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[]);
88 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
93 for (i = 0; i < hash_size; i++) {
94 hash ^= hash_start[i];
100 /*********************** tlb specific functions ***************************/
102 static inline void _lock_tx_hashtbl_bh(struct bonding *bond)
104 spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
107 static inline void _unlock_tx_hashtbl_bh(struct bonding *bond)
109 spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
112 static inline void _lock_tx_hashtbl(struct bonding *bond)
114 spin_lock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
117 static inline void _unlock_tx_hashtbl(struct bonding *bond)
119 spin_unlock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
122 /* Caller must hold tx_hashtbl lock */
123 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
126 entry->load_history = 1 + entry->tx_bytes /
127 BOND_TLB_REBALANCE_INTERVAL;
131 entry->tx_slave = NULL;
132 entry->next = TLB_NULL_INDEX;
133 entry->prev = TLB_NULL_INDEX;
136 static inline void tlb_init_slave(struct slave *slave)
138 SLAVE_TLB_INFO(slave).load = 0;
139 SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
142 /* Caller must hold bond lock for read, BH disabled */
143 static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
146 struct tlb_client_info *tx_hash_table;
149 /* clear slave from tx_hashtbl */
150 tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
152 /* skip this if we've already freed the tx hash table */
154 index = SLAVE_TLB_INFO(slave).head;
155 while (index != TLB_NULL_INDEX) {
156 u32 next_index = tx_hash_table[index].next;
157 tlb_init_table_entry(&tx_hash_table[index], save_load);
162 tlb_init_slave(slave);
165 /* Caller must hold bond lock for read */
166 static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
169 _lock_tx_hashtbl_bh(bond);
170 __tlb_clear_slave(bond, slave, save_load);
171 _unlock_tx_hashtbl_bh(bond);
174 /* Must be called before starting the monitor timer */
175 static int tlb_initialize(struct bonding *bond)
177 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
178 int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
179 struct tlb_client_info *new_hashtbl;
182 new_hashtbl = kzalloc(size, GFP_KERNEL);
186 _lock_tx_hashtbl_bh(bond);
188 bond_info->tx_hashtbl = new_hashtbl;
190 for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
191 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
194 _unlock_tx_hashtbl_bh(bond);
199 /* Must be called only after all slaves have been released */
200 static void tlb_deinitialize(struct bonding *bond)
202 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
204 _lock_tx_hashtbl_bh(bond);
206 kfree(bond_info->tx_hashtbl);
207 bond_info->tx_hashtbl = NULL;
209 _unlock_tx_hashtbl_bh(bond);
212 static long long compute_gap(struct slave *slave)
214 return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
215 (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
218 /* Caller must hold bond lock for read */
219 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
221 struct slave *slave, *least_loaded;
228 /* Find the slave with the largest gap */
229 bond_for_each_slave(bond, slave, i) {
230 if (SLAVE_IS_OK(slave)) {
231 long long gap = compute_gap(slave);
234 least_loaded = slave;
243 static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
246 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
247 struct tlb_client_info *hash_table;
248 struct slave *assigned_slave;
250 hash_table = bond_info->tx_hashtbl;
251 assigned_slave = hash_table[hash_index].tx_slave;
252 if (!assigned_slave) {
253 assigned_slave = tlb_get_least_loaded_slave(bond);
255 if (assigned_slave) {
256 struct tlb_slave_info *slave_info =
257 &(SLAVE_TLB_INFO(assigned_slave));
258 u32 next_index = slave_info->head;
260 hash_table[hash_index].tx_slave = assigned_slave;
261 hash_table[hash_index].next = next_index;
262 hash_table[hash_index].prev = TLB_NULL_INDEX;
264 if (next_index != TLB_NULL_INDEX) {
265 hash_table[next_index].prev = hash_index;
268 slave_info->head = hash_index;
270 hash_table[hash_index].load_history;
274 if (assigned_slave) {
275 hash_table[hash_index].tx_bytes += skb_len;
278 return assigned_slave;
281 /* Caller must hold bond lock for read */
282 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
285 struct slave *tx_slave;
287 * We don't need to disable softirq here, becase
288 * tlb_choose_channel() is only called by bond_alb_xmit()
289 * which already has softirq disabled.
291 _lock_tx_hashtbl(bond);
292 tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
293 _unlock_tx_hashtbl(bond);
297 /*********************** rlb specific functions ***************************/
298 static inline void _lock_rx_hashtbl_bh(struct bonding *bond)
300 spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
303 static inline void _unlock_rx_hashtbl_bh(struct bonding *bond)
305 spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
308 static inline void _lock_rx_hashtbl(struct bonding *bond)
310 spin_lock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
313 static inline void _unlock_rx_hashtbl(struct bonding *bond)
315 spin_unlock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
318 /* when an ARP REPLY is received from a client update its info
321 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
323 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
324 struct rlb_client_info *client_info;
327 _lock_rx_hashtbl_bh(bond);
329 hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
330 client_info = &(bond_info->rx_hashtbl[hash_index]);
332 if ((client_info->assigned) &&
333 (client_info->ip_src == arp->ip_dst) &&
334 (client_info->ip_dst == arp->ip_src) &&
335 (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) {
336 /* update the clients MAC address */
337 memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
338 client_info->ntt = 1;
339 bond_info->rx_ntt = 1;
342 _unlock_rx_hashtbl_bh(bond);
345 static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
348 struct arp_pkt *arp, _arp;
350 if (skb->protocol != cpu_to_be16(ETH_P_ARP))
353 arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp);
357 if (arp->op_code == htons(ARPOP_REPLY)) {
358 /* update rx hash table for this ARP */
359 rlb_update_entry_from_arp(bond, arp);
360 pr_debug("Server received an ARP Reply from client\n");
363 return RX_HANDLER_ANOTHER;
366 /* Caller must hold bond lock for read */
367 static struct slave *rlb_next_rx_slave(struct bonding *bond)
369 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
370 struct slave *rx_slave, *slave, *start_at;
373 if (bond_info->next_rx_slave) {
374 start_at = bond_info->next_rx_slave;
376 start_at = bond->first_slave;
381 bond_for_each_slave_from(bond, slave, i, start_at) {
382 if (SLAVE_IS_OK(slave)) {
385 } else if (slave->speed > rx_slave->speed) {
392 bond_info->next_rx_slave = rx_slave->next;
398 /* teach the switch the mac of a disabled slave
399 * on the primary for fault tolerance
401 * Caller must hold bond->curr_slave_lock for write or bond lock for write
403 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
405 if (!bond->curr_active_slave) {
409 if (!bond->alb_info.primary_is_promisc) {
410 if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1))
411 bond->alb_info.primary_is_promisc = 1;
413 bond->alb_info.primary_is_promisc = 0;
416 bond->alb_info.rlb_promisc_timeout_counter = 0;
418 alb_send_learning_packets(bond->curr_active_slave, addr);
421 /* slave being removed should not be active at this point
423 * Caller must hold bond lock for read
425 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
427 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
428 struct rlb_client_info *rx_hash_table;
429 u32 index, next_index;
431 /* clear slave from rx_hashtbl */
432 _lock_rx_hashtbl_bh(bond);
434 rx_hash_table = bond_info->rx_hashtbl;
435 index = bond_info->rx_hashtbl_head;
436 for (; index != RLB_NULL_INDEX; index = next_index) {
437 next_index = rx_hash_table[index].next;
438 if (rx_hash_table[index].slave == slave) {
439 struct slave *assigned_slave = rlb_next_rx_slave(bond);
441 if (assigned_slave) {
442 rx_hash_table[index].slave = assigned_slave;
443 if (!ether_addr_equal_64bits(rx_hash_table[index].mac_dst,
445 bond_info->rx_hashtbl[index].ntt = 1;
446 bond_info->rx_ntt = 1;
447 /* A slave has been removed from the
448 * table because it is either disabled
449 * or being released. We must retry the
450 * update to avoid clients from not
451 * being updated & disconnecting when
454 bond_info->rlb_update_retry_counter =
457 } else { /* there is no active slave */
458 rx_hash_table[index].slave = NULL;
463 _unlock_rx_hashtbl_bh(bond);
465 write_lock_bh(&bond->curr_slave_lock);
467 if (slave != bond->curr_active_slave) {
468 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
471 write_unlock_bh(&bond->curr_slave_lock);
474 static void rlb_update_client(struct rlb_client_info *client_info)
478 if (!client_info->slave) {
482 for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
485 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
487 client_info->slave->dev,
489 client_info->mac_dst,
490 client_info->slave->dev->dev_addr,
491 client_info->mac_dst);
493 pr_err("%s: Error: failed to create an ARP packet\n",
494 client_info->slave->dev->master->name);
498 skb->dev = client_info->slave->dev;
500 if (client_info->tag) {
501 skb = vlan_put_tag(skb, client_info->vlan_id);
503 pr_err("%s: Error: failed to insert VLAN tag\n",
504 client_info->slave->dev->master->name);
513 /* sends ARP REPLIES that update the clients that need updating */
514 static void rlb_update_rx_clients(struct bonding *bond)
516 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
517 struct rlb_client_info *client_info;
520 _lock_rx_hashtbl_bh(bond);
522 hash_index = bond_info->rx_hashtbl_head;
523 for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
524 client_info = &(bond_info->rx_hashtbl[hash_index]);
525 if (client_info->ntt) {
526 rlb_update_client(client_info);
527 if (bond_info->rlb_update_retry_counter == 0) {
528 client_info->ntt = 0;
533 /* do not update the entries again until this counter is zero so that
534 * not to confuse the clients.
536 bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
538 _unlock_rx_hashtbl_bh(bond);
541 /* The slave was assigned a new mac address - update the clients */
542 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
544 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
545 struct rlb_client_info *client_info;
549 _lock_rx_hashtbl_bh(bond);
551 hash_index = bond_info->rx_hashtbl_head;
552 for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
553 client_info = &(bond_info->rx_hashtbl[hash_index]);
555 if ((client_info->slave == slave) &&
556 !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
557 client_info->ntt = 1;
562 // update the team's flag only after the whole iteration
564 bond_info->rx_ntt = 1;
566 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
569 _unlock_rx_hashtbl_bh(bond);
572 /* mark all clients using src_ip to be updated */
573 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
575 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
576 struct rlb_client_info *client_info;
579 _lock_rx_hashtbl(bond);
581 hash_index = bond_info->rx_hashtbl_head;
582 for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
583 client_info = &(bond_info->rx_hashtbl[hash_index]);
585 if (!client_info->slave) {
586 pr_err("%s: Error: found a client with no channel in the client's hash table\n",
590 /*update all clients using this src_ip, that are not assigned
591 * to the team's address (curr_active_slave) and have a known
592 * unicast mac address.
594 if ((client_info->ip_src == src_ip) &&
595 !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
596 bond->dev->dev_addr) &&
597 !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
598 client_info->ntt = 1;
599 bond_info->rx_ntt = 1;
603 _unlock_rx_hashtbl(bond);
606 /* Caller must hold both bond and ptr locks for read */
607 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
609 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
610 struct arp_pkt *arp = arp_pkt(skb);
611 struct slave *assigned_slave;
612 struct rlb_client_info *client_info;
615 _lock_rx_hashtbl(bond);
617 hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
618 client_info = &(bond_info->rx_hashtbl[hash_index]);
620 if (client_info->assigned) {
621 if ((client_info->ip_src == arp->ip_src) &&
622 (client_info->ip_dst == arp->ip_dst)) {
623 /* the entry is already assigned to this client */
624 if (!ether_addr_equal_64bits(arp->mac_dst, mac_bcast)) {
625 /* update mac address from arp */
626 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
629 assigned_slave = client_info->slave;
630 if (assigned_slave) {
631 _unlock_rx_hashtbl(bond);
632 return assigned_slave;
635 /* the entry is already assigned to some other client,
636 * move the old client to primary (curr_active_slave) so
637 * that the new client can be assigned to this entry.
639 if (bond->curr_active_slave &&
640 client_info->slave != bond->curr_active_slave) {
641 client_info->slave = bond->curr_active_slave;
642 rlb_update_client(client_info);
646 /* assign a new slave */
647 assigned_slave = rlb_next_rx_slave(bond);
649 if (assigned_slave) {
650 client_info->ip_src = arp->ip_src;
651 client_info->ip_dst = arp->ip_dst;
652 /* arp->mac_dst is broadcast for arp reqeusts.
653 * will be updated with clients actual unicast mac address
654 * upon receiving an arp reply.
656 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
657 client_info->slave = assigned_slave;
659 if (!ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
660 client_info->ntt = 1;
661 bond->alb_info.rx_ntt = 1;
663 client_info->ntt = 0;
666 if (bond_vlan_used(bond)) {
667 if (!vlan_get_tag(skb, &client_info->vlan_id))
668 client_info->tag = 1;
671 if (!client_info->assigned) {
672 u32 prev_tbl_head = bond_info->rx_hashtbl_head;
673 bond_info->rx_hashtbl_head = hash_index;
674 client_info->next = prev_tbl_head;
675 if (prev_tbl_head != RLB_NULL_INDEX) {
676 bond_info->rx_hashtbl[prev_tbl_head].prev =
679 client_info->assigned = 1;
683 _unlock_rx_hashtbl(bond);
685 return assigned_slave;
688 /* chooses (and returns) transmit channel for arp reply
689 * does not choose channel for other arp types since they are
690 * sent on the curr_active_slave
692 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
694 struct arp_pkt *arp = arp_pkt(skb);
695 struct slave *tx_slave = NULL;
697 if (arp->op_code == htons(ARPOP_REPLY)) {
698 /* the arp must be sent on the selected
701 tx_slave = rlb_choose_channel(skb, bond);
703 memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
705 pr_debug("Server sent ARP Reply packet\n");
706 } else if (arp->op_code == htons(ARPOP_REQUEST)) {
707 /* Create an entry in the rx_hashtbl for this client as a
709 * When the arp reply is received the entry will be updated
710 * with the correct unicast address of the client.
712 rlb_choose_channel(skb, bond);
714 /* The ARP reply packets must be delayed so that
715 * they can cancel out the influence of the ARP request.
717 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
719 /* arp requests are broadcast and are sent on the primary
720 * the arp request will collapse all clients on the subnet to
721 * the primary slave. We must register these clients to be
722 * updated with their assigned mac.
724 rlb_req_update_subnet_clients(bond, arp->ip_src);
725 pr_debug("Server sent ARP Request packet\n");
731 /* Caller must hold bond lock for read */
732 static void rlb_rebalance(struct bonding *bond)
734 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
735 struct slave *assigned_slave;
736 struct rlb_client_info *client_info;
740 _lock_rx_hashtbl_bh(bond);
743 hash_index = bond_info->rx_hashtbl_head;
744 for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
745 client_info = &(bond_info->rx_hashtbl[hash_index]);
746 assigned_slave = rlb_next_rx_slave(bond);
747 if (assigned_slave && (client_info->slave != assigned_slave)) {
748 client_info->slave = assigned_slave;
749 client_info->ntt = 1;
754 /* update the team's flag only after the whole iteration */
756 bond_info->rx_ntt = 1;
758 _unlock_rx_hashtbl_bh(bond);
761 /* Caller must hold rx_hashtbl lock */
762 static void rlb_init_table_entry(struct rlb_client_info *entry)
764 memset(entry, 0, sizeof(struct rlb_client_info));
765 entry->next = RLB_NULL_INDEX;
766 entry->prev = RLB_NULL_INDEX;
769 static int rlb_initialize(struct bonding *bond)
771 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
772 struct rlb_client_info *new_hashtbl;
773 int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
776 new_hashtbl = kmalloc(size, GFP_KERNEL);
780 _lock_rx_hashtbl_bh(bond);
782 bond_info->rx_hashtbl = new_hashtbl;
784 bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
786 for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
787 rlb_init_table_entry(bond_info->rx_hashtbl + i);
790 _unlock_rx_hashtbl_bh(bond);
792 /* register to receive ARPs */
793 bond->recv_probe = rlb_arp_recv;
798 static void rlb_deinitialize(struct bonding *bond)
800 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
802 _lock_rx_hashtbl_bh(bond);
804 kfree(bond_info->rx_hashtbl);
805 bond_info->rx_hashtbl = NULL;
806 bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
808 _unlock_rx_hashtbl_bh(bond);
811 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
813 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
816 _lock_rx_hashtbl_bh(bond);
818 curr_index = bond_info->rx_hashtbl_head;
819 while (curr_index != RLB_NULL_INDEX) {
820 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
821 u32 next_index = bond_info->rx_hashtbl[curr_index].next;
822 u32 prev_index = bond_info->rx_hashtbl[curr_index].prev;
824 if (curr->tag && (curr->vlan_id == vlan_id)) {
825 if (curr_index == bond_info->rx_hashtbl_head) {
826 bond_info->rx_hashtbl_head = next_index;
828 if (prev_index != RLB_NULL_INDEX) {
829 bond_info->rx_hashtbl[prev_index].next = next_index;
831 if (next_index != RLB_NULL_INDEX) {
832 bond_info->rx_hashtbl[next_index].prev = prev_index;
835 rlb_init_table_entry(curr);
838 curr_index = next_index;
841 _unlock_rx_hashtbl_bh(bond);
844 /*********************** tlb/rlb shared functions *********************/
846 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
848 struct bonding *bond = bond_get_bond_by_slave(slave);
849 struct learning_pkt pkt;
850 int size = sizeof(struct learning_pkt);
853 memset(&pkt, 0, size);
854 memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
855 memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
856 pkt.type = cpu_to_be16(ETH_P_LOOP);
858 for (i = 0; i < MAX_LP_BURST; i++) {
862 skb = dev_alloc_skb(size);
867 data = skb_put(skb, size);
868 memcpy(data, &pkt, size);
870 skb_reset_mac_header(skb);
871 skb->network_header = skb->mac_header + ETH_HLEN;
872 skb->protocol = pkt.type;
873 skb->priority = TC_PRIO_CONTROL;
874 skb->dev = slave->dev;
876 if (bond_vlan_used(bond)) {
877 struct vlan_entry *vlan;
879 vlan = bond_next_vlan(bond,
880 bond->alb_info.current_alb_vlan);
882 bond->alb_info.current_alb_vlan = vlan;
888 skb = vlan_put_tag(skb, vlan->vlan_id);
890 pr_err("%s: Error: failed to insert VLAN tag\n",
900 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[])
902 struct net_device *dev = slave->dev;
903 struct sockaddr s_addr;
905 if (slave->bond->params.mode == BOND_MODE_TLB) {
906 memcpy(dev->dev_addr, addr, dev->addr_len);
910 /* for rlb each slave must have a unique hw mac addresses so that */
911 /* each slave will receive packets destined to a different mac */
912 memcpy(s_addr.sa_data, addr, dev->addr_len);
913 s_addr.sa_family = dev->type;
914 if (dev_set_mac_address(dev, &s_addr)) {
915 pr_err("%s: Error: dev_set_mac_address of dev %s failed!\n"
916 "ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
917 dev->master->name, dev->name);
924 * Swap MAC addresses between two slaves.
926 * Called with RTNL held, and no other locks.
930 static void alb_swap_mac_addr(struct bonding *bond, struct slave *slave1, struct slave *slave2)
932 u8 tmp_mac_addr[ETH_ALEN];
934 memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
935 alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr);
936 alb_set_slave_mac_addr(slave2, tmp_mac_addr);
941 * Send learning packets after MAC address swap.
943 * Called with RTNL and no other locks
945 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
946 struct slave *slave2)
948 int slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
949 struct slave *disabled_slave = NULL;
953 /* fasten the change in the switch */
954 if (SLAVE_IS_OK(slave1)) {
955 alb_send_learning_packets(slave1, slave1->dev->dev_addr);
956 if (bond->alb_info.rlb_enabled) {
957 /* inform the clients that the mac address
960 rlb_req_update_slave_clients(bond, slave1);
963 disabled_slave = slave1;
966 if (SLAVE_IS_OK(slave2)) {
967 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
968 if (bond->alb_info.rlb_enabled) {
969 /* inform the clients that the mac address
972 rlb_req_update_slave_clients(bond, slave2);
975 disabled_slave = slave2;
978 if (bond->alb_info.rlb_enabled && slaves_state_differ) {
979 /* A disabled slave was assigned an active mac addr */
980 rlb_teach_disabled_mac_on_primary(bond,
981 disabled_slave->dev->dev_addr);
986 * alb_change_hw_addr_on_detach
987 * @bond: bonding we're working on
988 * @slave: the slave that was just detached
990 * We assume that @slave was already detached from the slave list.
992 * If @slave's permanent hw address is different both from its current
993 * address and from @bond's address, then somewhere in the bond there's
994 * a slave that has @slave's permanet address as its current address.
995 * We'll make sure that that slave no longer uses @slave's permanent address.
997 * Caller must hold RTNL and no other locks
999 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1004 perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1005 slave->dev->dev_addr);
1006 perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1007 bond->dev->dev_addr);
1009 if (perm_curr_diff && perm_bond_diff) {
1010 struct slave *tmp_slave;
1013 bond_for_each_slave(bond, tmp_slave, i) {
1014 if (ether_addr_equal_64bits(slave->perm_hwaddr,
1015 tmp_slave->dev->dev_addr)) {
1022 /* locking: needs RTNL and nothing else */
1023 alb_swap_mac_addr(bond, slave, tmp_slave);
1024 alb_fasten_mac_swap(bond, slave, tmp_slave);
1030 * alb_handle_addr_collision_on_attach
1031 * @bond: bonding we're working on
1032 * @slave: the slave that was just attached
1034 * checks uniqueness of slave's mac address and handles the case the
1035 * new slave uses the bonds mac address.
1037 * If the permanent hw address of @slave is @bond's hw address, we need to
1038 * find a different hw address to give @slave, that isn't in use by any other
1039 * slave in the bond. This address must be, of course, one of the permanent
1040 * addresses of the other slaves.
1042 * We go over the slave list, and for each slave there we compare its
1043 * permanent hw address with the current address of all the other slaves.
1044 * If no match was found, then we've found a slave with a permanent address
1045 * that isn't used by any other slave in the bond, so we can assign it to
1048 * assumption: this function is called before @slave is attached to the
1051 * caller must hold the bond lock for write since the mac addresses are compared
1052 * and may be swapped.
1054 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1056 struct slave *tmp_slave1, *tmp_slave2, *free_mac_slave;
1057 struct slave *has_bond_addr = bond->curr_active_slave;
1058 int i, j, found = 0;
1060 if (bond->slave_cnt == 0) {
1061 /* this is the first slave */
1065 /* if slave's mac address differs from bond's mac address
1066 * check uniqueness of slave's mac address against the other
1067 * slaves in the bond.
1069 if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1070 bond_for_each_slave(bond, tmp_slave1, i) {
1071 if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1072 slave->dev->dev_addr)) {
1081 /* Try setting slave mac to bond address and fall-through
1082 to code handling that situation below... */
1083 alb_set_slave_mac_addr(slave, bond->dev->dev_addr);
1086 /* The slave's address is equal to the address of the bond.
1087 * Search for a spare address in the bond for this slave.
1089 free_mac_slave = NULL;
1091 bond_for_each_slave(bond, tmp_slave1, i) {
1093 bond_for_each_slave(bond, tmp_slave2, j) {
1094 if (ether_addr_equal_64bits(tmp_slave1->perm_hwaddr,
1095 tmp_slave2->dev->dev_addr)) {
1102 /* no slave has tmp_slave1's perm addr
1105 free_mac_slave = tmp_slave1;
1109 if (!has_bond_addr) {
1110 if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1111 bond->dev->dev_addr)) {
1113 has_bond_addr = tmp_slave1;
1118 if (free_mac_slave) {
1119 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr);
1121 pr_warning("%s: Warning: the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1122 bond->dev->name, slave->dev->name,
1123 free_mac_slave->dev->name);
1125 } else if (has_bond_addr) {
1126 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",
1127 bond->dev->name, slave->dev->name);
1135 * alb_set_mac_address
1139 * In TLB mode all slaves are configured to the bond's hw address, but set
1140 * their dev_addr field to different addresses (based on their permanent hw
1143 * For each slave, this function sets the interface to the new address and then
1144 * changes its dev_addr field to its previous value.
1146 * Unwinding assumes bond's mac address has not yet changed.
1148 static int alb_set_mac_address(struct bonding *bond, void *addr)
1151 struct slave *slave, *stop_at;
1152 char tmp_addr[ETH_ALEN];
1156 if (bond->alb_info.rlb_enabled) {
1160 bond_for_each_slave(bond, slave, i) {
1161 /* save net_device's current hw address */
1162 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1164 res = dev_set_mac_address(slave->dev, addr);
1166 /* restore net_device's hw address */
1167 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1176 memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1177 sa.sa_family = bond->dev->type;
1179 /* unwind from head to the slave that failed */
1181 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
1182 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1183 dev_set_mac_address(slave->dev, &sa);
1184 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1190 /************************ exported alb funcions ************************/
1192 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1196 res = tlb_initialize(bond);
1202 bond->alb_info.rlb_enabled = 1;
1203 /* initialize rlb */
1204 res = rlb_initialize(bond);
1206 tlb_deinitialize(bond);
1210 bond->alb_info.rlb_enabled = 0;
1216 void bond_alb_deinitialize(struct bonding *bond)
1218 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1220 tlb_deinitialize(bond);
1222 if (bond_info->rlb_enabled) {
1223 rlb_deinitialize(bond);
1227 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1229 struct bonding *bond = netdev_priv(bond_dev);
1230 struct ethhdr *eth_data;
1231 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1232 struct slave *tx_slave = NULL;
1233 static const __be32 ip_bcast = htonl(0xffffffff);
1235 int do_tx_balance = 1;
1237 const u8 *hash_start = NULL;
1239 struct ipv6hdr *ip6hdr;
1241 skb_reset_mac_header(skb);
1242 eth_data = eth_hdr(skb);
1244 /* make sure that the curr_active_slave do not change during tx
1246 read_lock(&bond->curr_slave_lock);
1248 switch (ntohs(skb->protocol)) {
1250 const struct iphdr *iph = ip_hdr(skb);
1252 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast) ||
1253 (iph->daddr == ip_bcast) ||
1254 (iph->protocol == IPPROTO_IGMP)) {
1258 hash_start = (char *)&(iph->daddr);
1259 hash_size = sizeof(iph->daddr);
1263 /* IPv6 doesn't really use broadcast mac address, but leave
1264 * that here just in case.
1266 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast)) {
1271 /* IPv6 uses all-nodes multicast as an equivalent to
1272 * broadcasts in IPv4.
1274 if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1279 /* Additianally, DAD probes should not be tx-balanced as that
1280 * will lead to false positives for duplicate addresses and
1281 * prevent address configuration from working.
1283 ip6hdr = ipv6_hdr(skb);
1284 if (ipv6_addr_any(&ip6hdr->saddr)) {
1289 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1290 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1293 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1294 /* something is wrong with this packet */
1299 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1300 /* The only protocol worth balancing in
1301 * this family since it has an "ARP" like
1308 hash_start = (char*)eth_data->h_dest;
1309 hash_size = ETH_ALEN;
1313 if (bond_info->rlb_enabled) {
1314 tx_slave = rlb_arp_xmit(skb, bond);
1322 if (do_tx_balance) {
1323 hash_index = _simple_hash(hash_start, hash_size);
1324 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1328 /* unbalanced or unassigned, send through primary */
1329 tx_slave = bond->curr_active_slave;
1330 bond_info->unbalanced_load += skb->len;
1333 if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1334 if (tx_slave != bond->curr_active_slave) {
1335 memcpy(eth_data->h_source,
1336 tx_slave->dev->dev_addr,
1340 res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1343 _lock_tx_hashtbl(bond);
1344 __tlb_clear_slave(bond, tx_slave, 0);
1345 _unlock_tx_hashtbl(bond);
1349 read_unlock(&bond->curr_slave_lock);
1352 /* no suitable interface, frame not sent */
1355 return NETDEV_TX_OK;
1358 void bond_alb_monitor(struct work_struct *work)
1360 struct bonding *bond = container_of(work, struct bonding,
1362 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1363 struct slave *slave;
1366 read_lock(&bond->lock);
1368 if (bond->slave_cnt == 0) {
1369 bond_info->tx_rebalance_counter = 0;
1370 bond_info->lp_counter = 0;
1374 bond_info->tx_rebalance_counter++;
1375 bond_info->lp_counter++;
1377 /* send learning packets */
1378 if (bond_info->lp_counter >= BOND_ALB_LP_TICKS) {
1379 /* change of curr_active_slave involves swapping of mac addresses.
1380 * in order to avoid this swapping from happening while
1381 * sending the learning packets, the curr_slave_lock must be held for
1384 read_lock(&bond->curr_slave_lock);
1386 bond_for_each_slave(bond, slave, i) {
1387 alb_send_learning_packets(slave, slave->dev->dev_addr);
1390 read_unlock(&bond->curr_slave_lock);
1392 bond_info->lp_counter = 0;
1395 /* rebalance tx traffic */
1396 if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1398 read_lock(&bond->curr_slave_lock);
1400 bond_for_each_slave(bond, slave, i) {
1401 tlb_clear_slave(bond, slave, 1);
1402 if (slave == bond->curr_active_slave) {
1403 SLAVE_TLB_INFO(slave).load =
1404 bond_info->unbalanced_load /
1405 BOND_TLB_REBALANCE_INTERVAL;
1406 bond_info->unbalanced_load = 0;
1410 read_unlock(&bond->curr_slave_lock);
1412 bond_info->tx_rebalance_counter = 0;
1415 /* handle rlb stuff */
1416 if (bond_info->rlb_enabled) {
1417 if (bond_info->primary_is_promisc &&
1418 (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1421 * dev_set_promiscuity requires rtnl and
1422 * nothing else. Avoid race with bond_close.
1424 read_unlock(&bond->lock);
1425 if (!rtnl_trylock()) {
1426 read_lock(&bond->lock);
1430 bond_info->rlb_promisc_timeout_counter = 0;
1432 /* If the primary was set to promiscuous mode
1433 * because a slave was disabled then
1434 * it can now leave promiscuous mode.
1436 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1437 bond_info->primary_is_promisc = 0;
1440 read_lock(&bond->lock);
1443 if (bond_info->rlb_rebalance) {
1444 bond_info->rlb_rebalance = 0;
1445 rlb_rebalance(bond);
1448 /* check if clients need updating */
1449 if (bond_info->rx_ntt) {
1450 if (bond_info->rlb_update_delay_counter) {
1451 --bond_info->rlb_update_delay_counter;
1453 rlb_update_rx_clients(bond);
1454 if (bond_info->rlb_update_retry_counter) {
1455 --bond_info->rlb_update_retry_counter;
1457 bond_info->rx_ntt = 0;
1464 queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1466 read_unlock(&bond->lock);
1469 /* assumption: called before the slave is attached to the bond
1470 * and not locked by the bond lock
1472 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1476 res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr);
1481 /* caller must hold the bond lock for write since the mac addresses
1482 * are compared and may be swapped.
1484 read_lock(&bond->lock);
1486 res = alb_handle_addr_collision_on_attach(bond, slave);
1488 read_unlock(&bond->lock);
1494 tlb_init_slave(slave);
1496 /* order a rebalance ASAP */
1497 bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1499 if (bond->alb_info.rlb_enabled) {
1500 bond->alb_info.rlb_rebalance = 1;
1507 * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1510 * Caller must hold RTNL and no other locks
1512 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1514 if (bond->slave_cnt > 1) {
1515 alb_change_hw_addr_on_detach(bond, slave);
1518 tlb_clear_slave(bond, slave, 0);
1520 if (bond->alb_info.rlb_enabled) {
1521 bond->alb_info.next_rx_slave = NULL;
1522 rlb_clear_slave(bond, slave);
1526 /* Caller must hold bond lock for read */
1527 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1529 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1531 if (link == BOND_LINK_DOWN) {
1532 tlb_clear_slave(bond, slave, 0);
1533 if (bond->alb_info.rlb_enabled) {
1534 rlb_clear_slave(bond, slave);
1536 } else if (link == BOND_LINK_UP) {
1537 /* order a rebalance ASAP */
1538 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1539 if (bond->alb_info.rlb_enabled) {
1540 bond->alb_info.rlb_rebalance = 1;
1541 /* If the updelay module parameter is smaller than the
1542 * forwarding delay of the switch the rebalance will
1543 * not work because the rebalance arp replies will
1544 * not be forwarded to the clients..
1551 * bond_alb_handle_active_change - assign new curr_active_slave
1552 * @bond: our bonding struct
1553 * @new_slave: new slave to assign
1555 * Set the bond->curr_active_slave to @new_slave and handle
1556 * mac address swapping and promiscuity changes as needed.
1558 * If new_slave is NULL, caller must hold curr_slave_lock or
1559 * bond->lock for write.
1561 * If new_slave is not NULL, caller must hold RTNL, bond->lock for
1562 * read and curr_slave_lock for write. Processing here may sleep, so
1563 * no other locks may be held.
1565 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1566 __releases(&bond->curr_slave_lock)
1567 __releases(&bond->lock)
1568 __acquires(&bond->lock)
1569 __acquires(&bond->curr_slave_lock)
1571 struct slave *swap_slave;
1574 if (bond->curr_active_slave == new_slave) {
1578 if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1579 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1580 bond->alb_info.primary_is_promisc = 0;
1581 bond->alb_info.rlb_promisc_timeout_counter = 0;
1584 swap_slave = bond->curr_active_slave;
1585 bond->curr_active_slave = new_slave;
1587 if (!new_slave || (bond->slave_cnt == 0)) {
1591 /* set the new curr_active_slave to the bonds mac address
1592 * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1595 struct slave *tmp_slave;
1596 /* find slave that is holding the bond's mac address */
1597 bond_for_each_slave(bond, tmp_slave, i) {
1598 if (ether_addr_equal_64bits(tmp_slave->dev->dev_addr,
1599 bond->dev->dev_addr)) {
1600 swap_slave = tmp_slave;
1607 * Arrange for swap_slave and new_slave to temporarily be
1608 * ignored so we can mess with their MAC addresses without
1609 * fear of interference from transmit activity.
1612 tlb_clear_slave(bond, swap_slave, 1);
1614 tlb_clear_slave(bond, new_slave, 1);
1616 write_unlock_bh(&bond->curr_slave_lock);
1617 read_unlock(&bond->lock);
1621 /* curr_active_slave must be set before calling alb_swap_mac_addr */
1623 /* swap mac address */
1624 alb_swap_mac_addr(bond, swap_slave, new_slave);
1626 /* set the new_slave to the bond mac address */
1627 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr);
1631 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1632 read_lock(&bond->lock);
1634 read_lock(&bond->lock);
1635 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1638 write_lock_bh(&bond->curr_slave_lock);
1644 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1645 __acquires(&bond->lock)
1646 __releases(&bond->lock)
1648 struct bonding *bond = netdev_priv(bond_dev);
1649 struct sockaddr *sa = addr;
1650 struct slave *slave, *swap_slave;
1654 if (!is_valid_ether_addr(sa->sa_data)) {
1655 return -EADDRNOTAVAIL;
1658 res = alb_set_mac_address(bond, addr);
1663 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1665 /* If there is no curr_active_slave there is nothing else to do.
1666 * Otherwise we'll need to pass the new address to it and handle
1669 if (!bond->curr_active_slave) {
1675 bond_for_each_slave(bond, slave, i) {
1676 if (ether_addr_equal_64bits(slave->dev->dev_addr,
1677 bond_dev->dev_addr)) {
1684 alb_swap_mac_addr(bond, swap_slave, bond->curr_active_slave);
1685 alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave);
1687 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr);
1689 read_lock(&bond->lock);
1690 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1691 if (bond->alb_info.rlb_enabled) {
1692 /* inform clients mac address has changed */
1693 rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1695 read_unlock(&bond->lock);
1701 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1703 if (bond->alb_info.current_alb_vlan &&
1704 (bond->alb_info.current_alb_vlan->vlan_id == vlan_id)) {
1705 bond->alb_info.current_alb_vlan = NULL;
1708 if (bond->alb_info.rlb_enabled) {
1709 rlb_clear_vlan(bond, vlan_id);