balance ARPs that do not originate locally
729 * (e.g.,arrive via a bridge).
731 if (!bond_slave_has_mac(bond, arp->mac_src))
734 if (arp->op_code == htons(ARPOP_REPLY)) {
735 /* the arp must be sent on the selected
738 tx_slave = rlb_choose_channel(skb, bond);
740 memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
742 pr_debug("Server sent ARP Reply packet\n");
743 } else if (arp->op_code == htons(ARPOP_REQUEST)) {
744 /* Create an entry in the rx_hashtbl for this client as a
746 * When the arp reply is received the entry will be updated
747 * with the correct unicast address of the client.
749 rlb_choose_channel(skb, bond);
751 /* The ARP reply packets must be delayed so that
752 * they can cancel out the influence of the ARP request.
754 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
756 /* arp requests are broadcast and are sent on the primary
757 * the arp request will collapse all clients on the subnet to
758 * the primary slave. We must register these clients to be
759 * updated with their assigned mac.
761 rlb_req_update_subnet_clients(bond, arp->ip_src);
762 pr_debug("Server sent ARP Request packet\n");
768 /* Caller must hold bond lock for read */
769 static void rlb_rebalance(struct bonding *bond)
771 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
772 struct slave *assigned_slave;
773 struct rlb_client_info *client_info;
777 _lock_rx_hashtbl_bh(bond);
780 hash_index = bond_info->rx_hashtbl_used_head;
781 for (; hash_index != RLB_NULL_INDEX;
782 hash_index = client_info->used_next) {
783 client_info = &(bond_info->rx_hashtbl[hash_index]);
784 assigned_slave = rlb_next_rx_slave(bond);
785 if (assigned_slave && (client_info->slave != assigned_slave)) {
786 client_info->slave = assigned_slave;
787 client_info->ntt = 1;
792 /* update the team's flag only after the whole iteration */
794 bond_info->rx_ntt = 1;
796 _unlock_rx_hashtbl_bh(bond);
799 /* Caller must hold rx_hashtbl lock */
800 static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
802 entry->used_next = RLB_NULL_INDEX;
803 entry->used_prev = RLB_NULL_INDEX;
808 static void rlb_init_table_entry_src(struct rlb_client_info *entry)
810 entry->src_first = RLB_NULL_INDEX;
811 entry->src_prev = RLB_NULL_INDEX;
812 entry->src_next = RLB_NULL_INDEX;
815 static void rlb_init_table_entry(struct rlb_client_info *entry)
817 memset(entry, 0, sizeof(struct rlb_client_info));
818 rlb_init_table_entry_dst(entry);
819 rlb_init_table_entry_src(entry);
822 static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
824 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
825 u32 next_index = bond_info->rx_hashtbl[index].used_next;
826 u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
828 if (index == bond_info->rx_hashtbl_used_head)
829 bond_info->rx_hashtbl_used_head = next_index;
830 if (prev_index != RLB_NULL_INDEX)
831 bond_info->rx_hashtbl[prev_index].used_next = next_index;
832 if (next_index != RLB_NULL_INDEX)
833 bond_info->rx_hashtbl[next_index].used_prev = prev_index;
836 /* unlink a rlb hash table entry from the src list */
837 static void rlb_src_unlink(struct bonding *bond, u32 index)
839 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
840 u32 next_index = bond_info->rx_hashtbl[index].src_next;
841 u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
843 bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
844 bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
846 if (next_index != RLB_NULL_INDEX)
847 bond_info->rx_hashtbl[next_index].src_prev = prev_index;
849 if (prev_index == RLB_NULL_INDEX)
852 /* is prev_index pointing to the head of this list? */
853 if (bond_info->rx_hashtbl[prev_index].src_first == index)
854 bond_info->rx_hashtbl[prev_index].src_first = next_index;
856 bond_info->rx_hashtbl[prev_index].src_next = next_index;
860 static void rlb_delete_table_entry(struct bonding *bond, u32 index)
862 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
863 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
865 rlb_delete_table_entry_dst(bond, index);
866 rlb_init_table_entry_dst(entry);
868 rlb_src_unlink(bond, index);
871 /* add the rx_hashtbl[ip_dst_hash] entry to the list
872 * of entries with identical ip_src_hash
874 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
876 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
879 bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
880 next = bond_info->rx_hashtbl[ip_src_hash].src_first;
881 bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
882 if (next != RLB_NULL_INDEX)
883 bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
884 bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
887 /* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does
888 * not match arp->mac_src */
889 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
891 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
892 u32 ip_src_hash = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
895 _lock_rx_hashtbl_bh(bond);
897 index = bond_info->rx_hashtbl[ip_src_hash].src_first;
898 while (index != RLB_NULL_INDEX) {
899 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
900 u32 next_index = entry->src_next;
901 if (entry->ip_src == arp->ip_src &&
902 !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
903 rlb_delete_table_entry(bond, index);
906 _unlock_rx_hashtbl_bh(bond);
909 static int rlb_initialize(struct bonding *bond)
911 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
912 struct rlb_client_info *new_hashtbl;
913 int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
916 new_hashtbl = kmalloc(size, GFP_KERNEL);
920 _lock_rx_hashtbl_bh(bond);
922 bond_info->rx_hashtbl = new_hashtbl;
924 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
926 for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
927 rlb_init_table_entry(bond_info->rx_hashtbl + i);
930 _unlock_rx_hashtbl_bh(bond);
932 /* register to receive ARPs */
933 bond->recv_probe = rlb_arp_recv;
938 static void rlb_deinitialize(struct bonding *bond)
940 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
942 _lock_rx_hashtbl_bh(bond);
944 kfree(bond_info->rx_hashtbl);
945 bond_info->rx_hashtbl = NULL;
946 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
948 _unlock_rx_hashtbl_bh(bond);
951 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
953 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
956 _lock_rx_hashtbl_bh(bond);
958 curr_index = bond_info->rx_hashtbl_used_head;
959 while (curr_index != RLB_NULL_INDEX) {
960 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
961 u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
963 if (curr->tag && (curr->vlan_id == vlan_id))
964 rlb_delete_table_entry(bond, curr_index);
966 curr_index = next_index;
969 _unlock_rx_hashtbl_bh(bond);
972 /*********************** tlb/rlb shared functions *********************/
974 static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
977 struct learning_pkt pkt;
979 int size = sizeof(struct learning_pkt);
982 memset(&pkt, 0, size);
983 memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
984 memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
985 pkt.type = cpu_to_be16(ETH_P_LOOP);
987 skb = dev_alloc_skb(size);
991 data = skb_put(skb, size);
992 memcpy(data, &pkt, size);
994 skb_reset_mac_header(skb);
995 skb->network_header = skb->mac_header + ETH_HLEN;
996 skb->protocol = pkt.type;
997 skb->priority = TC_PRIO_CONTROL;
998 skb->dev = slave->dev;
1001 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vid);
1003 pr_err("%s: Error: failed to insert VLAN tag\n",
1004 slave->bond->dev->name);
1009 dev_queue_xmit(skb);
1013 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
1015 struct bonding *bond = bond_get_bond_by_slave(slave);
1016 struct net_device *upper;
1017 struct list_head *iter;
1020 alb_send_lp_vid(slave, mac_addr, 0);
1022 /* loop through vlans and send one packet for each */
1024 netdev_for_each_upper_dev_rcu(bond->dev, upper, iter) {
1025 if (upper->priv_flags & IFF_802_1Q_VLAN)
1026 alb_send_lp_vid(slave, mac_addr,
1027 vlan_dev_vlan_id(upper));
1032 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[])
1034 struct net_device *dev = slave->dev;
1035 struct sockaddr s_addr;
1037 if (slave->bond->params.mode == BOND_MODE_TLB) {
1038 memcpy(dev->dev_addr, addr, dev->addr_len);
1042 /* for rlb each slave must have a unique hw mac addresses so that */
1043 /* each slave will receive packets destined to a different mac */
1044 memcpy(s_addr.sa_data, addr, dev->addr_len);
1045 s_addr.sa_family = dev->type;
1046 if (dev_set_mac_address(dev, &s_addr)) {
1047 pr_err("%s: Error: dev_set_mac_address of dev %s failed!\n"
1048 "ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
1049 slave->bond->dev->name, dev->name);
1056 * Swap MAC addresses between two slaves.
1058 * Called with RTNL held, and no other locks.
1062 static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1064 u8 tmp_mac_addr[ETH_ALEN];
1066 memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
1067 alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr);
1068 alb_set_slave_mac_addr(slave2, tmp_mac_addr);
1073 * Send learning packets after MAC address swap.
1075 * Called with RTNL and no other locks
1077 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1078 struct slave *slave2)
1080 int slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
1081 struct slave *disabled_slave = NULL;
1085 /* fasten the change in the switch */
1086 if (SLAVE_IS_OK(slave1)) {
1087 alb_send_learning_packets(slave1, slave1->dev->dev_addr);
1088 if (bond->alb_info.rlb_enabled) {
1089 /* inform the clients that the mac address
1092 rlb_req_update_slave_clients(bond, slave1);
1095 disabled_slave = slave1;
1098 if (SLAVE_IS_OK(slave2)) {
1099 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
1100 if (bond->alb_info.rlb_enabled) {
1101 /* inform the clients that the mac address
1104 rlb_req_update_slave_clients(bond, slave2);
1107 disabled_slave = slave2;
1110 if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1111 /* A disabled slave was assigned an active mac addr */
1112 rlb_teach_disabled_mac_on_primary(bond,
1113 disabled_slave->dev->dev_addr);
1118 * alb_change_hw_addr_on_detach
1119 * @bond: bonding we're working on
1120 * @slave: the slave that was just detached
1122 * We assume that @slave was already detached from the slave list.
1124 * If @slave's permanent hw address is different both from its current
1125 * address and from @bond's address, then somewhere in the bond there's
1126 * a slave that has @slave's permanet address as its current address.
1127 * We'll make sure that that slave no longer uses @slave's permanent address.
1129 * Caller must hold RTNL and no other locks
1131 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1135 struct slave *found_slave;
1137 perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1138 slave->dev->dev_addr);
1139 perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1140 bond->dev->dev_addr);
1142 if (perm_curr_diff && perm_bond_diff) {
1143 found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1146 /* locking: needs RTNL and nothing else */
1147 alb_swap_mac_addr(slave, found_slave);
1148 alb_fasten_mac_swap(bond, slave, found_slave);
1154 * alb_handle_addr_collision_on_attach
1155 * @bond: bonding we're working on
1156 * @slave: the slave that was just attached
1158 * checks uniqueness of slave's mac address and handles the case the
1159 * new slave uses the bonds mac address.
1161 * If the permanent hw address of @slave is @bond's hw address, we need to
1162 * find a different hw address to give @slave, that isn't in use by any other
1163 * slave in the bond. This address must be, of course, one of the permanent
1164 * addresses of the other slaves.
1166 * We go over the slave list, and for each slave there we compare its
1167 * permanent hw address with the current address of all the other slaves.
1168 * If no match was found, then we've found a slave with a permanent address
1169 * that isn't used by any other slave in the bond, so we can assign it to
1172 * assumption: this function is called before @slave is attached to the
1175 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1177 struct slave *tmp_slave1, *free_mac_slave = NULL;
1178 struct slave *has_bond_addr = bond->curr_active_slave;
1180 if (list_empty(&bond->slave_list)) {
1181 /* this is the first slave */
1185 /* if slave's mac address differs from bond's mac address
1186 * check uniqueness of slave's mac address against the other
1187 * slaves in the bond.
1189 if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1190 if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1193 /* Try setting slave mac to bond address and fall-through
1194 to code handling that situation below... */
1195 alb_set_slave_mac_addr(slave, bond->dev->dev_addr);
1198 /* The slave's address is equal to the address of the bond.
1199 * Search for a spare address in the bond for this slave.
1201 bond_for_each_slave(bond, tmp_slave1) {
1202 if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1203 /* no slave has tmp_slave1's perm addr
1206 free_mac_slave = tmp_slave1;
1210 if (!has_bond_addr) {
1211 if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1212 bond->dev->dev_addr)) {
1214 has_bond_addr = tmp_slave1;
1219 if (free_mac_slave) {
1220 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr);
1222 pr_warning("%s: Warning: the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1223 bond->dev->name, slave->dev->name,
1224 free_mac_slave->dev->name);
1226 } else if (has_bond_addr) {
1227 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",
1228 bond->dev->name, slave->dev->name);
1236 * alb_set_mac_address
1240 * In TLB mode all slaves are configured to the bond's hw address, but set
1241 * their dev_addr field to different addresses (based on their permanent hw
1244 * For each slave, this function sets the interface to the new address and then
1245 * changes its dev_addr field to its previous value.
1247 * Unwinding assumes bond's mac address has not yet changed.
1249 static int alb_set_mac_address(struct bonding *bond, void *addr)
1251 char tmp_addr[ETH_ALEN];
1252 struct slave *slave;
1256 if (bond->alb_info.rlb_enabled)
1259 bond_for_each_slave(bond, slave) {
1260 /* save net_device's current hw address */
1261 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1263 res = dev_set_mac_address(slave->dev, addr);
1265 /* restore net_device's hw address */
1266 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1275 memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1276 sa.sa_family = bond->dev->type;
1278 /* unwind from head to the slave that failed */
1279 bond_for_each_slave_continue_reverse(bond, slave) {
1280 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1281 dev_set_mac_address(slave->dev, &sa);
1282 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1288 /************************ exported alb funcions ************************/
1290 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1294 res = tlb_initialize(bond);
1300 bond->alb_info.rlb_enabled = 1;
1301 /* initialize rlb */
1302 res = rlb_initialize(bond);
1304 tlb_deinitialize(bond);
1308 bond->alb_info.rlb_enabled = 0;
1314 void bond_alb_deinitialize(struct bonding *bond)
1316 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1318 tlb_deinitialize(bond);
1320 if (bond_info->rlb_enabled) {
1321 rlb_deinitialize(bond);
1325 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1327 struct bonding *bond = netdev_priv(bond_dev);
1328 struct ethhdr *eth_data;
1329 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1330 struct slave *tx_slave = NULL;
1331 static const __be32 ip_bcast = htonl(0xffffffff);
1333 int do_tx_balance = 1;
1335 const u8 *hash_start = NULL;
1337 struct ipv6hdr *ip6hdr;
1339 skb_reset_mac_header(skb);
1340 eth_data = eth_hdr(skb);
1342 /* make sure that the curr_active_slave do not change during tx
1344 read_lock(&bond->lock);
1345 read_lock(&bond->curr_slave_lock);
1347 switch (ntohs(skb->protocol)) {
1349 const struct iphdr *iph = ip_hdr(skb);
1351 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast) ||
1352 (iph->daddr == ip_bcast) ||
1353 (iph->protocol == IPPROTO_IGMP)) {
1357 hash_start = (char *)&(iph->daddr);
1358 hash_size = sizeof(iph->daddr);
1362 /* IPv6 doesn't really use broadcast mac address, but leave
1363 * that here just in case.
1365 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast)) {
1370 /* IPv6 uses all-nodes multicast as an equivalent to
1371 * broadcasts in IPv4.
1373 if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1378 /* Additianally, DAD probes should not be tx-balanced as that
1379 * will lead to false positives for duplicate addresses and
1380 * prevent address configuration from working.
1382 ip6hdr = ipv6_hdr(skb);
1383 if (ipv6_addr_any(&ip6hdr->saddr)) {
1388 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1389 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1392 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1393 /* something is wrong with this packet */
1398 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1399 /* The only protocol worth balancing in
1400 * this family since it has an "ARP" like
1407 hash_start = (char*)eth_data->h_dest;
1408 hash_size = ETH_ALEN;
1412 if (bond_info->rlb_enabled) {
1413 tx_slave = rlb_arp_xmit(skb, bond);
1421 if (do_tx_balance) {
1422 hash_index = _simple_hash(hash_start, hash_size);
1423 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1427 /* unbalanced or unassigned, send through primary */
1428 tx_slave = bond->curr_active_slave;
1429 bond_info->unbalanced_load += skb->len;
1432 if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1433 if (tx_slave != bond->curr_active_slave) {
1434 memcpy(eth_data->h_source,
1435 tx_slave->dev->dev_addr,
1439 res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1442 _lock_tx_hashtbl(bond);
1443 __tlb_clear_slave(bond, tx_slave, 0);
1444 _unlock_tx_hashtbl(bond);
1448 read_unlock(&bond->curr_slave_lock);
1449 read_unlock(&bond->lock);
1451 /* no suitable interface, frame not sent */
1455 return NETDEV_TX_OK;
1458 void bond_alb_monitor(struct work_struct *work)
1460 struct bonding *bond = container_of(work, struct bonding,
1462 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1463 struct slave *slave;
1465 read_lock(&bond->lock);
1467 if (list_empty(&bond->slave_list)) {
1468 bond_info->tx_rebalance_counter = 0;
1469 bond_info->lp_counter = 0;
1473 bond_info->tx_rebalance_counter++;
1474 bond_info->lp_counter++;
1476 /* send learning packets */
1477 if (bond_info->lp_counter >= BOND_ALB_LP_TICKS) {
1478 /* change of curr_active_slave involves swapping of mac addresses.
1479 * in order to avoid this swapping from happening while
1480 * sending the learning packets, the curr_slave_lock must be held for
1483 read_lock(&bond->curr_slave_lock);
1485 bond_for_each_slave(bond, slave)
1486 alb_send_learning_packets(slave, slave->dev->dev_addr);
1488 read_unlock(&bond->curr_slave_lock);
1490 bond_info->lp_counter = 0;
1493 /* rebalance tx traffic */
1494 if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1496 read_lock(&bond->curr_slave_lock);
1498 bond_for_each_slave(bond, slave) {
1499 tlb_clear_slave(bond, slave, 1);
1500 if (slave == bond->curr_active_slave) {
1501 SLAVE_TLB_INFO(slave).load =
1502 bond_info->unbalanced_load /
1503 BOND_TLB_REBALANCE_INTERVAL;
1504 bond_info->unbalanced_load = 0;
1508 read_unlock(&bond->curr_slave_lock);
1510 bond_info->tx_rebalance_counter = 0;
1513 /* handle rlb stuff */
1514 if (bond_info->rlb_enabled) {
1515 if (bond_info->primary_is_promisc &&
1516 (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1519 * dev_set_promiscuity requires rtnl and
1520 * nothing else. Avoid race with bond_close.
1522 read_unlock(&bond->lock);
1523 if (!rtnl_trylock()) {
1524 read_lock(&bond->lock);
1528 bond_info->rlb_promisc_timeout_counter = 0;
1530 /* If the primary was set to promiscuous mode
1531 * because a slave was disabled then
1532 * it can now leave promiscuous mode.
1534 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1535 bond_info->primary_is_promisc = 0;
1538 read_lock(&bond->lock);
1541 if (bond_info->rlb_rebalance) {
1542 bond_info->rlb_rebalance = 0;
1543 rlb_rebalance(bond);
1546 /* check if clients need updating */
1547 if (bond_info->rx_ntt) {
1548 if (bond_info->rlb_update_delay_counter) {
1549 --bond_info->rlb_update_delay_counter;
1551 rlb_update_rx_clients(bond);
1552 if (bond_info->rlb_update_retry_counter) {
1553 --bond_info->rlb_update_retry_counter;
1555 bond_info->rx_ntt = 0;
1562 queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1564 read_unlock(&bond->lock);
1567 /* assumption: called before the slave is attached to the bond
1568 * and not locked by the bond lock
1570 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1574 res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr);
1579 res = alb_handle_addr_collision_on_attach(bond, slave);
1584 tlb_init_slave(slave);
1586 /* order a rebalance ASAP */
1587 bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1589 if (bond->alb_info.rlb_enabled) {
1590 bond->alb_info.rlb_rebalance = 1;
1597 * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1600 * Caller must hold RTNL and no other locks
1602 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1604 if (!list_empty(&bond->slave_list))
1605 alb_change_hw_addr_on_detach(bond, slave);
1607 tlb_clear_slave(bond, slave, 0);
1609 if (bond->alb_info.rlb_enabled) {
1610 bond->alb_info.next_rx_slave = NULL;
1611 rlb_clear_slave(bond, slave);
1615 /* Caller must hold bond lock for read */
1616 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1618 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1620 if (link == BOND_LINK_DOWN) {
1621 tlb_clear_slave(bond, slave, 0);
1622 if (bond->alb_info.rlb_enabled) {
1623 rlb_clear_slave(bond, slave);
1625 } else if (link == BOND_LINK_UP) {
1626 /* order a rebalance ASAP */
1627 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1628 if (bond->alb_info.rlb_enabled) {
1629 bond->alb_info.rlb_rebalance = 1;
1630 /* If the updelay module parameter is smaller than the
1631 * forwarding delay of the switch the rebalance will
1632 * not work because the rebalance arp replies will
1633 * not be forwarded to the clients..
1640 * bond_alb_handle_active_change - assign new curr_active_slave
1641 * @bond: our bonding struct
1642 * @new_slave: new slave to assign
1644 * Set the bond->curr_active_slave to @new_slave and handle
1645 * mac address swapping and promiscuity changes as needed.
1647 * If new_slave is NULL, caller must hold curr_slave_lock or
1648 * bond->lock for write.
1650 * If new_slave is not NULL, caller must hold RTNL, bond->lock for
1651 * read and curr_slave_lock for write. Processing here may sleep, so
1652 * no other locks may be held.
1654 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1655 __releases(&bond->curr_slave_lock)
1656 __releases(&bond->lock)
1657 __acquires(&bond->lock)
1658 __acquires(&bond->curr_slave_lock)
1660 struct slave *swap_slave;
1662 if (bond->curr_active_slave == new_slave)
1665 if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1666 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1667 bond->alb_info.primary_is_promisc = 0;
1668 bond->alb_info.rlb_promisc_timeout_counter = 0;
1671 swap_slave = bond->curr_active_slave;
1672 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1674 if (!new_slave || list_empty(&bond->slave_list))
1677 /* set the new curr_active_slave to the bonds mac address
1678 * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1681 swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1684 * Arrange for swap_slave and new_slave to temporarily be
1685 * ignored so we can mess with their MAC addresses without
1686 * fear of interference from transmit activity.
1689 tlb_clear_slave(bond, swap_slave, 1);
1690 tlb_clear_slave(bond, new_slave, 1);
1692 write_unlock_bh(&bond->curr_slave_lock);
1693 read_unlock(&bond->lock);
1697 /* curr_active_slave must be set before calling alb_swap_mac_addr */
1699 /* swap mac address */
1700 alb_swap_mac_addr(swap_slave, new_slave);
1701 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1702 read_lock(&bond->lock);
1704 /* set the new_slave to the bond mac address */
1705 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr);
1706 read_lock(&bond->lock);
1707 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1710 write_lock_bh(&bond->curr_slave_lock);
1716 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1717 __acquires(&bond->lock)
1718 __releases(&bond->lock)
1720 struct bonding *bond = netdev_priv(bond_dev);
1721 struct sockaddr *sa = addr;
1722 struct slave *swap_slave;
1725 if (!is_valid_ether_addr(sa->sa_data)) {
1726 return -EADDRNOTAVAIL;
1729 res = alb_set_mac_address(bond, addr);
1734 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1736 /* If there is no curr_active_slave there is nothing else to do.
1737 * Otherwise we'll need to pass the new address to it and handle
1740 if (!bond->curr_active_slave) {
1744 swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1747 alb_swap_mac_addr(swap_slave, bond->curr_active_slave);
1748 alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave);
1750 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr);
1752 read_lock(&bond->lock);
1753 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1754 if (bond->alb_info.rlb_enabled) {
1755 /* inform clients mac address has changed */
1756 rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1758 read_unlock(&bond->lock);
1764 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1766 if (bond->alb_info.rlb_enabled) {
1767 rlb_clear_vlan(bond, vlan_id);