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(struct bonding *bond)
104 spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
107 static inline void _unlock_tx_hashtbl(struct bonding *bond)
109 spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
112 /* Caller must hold tx_hashtbl lock */
113 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
116 entry->load_history = 1 + entry->tx_bytes /
117 BOND_TLB_REBALANCE_INTERVAL;
121 entry->tx_slave = NULL;
122 entry->next = TLB_NULL_INDEX;
123 entry->prev = TLB_NULL_INDEX;
126 static inline void tlb_init_slave(struct slave *slave)
128 SLAVE_TLB_INFO(slave).load = 0;
129 SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
132 /* Caller must hold bond lock for read */
133 static void tlb_clear_slave(struct bonding *bond, struct slave *slave, int save_load)
135 struct tlb_client_info *tx_hash_table;
138 _lock_tx_hashtbl(bond);
140 /* clear slave from tx_hashtbl */
141 tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
143 /* skip this if we've already freed the tx hash table */
145 index = SLAVE_TLB_INFO(slave).head;
146 while (index != TLB_NULL_INDEX) {
147 u32 next_index = tx_hash_table[index].next;
148 tlb_init_table_entry(&tx_hash_table[index], save_load);
153 tlb_init_slave(slave);
155 _unlock_tx_hashtbl(bond);
158 /* Must be called before starting the monitor timer */
159 static int tlb_initialize(struct bonding *bond)
161 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
162 int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
163 struct tlb_client_info *new_hashtbl;
166 spin_lock_init(&(bond_info->tx_hashtbl_lock));
168 new_hashtbl = kzalloc(size, GFP_KERNEL);
170 pr_err("%s: Error: Failed to allocate TLB hash table\n",
174 _lock_tx_hashtbl(bond);
176 bond_info->tx_hashtbl = new_hashtbl;
178 for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
179 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 1);
182 _unlock_tx_hashtbl(bond);
187 /* Must be called only after all slaves have been released */
188 static void tlb_deinitialize(struct bonding *bond)
190 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
192 _lock_tx_hashtbl(bond);
194 kfree(bond_info->tx_hashtbl);
195 bond_info->tx_hashtbl = NULL;
197 _unlock_tx_hashtbl(bond);
200 static long long compute_gap(struct slave *slave)
202 return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
203 (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
206 /* Caller must hold bond lock for read */
207 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
209 struct slave *slave, *least_loaded;
216 /* Find the slave with the largest gap */
217 bond_for_each_slave(bond, slave, i) {
218 if (SLAVE_IS_OK(slave)) {
219 long long gap = compute_gap(slave);
222 least_loaded = slave;
231 /* Caller must hold bond lock for read */
232 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index, u32 skb_len)
234 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
235 struct tlb_client_info *hash_table;
236 struct slave *assigned_slave;
238 _lock_tx_hashtbl(bond);
240 hash_table = bond_info->tx_hashtbl;
241 assigned_slave = hash_table[hash_index].tx_slave;
242 if (!assigned_slave) {
243 assigned_slave = tlb_get_least_loaded_slave(bond);
245 if (assigned_slave) {
246 struct tlb_slave_info *slave_info =
247 &(SLAVE_TLB_INFO(assigned_slave));
248 u32 next_index = slave_info->head;
250 hash_table[hash_index].tx_slave = assigned_slave;
251 hash_table[hash_index].next = next_index;
252 hash_table[hash_index].prev = TLB_NULL_INDEX;
254 if (next_index != TLB_NULL_INDEX) {
255 hash_table[next_index].prev = hash_index;
258 slave_info->head = hash_index;
260 hash_table[hash_index].load_history;
264 if (assigned_slave) {
265 hash_table[hash_index].tx_bytes += skb_len;
268 _unlock_tx_hashtbl(bond);
270 return assigned_slave;
273 /*********************** rlb specific functions ***************************/
274 static inline void _lock_rx_hashtbl(struct bonding *bond)
276 spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
279 static inline void _unlock_rx_hashtbl(struct bonding *bond)
281 spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
284 /* when an ARP REPLY is received from a client update its info
287 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
289 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
290 struct rlb_client_info *client_info;
293 _lock_rx_hashtbl(bond);
295 hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
296 client_info = &(bond_info->rx_hashtbl[hash_index]);
298 if ((client_info->assigned) &&
299 (client_info->ip_src == arp->ip_dst) &&
300 (client_info->ip_dst == arp->ip_src) &&
301 (compare_ether_addr_64bits(client_info->mac_dst, arp->mac_src))) {
302 /* update the clients MAC address */
303 memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
304 client_info->ntt = 1;
305 bond_info->rx_ntt = 1;
308 _unlock_rx_hashtbl(bond);
311 static int rlb_arp_recv(struct sk_buff *skb, struct net_device *bond_dev, struct packet_type *ptype, struct net_device *orig_dev)
313 struct bonding *bond;
314 struct arp_pkt *arp = (struct arp_pkt *)skb->data;
315 int res = NET_RX_DROP;
317 while (bond_dev->priv_flags & IFF_802_1Q_VLAN)
318 bond_dev = vlan_dev_real_dev(bond_dev);
320 if (!(bond_dev->priv_flags & IFF_BONDING) ||
321 !(bond_dev->flags & IFF_MASTER))
325 pr_debug("Packet has no ARP data\n");
329 if (!pskb_may_pull(skb, arp_hdr_len(bond_dev)))
332 if (skb->len < sizeof(struct arp_pkt)) {
333 pr_debug("Packet is too small to be an ARP\n");
337 if (arp->op_code == htons(ARPOP_REPLY)) {
338 /* update rx hash table for this ARP */
339 bond = netdev_priv(bond_dev);
340 rlb_update_entry_from_arp(bond, arp);
341 pr_debug("Server received an ARP Reply from client\n");
344 res = NET_RX_SUCCESS;
352 /* Caller must hold bond lock for read */
353 static struct slave *rlb_next_rx_slave(struct bonding *bond)
355 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
356 struct slave *rx_slave, *slave, *start_at;
359 if (bond_info->next_rx_slave) {
360 start_at = bond_info->next_rx_slave;
362 start_at = bond->first_slave;
367 bond_for_each_slave_from(bond, slave, i, start_at) {
368 if (SLAVE_IS_OK(slave)) {
371 } else if (slave->speed > rx_slave->speed) {
378 bond_info->next_rx_slave = rx_slave->next;
384 /* teach the switch the mac of a disabled slave
385 * on the primary for fault tolerance
387 * Caller must hold bond->curr_slave_lock for write or bond lock for write
389 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
391 if (!bond->curr_active_slave) {
395 if (!bond->alb_info.primary_is_promisc) {
396 if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1))
397 bond->alb_info.primary_is_promisc = 1;
399 bond->alb_info.primary_is_promisc = 0;
402 bond->alb_info.rlb_promisc_timeout_counter = 0;
404 alb_send_learning_packets(bond->curr_active_slave, addr);
407 /* slave being removed should not be active at this point
409 * Caller must hold bond lock for read
411 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
413 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
414 struct rlb_client_info *rx_hash_table;
415 u32 index, next_index;
417 /* clear slave from rx_hashtbl */
418 _lock_rx_hashtbl(bond);
420 rx_hash_table = bond_info->rx_hashtbl;
421 index = bond_info->rx_hashtbl_head;
422 for (; index != RLB_NULL_INDEX; index = next_index) {
423 next_index = rx_hash_table[index].next;
424 if (rx_hash_table[index].slave == slave) {
425 struct slave *assigned_slave = rlb_next_rx_slave(bond);
427 if (assigned_slave) {
428 rx_hash_table[index].slave = assigned_slave;
429 if (compare_ether_addr_64bits(rx_hash_table[index].mac_dst,
431 bond_info->rx_hashtbl[index].ntt = 1;
432 bond_info->rx_ntt = 1;
433 /* A slave has been removed from the
434 * table because it is either disabled
435 * or being released. We must retry the
436 * update to avoid clients from not
437 * being updated & disconnecting when
440 bond_info->rlb_update_retry_counter =
443 } else { /* there is no active slave */
444 rx_hash_table[index].slave = NULL;
449 _unlock_rx_hashtbl(bond);
451 write_lock_bh(&bond->curr_slave_lock);
453 if (slave != bond->curr_active_slave) {
454 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
457 write_unlock_bh(&bond->curr_slave_lock);
460 static void rlb_update_client(struct rlb_client_info *client_info)
464 if (!client_info->slave) {
468 for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
471 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
473 client_info->slave->dev,
475 client_info->mac_dst,
476 client_info->slave->dev->dev_addr,
477 client_info->mac_dst);
479 pr_err("%s: Error: failed to create an ARP packet\n",
480 client_info->slave->dev->master->name);
484 skb->dev = client_info->slave->dev;
486 if (client_info->tag) {
487 skb = vlan_put_tag(skb, client_info->vlan_id);
489 pr_err("%s: Error: failed to insert VLAN tag\n",
490 client_info->slave->dev->master->name);
499 /* sends ARP REPLIES that update the clients that need updating */
500 static void rlb_update_rx_clients(struct bonding *bond)
502 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
503 struct rlb_client_info *client_info;
506 _lock_rx_hashtbl(bond);
508 hash_index = bond_info->rx_hashtbl_head;
509 for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
510 client_info = &(bond_info->rx_hashtbl[hash_index]);
511 if (client_info->ntt) {
512 rlb_update_client(client_info);
513 if (bond_info->rlb_update_retry_counter == 0) {
514 client_info->ntt = 0;
519 /* do not update the entries again until this counter is zero so that
520 * not to confuse the clients.
522 bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
524 _unlock_rx_hashtbl(bond);
527 /* The slave was assigned a new mac address - update the clients */
528 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
530 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
531 struct rlb_client_info *client_info;
535 _lock_rx_hashtbl(bond);
537 hash_index = bond_info->rx_hashtbl_head;
538 for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
539 client_info = &(bond_info->rx_hashtbl[hash_index]);
541 if ((client_info->slave == slave) &&
542 compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
543 client_info->ntt = 1;
548 // update the team's flag only after the whole iteration
550 bond_info->rx_ntt = 1;
552 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
555 _unlock_rx_hashtbl(bond);
558 /* mark all clients using src_ip to be updated */
559 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
561 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
562 struct rlb_client_info *client_info;
565 _lock_rx_hashtbl(bond);
567 hash_index = bond_info->rx_hashtbl_head;
568 for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
569 client_info = &(bond_info->rx_hashtbl[hash_index]);
571 if (!client_info->slave) {
572 pr_err("%s: Error: found a client with no channel in the client's hash table\n",
576 /*update all clients using this src_ip, that are not assigned
577 * to the team's address (curr_active_slave) and have a known
578 * unicast mac address.
580 if ((client_info->ip_src == src_ip) &&
581 compare_ether_addr_64bits(client_info->slave->dev->dev_addr,
582 bond->dev->dev_addr) &&
583 compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
584 client_info->ntt = 1;
585 bond_info->rx_ntt = 1;
589 _unlock_rx_hashtbl(bond);
592 /* Caller must hold both bond and ptr locks for read */
593 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
595 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
596 struct arp_pkt *arp = arp_pkt(skb);
597 struct slave *assigned_slave;
598 struct rlb_client_info *client_info;
601 _lock_rx_hashtbl(bond);
603 hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_src));
604 client_info = &(bond_info->rx_hashtbl[hash_index]);
606 if (client_info->assigned) {
607 if ((client_info->ip_src == arp->ip_src) &&
608 (client_info->ip_dst == arp->ip_dst)) {
609 /* the entry is already assigned to this client */
610 if (compare_ether_addr_64bits(arp->mac_dst, mac_bcast)) {
611 /* update mac address from arp */
612 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
615 assigned_slave = client_info->slave;
616 if (assigned_slave) {
617 _unlock_rx_hashtbl(bond);
618 return assigned_slave;
621 /* the entry is already assigned to some other client,
622 * move the old client to primary (curr_active_slave) so
623 * that the new client can be assigned to this entry.
625 if (bond->curr_active_slave &&
626 client_info->slave != bond->curr_active_slave) {
627 client_info->slave = bond->curr_active_slave;
628 rlb_update_client(client_info);
632 /* assign a new slave */
633 assigned_slave = rlb_next_rx_slave(bond);
635 if (assigned_slave) {
636 client_info->ip_src = arp->ip_src;
637 client_info->ip_dst = arp->ip_dst;
638 /* arp->mac_dst is broadcast for arp reqeusts.
639 * will be updated with clients actual unicast mac address
640 * upon receiving an arp reply.
642 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
643 client_info->slave = assigned_slave;
645 if (compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
646 client_info->ntt = 1;
647 bond->alb_info.rx_ntt = 1;
649 client_info->ntt = 0;
653 if (!vlan_get_tag(skb, &client_info->vlan_id))
654 client_info->tag = 1;
657 if (!client_info->assigned) {
658 u32 prev_tbl_head = bond_info->rx_hashtbl_head;
659 bond_info->rx_hashtbl_head = hash_index;
660 client_info->next = prev_tbl_head;
661 if (prev_tbl_head != RLB_NULL_INDEX) {
662 bond_info->rx_hashtbl[prev_tbl_head].prev =
665 client_info->assigned = 1;
669 _unlock_rx_hashtbl(bond);
671 return assigned_slave;
674 /* chooses (and returns) transmit channel for arp reply
675 * does not choose channel for other arp types since they are
676 * sent on the curr_active_slave
678 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
680 struct arp_pkt *arp = arp_pkt(skb);
681 struct slave *tx_slave = NULL;
683 if (arp->op_code == htons(ARPOP_REPLY)) {
684 /* the arp must be sent on the selected
687 tx_slave = rlb_choose_channel(skb, bond);
689 memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
691 pr_debug("Server sent ARP Reply packet\n");
692 } else if (arp->op_code == htons(ARPOP_REQUEST)) {
693 /* Create an entry in the rx_hashtbl for this client as a
695 * When the arp reply is received the entry will be updated
696 * with the correct unicast address of the client.
698 rlb_choose_channel(skb, bond);
700 /* The ARP relpy packets must be delayed so that
701 * they can cancel out the influence of the ARP request.
703 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
705 /* arp requests are broadcast and are sent on the primary
706 * the arp request will collapse all clients on the subnet to
707 * the primary slave. We must register these clients to be
708 * updated with their assigned mac.
710 rlb_req_update_subnet_clients(bond, arp->ip_src);
711 pr_debug("Server sent ARP Request packet\n");
717 /* Caller must hold bond lock for read */
718 static void rlb_rebalance(struct bonding *bond)
720 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
721 struct slave *assigned_slave;
722 struct rlb_client_info *client_info;
726 _lock_rx_hashtbl(bond);
729 hash_index = bond_info->rx_hashtbl_head;
730 for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
731 client_info = &(bond_info->rx_hashtbl[hash_index]);
732 assigned_slave = rlb_next_rx_slave(bond);
733 if (assigned_slave && (client_info->slave != assigned_slave)) {
734 client_info->slave = assigned_slave;
735 client_info->ntt = 1;
740 /* update the team's flag only after the whole iteration */
742 bond_info->rx_ntt = 1;
744 _unlock_rx_hashtbl(bond);
747 /* Caller must hold rx_hashtbl lock */
748 static void rlb_init_table_entry(struct rlb_client_info *entry)
750 memset(entry, 0, sizeof(struct rlb_client_info));
751 entry->next = RLB_NULL_INDEX;
752 entry->prev = RLB_NULL_INDEX;
755 static int rlb_initialize(struct bonding *bond)
757 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
758 struct packet_type *pk_type = &(BOND_ALB_INFO(bond).rlb_pkt_type);
759 struct rlb_client_info *new_hashtbl;
760 int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
763 spin_lock_init(&(bond_info->rx_hashtbl_lock));
765 new_hashtbl = kmalloc(size, GFP_KERNEL);
767 pr_err("%s: Error: Failed to allocate RLB hash table\n",
771 _lock_rx_hashtbl(bond);
773 bond_info->rx_hashtbl = new_hashtbl;
775 bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
777 for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
778 rlb_init_table_entry(bond_info->rx_hashtbl + i);
781 _unlock_rx_hashtbl(bond);
783 /*initialize packet type*/
784 pk_type->type = cpu_to_be16(ETH_P_ARP);
785 pk_type->dev = bond->dev;
786 pk_type->func = rlb_arp_recv;
788 /* register to receive ARPs */
789 dev_add_pack(pk_type);
794 static void rlb_deinitialize(struct bonding *bond)
796 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
798 dev_remove_pack(&(bond_info->rlb_pkt_type));
800 _lock_rx_hashtbl(bond);
802 kfree(bond_info->rx_hashtbl);
803 bond_info->rx_hashtbl = NULL;
804 bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
806 _unlock_rx_hashtbl(bond);
809 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
811 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
814 _lock_rx_hashtbl(bond);
816 curr_index = bond_info->rx_hashtbl_head;
817 while (curr_index != RLB_NULL_INDEX) {
818 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
819 u32 next_index = bond_info->rx_hashtbl[curr_index].next;
820 u32 prev_index = bond_info->rx_hashtbl[curr_index].prev;
822 if (curr->tag && (curr->vlan_id == vlan_id)) {
823 if (curr_index == bond_info->rx_hashtbl_head) {
824 bond_info->rx_hashtbl_head = next_index;
826 if (prev_index != RLB_NULL_INDEX) {
827 bond_info->rx_hashtbl[prev_index].next = next_index;
829 if (next_index != RLB_NULL_INDEX) {
830 bond_info->rx_hashtbl[next_index].prev = prev_index;
833 rlb_init_table_entry(curr);
836 curr_index = next_index;
839 _unlock_rx_hashtbl(bond);
842 /*********************** tlb/rlb shared functions *********************/
844 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
846 struct bonding *bond = bond_get_bond_by_slave(slave);
847 struct learning_pkt pkt;
848 int size = sizeof(struct learning_pkt);
851 memset(&pkt, 0, size);
852 memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
853 memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
854 pkt.type = cpu_to_be16(ETH_P_LOOP);
856 for (i = 0; i < MAX_LP_BURST; i++) {
860 skb = dev_alloc_skb(size);
865 data = skb_put(skb, size);
866 memcpy(data, &pkt, size);
868 skb_reset_mac_header(skb);
869 skb->network_header = skb->mac_header + ETH_HLEN;
870 skb->protocol = pkt.type;
871 skb->priority = TC_PRIO_CONTROL;
872 skb->dev = slave->dev;
875 struct vlan_entry *vlan;
877 vlan = bond_next_vlan(bond,
878 bond->alb_info.current_alb_vlan);
880 bond->alb_info.current_alb_vlan = vlan;
886 skb = vlan_put_tag(skb, vlan->vlan_id);
888 pr_err("%s: Error: failed to insert VLAN tag\n",
898 /* hw is a boolean parameter that determines whether we should try and
899 * set the hw address of the device as well as the hw address of the
902 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[], int hw)
904 struct net_device *dev = slave->dev;
905 struct sockaddr s_addr;
908 memcpy(dev->dev_addr, addr, dev->addr_len);
912 /* for rlb each slave must have a unique hw mac addresses so that */
913 /* each slave will receive packets destined to a different mac */
914 memcpy(s_addr.sa_data, addr, dev->addr_len);
915 s_addr.sa_family = dev->type;
916 if (dev_set_mac_address(dev, &s_addr)) {
917 pr_err("%s: Error: dev_set_mac_address of dev %s failed!\n"
918 "ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
919 dev->master->name, dev->name);
926 * Swap MAC addresses between two slaves.
928 * Called with RTNL held, and no other locks.
932 static void alb_swap_mac_addr(struct bonding *bond, struct slave *slave1, struct slave *slave2)
934 u8 tmp_mac_addr[ETH_ALEN];
936 memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
937 alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr, bond->alb_info.rlb_enabled);
938 alb_set_slave_mac_addr(slave2, tmp_mac_addr, bond->alb_info.rlb_enabled);
943 * Send learning packets after MAC address swap.
945 * Called with RTNL and no other locks
947 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
948 struct slave *slave2)
950 int slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
951 struct slave *disabled_slave = NULL;
955 /* fasten the change in the switch */
956 if (SLAVE_IS_OK(slave1)) {
957 alb_send_learning_packets(slave1, slave1->dev->dev_addr);
958 if (bond->alb_info.rlb_enabled) {
959 /* inform the clients that the mac address
962 rlb_req_update_slave_clients(bond, slave1);
965 disabled_slave = slave1;
968 if (SLAVE_IS_OK(slave2)) {
969 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
970 if (bond->alb_info.rlb_enabled) {
971 /* inform the clients that the mac address
974 rlb_req_update_slave_clients(bond, slave2);
977 disabled_slave = slave2;
980 if (bond->alb_info.rlb_enabled && slaves_state_differ) {
981 /* A disabled slave was assigned an active mac addr */
982 rlb_teach_disabled_mac_on_primary(bond,
983 disabled_slave->dev->dev_addr);
988 * alb_change_hw_addr_on_detach
989 * @bond: bonding we're working on
990 * @slave: the slave that was just detached
992 * We assume that @slave was already detached from the slave list.
994 * If @slave's permanent hw address is different both from its current
995 * address and from @bond's address, then somewhere in the bond there's
996 * a slave that has @slave's permanet address as its current address.
997 * We'll make sure that that slave no longer uses @slave's permanent address.
999 * Caller must hold RTNL and no other locks
1001 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1006 perm_curr_diff = compare_ether_addr_64bits(slave->perm_hwaddr,
1007 slave->dev->dev_addr);
1008 perm_bond_diff = compare_ether_addr_64bits(slave->perm_hwaddr,
1009 bond->dev->dev_addr);
1011 if (perm_curr_diff && perm_bond_diff) {
1012 struct slave *tmp_slave;
1015 bond_for_each_slave(bond, tmp_slave, i) {
1016 if (!compare_ether_addr_64bits(slave->perm_hwaddr,
1017 tmp_slave->dev->dev_addr)) {
1024 /* locking: needs RTNL and nothing else */
1025 alb_swap_mac_addr(bond, slave, tmp_slave);
1026 alb_fasten_mac_swap(bond, slave, tmp_slave);
1032 * alb_handle_addr_collision_on_attach
1033 * @bond: bonding we're working on
1034 * @slave: the slave that was just attached
1036 * checks uniqueness of slave's mac address and handles the case the
1037 * new slave uses the bonds mac address.
1039 * If the permanent hw address of @slave is @bond's hw address, we need to
1040 * find a different hw address to give @slave, that isn't in use by any other
1041 * slave in the bond. This address must be, of course, one of the premanent
1042 * addresses of the other slaves.
1044 * We go over the slave list, and for each slave there we compare its
1045 * permanent hw address with the current address of all the other slaves.
1046 * If no match was found, then we've found a slave with a permanent address
1047 * that isn't used by any other slave in the bond, so we can assign it to
1050 * assumption: this function is called before @slave is attached to the
1053 * caller must hold the bond lock for write since the mac addresses are compared
1054 * and may be swapped.
1056 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1058 struct slave *tmp_slave1, *tmp_slave2, *free_mac_slave;
1059 struct slave *has_bond_addr = bond->curr_active_slave;
1060 int i, j, found = 0;
1062 if (bond->slave_cnt == 0) {
1063 /* this is the first slave */
1067 /* if slave's mac address differs from bond's mac address
1068 * check uniqueness of slave's mac address against the other
1069 * slaves in the bond.
1071 if (compare_ether_addr_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1072 bond_for_each_slave(bond, tmp_slave1, i) {
1073 if (!compare_ether_addr_64bits(tmp_slave1->dev->dev_addr,
1074 slave->dev->dev_addr)) {
1083 /* Try setting slave mac to bond address and fall-through
1084 to code handling that situation below... */
1085 alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1086 bond->alb_info.rlb_enabled);
1089 /* The slave's address is equal to the address of the bond.
1090 * Search for a spare address in the bond for this slave.
1092 free_mac_slave = NULL;
1094 bond_for_each_slave(bond, tmp_slave1, i) {
1096 bond_for_each_slave(bond, tmp_slave2, j) {
1097 if (!compare_ether_addr_64bits(tmp_slave1->perm_hwaddr,
1098 tmp_slave2->dev->dev_addr)) {
1105 /* no slave has tmp_slave1's perm addr
1108 free_mac_slave = tmp_slave1;
1112 if (!has_bond_addr) {
1113 if (!compare_ether_addr_64bits(tmp_slave1->dev->dev_addr,
1114 bond->dev->dev_addr)) {
1116 has_bond_addr = tmp_slave1;
1121 if (free_mac_slave) {
1122 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1123 bond->alb_info.rlb_enabled);
1125 pr_warning("%s: Warning: the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1126 bond->dev->name, slave->dev->name,
1127 free_mac_slave->dev->name);
1129 } else if (has_bond_addr) {
1130 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",
1131 bond->dev->name, slave->dev->name);
1139 * alb_set_mac_address
1143 * In TLB mode all slaves are configured to the bond's hw address, but set
1144 * their dev_addr field to different addresses (based on their permanent hw
1147 * For each slave, this function sets the interface to the new address and then
1148 * changes its dev_addr field to its previous value.
1150 * Unwinding assumes bond's mac address has not yet changed.
1152 static int alb_set_mac_address(struct bonding *bond, void *addr)
1155 struct slave *slave, *stop_at;
1156 char tmp_addr[ETH_ALEN];
1160 if (bond->alb_info.rlb_enabled) {
1164 bond_for_each_slave(bond, slave, i) {
1165 /* save net_device's current hw address */
1166 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1168 res = dev_set_mac_address(slave->dev, addr);
1170 /* restore net_device's hw address */
1171 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1180 memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1181 sa.sa_family = bond->dev->type;
1183 /* unwind from head to the slave that failed */
1185 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
1186 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1187 dev_set_mac_address(slave->dev, &sa);
1188 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1194 /************************ exported alb funcions ************************/
1196 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1200 res = tlb_initialize(bond);
1206 bond->alb_info.rlb_enabled = 1;
1207 /* initialize rlb */
1208 res = rlb_initialize(bond);
1210 tlb_deinitialize(bond);
1214 bond->alb_info.rlb_enabled = 0;
1220 void bond_alb_deinitialize(struct bonding *bond)
1222 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1224 tlb_deinitialize(bond);
1226 if (bond_info->rlb_enabled) {
1227 rlb_deinitialize(bond);
1231 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1233 struct bonding *bond = netdev_priv(bond_dev);
1234 struct ethhdr *eth_data;
1235 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1236 struct slave *tx_slave = NULL;
1237 static const __be32 ip_bcast = htonl(0xffffffff);
1239 int do_tx_balance = 1;
1241 const u8 *hash_start = NULL;
1243 struct ipv6hdr *ip6hdr;
1245 skb_reset_mac_header(skb);
1246 eth_data = eth_hdr(skb);
1248 /* make sure that the curr_active_slave and the slaves list do
1249 * not change during tx
1251 read_lock(&bond->lock);
1252 read_lock(&bond->curr_slave_lock);
1254 if (!BOND_IS_OK(bond)) {
1258 switch (ntohs(skb->protocol)) {
1260 const struct iphdr *iph = ip_hdr(skb);
1262 if (!compare_ether_addr_64bits(eth_data->h_dest, mac_bcast) ||
1263 (iph->daddr == ip_bcast) ||
1264 (iph->protocol == IPPROTO_IGMP)) {
1268 hash_start = (char *)&(iph->daddr);
1269 hash_size = sizeof(iph->daddr);
1273 /* IPv6 doesn't really use broadcast mac address, but leave
1274 * that here just in case.
1276 if (!compare_ether_addr_64bits(eth_data->h_dest, mac_bcast)) {
1281 /* IPv6 uses all-nodes multicast as an equivalent to
1282 * broadcasts in IPv4.
1284 if (!compare_ether_addr_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1289 /* Additianally, DAD probes should not be tx-balanced as that
1290 * will lead to false positives for duplicate addresses and
1291 * prevent address configuration from working.
1293 ip6hdr = ipv6_hdr(skb);
1294 if (ipv6_addr_any(&ip6hdr->saddr)) {
1299 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1300 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1303 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1304 /* something is wrong with this packet */
1309 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1310 /* The only protocol worth balancing in
1311 * this family since it has an "ARP" like
1318 hash_start = (char*)eth_data->h_dest;
1319 hash_size = ETH_ALEN;
1323 if (bond_info->rlb_enabled) {
1324 tx_slave = rlb_arp_xmit(skb, bond);
1332 if (do_tx_balance) {
1333 hash_index = _simple_hash(hash_start, hash_size);
1334 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1338 /* unbalanced or unassigned, send through primary */
1339 tx_slave = bond->curr_active_slave;
1340 bond_info->unbalanced_load += skb->len;
1343 if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1344 if (tx_slave != bond->curr_active_slave) {
1345 memcpy(eth_data->h_source,
1346 tx_slave->dev->dev_addr,
1350 res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1353 tlb_clear_slave(bond, tx_slave, 0);
1359 /* no suitable interface, frame not sent */
1362 read_unlock(&bond->curr_slave_lock);
1363 read_unlock(&bond->lock);
1364 return NETDEV_TX_OK;
1367 void bond_alb_monitor(struct work_struct *work)
1369 struct bonding *bond = container_of(work, struct bonding,
1371 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1372 struct slave *slave;
1375 read_lock(&bond->lock);
1377 if (bond->kill_timers) {
1381 if (bond->slave_cnt == 0) {
1382 bond_info->tx_rebalance_counter = 0;
1383 bond_info->lp_counter = 0;
1387 bond_info->tx_rebalance_counter++;
1388 bond_info->lp_counter++;
1390 /* send learning packets */
1391 if (bond_info->lp_counter >= BOND_ALB_LP_TICKS) {
1392 /* change of curr_active_slave involves swapping of mac addresses.
1393 * in order to avoid this swapping from happening while
1394 * sending the learning packets, the curr_slave_lock must be held for
1397 read_lock(&bond->curr_slave_lock);
1399 bond_for_each_slave(bond, slave, i) {
1400 alb_send_learning_packets(slave, slave->dev->dev_addr);
1403 read_unlock(&bond->curr_slave_lock);
1405 bond_info->lp_counter = 0;
1408 /* rebalance tx traffic */
1409 if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1411 read_lock(&bond->curr_slave_lock);
1413 bond_for_each_slave(bond, slave, i) {
1414 tlb_clear_slave(bond, slave, 1);
1415 if (slave == bond->curr_active_slave) {
1416 SLAVE_TLB_INFO(slave).load =
1417 bond_info->unbalanced_load /
1418 BOND_TLB_REBALANCE_INTERVAL;
1419 bond_info->unbalanced_load = 0;
1423 read_unlock(&bond->curr_slave_lock);
1425 bond_info->tx_rebalance_counter = 0;
1428 /* handle rlb stuff */
1429 if (bond_info->rlb_enabled) {
1430 if (bond_info->primary_is_promisc &&
1431 (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1434 * dev_set_promiscuity requires rtnl and
1437 read_unlock(&bond->lock);
1440 bond_info->rlb_promisc_timeout_counter = 0;
1442 /* If the primary was set to promiscuous mode
1443 * because a slave was disabled then
1444 * it can now leave promiscuous mode.
1446 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1447 bond_info->primary_is_promisc = 0;
1450 read_lock(&bond->lock);
1453 if (bond_info->rlb_rebalance) {
1454 bond_info->rlb_rebalance = 0;
1455 rlb_rebalance(bond);
1458 /* check if clients need updating */
1459 if (bond_info->rx_ntt) {
1460 if (bond_info->rlb_update_delay_counter) {
1461 --bond_info->rlb_update_delay_counter;
1463 rlb_update_rx_clients(bond);
1464 if (bond_info->rlb_update_retry_counter) {
1465 --bond_info->rlb_update_retry_counter;
1467 bond_info->rx_ntt = 0;
1474 queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1476 read_unlock(&bond->lock);
1479 /* assumption: called before the slave is attached to the bond
1480 * and not locked by the bond lock
1482 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1486 res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1487 bond->alb_info.rlb_enabled);
1492 /* caller must hold the bond lock for write since the mac addresses
1493 * are compared and may be swapped.
1495 read_lock(&bond->lock);
1497 res = alb_handle_addr_collision_on_attach(bond, slave);
1499 read_unlock(&bond->lock);
1505 tlb_init_slave(slave);
1507 /* order a rebalance ASAP */
1508 bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1510 if (bond->alb_info.rlb_enabled) {
1511 bond->alb_info.rlb_rebalance = 1;
1518 * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1521 * Caller must hold RTNL and no other locks
1523 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1525 if (bond->slave_cnt > 1) {
1526 alb_change_hw_addr_on_detach(bond, slave);
1529 tlb_clear_slave(bond, slave, 0);
1531 if (bond->alb_info.rlb_enabled) {
1532 bond->alb_info.next_rx_slave = NULL;
1533 rlb_clear_slave(bond, slave);
1537 /* Caller must hold bond lock for read */
1538 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1540 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1542 if (link == BOND_LINK_DOWN) {
1543 tlb_clear_slave(bond, slave, 0);
1544 if (bond->alb_info.rlb_enabled) {
1545 rlb_clear_slave(bond, slave);
1547 } else if (link == BOND_LINK_UP) {
1548 /* order a rebalance ASAP */
1549 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1550 if (bond->alb_info.rlb_enabled) {
1551 bond->alb_info.rlb_rebalance = 1;
1552 /* If the updelay module parameter is smaller than the
1553 * forwarding delay of the switch the rebalance will
1554 * not work because the rebalance arp replies will
1555 * not be forwarded to the clients..
1562 * bond_alb_handle_active_change - assign new curr_active_slave
1563 * @bond: our bonding struct
1564 * @new_slave: new slave to assign
1566 * Set the bond->curr_active_slave to @new_slave and handle
1567 * mac address swapping and promiscuity changes as needed.
1569 * If new_slave is NULL, caller must hold curr_slave_lock or
1570 * bond->lock for write.
1572 * If new_slave is not NULL, caller must hold RTNL, bond->lock for
1573 * read and curr_slave_lock for write. Processing here may sleep, so
1574 * no other locks may be held.
1576 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1577 __releases(&bond->curr_slave_lock)
1578 __releases(&bond->lock)
1579 __acquires(&bond->lock)
1580 __acquires(&bond->curr_slave_lock)
1582 struct slave *swap_slave;
1585 if (bond->curr_active_slave == new_slave) {
1589 if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1590 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1591 bond->alb_info.primary_is_promisc = 0;
1592 bond->alb_info.rlb_promisc_timeout_counter = 0;
1595 swap_slave = bond->curr_active_slave;
1596 bond->curr_active_slave = new_slave;
1598 if (!new_slave || (bond->slave_cnt == 0)) {
1602 /* set the new curr_active_slave to the bonds mac address
1603 * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1606 struct slave *tmp_slave;
1607 /* find slave that is holding the bond's mac address */
1608 bond_for_each_slave(bond, tmp_slave, i) {
1609 if (!compare_ether_addr_64bits(tmp_slave->dev->dev_addr,
1610 bond->dev->dev_addr)) {
1611 swap_slave = tmp_slave;
1618 * Arrange for swap_slave and new_slave to temporarily be
1619 * ignored so we can mess with their MAC addresses without
1620 * fear of interference from transmit activity.
1623 tlb_clear_slave(bond, swap_slave, 1);
1625 tlb_clear_slave(bond, new_slave, 1);
1627 write_unlock_bh(&bond->curr_slave_lock);
1628 read_unlock(&bond->lock);
1632 /* curr_active_slave must be set before calling alb_swap_mac_addr */
1634 /* swap mac address */
1635 alb_swap_mac_addr(bond, swap_slave, new_slave);
1637 /* set the new_slave to the bond mac address */
1638 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1639 bond->alb_info.rlb_enabled);
1643 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1644 read_lock(&bond->lock);
1646 read_lock(&bond->lock);
1647 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1650 write_lock_bh(&bond->curr_slave_lock);
1656 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1657 __acquires(&bond->lock)
1658 __releases(&bond->lock)
1660 struct bonding *bond = netdev_priv(bond_dev);
1661 struct sockaddr *sa = addr;
1662 struct slave *slave, *swap_slave;
1666 if (!is_valid_ether_addr(sa->sa_data)) {
1667 return -EADDRNOTAVAIL;
1670 res = alb_set_mac_address(bond, addr);
1675 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1677 /* If there is no curr_active_slave there is nothing else to do.
1678 * Otherwise we'll need to pass the new address to it and handle
1681 if (!bond->curr_active_slave) {
1687 bond_for_each_slave(bond, slave, i) {
1688 if (!compare_ether_addr_64bits(slave->dev->dev_addr,
1689 bond_dev->dev_addr)) {
1696 alb_swap_mac_addr(bond, swap_slave, bond->curr_active_slave);
1697 alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave);
1699 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr,
1700 bond->alb_info.rlb_enabled);
1702 read_lock(&bond->lock);
1703 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1704 if (bond->alb_info.rlb_enabled) {
1705 /* inform clients mac address has changed */
1706 rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1708 read_unlock(&bond->lock);
1714 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1716 if (bond->alb_info.current_alb_vlan &&
1717 (bond->alb_info.current_alb_vlan->vlan_id == vlan_id)) {
1718 bond->alb_info.current_alb_vlan = NULL;
1721 if (bond->alb_info.rlb_enabled) {
1722 rlb_clear_vlan(bond, vlan_id);