2 * Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of the GNU General Public License as published by the
6 * Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful, but
10 * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
11 * or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, see <http://www.gnu.org/licenses/>.
17 * The full GNU General Public License is included in this distribution in the
18 * file called LICENSE.
22 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
24 #include <linux/skbuff.h>
25 #include <linux/netdevice.h>
26 #include <linux/etherdevice.h>
27 #include <linux/pkt_sched.h>
28 #include <linux/spinlock.h>
29 #include <linux/slab.h>
30 #include <linux/timer.h>
32 #include <linux/ipv6.h>
33 #include <linux/if_arp.h>
34 #include <linux/if_ether.h>
35 #include <linux/if_bonding.h>
36 #include <linux/if_vlan.h>
41 #include <asm/byteorder.h>
47 #ifndef __long_aligned
48 #define __long_aligned __attribute__((aligned((sizeof(long)))))
50 static const u8 mac_bcast[ETH_ALEN] __long_aligned = {
51 0xff, 0xff, 0xff, 0xff, 0xff, 0xff
53 static const u8 mac_v6_allmcast[ETH_ALEN] __long_aligned = {
54 0x33, 0x33, 0x00, 0x00, 0x00, 0x01
56 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
63 u8 padding[ETH_ZLEN - ETH_HLEN];
68 __be16 prot_addr_space;
72 u8 mac_src[ETH_ALEN]; /* sender hardware address */
73 __be32 ip_src; /* sender IP address */
74 u8 mac_dst[ETH_ALEN]; /* target hardware address */
75 __be32 ip_dst; /* target IP address */
79 static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb)
81 return (struct arp_pkt *)skb_network_header(skb);
84 /* Forward declaration */
85 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];
103 /*********************** tlb specific functions ***************************/
105 static inline void _lock_tx_hashtbl_bh(struct bonding *bond)
107 spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
110 static inline void _unlock_tx_hashtbl_bh(struct bonding *bond)
112 spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
115 static inline void _lock_tx_hashtbl(struct bonding *bond)
117 spin_lock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
120 static inline void _unlock_tx_hashtbl(struct bonding *bond)
122 spin_unlock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
125 /* Caller must hold tx_hashtbl lock */
126 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
129 entry->load_history = 1 + entry->tx_bytes /
130 BOND_TLB_REBALANCE_INTERVAL;
134 entry->tx_slave = NULL;
135 entry->next = TLB_NULL_INDEX;
136 entry->prev = TLB_NULL_INDEX;
139 static inline void tlb_init_slave(struct slave *slave)
141 SLAVE_TLB_INFO(slave).load = 0;
142 SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
145 /* Caller must hold bond lock for read, BH disabled */
146 static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
149 struct tlb_client_info *tx_hash_table;
152 /* clear slave from tx_hashtbl */
153 tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
155 /* skip this if we've already freed the tx hash table */
157 index = SLAVE_TLB_INFO(slave).head;
158 while (index != TLB_NULL_INDEX) {
159 u32 next_index = tx_hash_table[index].next;
160 tlb_init_table_entry(&tx_hash_table[index], save_load);
165 tlb_init_slave(slave);
168 /* Caller must hold bond lock for read */
169 static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
172 _lock_tx_hashtbl_bh(bond);
173 __tlb_clear_slave(bond, slave, save_load);
174 _unlock_tx_hashtbl_bh(bond);
177 /* Must be called before starting the monitor timer */
178 static int tlb_initialize(struct bonding *bond)
180 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
181 int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
182 struct tlb_client_info *new_hashtbl;
185 new_hashtbl = kzalloc(size, GFP_KERNEL);
189 _lock_tx_hashtbl_bh(bond);
191 bond_info->tx_hashtbl = new_hashtbl;
193 for (i = 0; i < TLB_HASH_TABLE_SIZE; i++)
194 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
196 _unlock_tx_hashtbl_bh(bond);
201 /* Must be called only after all slaves have been released */
202 static void tlb_deinitialize(struct bonding *bond)
204 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
206 _lock_tx_hashtbl_bh(bond);
208 kfree(bond_info->tx_hashtbl);
209 bond_info->tx_hashtbl = NULL;
211 _unlock_tx_hashtbl_bh(bond);
214 static long long compute_gap(struct slave *slave)
216 return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
217 (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
220 /* Caller must hold bond lock for read */
221 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
223 struct slave *slave, *least_loaded;
224 struct list_head *iter;
230 /* Find the slave with the largest gap */
231 bond_for_each_slave_rcu(bond, slave, iter) {
232 if (bond_slave_can_tx(slave)) {
233 long long gap = compute_gap(slave);
236 least_loaded = slave;
245 static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
248 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
249 struct tlb_client_info *hash_table;
250 struct slave *assigned_slave;
252 hash_table = bond_info->tx_hashtbl;
253 assigned_slave = hash_table[hash_index].tx_slave;
254 if (!assigned_slave) {
255 assigned_slave = tlb_get_least_loaded_slave(bond);
257 if (assigned_slave) {
258 struct tlb_slave_info *slave_info =
259 &(SLAVE_TLB_INFO(assigned_slave));
260 u32 next_index = slave_info->head;
262 hash_table[hash_index].tx_slave = assigned_slave;
263 hash_table[hash_index].next = next_index;
264 hash_table[hash_index].prev = TLB_NULL_INDEX;
266 if (next_index != TLB_NULL_INDEX)
267 hash_table[next_index].prev = hash_index;
269 slave_info->head = hash_index;
271 hash_table[hash_index].load_history;
276 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 ether_addr_copy(client_info->mac_dst, arp->mac_src);
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 /* We received an ARP from arp->ip_src.
358 * We might have used this IP address previously (on the bonding host
359 * itself or on a system that is bridged together with the bond).
360 * However, if arp->mac_src is different than what is stored in
361 * rx_hashtbl, some other host is now using the IP and we must prevent
362 * sending out client updates with this IP address and the old MAC
364 * Clean up all hash table entries that have this address as ip_src but
365 * have a different mac_src.
367 rlb_purge_src_ip(bond, arp);
369 if (arp->op_code == htons(ARPOP_REPLY)) {
370 /* update rx hash table for this ARP */
371 rlb_update_entry_from_arp(bond, arp);
372 pr_debug("Server received an ARP Reply from client\n");
375 return RX_HANDLER_ANOTHER;
378 /* Caller must hold bond lock for read */
379 static struct slave *rlb_next_rx_slave(struct bonding *bond)
381 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
382 struct slave *before = NULL, *rx_slave = NULL, *slave;
383 struct list_head *iter;
386 bond_for_each_slave(bond, slave, iter) {
387 if (!bond_slave_can_tx(slave))
390 if (!before || before->speed < slave->speed)
393 if (!rx_slave || rx_slave->speed < slave->speed)
396 if (slave == bond_info->rx_slave)
399 /* we didn't find anything after the current or we have something
400 * better before and up to the current slave
402 if (!rx_slave || (before && rx_slave->speed < before->speed))
406 bond_info->rx_slave = rx_slave;
411 /* Caller must hold rcu_read_lock() for read */
412 static struct slave *__rlb_next_rx_slave(struct bonding *bond)
414 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
415 struct slave *before = NULL, *rx_slave = NULL, *slave;
416 struct list_head *iter;
419 bond_for_each_slave_rcu(bond, slave, iter) {
420 if (!bond_slave_can_tx(slave))
423 if (!before || before->speed < slave->speed)
426 if (!rx_slave || rx_slave->speed < slave->speed)
429 if (slave == bond_info->rx_slave)
432 /* we didn't find anything after the current or we have something
433 * better before and up to the current slave
435 if (!rx_slave || (before && rx_slave->speed < before->speed))
439 bond_info->rx_slave = rx_slave;
444 /* teach the switch the mac of a disabled slave
445 * on the primary for fault tolerance
447 * Caller must hold bond->curr_slave_lock for write or bond lock for write
449 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
451 if (!bond->curr_active_slave)
454 if (!bond->alb_info.primary_is_promisc) {
455 if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1))
456 bond->alb_info.primary_is_promisc = 1;
458 bond->alb_info.primary_is_promisc = 0;
461 bond->alb_info.rlb_promisc_timeout_counter = 0;
463 alb_send_learning_packets(bond->curr_active_slave, addr, true);
466 /* slave being removed should not be active at this point
468 * Caller must hold rtnl.
470 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
472 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
473 struct rlb_client_info *rx_hash_table;
474 u32 index, next_index;
476 /* clear slave from rx_hashtbl */
477 _lock_rx_hashtbl_bh(bond);
479 rx_hash_table = bond_info->rx_hashtbl;
480 index = bond_info->rx_hashtbl_used_head;
481 for (; index != RLB_NULL_INDEX; index = next_index) {
482 next_index = rx_hash_table[index].used_next;
483 if (rx_hash_table[index].slave == slave) {
484 struct slave *assigned_slave = rlb_next_rx_slave(bond);
486 if (assigned_slave) {
487 rx_hash_table[index].slave = assigned_slave;
488 if (!ether_addr_equal_64bits(rx_hash_table[index].mac_dst,
490 bond_info->rx_hashtbl[index].ntt = 1;
491 bond_info->rx_ntt = 1;
492 /* A slave has been removed from the
493 * table because it is either disabled
494 * or being released. We must retry the
495 * update to avoid clients from not
496 * being updated & disconnecting when
499 bond_info->rlb_update_retry_counter =
502 } else { /* there is no active slave */
503 rx_hash_table[index].slave = NULL;
508 _unlock_rx_hashtbl_bh(bond);
510 write_lock_bh(&bond->curr_slave_lock);
512 if (slave != bond->curr_active_slave)
513 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
515 write_unlock_bh(&bond->curr_slave_lock);
518 static void rlb_update_client(struct rlb_client_info *client_info)
522 if (!client_info->slave)
525 for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
528 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
530 client_info->slave->dev,
532 client_info->mac_dst,
533 client_info->slave->dev->dev_addr,
534 client_info->mac_dst);
536 pr_err("%s: Error: failed to create an ARP packet\n",
537 client_info->slave->bond->dev->name);
541 skb->dev = client_info->slave->dev;
543 if (client_info->vlan_id) {
544 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), client_info->vlan_id);
546 pr_err("%s: Error: failed to insert VLAN tag\n",
547 client_info->slave->bond->dev->name);
556 /* sends ARP REPLIES that update the clients that need updating */
557 static void rlb_update_rx_clients(struct bonding *bond)
559 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
560 struct rlb_client_info *client_info;
563 _lock_rx_hashtbl_bh(bond);
565 hash_index = bond_info->rx_hashtbl_used_head;
566 for (; hash_index != RLB_NULL_INDEX;
567 hash_index = client_info->used_next) {
568 client_info = &(bond_info->rx_hashtbl[hash_index]);
569 if (client_info->ntt) {
570 rlb_update_client(client_info);
571 if (bond_info->rlb_update_retry_counter == 0)
572 client_info->ntt = 0;
576 /* do not update the entries again until this counter is zero so that
577 * not to confuse the clients.
579 bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
581 _unlock_rx_hashtbl_bh(bond);
584 /* The slave was assigned a new mac address - update the clients */
585 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
587 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
588 struct rlb_client_info *client_info;
592 _lock_rx_hashtbl_bh(bond);
594 hash_index = bond_info->rx_hashtbl_used_head;
595 for (; hash_index != RLB_NULL_INDEX;
596 hash_index = client_info->used_next) {
597 client_info = &(bond_info->rx_hashtbl[hash_index]);
599 if ((client_info->slave == slave) &&
600 !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
601 client_info->ntt = 1;
606 /* update the team's flag only after the whole iteration */
608 bond_info->rx_ntt = 1;
609 /* fasten the change */
610 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
613 _unlock_rx_hashtbl_bh(bond);
616 /* mark all clients using src_ip to be updated */
617 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
619 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
620 struct rlb_client_info *client_info;
623 _lock_rx_hashtbl(bond);
625 hash_index = bond_info->rx_hashtbl_used_head;
626 for (; hash_index != RLB_NULL_INDEX;
627 hash_index = client_info->used_next) {
628 client_info = &(bond_info->rx_hashtbl[hash_index]);
630 if (!client_info->slave) {
631 pr_err("%s: Error: found a client with no channel in the client's hash table\n",
635 /*update all clients using this src_ip, that are not assigned
636 * to the team's address (curr_active_slave) and have a known
637 * unicast mac address.
639 if ((client_info->ip_src == src_ip) &&
640 !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
641 bond->dev->dev_addr) &&
642 !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
643 client_info->ntt = 1;
644 bond_info->rx_ntt = 1;
648 _unlock_rx_hashtbl(bond);
651 /* Caller must hold both bond and ptr locks for read */
652 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
654 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
655 struct arp_pkt *arp = arp_pkt(skb);
656 struct slave *assigned_slave, *curr_active_slave;
657 struct rlb_client_info *client_info;
660 _lock_rx_hashtbl(bond);
662 curr_active_slave = rcu_dereference(bond->curr_active_slave);
664 hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
665 client_info = &(bond_info->rx_hashtbl[hash_index]);
667 if (client_info->assigned) {
668 if ((client_info->ip_src == arp->ip_src) &&
669 (client_info->ip_dst == arp->ip_dst)) {
670 /* the entry is already assigned to this client */
671 if (!ether_addr_equal_64bits(arp->mac_dst, mac_bcast)) {
672 /* update mac address from arp */
673 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
675 ether_addr_copy(client_info->mac_src, arp->mac_src);
677 assigned_slave = client_info->slave;
678 if (assigned_slave) {
679 _unlock_rx_hashtbl(bond);
680 return assigned_slave;
683 /* the entry is already assigned to some other client,
684 * move the old client to primary (curr_active_slave) so
685 * that the new client can be assigned to this entry.
687 if (bond->curr_active_slave &&
688 client_info->slave != curr_active_slave) {
689 client_info->slave = curr_active_slave;
690 rlb_update_client(client_info);
694 /* assign a new slave */
695 assigned_slave = __rlb_next_rx_slave(bond);
697 if (assigned_slave) {
698 if (!(client_info->assigned &&
699 client_info->ip_src == arp->ip_src)) {
700 /* ip_src is going to be updated,
701 * fix the src hash list
703 u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
704 sizeof(arp->ip_src));
705 rlb_src_unlink(bond, hash_index);
706 rlb_src_link(bond, hash_src, hash_index);
709 client_info->ip_src = arp->ip_src;
710 client_info->ip_dst = arp->ip_dst;
711 /* arp->mac_dst is broadcast for arp reqeusts.
712 * will be updated with clients actual unicast mac address
713 * upon receiving an arp reply.
715 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
716 ether_addr_copy(client_info->mac_src, arp->mac_src);
717 client_info->slave = assigned_slave;
719 if (!ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
720 client_info->ntt = 1;
721 bond->alb_info.rx_ntt = 1;
723 client_info->ntt = 0;
726 if (vlan_get_tag(skb, &client_info->vlan_id))
727 client_info->vlan_id = 0;
729 if (!client_info->assigned) {
730 u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
731 bond_info->rx_hashtbl_used_head = hash_index;
732 client_info->used_next = prev_tbl_head;
733 if (prev_tbl_head != RLB_NULL_INDEX) {
734 bond_info->rx_hashtbl[prev_tbl_head].used_prev =
737 client_info->assigned = 1;
741 _unlock_rx_hashtbl(bond);
743 return assigned_slave;
746 /* chooses (and returns) transmit channel for arp reply
747 * does not choose channel for other arp types since they are
748 * sent on the curr_active_slave
750 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
752 struct arp_pkt *arp = arp_pkt(skb);
753 struct slave *tx_slave = NULL;
755 /* Don't modify or load balance ARPs that do not originate locally
756 * (e.g.,arrive via a bridge).
758 if (!bond_slave_has_mac_rcu(bond, arp->mac_src))
761 if (arp->op_code == htons(ARPOP_REPLY)) {
762 /* the arp must be sent on the selected
765 tx_slave = rlb_choose_channel(skb, bond);
767 ether_addr_copy(arp->mac_src, tx_slave->dev->dev_addr);
768 pr_debug("Server sent ARP Reply packet\n");
769 } else if (arp->op_code == htons(ARPOP_REQUEST)) {
770 /* Create an entry in the rx_hashtbl for this client as a
772 * When the arp reply is received the entry will be updated
773 * with the correct unicast address of the client.
775 rlb_choose_channel(skb, bond);
777 /* The ARP reply packets must be delayed so that
778 * they can cancel out the influence of the ARP request.
780 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
782 /* arp requests are broadcast and are sent on the primary
783 * the arp request will collapse all clients on the subnet to
784 * the primary slave. We must register these clients to be
785 * updated with their assigned mac.
787 rlb_req_update_subnet_clients(bond, arp->ip_src);
788 pr_debug("Server sent ARP Request packet\n");
794 /* Caller must hold bond lock for read */
795 static void rlb_rebalance(struct bonding *bond)
797 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
798 struct slave *assigned_slave;
799 struct rlb_client_info *client_info;
803 _lock_rx_hashtbl_bh(bond);
806 hash_index = bond_info->rx_hashtbl_used_head;
807 for (; hash_index != RLB_NULL_INDEX;
808 hash_index = client_info->used_next) {
809 client_info = &(bond_info->rx_hashtbl[hash_index]);
810 assigned_slave = __rlb_next_rx_slave(bond);
811 if (assigned_slave && (client_info->slave != assigned_slave)) {
812 client_info->slave = assigned_slave;
813 client_info->ntt = 1;
818 /* update the team's flag only after the whole iteration */
820 bond_info->rx_ntt = 1;
821 _unlock_rx_hashtbl_bh(bond);
824 /* Caller must hold rx_hashtbl lock */
825 static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
827 entry->used_next = RLB_NULL_INDEX;
828 entry->used_prev = RLB_NULL_INDEX;
833 static void rlb_init_table_entry_src(struct rlb_client_info *entry)
835 entry->src_first = RLB_NULL_INDEX;
836 entry->src_prev = RLB_NULL_INDEX;
837 entry->src_next = RLB_NULL_INDEX;
840 static void rlb_init_table_entry(struct rlb_client_info *entry)
842 memset(entry, 0, sizeof(struct rlb_client_info));
843 rlb_init_table_entry_dst(entry);
844 rlb_init_table_entry_src(entry);
847 static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
849 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
850 u32 next_index = bond_info->rx_hashtbl[index].used_next;
851 u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
853 if (index == bond_info->rx_hashtbl_used_head)
854 bond_info->rx_hashtbl_used_head = next_index;
855 if (prev_index != RLB_NULL_INDEX)
856 bond_info->rx_hashtbl[prev_index].used_next = next_index;
857 if (next_index != RLB_NULL_INDEX)
858 bond_info->rx_hashtbl[next_index].used_prev = prev_index;
861 /* unlink a rlb hash table entry from the src list */
862 static void rlb_src_unlink(struct bonding *bond, u32 index)
864 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
865 u32 next_index = bond_info->rx_hashtbl[index].src_next;
866 u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
868 bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
869 bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
871 if (next_index != RLB_NULL_INDEX)
872 bond_info->rx_hashtbl[next_index].src_prev = prev_index;
874 if (prev_index == RLB_NULL_INDEX)
877 /* is prev_index pointing to the head of this list? */
878 if (bond_info->rx_hashtbl[prev_index].src_first == index)
879 bond_info->rx_hashtbl[prev_index].src_first = next_index;
881 bond_info->rx_hashtbl[prev_index].src_next = next_index;
885 static void rlb_delete_table_entry(struct bonding *bond, u32 index)
887 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
888 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
890 rlb_delete_table_entry_dst(bond, index);
891 rlb_init_table_entry_dst(entry);
893 rlb_src_unlink(bond, index);
896 /* add the rx_hashtbl[ip_dst_hash] entry to the list
897 * of entries with identical ip_src_hash
899 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
901 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
904 bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
905 next = bond_info->rx_hashtbl[ip_src_hash].src_first;
906 bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
907 if (next != RLB_NULL_INDEX)
908 bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
909 bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
912 /* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does
913 * not match arp->mac_src */
914 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
916 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
917 u32 ip_src_hash = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
920 _lock_rx_hashtbl_bh(bond);
922 index = bond_info->rx_hashtbl[ip_src_hash].src_first;
923 while (index != RLB_NULL_INDEX) {
924 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
925 u32 next_index = entry->src_next;
926 if (entry->ip_src == arp->ip_src &&
927 !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
928 rlb_delete_table_entry(bond, index);
931 _unlock_rx_hashtbl_bh(bond);
934 static int rlb_initialize(struct bonding *bond)
936 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
937 struct rlb_client_info *new_hashtbl;
938 int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
941 new_hashtbl = kmalloc(size, GFP_KERNEL);
945 _lock_rx_hashtbl_bh(bond);
947 bond_info->rx_hashtbl = new_hashtbl;
949 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
951 for (i = 0; i < RLB_HASH_TABLE_SIZE; i++)
952 rlb_init_table_entry(bond_info->rx_hashtbl + i);
954 _unlock_rx_hashtbl_bh(bond);
956 /* register to receive ARPs */
957 bond->recv_probe = rlb_arp_recv;
962 static void rlb_deinitialize(struct bonding *bond)
964 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
966 _lock_rx_hashtbl_bh(bond);
968 kfree(bond_info->rx_hashtbl);
969 bond_info->rx_hashtbl = NULL;
970 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
972 _unlock_rx_hashtbl_bh(bond);
975 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
977 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
980 _lock_rx_hashtbl_bh(bond);
982 curr_index = bond_info->rx_hashtbl_used_head;
983 while (curr_index != RLB_NULL_INDEX) {
984 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
985 u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
987 if (curr->vlan_id == vlan_id)
988 rlb_delete_table_entry(bond, curr_index);
990 curr_index = next_index;
993 _unlock_rx_hashtbl_bh(bond);
996 /*********************** tlb/rlb shared functions *********************/
998 static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
999 __be16 vlan_proto, u16 vid)
1001 struct learning_pkt pkt;
1002 struct sk_buff *skb;
1003 int size = sizeof(struct learning_pkt);
1006 memset(&pkt, 0, size);
1007 ether_addr_copy(pkt.mac_dst, mac_addr);
1008 ether_addr_copy(pkt.mac_src, mac_addr);
1009 pkt.type = cpu_to_be16(ETH_P_LOOPBACK);
1011 skb = dev_alloc_skb(size);
1015 data = skb_put(skb, size);
1016 memcpy(data, &pkt, size);
1018 skb_reset_mac_header(skb);
1019 skb->network_header = skb->mac_header + ETH_HLEN;
1020 skb->protocol = pkt.type;
1021 skb->priority = TC_PRIO_CONTROL;
1022 skb->dev = slave->dev;
1025 skb = vlan_put_tag(skb, vlan_proto, vid);
1027 pr_err("%s: Error: failed to insert VLAN tag\n",
1028 slave->bond->dev->name);
1033 dev_queue_xmit(skb);
1036 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
1039 struct bonding *bond = bond_get_bond_by_slave(slave);
1040 struct net_device *upper;
1041 struct list_head *iter;
1044 alb_send_lp_vid(slave, mac_addr, 0, 0);
1046 /* loop through vlans and send one packet for each */
1048 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
1049 if (is_vlan_dev(upper) && vlan_get_encap_level(upper) == 0) {
1051 ether_addr_equal_64bits(mac_addr,
1053 alb_send_lp_vid(slave, mac_addr,
1054 vlan_dev_vlan_proto(upper),
1055 vlan_dev_vlan_id(upper));
1056 } else if (!strict_match) {
1057 alb_send_lp_vid(slave, upper->dev_addr,
1058 vlan_dev_vlan_proto(upper),
1059 vlan_dev_vlan_id(upper));
1066 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[])
1068 struct net_device *dev = slave->dev;
1069 struct sockaddr s_addr;
1071 if (BOND_MODE(slave->bond) == BOND_MODE_TLB) {
1072 memcpy(dev->dev_addr, addr, dev->addr_len);
1076 /* for rlb each slave must have a unique hw mac addresses so that */
1077 /* each slave will receive packets destined to a different mac */
1078 memcpy(s_addr.sa_data, addr, dev->addr_len);
1079 s_addr.sa_family = dev->type;
1080 if (dev_set_mac_address(dev, &s_addr)) {
1081 pr_err("%s: Error: dev_set_mac_address of dev %s failed!\n"
1082 "ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
1083 slave->bond->dev->name, dev->name);
1090 * Swap MAC addresses between two slaves.
1092 * Called with RTNL held, and no other locks.
1096 static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1098 u8 tmp_mac_addr[ETH_ALEN];
1100 ether_addr_copy(tmp_mac_addr, slave1->dev->dev_addr);
1101 alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr);
1102 alb_set_slave_mac_addr(slave2, tmp_mac_addr);
1107 * Send learning packets after MAC address swap.
1109 * Called with RTNL and no other locks
1111 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1112 struct slave *slave2)
1114 int slaves_state_differ = (bond_slave_can_tx(slave1) != bond_slave_can_tx(slave2));
1115 struct slave *disabled_slave = NULL;
1119 /* fasten the change in the switch */
1120 if (bond_slave_can_tx(slave1)) {
1121 alb_send_learning_packets(slave1, slave1->dev->dev_addr, false);
1122 if (bond->alb_info.rlb_enabled) {
1123 /* inform the clients that the mac address
1126 rlb_req_update_slave_clients(bond, slave1);
1129 disabled_slave = slave1;
1132 if (bond_slave_can_tx(slave2)) {
1133 alb_send_learning_packets(slave2, slave2->dev->dev_addr, false);
1134 if (bond->alb_info.rlb_enabled) {
1135 /* inform the clients that the mac address
1138 rlb_req_update_slave_clients(bond, slave2);
1141 disabled_slave = slave2;
1144 if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1145 /* A disabled slave was assigned an active mac addr */
1146 rlb_teach_disabled_mac_on_primary(bond,
1147 disabled_slave->dev->dev_addr);
1152 * alb_change_hw_addr_on_detach
1153 * @bond: bonding we're working on
1154 * @slave: the slave that was just detached
1156 * We assume that @slave was already detached from the slave list.
1158 * If @slave's permanent hw address is different both from its current
1159 * address and from @bond's address, then somewhere in the bond there's
1160 * a slave that has @slave's permanet address as its current address.
1161 * We'll make sure that that slave no longer uses @slave's permanent address.
1163 * Caller must hold RTNL and no other locks
1165 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1169 struct slave *found_slave;
1171 perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1172 slave->dev->dev_addr);
1173 perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1174 bond->dev->dev_addr);
1176 if (perm_curr_diff && perm_bond_diff) {
1177 found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1180 /* locking: needs RTNL and nothing else */
1181 alb_swap_mac_addr(slave, found_slave);
1182 alb_fasten_mac_swap(bond, slave, found_slave);
1188 * alb_handle_addr_collision_on_attach
1189 * @bond: bonding we're working on
1190 * @slave: the slave that was just attached
1192 * checks uniqueness of slave's mac address and handles the case the
1193 * new slave uses the bonds mac address.
1195 * If the permanent hw address of @slave is @bond's hw address, we need to
1196 * find a different hw address to give @slave, that isn't in use by any other
1197 * slave in the bond. This address must be, of course, one of the permanent
1198 * addresses of the other slaves.
1200 * We go over the slave list, and for each slave there we compare its
1201 * permanent hw address with the current address of all the other slaves.
1202 * If no match was found, then we've found a slave with a permanent address
1203 * that isn't used by any other slave in the bond, so we can assign it to
1206 * assumption: this function is called before @slave is attached to the
1209 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1211 struct slave *has_bond_addr = bond->curr_active_slave;
1212 struct slave *tmp_slave1, *free_mac_slave = NULL;
1213 struct list_head *iter;
1215 if (!bond_has_slaves(bond)) {
1216 /* this is the first slave */
1220 /* if slave's mac address differs from bond's mac address
1221 * check uniqueness of slave's mac address against the other
1222 * slaves in the bond.
1224 if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1225 if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1228 /* Try setting slave mac to bond address and fall-through
1229 to code handling that situation below... */
1230 alb_set_slave_mac_addr(slave, bond->dev->dev_addr);
1233 /* The slave's address is equal to the address of the bond.
1234 * Search for a spare address in the bond for this slave.
1236 bond_for_each_slave(bond, tmp_slave1, iter) {
1237 if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1238 /* no slave has tmp_slave1's perm addr
1241 free_mac_slave = tmp_slave1;
1245 if (!has_bond_addr) {
1246 if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1247 bond->dev->dev_addr)) {
1249 has_bond_addr = tmp_slave1;
1254 if (free_mac_slave) {
1255 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr);
1257 pr_warn("%s: Warning: the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1258 bond->dev->name, slave->dev->name,
1259 free_mac_slave->dev->name);
1261 } else if (has_bond_addr) {
1262 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",
1263 bond->dev->name, slave->dev->name);
1271 * alb_set_mac_address
1275 * In TLB mode all slaves are configured to the bond's hw address, but set
1276 * their dev_addr field to different addresses (based on their permanent hw
1279 * For each slave, this function sets the interface to the new address and then
1280 * changes its dev_addr field to its previous value.
1282 * Unwinding assumes bond's mac address has not yet changed.
1284 static int alb_set_mac_address(struct bonding *bond, void *addr)
1286 struct slave *slave, *rollback_slave;
1287 struct list_head *iter;
1289 char tmp_addr[ETH_ALEN];
1292 if (bond->alb_info.rlb_enabled)
1295 bond_for_each_slave(bond, slave, iter) {
1296 /* save net_device's current hw address */
1297 ether_addr_copy(tmp_addr, slave->dev->dev_addr);
1299 res = dev_set_mac_address(slave->dev, addr);
1301 /* restore net_device's hw address */
1302 ether_addr_copy(slave->dev->dev_addr, tmp_addr);
1311 memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1312 sa.sa_family = bond->dev->type;
1314 /* unwind from head to the slave that failed */
1315 bond_for_each_slave(bond, rollback_slave, iter) {
1316 if (rollback_slave == slave)
1318 ether_addr_copy(tmp_addr, rollback_slave->dev->dev_addr);
1319 dev_set_mac_address(rollback_slave->dev, &sa);
1320 ether_addr_copy(rollback_slave->dev->dev_addr, tmp_addr);
1326 /************************ exported alb funcions ************************/
1328 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1332 res = tlb_initialize(bond);
1337 bond->alb_info.rlb_enabled = 1;
1338 /* initialize rlb */
1339 res = rlb_initialize(bond);
1341 tlb_deinitialize(bond);
1345 bond->alb_info.rlb_enabled = 0;
1351 void bond_alb_deinitialize(struct bonding *bond)
1353 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1355 tlb_deinitialize(bond);
1357 if (bond_info->rlb_enabled)
1358 rlb_deinitialize(bond);
1361 static int bond_do_alb_xmit(struct sk_buff *skb, struct bonding *bond,
1362 struct slave *tx_slave)
1364 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1365 struct ethhdr *eth_data = eth_hdr(skb);
1368 /* unbalanced or unassigned, send through primary */
1369 tx_slave = rcu_dereference(bond->curr_active_slave);
1370 if (bond->params.tlb_dynamic_lb)
1371 bond_info->unbalanced_load += skb->len;
1374 if (tx_slave && bond_slave_can_tx(tx_slave)) {
1375 if (tx_slave != rcu_dereference(bond->curr_active_slave)) {
1376 ether_addr_copy(eth_data->h_source,
1377 tx_slave->dev->dev_addr);
1380 bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1384 if (tx_slave && bond->params.tlb_dynamic_lb) {
1385 _lock_tx_hashtbl(bond);
1386 __tlb_clear_slave(bond, tx_slave, 0);
1387 _unlock_tx_hashtbl(bond);
1390 /* no suitable interface, frame not sent */
1391 dev_kfree_skb_any(skb);
1393 return NETDEV_TX_OK;
1396 int bond_tlb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1398 struct bonding *bond = netdev_priv(bond_dev);
1399 struct ethhdr *eth_data;
1400 struct slave *tx_slave = NULL;
1403 skb_reset_mac_header(skb);
1404 eth_data = eth_hdr(skb);
1406 /* Do not TX balance any multicast or broadcast */
1407 if (!is_multicast_ether_addr(eth_data->h_dest)) {
1408 switch (skb->protocol) {
1409 case htons(ETH_P_IP):
1410 case htons(ETH_P_IPX):
1411 /* In case of IPX, it will falback to L2 hash */
1412 case htons(ETH_P_IPV6):
1413 hash_index = bond_xmit_hash(bond, skb);
1414 if (bond->params.tlb_dynamic_lb) {
1415 tx_slave = tlb_choose_channel(bond,
1419 struct list_head *iter;
1420 int idx = hash_index % bond->slave_cnt;
1422 bond_for_each_slave_rcu(bond, tx_slave, iter)
1429 return bond_do_alb_xmit(skb, bond, tx_slave);
1432 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1434 struct bonding *bond = netdev_priv(bond_dev);
1435 struct ethhdr *eth_data;
1436 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1437 struct slave *tx_slave = NULL;
1438 static const __be32 ip_bcast = htonl(0xffffffff);
1440 bool do_tx_balance = true;
1442 const u8 *hash_start = NULL;
1443 struct ipv6hdr *ip6hdr;
1445 skb_reset_mac_header(skb);
1446 eth_data = eth_hdr(skb);
1448 switch (ntohs(skb->protocol)) {
1450 const struct iphdr *iph = ip_hdr(skb);
1452 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast) ||
1453 (iph->daddr == ip_bcast) ||
1454 (iph->protocol == IPPROTO_IGMP)) {
1455 do_tx_balance = false;
1458 hash_start = (char *)&(iph->daddr);
1459 hash_size = sizeof(iph->daddr);
1463 /* IPv6 doesn't really use broadcast mac address, but leave
1464 * that here just in case.
1466 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast)) {
1467 do_tx_balance = false;
1471 /* IPv6 uses all-nodes multicast as an equivalent to
1472 * broadcasts in IPv4.
1474 if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1475 do_tx_balance = false;
1479 /* Additianally, DAD probes should not be tx-balanced as that
1480 * will lead to false positives for duplicate addresses and
1481 * prevent address configuration from working.
1483 ip6hdr = ipv6_hdr(skb);
1484 if (ipv6_addr_any(&ip6hdr->saddr)) {
1485 do_tx_balance = false;
1489 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1490 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1493 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1494 /* something is wrong with this packet */
1495 do_tx_balance = false;
1499 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1500 /* The only protocol worth balancing in
1501 * this family since it has an "ARP" like
1504 do_tx_balance = false;
1508 hash_start = (char *)eth_data->h_dest;
1509 hash_size = ETH_ALEN;
1512 do_tx_balance = false;
1513 if (bond_info->rlb_enabled)
1514 tx_slave = rlb_arp_xmit(skb, bond);
1517 do_tx_balance = false;
1521 if (do_tx_balance) {
1522 hash_index = _simple_hash(hash_start, hash_size);
1523 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1526 return bond_do_alb_xmit(skb, bond, tx_slave);
1529 void bond_alb_monitor(struct work_struct *work)
1531 struct bonding *bond = container_of(work, struct bonding,
1533 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1534 struct list_head *iter;
1535 struct slave *slave;
1537 if (!bond_has_slaves(bond)) {
1538 bond_info->tx_rebalance_counter = 0;
1539 bond_info->lp_counter = 0;
1545 bond_info->tx_rebalance_counter++;
1546 bond_info->lp_counter++;
1548 /* send learning packets */
1549 if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1552 /* change of curr_active_slave involves swapping of mac addresses.
1553 * in order to avoid this swapping from happening while
1554 * sending the learning packets, the curr_slave_lock must be held for
1557 read_lock(&bond->curr_slave_lock);
1559 bond_for_each_slave_rcu(bond, slave, iter) {
1560 /* If updating current_active, use all currently
1561 * user mac addreses (!strict_match). Otherwise, only
1562 * use mac of the slave device.
1564 strict_match = (slave != bond->curr_active_slave);
1565 alb_send_learning_packets(slave, slave->dev->dev_addr,
1569 read_unlock(&bond->curr_slave_lock);
1571 bond_info->lp_counter = 0;
1574 /* rebalance tx traffic */
1575 if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1577 read_lock(&bond->curr_slave_lock);
1579 bond_for_each_slave_rcu(bond, slave, iter) {
1580 tlb_clear_slave(bond, slave, 1);
1581 if (slave == bond->curr_active_slave) {
1582 SLAVE_TLB_INFO(slave).load =
1583 bond_info->unbalanced_load /
1584 BOND_TLB_REBALANCE_INTERVAL;
1585 bond_info->unbalanced_load = 0;
1589 read_unlock(&bond->curr_slave_lock);
1591 bond_info->tx_rebalance_counter = 0;
1594 /* handle rlb stuff */
1595 if (bond_info->rlb_enabled) {
1596 if (bond_info->primary_is_promisc &&
1597 (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1600 * dev_set_promiscuity requires rtnl and
1601 * nothing else. Avoid race with bond_close.
1604 if (!rtnl_trylock())
1607 bond_info->rlb_promisc_timeout_counter = 0;
1609 /* If the primary was set to promiscuous mode
1610 * because a slave was disabled then
1611 * it can now leave promiscuous mode.
1613 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1614 bond_info->primary_is_promisc = 0;
1620 if (bond_info->rlb_rebalance) {
1621 bond_info->rlb_rebalance = 0;
1622 rlb_rebalance(bond);
1625 /* check if clients need updating */
1626 if (bond_info->rx_ntt) {
1627 if (bond_info->rlb_update_delay_counter) {
1628 --bond_info->rlb_update_delay_counter;
1630 rlb_update_rx_clients(bond);
1631 if (bond_info->rlb_update_retry_counter)
1632 --bond_info->rlb_update_retry_counter;
1634 bond_info->rx_ntt = 0;
1640 queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1643 /* assumption: called before the slave is attached to the bond
1644 * and not locked by the bond lock
1646 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1650 res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr);
1654 res = alb_handle_addr_collision_on_attach(bond, slave);
1658 tlb_init_slave(slave);
1660 /* order a rebalance ASAP */
1661 bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1663 if (bond->alb_info.rlb_enabled)
1664 bond->alb_info.rlb_rebalance = 1;
1670 * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1673 * Caller must hold RTNL and no other locks
1675 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1677 if (bond_has_slaves(bond))
1678 alb_change_hw_addr_on_detach(bond, slave);
1680 tlb_clear_slave(bond, slave, 0);
1682 if (bond->alb_info.rlb_enabled) {
1683 bond->alb_info.rx_slave = NULL;
1684 rlb_clear_slave(bond, slave);
1688 /* Caller must hold bond lock for read */
1689 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1691 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1693 if (link == BOND_LINK_DOWN) {
1694 tlb_clear_slave(bond, slave, 0);
1695 if (bond->alb_info.rlb_enabled)
1696 rlb_clear_slave(bond, slave);
1697 } else if (link == BOND_LINK_UP) {
1698 /* order a rebalance ASAP */
1699 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1700 if (bond->alb_info.rlb_enabled) {
1701 bond->alb_info.rlb_rebalance = 1;
1702 /* If the updelay module parameter is smaller than the
1703 * forwarding delay of the switch the rebalance will
1704 * not work because the rebalance arp replies will
1705 * not be forwarded to the clients..
1712 * bond_alb_handle_active_change - assign new curr_active_slave
1713 * @bond: our bonding struct
1714 * @new_slave: new slave to assign
1716 * Set the bond->curr_active_slave to @new_slave and handle
1717 * mac address swapping and promiscuity changes as needed.
1719 * If new_slave is NULL, caller must hold curr_slave_lock or
1720 * bond->lock for write.
1722 * If new_slave is not NULL, caller must hold RTNL, curr_slave_lock
1723 * for write. Processing here may sleep, so no other locks may be held.
1725 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1726 __releases(&bond->curr_slave_lock)
1727 __acquires(&bond->curr_slave_lock)
1729 struct slave *swap_slave;
1731 if (bond->curr_active_slave == new_slave)
1734 if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1735 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1736 bond->alb_info.primary_is_promisc = 0;
1737 bond->alb_info.rlb_promisc_timeout_counter = 0;
1740 swap_slave = bond->curr_active_slave;
1741 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1743 if (!new_slave || !bond_has_slaves(bond))
1746 /* set the new curr_active_slave to the bonds mac address
1747 * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1750 swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1753 * Arrange for swap_slave and new_slave to temporarily be
1754 * ignored so we can mess with their MAC addresses without
1755 * fear of interference from transmit activity.
1758 tlb_clear_slave(bond, swap_slave, 1);
1759 tlb_clear_slave(bond, new_slave, 1);
1761 write_unlock_bh(&bond->curr_slave_lock);
1765 /* in TLB mode, the slave might flip down/up with the old dev_addr,
1766 * and thus filter bond->dev_addr's packets, so force bond's mac
1768 if (BOND_MODE(bond) == BOND_MODE_TLB) {
1770 u8 tmp_addr[ETH_ALEN];
1772 ether_addr_copy(tmp_addr, new_slave->dev->dev_addr);
1774 memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1775 sa.sa_family = bond->dev->type;
1776 /* we don't care if it can't change its mac, best effort */
1777 dev_set_mac_address(new_slave->dev, &sa);
1779 ether_addr_copy(new_slave->dev->dev_addr, tmp_addr);
1782 /* curr_active_slave must be set before calling alb_swap_mac_addr */
1784 /* swap mac address */
1785 alb_swap_mac_addr(swap_slave, new_slave);
1786 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1788 /* set the new_slave to the bond mac address */
1789 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr);
1790 alb_send_learning_packets(new_slave, bond->dev->dev_addr,
1794 write_lock_bh(&bond->curr_slave_lock);
1800 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1801 __acquires(&bond->lock)
1802 __releases(&bond->lock)
1804 struct bonding *bond = netdev_priv(bond_dev);
1805 struct sockaddr *sa = addr;
1806 struct slave *swap_slave;
1809 if (!is_valid_ether_addr(sa->sa_data))
1810 return -EADDRNOTAVAIL;
1812 res = alb_set_mac_address(bond, addr);
1816 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1818 /* If there is no curr_active_slave there is nothing else to do.
1819 * Otherwise we'll need to pass the new address to it and handle
1822 if (!bond->curr_active_slave)
1825 swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1828 alb_swap_mac_addr(swap_slave, bond->curr_active_slave);
1829 alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave);
1831 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr);
1833 read_lock(&bond->lock);
1834 alb_send_learning_packets(bond->curr_active_slave,
1835 bond_dev->dev_addr, false);
1836 if (bond->alb_info.rlb_enabled) {
1837 /* inform clients mac address has changed */
1838 rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1840 read_unlock(&bond->lock);
1846 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1848 if (bond->alb_info.rlb_enabled)
1849 rlb_clear_vlan(bond, vlan_id);