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.
24 * 2003/06/25 - Shmulik Hen <shmulik.hen at intel dot com>
25 * - Fixed signed/unsigned calculation errors that caused load sharing
26 * to collapse to one slave under very heavy UDP Tx stress.
28 * 2003/08/06 - Amir Noam <amir.noam at intel dot com>
29 * - Add support for setting bond's MAC address with special
30 * handling required for ALB/TLB.
32 * 2003/12/01 - Shmulik Hen <shmulik.hen at intel dot com>
33 * - Code cleanup and style changes
35 * 2003/12/30 - Amir Noam <amir.noam at intel dot com>
36 * - Fixed: Cannot remove and re-enslave the original active slave.
38 * 2004/01/14 - Shmulik Hen <shmulik.hen at intel dot com>
39 * - Add capability to tag self generated packets in ALB/TLB modes.
41 * 2005/12/02 - Michael O'Donnell <Michael.ODonnell at stratus dot com>
42 * - Stratus88746: tlb_clear_slave() must tlb_init_slave() while locked.
45 //#define BONDING_DEBUG 1
47 #include <linux/skbuff.h>
48 #include <linux/netdevice.h>
49 #include <linux/etherdevice.h>
50 #include <linux/pkt_sched.h>
51 #include <linux/spinlock.h>
52 #include <linux/slab.h>
53 #include <linux/timer.h>
55 #include <linux/ipv6.h>
56 #include <linux/if_arp.h>
57 #include <linux/if_ether.h>
58 #include <linux/if_bonding.h>
59 #include <linux/if_vlan.h>
62 #include <asm/byteorder.h>
67 #define ALB_TIMER_TICKS_PER_SEC 10 /* should be a divisor of HZ */
68 #define BOND_TLB_REBALANCE_INTERVAL 10 /* In seconds, periodic re-balancing.
69 * Used for division - never set
72 #define BOND_ALB_LP_INTERVAL 1 /* In seconds, periodic send of
73 * learning packets to the switch
76 #define BOND_TLB_REBALANCE_TICKS (BOND_TLB_REBALANCE_INTERVAL \
77 * ALB_TIMER_TICKS_PER_SEC)
79 #define BOND_ALB_LP_TICKS (BOND_ALB_LP_INTERVAL \
80 * ALB_TIMER_TICKS_PER_SEC)
82 #define TLB_HASH_TABLE_SIZE 256 /* The size of the clients hash table.
83 * Note that this value MUST NOT be smaller
84 * because the key hash table is BYTE wide !
88 #define TLB_NULL_INDEX 0xffffffff
89 #define MAX_LP_BURST 3
92 #define RLB_HASH_TABLE_SIZE 256
93 #define RLB_NULL_INDEX 0xffffffff
94 #define RLB_UPDATE_DELAY 2*ALB_TIMER_TICKS_PER_SEC /* 2 seconds */
95 #define RLB_ARP_BURST_SIZE 2
96 #define RLB_UPDATE_RETRY 3 /* 3-ticks - must be smaller than the rlb
97 * rebalance interval (5 min).
99 /* RLB_PROMISC_TIMEOUT = 10 sec equals the time that the current slave is
100 * promiscuous after failover
102 #define RLB_PROMISC_TIMEOUT 10*ALB_TIMER_TICKS_PER_SEC
104 static const u8 mac_bcast
[ETH_ALEN
] = {0xff,0xff,0xff,0xff,0xff,0xff};
105 static const int alb_delta_in_ticks
= HZ
/ ALB_TIMER_TICKS_PER_SEC
;
108 struct learning_pkt
{
109 u8 mac_dst
[ETH_ALEN
];
110 u8 mac_src
[ETH_ALEN
];
112 u8 padding
[ETH_ZLEN
- ETH_HLEN
];
121 u8 mac_src
[ETH_ALEN
]; /* sender hardware address */
122 u32 ip_src
; /* sender IP address */
123 u8 mac_dst
[ETH_ALEN
]; /* target hardware address */
124 u32 ip_dst
; /* target IP address */
128 /* Forward declaration */
129 static void alb_send_learning_packets(struct slave
*slave
, u8 mac_addr
[]);
131 static inline u8
_simple_hash(u8
*hash_start
, int hash_size
)
136 for (i
= 0; i
< hash_size
; i
++) {
137 hash
^= hash_start
[i
];
143 /*********************** tlb specific functions ***************************/
145 static inline void _lock_tx_hashtbl(struct bonding
*bond
)
147 spin_lock(&(BOND_ALB_INFO(bond
).tx_hashtbl_lock
));
150 static inline void _unlock_tx_hashtbl(struct bonding
*bond
)
152 spin_unlock(&(BOND_ALB_INFO(bond
).tx_hashtbl_lock
));
155 /* Caller must hold tx_hashtbl lock */
156 static inline void tlb_init_table_entry(struct tlb_client_info
*entry
, int save_load
)
159 entry
->load_history
= 1 + entry
->tx_bytes
/
160 BOND_TLB_REBALANCE_INTERVAL
;
164 entry
->tx_slave
= NULL
;
165 entry
->next
= TLB_NULL_INDEX
;
166 entry
->prev
= TLB_NULL_INDEX
;
169 static inline void tlb_init_slave(struct slave
*slave
)
171 SLAVE_TLB_INFO(slave
).load
= 0;
172 SLAVE_TLB_INFO(slave
).head
= TLB_NULL_INDEX
;
175 /* Caller must hold bond lock for read */
176 static void tlb_clear_slave(struct bonding
*bond
, struct slave
*slave
, int save_load
)
178 struct tlb_client_info
*tx_hash_table
;
181 _lock_tx_hashtbl(bond
);
183 /* clear slave from tx_hashtbl */
184 tx_hash_table
= BOND_ALB_INFO(bond
).tx_hashtbl
;
187 index
= SLAVE_TLB_INFO(slave
).head
;
188 while (index
!= TLB_NULL_INDEX
) {
189 u32 next_index
= tx_hash_table
[index
].next
;
190 tlb_init_table_entry(&tx_hash_table
[index
], save_load
);
195 tlb_init_slave(slave
); /* Stratus88746: do this before unlocking */
197 _unlock_tx_hashtbl(bond
);
200 /* Must be called before starting the monitor timer */
201 static int tlb_initialize(struct bonding
*bond
)
203 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
204 int size
= TLB_HASH_TABLE_SIZE
* sizeof(struct tlb_client_info
);
207 spin_lock_init(&(bond_info
->tx_hashtbl_lock
));
209 _lock_tx_hashtbl(bond
);
211 bond_info
->tx_hashtbl
= kmalloc(size
, GFP_KERNEL
);
212 if (!bond_info
->tx_hashtbl
) {
213 printk(KERN_ERR DRV_NAME
214 ": Error: %s: Failed to allocate TLB hash table\n",
216 _unlock_tx_hashtbl(bond
);
220 memset(bond_info
->tx_hashtbl
, 0, size
);
222 for (i
= 0; i
< TLB_HASH_TABLE_SIZE
; i
++) {
223 tlb_init_table_entry(&bond_info
->tx_hashtbl
[i
], 1);
226 _unlock_tx_hashtbl(bond
);
231 /* Must be called only after all slaves have been released */
232 static void tlb_deinitialize(struct bonding
*bond
)
234 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
236 _lock_tx_hashtbl(bond
);
238 kfree(bond_info
->tx_hashtbl
);
239 bond_info
->tx_hashtbl
= NULL
;
241 _unlock_tx_hashtbl(bond
);
244 /* Caller must hold bond lock for read */
245 static struct slave
*tlb_get_least_loaded_slave(struct bonding
*bond
)
247 struct slave
*slave
, *least_loaded
;
251 /* Find the first enabled slave */
252 bond_for_each_slave(bond
, slave
, i
) {
253 if (SLAVE_IS_OK(slave
)) {
263 least_loaded
= slave
;
264 max_gap
= (s64
)(slave
->speed
<< 20) - /* Convert to Megabit per sec */
265 (s64
)(SLAVE_TLB_INFO(slave
).load
<< 3); /* Bytes to bits */
267 /* Find the slave with the largest gap */
268 bond_for_each_slave_from(bond
, slave
, i
, least_loaded
) {
269 if (SLAVE_IS_OK(slave
)) {
270 s64 gap
= (s64
)(slave
->speed
<< 20) -
271 (s64
)(SLAVE_TLB_INFO(slave
).load
<< 3);
273 least_loaded
= slave
;
282 /* Caller must hold bond lock for read */
283 struct slave
*tlb_choose_channel(struct bonding
*bond
, u32 hash_index
, u32 skb_len
)
285 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
286 struct tlb_client_info
*hash_table
;
287 struct slave
*assigned_slave
;
289 _lock_tx_hashtbl(bond
);
291 hash_table
= bond_info
->tx_hashtbl
;
292 assigned_slave
= hash_table
[hash_index
].tx_slave
;
293 if (!assigned_slave
) {
294 assigned_slave
= tlb_get_least_loaded_slave(bond
);
296 if (assigned_slave
) {
297 struct tlb_slave_info
*slave_info
=
298 &(SLAVE_TLB_INFO(assigned_slave
));
299 u32 next_index
= slave_info
->head
;
301 hash_table
[hash_index
].tx_slave
= assigned_slave
;
302 hash_table
[hash_index
].next
= next_index
;
303 hash_table
[hash_index
].prev
= TLB_NULL_INDEX
;
305 if (next_index
!= TLB_NULL_INDEX
) {
306 hash_table
[next_index
].prev
= hash_index
;
309 slave_info
->head
= hash_index
;
311 hash_table
[hash_index
].load_history
;
315 if (assigned_slave
) {
316 hash_table
[hash_index
].tx_bytes
+= skb_len
;
319 _unlock_tx_hashtbl(bond
);
321 return assigned_slave
;
324 /*********************** rlb specific functions ***************************/
325 static inline void _lock_rx_hashtbl(struct bonding
*bond
)
327 spin_lock(&(BOND_ALB_INFO(bond
).rx_hashtbl_lock
));
330 static inline void _unlock_rx_hashtbl(struct bonding
*bond
)
332 spin_unlock(&(BOND_ALB_INFO(bond
).rx_hashtbl_lock
));
335 /* when an ARP REPLY is received from a client update its info
338 static void rlb_update_entry_from_arp(struct bonding
*bond
, struct arp_pkt
*arp
)
340 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
341 struct rlb_client_info
*client_info
;
344 _lock_rx_hashtbl(bond
);
346 hash_index
= _simple_hash((u8
*)&(arp
->ip_src
), sizeof(arp
->ip_src
));
347 client_info
= &(bond_info
->rx_hashtbl
[hash_index
]);
349 if ((client_info
->assigned
) &&
350 (client_info
->ip_src
== arp
->ip_dst
) &&
351 (client_info
->ip_dst
== arp
->ip_src
)) {
352 /* update the clients MAC address */
353 memcpy(client_info
->mac_dst
, arp
->mac_src
, ETH_ALEN
);
354 client_info
->ntt
= 1;
355 bond_info
->rx_ntt
= 1;
358 _unlock_rx_hashtbl(bond
);
361 static int rlb_arp_recv(struct sk_buff
*skb
, struct net_device
*bond_dev
, struct packet_type
*ptype
)
363 struct bonding
*bond
= bond_dev
->priv
;
364 struct arp_pkt
*arp
= (struct arp_pkt
*)skb
->data
;
365 int res
= NET_RX_DROP
;
367 if (!(bond_dev
->flags
& IFF_MASTER
)) {
372 dprintk("Packet has no ARP data\n");
376 if (skb
->len
< sizeof(struct arp_pkt
)) {
377 dprintk("Packet is too small to be an ARP\n");
381 if (arp
->op_code
== htons(ARPOP_REPLY
)) {
382 /* update rx hash table for this ARP */
383 rlb_update_entry_from_arp(bond
, arp
);
384 dprintk("Server received an ARP Reply from client\n");
387 res
= NET_RX_SUCCESS
;
395 /* Caller must hold bond lock for read */
396 static struct slave
*rlb_next_rx_slave(struct bonding
*bond
)
398 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
399 struct slave
*rx_slave
, *slave
, *start_at
;
402 if (bond_info
->next_rx_slave
) {
403 start_at
= bond_info
->next_rx_slave
;
405 start_at
= bond
->first_slave
;
410 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
411 if (SLAVE_IS_OK(slave
)) {
414 } else if (slave
->speed
> rx_slave
->speed
) {
421 bond_info
->next_rx_slave
= rx_slave
->next
;
427 /* teach the switch the mac of a disabled slave
428 * on the primary for fault tolerance
430 * Caller must hold bond->curr_slave_lock for write or bond lock for write
432 static void rlb_teach_disabled_mac_on_primary(struct bonding
*bond
, u8 addr
[])
434 if (!bond
->curr_active_slave
) {
438 if (!bond
->alb_info
.primary_is_promisc
) {
439 bond
->alb_info
.primary_is_promisc
= 1;
440 dev_set_promiscuity(bond
->curr_active_slave
->dev
, 1);
443 bond
->alb_info
.rlb_promisc_timeout_counter
= 0;
445 alb_send_learning_packets(bond
->curr_active_slave
, addr
);
448 /* slave being removed should not be active at this point
450 * Caller must hold bond lock for read
452 static void rlb_clear_slave(struct bonding
*bond
, struct slave
*slave
)
454 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
455 struct rlb_client_info
*rx_hash_table
;
456 u32 index
, next_index
;
458 /* clear slave from rx_hashtbl */
459 _lock_rx_hashtbl(bond
);
461 rx_hash_table
= bond_info
->rx_hashtbl
;
462 index
= bond_info
->rx_hashtbl_head
;
463 for (; index
!= RLB_NULL_INDEX
; index
= next_index
) {
464 next_index
= rx_hash_table
[index
].next
;
465 if (rx_hash_table
[index
].slave
== slave
) {
466 struct slave
*assigned_slave
= rlb_next_rx_slave(bond
);
468 if (assigned_slave
) {
469 rx_hash_table
[index
].slave
= assigned_slave
;
470 if (memcmp(rx_hash_table
[index
].mac_dst
,
471 mac_bcast
, ETH_ALEN
)) {
472 bond_info
->rx_hashtbl
[index
].ntt
= 1;
473 bond_info
->rx_ntt
= 1;
474 /* A slave has been removed from the
475 * table because it is either disabled
476 * or being released. We must retry the
477 * update to avoid clients from not
478 * being updated & disconnecting when
481 bond_info
->rlb_update_retry_counter
=
484 } else { /* there is no active slave */
485 rx_hash_table
[index
].slave
= NULL
;
490 _unlock_rx_hashtbl(bond
);
492 write_lock(&bond
->curr_slave_lock
);
494 if (slave
!= bond
->curr_active_slave
) {
495 rlb_teach_disabled_mac_on_primary(bond
, slave
->dev
->dev_addr
);
498 write_unlock(&bond
->curr_slave_lock
);
501 static void rlb_update_client(struct rlb_client_info
*client_info
)
505 if (!client_info
->slave
) {
509 for (i
= 0; i
< RLB_ARP_BURST_SIZE
; i
++) {
512 skb
= arp_create(ARPOP_REPLY
, ETH_P_ARP
,
514 client_info
->slave
->dev
,
516 client_info
->mac_dst
,
517 client_info
->slave
->dev
->dev_addr
,
518 client_info
->mac_dst
);
520 printk(KERN_ERR DRV_NAME
521 ": Error: failed to create an ARP packet\n");
525 skb
->dev
= client_info
->slave
->dev
;
527 if (client_info
->tag
) {
528 skb
= vlan_put_tag(skb
, client_info
->vlan_id
);
530 printk(KERN_ERR DRV_NAME
531 ": Error: failed to insert VLAN tag\n");
540 /* sends ARP REPLIES that update the clients that need updating */
541 static void rlb_update_rx_clients(struct bonding
*bond
)
543 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
544 struct rlb_client_info
*client_info
;
547 _lock_rx_hashtbl(bond
);
549 hash_index
= bond_info
->rx_hashtbl_head
;
550 for (; hash_index
!= RLB_NULL_INDEX
; hash_index
= client_info
->next
) {
551 client_info
= &(bond_info
->rx_hashtbl
[hash_index
]);
552 if (client_info
->ntt
) {
553 rlb_update_client(client_info
);
554 if (bond_info
->rlb_update_retry_counter
== 0) {
555 client_info
->ntt
= 0;
560 /* do not update the entries again untill this counter is zero so that
561 * not to confuse the clients.
563 bond_info
->rlb_update_delay_counter
= RLB_UPDATE_DELAY
;
565 _unlock_rx_hashtbl(bond
);
568 /* The slave was assigned a new mac address - update the clients */
569 static void rlb_req_update_slave_clients(struct bonding
*bond
, struct slave
*slave
)
571 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
572 struct rlb_client_info
*client_info
;
576 _lock_rx_hashtbl(bond
);
578 hash_index
= bond_info
->rx_hashtbl_head
;
579 for (; hash_index
!= RLB_NULL_INDEX
; hash_index
= client_info
->next
) {
580 client_info
= &(bond_info
->rx_hashtbl
[hash_index
]);
582 if ((client_info
->slave
== slave
) &&
583 memcmp(client_info
->mac_dst
, mac_bcast
, ETH_ALEN
)) {
584 client_info
->ntt
= 1;
589 // update the team's flag only after the whole iteration
591 bond_info
->rx_ntt
= 1;
593 bond_info
->rlb_update_retry_counter
= RLB_UPDATE_RETRY
;
596 _unlock_rx_hashtbl(bond
);
599 /* mark all clients using src_ip to be updated */
600 static void rlb_req_update_subnet_clients(struct bonding
*bond
, u32 src_ip
)
602 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
603 struct rlb_client_info
*client_info
;
606 _lock_rx_hashtbl(bond
);
608 hash_index
= bond_info
->rx_hashtbl_head
;
609 for (; hash_index
!= RLB_NULL_INDEX
; hash_index
= client_info
->next
) {
610 client_info
= &(bond_info
->rx_hashtbl
[hash_index
]);
612 if (!client_info
->slave
) {
613 printk(KERN_ERR DRV_NAME
614 ": Error: found a client with no channel in "
615 "the client's hash table\n");
618 /*update all clients using this src_ip, that are not assigned
619 * to the team's address (curr_active_slave) and have a known
620 * unicast mac address.
622 if ((client_info
->ip_src
== src_ip
) &&
623 memcmp(client_info
->slave
->dev
->dev_addr
,
624 bond
->dev
->dev_addr
, ETH_ALEN
) &&
625 memcmp(client_info
->mac_dst
, mac_bcast
, ETH_ALEN
)) {
626 client_info
->ntt
= 1;
627 bond_info
->rx_ntt
= 1;
631 _unlock_rx_hashtbl(bond
);
634 /* Caller must hold both bond and ptr locks for read */
635 struct slave
*rlb_choose_channel(struct sk_buff
*skb
, struct bonding
*bond
)
637 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
638 struct arp_pkt
*arp
= (struct arp_pkt
*)skb
->nh
.raw
;
639 struct slave
*assigned_slave
;
640 struct rlb_client_info
*client_info
;
643 _lock_rx_hashtbl(bond
);
645 hash_index
= _simple_hash((u8
*)&arp
->ip_dst
, sizeof(arp
->ip_src
));
646 client_info
= &(bond_info
->rx_hashtbl
[hash_index
]);
648 if (client_info
->assigned
) {
649 if ((client_info
->ip_src
== arp
->ip_src
) &&
650 (client_info
->ip_dst
== arp
->ip_dst
)) {
651 /* the entry is already assigned to this client */
652 if (memcmp(arp
->mac_dst
, mac_bcast
, ETH_ALEN
)) {
653 /* update mac address from arp */
654 memcpy(client_info
->mac_dst
, arp
->mac_dst
, ETH_ALEN
);
657 assigned_slave
= client_info
->slave
;
658 if (assigned_slave
) {
659 _unlock_rx_hashtbl(bond
);
660 return assigned_slave
;
663 /* the entry is already assigned to some other client,
664 * move the old client to primary (curr_active_slave) so
665 * that the new client can be assigned to this entry.
667 if (bond
->curr_active_slave
&&
668 client_info
->slave
!= bond
->curr_active_slave
) {
669 client_info
->slave
= bond
->curr_active_slave
;
670 rlb_update_client(client_info
);
674 /* assign a new slave */
675 assigned_slave
= rlb_next_rx_slave(bond
);
677 if (assigned_slave
) {
678 client_info
->ip_src
= arp
->ip_src
;
679 client_info
->ip_dst
= arp
->ip_dst
;
680 /* arp->mac_dst is broadcast for arp reqeusts.
681 * will be updated with clients actual unicast mac address
682 * upon receiving an arp reply.
684 memcpy(client_info
->mac_dst
, arp
->mac_dst
, ETH_ALEN
);
685 client_info
->slave
= assigned_slave
;
687 if (memcmp(client_info
->mac_dst
, mac_bcast
, ETH_ALEN
)) {
688 client_info
->ntt
= 1;
689 bond
->alb_info
.rx_ntt
= 1;
691 client_info
->ntt
= 0;
694 if (!list_empty(&bond
->vlan_list
)) {
695 unsigned short vlan_id
;
696 int res
= vlan_get_tag(skb
, &vlan_id
);
698 client_info
->tag
= 1;
699 client_info
->vlan_id
= vlan_id
;
703 if (!client_info
->assigned
) {
704 u32 prev_tbl_head
= bond_info
->rx_hashtbl_head
;
705 bond_info
->rx_hashtbl_head
= hash_index
;
706 client_info
->next
= prev_tbl_head
;
707 if (prev_tbl_head
!= RLB_NULL_INDEX
) {
708 bond_info
->rx_hashtbl
[prev_tbl_head
].prev
=
711 client_info
->assigned
= 1;
715 _unlock_rx_hashtbl(bond
);
717 return assigned_slave
;
720 /* chooses (and returns) transmit channel for arp reply
721 * does not choose channel for other arp types since they are
722 * sent on the curr_active_slave
724 static struct slave
*rlb_arp_xmit(struct sk_buff
*skb
, struct bonding
*bond
)
726 struct arp_pkt
*arp
= (struct arp_pkt
*)skb
->nh
.raw
;
727 struct slave
*tx_slave
= NULL
;
729 if (arp
->op_code
== __constant_htons(ARPOP_REPLY
)) {
730 /* the arp must be sent on the selected
733 tx_slave
= rlb_choose_channel(skb
, bond
);
735 memcpy(arp
->mac_src
,tx_slave
->dev
->dev_addr
, ETH_ALEN
);
737 dprintk("Server sent ARP Reply packet\n");
738 } else if (arp
->op_code
== __constant_htons(ARPOP_REQUEST
)) {
739 /* Create an entry in the rx_hashtbl for this client as a
741 * When the arp reply is received the entry will be updated
742 * with the correct unicast address of the client.
744 rlb_choose_channel(skb
, bond
);
746 /* The ARP relpy packets must be delayed so that
747 * they can cancel out the influence of the ARP request.
749 bond
->alb_info
.rlb_update_delay_counter
= RLB_UPDATE_DELAY
;
751 /* arp requests are broadcast and are sent on the primary
752 * the arp request will collapse all clients on the subnet to
753 * the primary slave. We must register these clients to be
754 * updated with their assigned mac.
756 rlb_req_update_subnet_clients(bond
, arp
->ip_src
);
757 dprintk("Server sent ARP Request packet\n");
763 /* Caller must hold bond lock for read */
764 static void rlb_rebalance(struct bonding
*bond
)
766 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
767 struct slave
*assigned_slave
;
768 struct rlb_client_info
*client_info
;
772 _lock_rx_hashtbl(bond
);
775 hash_index
= bond_info
->rx_hashtbl_head
;
776 for (; hash_index
!= RLB_NULL_INDEX
; hash_index
= client_info
->next
) {
777 client_info
= &(bond_info
->rx_hashtbl
[hash_index
]);
778 assigned_slave
= rlb_next_rx_slave(bond
);
779 if (assigned_slave
&& (client_info
->slave
!= assigned_slave
)) {
780 client_info
->slave
= assigned_slave
;
781 client_info
->ntt
= 1;
786 /* update the team's flag only after the whole iteration */
788 bond_info
->rx_ntt
= 1;
790 _unlock_rx_hashtbl(bond
);
793 /* Caller must hold rx_hashtbl lock */
794 static void rlb_init_table_entry(struct rlb_client_info
*entry
)
796 memset(entry
, 0, sizeof(struct rlb_client_info
));
797 entry
->next
= RLB_NULL_INDEX
;
798 entry
->prev
= RLB_NULL_INDEX
;
801 static int rlb_initialize(struct bonding
*bond
)
803 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
804 struct packet_type
*pk_type
= &(BOND_ALB_INFO(bond
).rlb_pkt_type
);
805 int size
= RLB_HASH_TABLE_SIZE
* sizeof(struct rlb_client_info
);
808 spin_lock_init(&(bond_info
->rx_hashtbl_lock
));
810 _lock_rx_hashtbl(bond
);
812 bond_info
->rx_hashtbl
= kmalloc(size
, GFP_KERNEL
);
813 if (!bond_info
->rx_hashtbl
) {
814 printk(KERN_ERR DRV_NAME
815 ": Error: %s: Failed to allocate RLB hash table\n",
817 _unlock_rx_hashtbl(bond
);
821 bond_info
->rx_hashtbl_head
= RLB_NULL_INDEX
;
823 for (i
= 0; i
< RLB_HASH_TABLE_SIZE
; i
++) {
824 rlb_init_table_entry(bond_info
->rx_hashtbl
+ i
);
827 _unlock_rx_hashtbl(bond
);
829 /*initialize packet type*/
830 pk_type
->type
= __constant_htons(ETH_P_ARP
);
831 pk_type
->dev
= bond
->dev
;
832 pk_type
->func
= rlb_arp_recv
;
833 pk_type
->data
= (void*)1; /* understand shared skbs */
835 /* register to receive ARPs */
836 dev_add_pack(pk_type
);
841 static void rlb_deinitialize(struct bonding
*bond
)
843 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
845 dev_remove_pack(&(bond_info
->rlb_pkt_type
));
847 _lock_rx_hashtbl(bond
);
849 kfree(bond_info
->rx_hashtbl
);
850 bond_info
->rx_hashtbl
= NULL
;
851 bond_info
->rx_hashtbl_head
= RLB_NULL_INDEX
;
853 _unlock_rx_hashtbl(bond
);
856 static void rlb_clear_vlan(struct bonding
*bond
, unsigned short vlan_id
)
858 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
861 _lock_rx_hashtbl(bond
);
863 curr_index
= bond_info
->rx_hashtbl_head
;
864 while (curr_index
!= RLB_NULL_INDEX
) {
865 struct rlb_client_info
*curr
= &(bond_info
->rx_hashtbl
[curr_index
]);
866 u32 next_index
= bond_info
->rx_hashtbl
[curr_index
].next
;
867 u32 prev_index
= bond_info
->rx_hashtbl
[curr_index
].prev
;
869 if (curr
->tag
&& (curr
->vlan_id
== vlan_id
)) {
870 if (curr_index
== bond_info
->rx_hashtbl_head
) {
871 bond_info
->rx_hashtbl_head
= next_index
;
873 if (prev_index
!= RLB_NULL_INDEX
) {
874 bond_info
->rx_hashtbl
[prev_index
].next
= next_index
;
876 if (next_index
!= RLB_NULL_INDEX
) {
877 bond_info
->rx_hashtbl
[next_index
].prev
= prev_index
;
880 rlb_init_table_entry(curr
);
883 curr_index
= next_index
;
886 _unlock_rx_hashtbl(bond
);
889 /*********************** tlb/rlb shared functions *********************/
891 static void alb_send_learning_packets(struct slave
*slave
, u8 mac_addr
[])
893 struct bonding
*bond
= bond_get_bond_by_slave(slave
);
894 struct learning_pkt pkt
;
895 int size
= sizeof(struct learning_pkt
);
898 memset(&pkt
, 0, size
);
899 memcpy(pkt
.mac_dst
, mac_addr
, ETH_ALEN
);
900 memcpy(pkt
.mac_src
, mac_addr
, ETH_ALEN
);
901 pkt
.type
= __constant_htons(ETH_P_LOOP
);
903 for (i
= 0; i
< MAX_LP_BURST
; i
++) {
907 skb
= dev_alloc_skb(size
);
912 data
= skb_put(skb
, size
);
913 memcpy(data
, &pkt
, size
);
916 skb
->nh
.raw
= data
+ ETH_HLEN
;
917 skb
->protocol
= pkt
.type
;
918 skb
->priority
= TC_PRIO_CONTROL
;
919 skb
->dev
= slave
->dev
;
921 if (!list_empty(&bond
->vlan_list
)) {
922 struct vlan_entry
*vlan
;
924 vlan
= bond_next_vlan(bond
,
925 bond
->alb_info
.current_alb_vlan
);
927 bond
->alb_info
.current_alb_vlan
= vlan
;
933 skb
= vlan_put_tag(skb
, vlan
->vlan_id
);
935 printk(KERN_ERR DRV_NAME
936 ": Error: failed to insert VLAN tag\n");
945 /* hw is a boolean parameter that determines whether we should try and
946 * set the hw address of the device as well as the hw address of the
949 static int alb_set_slave_mac_addr(struct slave
*slave
, u8 addr
[], int hw
)
951 struct net_device
*dev
= slave
->dev
;
952 struct sockaddr s_addr
;
955 memcpy(dev
->dev_addr
, addr
, dev
->addr_len
);
959 /* for rlb each slave must have a unique hw mac addresses so that */
960 /* each slave will receive packets destined to a different mac */
961 memcpy(s_addr
.sa_data
, addr
, dev
->addr_len
);
962 s_addr
.sa_family
= dev
->type
;
963 if (dev
->set_mac_address(dev
, &s_addr
)) {
964 printk(KERN_ERR DRV_NAME
965 ": Error: dev->set_mac_address of dev %s failed! ALB "
966 "mode requires that the base driver support setting "
967 "the hw address also when the network device's "
968 "interface is open\n",
975 /* Caller must hold bond lock for write or curr_slave_lock for write*/
976 static void alb_swap_mac_addr(struct bonding
*bond
, struct slave
*slave1
, struct slave
*slave2
)
978 struct slave
*disabled_slave
= NULL
;
979 u8 tmp_mac_addr
[ETH_ALEN
];
980 int slaves_state_differ
;
982 slaves_state_differ
= (SLAVE_IS_OK(slave1
) != SLAVE_IS_OK(slave2
));
984 memcpy(tmp_mac_addr
, slave1
->dev
->dev_addr
, ETH_ALEN
);
985 alb_set_slave_mac_addr(slave1
, slave2
->dev
->dev_addr
, bond
->alb_info
.rlb_enabled
);
986 alb_set_slave_mac_addr(slave2
, tmp_mac_addr
, bond
->alb_info
.rlb_enabled
);
988 /* fasten the change in the switch */
989 if (SLAVE_IS_OK(slave1
)) {
990 alb_send_learning_packets(slave1
, slave1
->dev
->dev_addr
);
991 if (bond
->alb_info
.rlb_enabled
) {
992 /* inform the clients that the mac address
995 rlb_req_update_slave_clients(bond
, slave1
);
998 disabled_slave
= slave1
;
1001 if (SLAVE_IS_OK(slave2
)) {
1002 alb_send_learning_packets(slave2
, slave2
->dev
->dev_addr
);
1003 if (bond
->alb_info
.rlb_enabled
) {
1004 /* inform the clients that the mac address
1007 rlb_req_update_slave_clients(bond
, slave2
);
1010 disabled_slave
= slave2
;
1013 if (bond
->alb_info
.rlb_enabled
&& slaves_state_differ
) {
1014 /* A disabled slave was assigned an active mac addr */
1015 rlb_teach_disabled_mac_on_primary(bond
,
1016 disabled_slave
->dev
->dev_addr
);
1021 * alb_change_hw_addr_on_detach
1022 * @bond: bonding we're working on
1023 * @slave: the slave that was just detached
1025 * We assume that @slave was already detached from the slave list.
1027 * If @slave's permanent hw address is different both from its current
1028 * address and from @bond's address, then somewhere in the bond there's
1029 * a slave that has @slave's permanet address as its current address.
1030 * We'll make sure that that slave no longer uses @slave's permanent address.
1032 * Caller must hold bond lock
1034 static void alb_change_hw_addr_on_detach(struct bonding
*bond
, struct slave
*slave
)
1039 perm_curr_diff
= memcmp(slave
->perm_hwaddr
,
1040 slave
->dev
->dev_addr
,
1042 perm_bond_diff
= memcmp(slave
->perm_hwaddr
,
1043 bond
->dev
->dev_addr
,
1046 if (perm_curr_diff
&& perm_bond_diff
) {
1047 struct slave
*tmp_slave
;
1050 bond_for_each_slave(bond
, tmp_slave
, i
) {
1051 if (!memcmp(slave
->perm_hwaddr
,
1052 tmp_slave
->dev
->dev_addr
,
1060 alb_swap_mac_addr(bond
, slave
, tmp_slave
);
1066 * alb_handle_addr_collision_on_attach
1067 * @bond: bonding we're working on
1068 * @slave: the slave that was just attached
1070 * checks uniqueness of slave's mac address and handles the case the
1071 * new slave uses the bonds mac address.
1073 * If the permanent hw address of @slave is @bond's hw address, we need to
1074 * find a different hw address to give @slave, that isn't in use by any other
1075 * slave in the bond. This address must be, of course, one of the premanent
1076 * addresses of the other slaves.
1078 * We go over the slave list, and for each slave there we compare its
1079 * permanent hw address with the current address of all the other slaves.
1080 * If no match was found, then we've found a slave with a permanent address
1081 * that isn't used by any other slave in the bond, so we can assign it to
1084 * assumption: this function is called before @slave is attached to the
1087 * caller must hold the bond lock for write since the mac addresses are compared
1088 * and may be swapped.
1090 static int alb_handle_addr_collision_on_attach(struct bonding
*bond
, struct slave
*slave
)
1092 struct slave
*tmp_slave1
, *tmp_slave2
, *free_mac_slave
;
1093 struct slave
*has_bond_addr
= bond
->curr_active_slave
;
1094 int i
, j
, found
= 0;
1096 if (bond
->slave_cnt
== 0) {
1097 /* this is the first slave */
1101 /* if slave's mac address differs from bond's mac address
1102 * check uniqueness of slave's mac address against the other
1103 * slaves in the bond.
1105 if (memcmp(slave
->perm_hwaddr
, bond
->dev
->dev_addr
, ETH_ALEN
)) {
1106 bond_for_each_slave(bond
, tmp_slave1
, i
) {
1107 if (!memcmp(tmp_slave1
->dev
->dev_addr
, slave
->dev
->dev_addr
,
1115 /* a slave was found that is using the mac address
1118 printk(KERN_ERR DRV_NAME
1119 ": Error: the hw address of slave %s is not "
1120 "unique - cannot enslave it!",
1128 /* The slave's address is equal to the address of the bond.
1129 * Search for a spare address in the bond for this slave.
1131 free_mac_slave
= NULL
;
1133 bond_for_each_slave(bond
, tmp_slave1
, i
) {
1135 bond_for_each_slave(bond
, tmp_slave2
, j
) {
1136 if (!memcmp(tmp_slave1
->perm_hwaddr
,
1137 tmp_slave2
->dev
->dev_addr
,
1145 /* no slave has tmp_slave1's perm addr
1148 free_mac_slave
= tmp_slave1
;
1152 if (!has_bond_addr
) {
1153 if (!memcmp(tmp_slave1
->dev
->dev_addr
,
1154 bond
->dev
->dev_addr
,
1157 has_bond_addr
= tmp_slave1
;
1162 if (free_mac_slave
) {
1163 alb_set_slave_mac_addr(slave
, free_mac_slave
->perm_hwaddr
,
1164 bond
->alb_info
.rlb_enabled
);
1166 printk(KERN_WARNING DRV_NAME
1167 ": Warning: the hw address of slave %s is in use by "
1168 "the bond; giving it the hw address of %s\n",
1169 slave
->dev
->name
, free_mac_slave
->dev
->name
);
1171 } else if (has_bond_addr
) {
1172 printk(KERN_ERR DRV_NAME
1173 ": Error: the hw address of slave %s is in use by the "
1174 "bond; couldn't find a slave with a free hw address to "
1175 "give it (this should not have happened)\n",
1184 * alb_set_mac_address
1188 * In TLB mode all slaves are configured to the bond's hw address, but set
1189 * their dev_addr field to different addresses (based on their permanent hw
1192 * For each slave, this function sets the interface to the new address and then
1193 * changes its dev_addr field to its previous value.
1195 * Unwinding assumes bond's mac address has not yet changed.
1197 static int alb_set_mac_address(struct bonding
*bond
, void *addr
)
1200 struct slave
*slave
, *stop_at
;
1201 char tmp_addr
[ETH_ALEN
];
1205 if (bond
->alb_info
.rlb_enabled
) {
1209 bond_for_each_slave(bond
, slave
, i
) {
1210 if (slave
->dev
->set_mac_address
== NULL
) {
1215 /* save net_device's current hw address */
1216 memcpy(tmp_addr
, slave
->dev
->dev_addr
, ETH_ALEN
);
1218 res
= slave
->dev
->set_mac_address(slave
->dev
, addr
);
1220 /* restore net_device's hw address */
1221 memcpy(slave
->dev
->dev_addr
, tmp_addr
, ETH_ALEN
);
1231 memcpy(sa
.sa_data
, bond
->dev
->dev_addr
, bond
->dev
->addr_len
);
1232 sa
.sa_family
= bond
->dev
->type
;
1234 /* unwind from head to the slave that failed */
1236 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
1237 memcpy(tmp_addr
, slave
->dev
->dev_addr
, ETH_ALEN
);
1238 slave
->dev
->set_mac_address(slave
->dev
, &sa
);
1239 memcpy(slave
->dev
->dev_addr
, tmp_addr
, ETH_ALEN
);
1245 /************************ exported alb funcions ************************/
1247 int bond_alb_initialize(struct bonding
*bond
, int rlb_enabled
)
1251 res
= tlb_initialize(bond
);
1257 bond
->alb_info
.rlb_enabled
= 1;
1258 /* initialize rlb */
1259 res
= rlb_initialize(bond
);
1261 tlb_deinitialize(bond
);
1269 void bond_alb_deinitialize(struct bonding
*bond
)
1271 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
1273 tlb_deinitialize(bond
);
1275 if (bond_info
->rlb_enabled
) {
1276 rlb_deinitialize(bond
);
1280 int bond_alb_xmit(struct sk_buff
*skb
, struct net_device
*bond_dev
)
1282 struct bonding
*bond
= bond_dev
->priv
;
1283 struct ethhdr
*eth_data
= (struct ethhdr
*)(skb
->mac
.raw
= skb
->data
);
1284 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
1285 struct slave
*tx_slave
= NULL
;
1286 static u32 ip_bcast
= 0xffffffff;
1288 int do_tx_balance
= 1;
1290 u8
*hash_start
= NULL
;
1293 /* make sure that the curr_active_slave and the slaves list do
1294 * not change during tx
1296 read_lock(&bond
->lock
);
1297 read_lock(&bond
->curr_slave_lock
);
1299 if (!BOND_IS_OK(bond
)) {
1303 switch (ntohs(skb
->protocol
)) {
1305 if ((memcmp(eth_data
->h_dest
, mac_bcast
, ETH_ALEN
) == 0) ||
1306 (skb
->nh
.iph
->daddr
== ip_bcast
)) {
1310 hash_start
= (char*)&(skb
->nh
.iph
->daddr
);
1311 hash_size
= sizeof(skb
->nh
.iph
->daddr
);
1314 if (memcmp(eth_data
->h_dest
, mac_bcast
, ETH_ALEN
) == 0) {
1319 hash_start
= (char*)&(skb
->nh
.ipv6h
->daddr
);
1320 hash_size
= sizeof(skb
->nh
.ipv6h
->daddr
);
1323 if (skb
->nh
.ipxh
->ipx_checksum
!=
1324 __constant_htons(IPX_NO_CHECKSUM
)) {
1325 /* something is wrong with this packet */
1330 if (skb
->nh
.ipxh
->ipx_type
!= IPX_TYPE_NCP
) {
1331 /* The only protocol worth balancing in
1332 * this family since it has an "ARP" like
1339 hash_start
= (char*)eth_data
->h_dest
;
1340 hash_size
= ETH_ALEN
;
1344 if (bond_info
->rlb_enabled
) {
1345 tx_slave
= rlb_arp_xmit(skb
, bond
);
1353 if (do_tx_balance
) {
1354 hash_index
= _simple_hash(hash_start
, hash_size
);
1355 tx_slave
= tlb_choose_channel(bond
, hash_index
, skb
->len
);
1359 /* unbalanced or unassigned, send through primary */
1360 tx_slave
= bond
->curr_active_slave
;
1361 bond_info
->unbalanced_load
+= skb
->len
;
1364 if (tx_slave
&& SLAVE_IS_OK(tx_slave
)) {
1365 if (tx_slave
!= bond
->curr_active_slave
) {
1366 memcpy(eth_data
->h_source
,
1367 tx_slave
->dev
->dev_addr
,
1371 res
= bond_dev_queue_xmit(bond
, skb
, tx_slave
->dev
);
1374 tlb_clear_slave(bond
, tx_slave
, 0);
1380 /* no suitable interface, frame not sent */
1383 read_unlock(&bond
->curr_slave_lock
);
1384 read_unlock(&bond
->lock
);
1388 void bond_alb_monitor(struct bonding
*bond
)
1390 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
1391 struct slave
*slave
;
1394 read_lock(&bond
->lock
);
1396 if (bond
->kill_timers
) {
1400 if (bond
->slave_cnt
== 0) {
1401 bond_info
->tx_rebalance_counter
= 0;
1402 bond_info
->lp_counter
= 0;
1406 bond_info
->tx_rebalance_counter
++;
1407 bond_info
->lp_counter
++;
1409 /* send learning packets */
1410 if (bond_info
->lp_counter
>= BOND_ALB_LP_TICKS
) {
1411 /* change of curr_active_slave involves swapping of mac addresses.
1412 * in order to avoid this swapping from happening while
1413 * sending the learning packets, the curr_slave_lock must be held for
1416 read_lock(&bond
->curr_slave_lock
);
1418 bond_for_each_slave(bond
, slave
, i
) {
1419 alb_send_learning_packets(slave
,slave
->dev
->dev_addr
);
1422 read_unlock(&bond
->curr_slave_lock
);
1424 bond_info
->lp_counter
= 0;
1427 /* rebalance tx traffic */
1428 if (bond_info
->tx_rebalance_counter
>= BOND_TLB_REBALANCE_TICKS
) {
1430 read_lock(&bond
->curr_slave_lock
);
1432 bond_for_each_slave(bond
, slave
, i
) {
1433 tlb_clear_slave(bond
, slave
, 1);
1434 if (slave
== bond
->curr_active_slave
) {
1435 SLAVE_TLB_INFO(slave
).load
=
1436 bond_info
->unbalanced_load
/
1437 BOND_TLB_REBALANCE_INTERVAL
;
1438 bond_info
->unbalanced_load
= 0;
1442 read_unlock(&bond
->curr_slave_lock
);
1444 bond_info
->tx_rebalance_counter
= 0;
1447 /* handle rlb stuff */
1448 if (bond_info
->rlb_enabled
) {
1449 /* the following code changes the promiscuity of the
1450 * the curr_active_slave. It needs to be locked with a
1451 * write lock to protect from other code that also
1452 * sets the promiscuity.
1454 write_lock(&bond
->curr_slave_lock
);
1456 if (bond_info
->primary_is_promisc
&&
1457 (++bond_info
->rlb_promisc_timeout_counter
>= RLB_PROMISC_TIMEOUT
)) {
1459 bond_info
->rlb_promisc_timeout_counter
= 0;
1461 /* If the primary was set to promiscuous mode
1462 * because a slave was disabled then
1463 * it can now leave promiscuous mode.
1465 dev_set_promiscuity(bond
->curr_active_slave
->dev
, -1);
1466 bond_info
->primary_is_promisc
= 0;
1469 write_unlock(&bond
->curr_slave_lock
);
1471 if (bond_info
->rlb_rebalance
) {
1472 bond_info
->rlb_rebalance
= 0;
1473 rlb_rebalance(bond
);
1476 /* check if clients need updating */
1477 if (bond_info
->rx_ntt
) {
1478 if (bond_info
->rlb_update_delay_counter
) {
1479 --bond_info
->rlb_update_delay_counter
;
1481 rlb_update_rx_clients(bond
);
1482 if (bond_info
->rlb_update_retry_counter
) {
1483 --bond_info
->rlb_update_retry_counter
;
1485 bond_info
->rx_ntt
= 0;
1492 mod_timer(&(bond_info
->alb_timer
), jiffies
+ alb_delta_in_ticks
);
1494 read_unlock(&bond
->lock
);
1497 /* assumption: called before the slave is attached to the bond
1498 * and not locked by the bond lock
1500 int bond_alb_init_slave(struct bonding
*bond
, struct slave
*slave
)
1504 res
= alb_set_slave_mac_addr(slave
, slave
->perm_hwaddr
,
1505 bond
->alb_info
.rlb_enabled
);
1510 /* caller must hold the bond lock for write since the mac addresses
1511 * are compared and may be swapped.
1513 write_lock_bh(&bond
->lock
);
1515 res
= alb_handle_addr_collision_on_attach(bond
, slave
);
1517 write_unlock_bh(&bond
->lock
);
1523 tlb_init_slave(slave
);
1525 /* order a rebalance ASAP */
1526 bond
->alb_info
.tx_rebalance_counter
= BOND_TLB_REBALANCE_TICKS
;
1528 if (bond
->alb_info
.rlb_enabled
) {
1529 bond
->alb_info
.rlb_rebalance
= 1;
1535 /* Caller must hold bond lock for write */
1536 void bond_alb_deinit_slave(struct bonding
*bond
, struct slave
*slave
)
1538 if (bond
->slave_cnt
> 1) {
1539 alb_change_hw_addr_on_detach(bond
, slave
);
1542 tlb_clear_slave(bond
, slave
, 0);
1544 if (bond
->alb_info
.rlb_enabled
) {
1545 bond
->alb_info
.next_rx_slave
= NULL
;
1546 rlb_clear_slave(bond
, slave
);
1550 /* Caller must hold bond lock for read */
1551 void bond_alb_handle_link_change(struct bonding
*bond
, struct slave
*slave
, char link
)
1553 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
1555 if (link
== BOND_LINK_DOWN
) {
1556 tlb_clear_slave(bond
, slave
, 0);
1557 if (bond
->alb_info
.rlb_enabled
) {
1558 rlb_clear_slave(bond
, slave
);
1560 } else if (link
== BOND_LINK_UP
) {
1561 /* order a rebalance ASAP */
1562 bond_info
->tx_rebalance_counter
= BOND_TLB_REBALANCE_TICKS
;
1563 if (bond
->alb_info
.rlb_enabled
) {
1564 bond
->alb_info
.rlb_rebalance
= 1;
1565 /* If the updelay module parameter is smaller than the
1566 * forwarding delay of the switch the rebalance will
1567 * not work because the rebalance arp replies will
1568 * not be forwarded to the clients..
1575 * bond_alb_handle_active_change - assign new curr_active_slave
1576 * @bond: our bonding struct
1577 * @new_slave: new slave to assign
1579 * Set the bond->curr_active_slave to @new_slave and handle
1580 * mac address swapping and promiscuity changes as needed.
1582 * Caller must hold bond curr_slave_lock for write (or bond lock for write)
1584 void bond_alb_handle_active_change(struct bonding
*bond
, struct slave
*new_slave
)
1586 struct slave
*swap_slave
;
1589 if (bond
->curr_active_slave
== new_slave
) {
1593 if (bond
->curr_active_slave
&& bond
->alb_info
.primary_is_promisc
) {
1594 dev_set_promiscuity(bond
->curr_active_slave
->dev
, -1);
1595 bond
->alb_info
.primary_is_promisc
= 0;
1596 bond
->alb_info
.rlb_promisc_timeout_counter
= 0;
1599 swap_slave
= bond
->curr_active_slave
;
1600 bond
->curr_active_slave
= new_slave
;
1602 if (!new_slave
|| (bond
->slave_cnt
== 0)) {
1606 /* set the new curr_active_slave to the bonds mac address
1607 * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1610 struct slave
*tmp_slave
;
1611 /* find slave that is holding the bond's mac address */
1612 bond_for_each_slave(bond
, tmp_slave
, i
) {
1613 if (!memcmp(tmp_slave
->dev
->dev_addr
,
1614 bond
->dev
->dev_addr
, ETH_ALEN
)) {
1615 swap_slave
= tmp_slave
;
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
,
1628 bond
->alb_info
.rlb_enabled
);
1629 /* fasten bond mac on new current slave */
1630 alb_send_learning_packets(new_slave
, bond
->dev
->dev_addr
);
1634 int bond_alb_set_mac_address(struct net_device
*bond_dev
, void *addr
)
1636 struct bonding
*bond
= bond_dev
->priv
;
1637 struct sockaddr
*sa
= addr
;
1638 struct slave
*slave
, *swap_slave
;
1642 if (!is_valid_ether_addr(sa
->sa_data
)) {
1643 return -EADDRNOTAVAIL
;
1646 res
= alb_set_mac_address(bond
, addr
);
1651 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
1653 /* If there is no curr_active_slave there is nothing else to do.
1654 * Otherwise we'll need to pass the new address to it and handle
1657 if (!bond
->curr_active_slave
) {
1663 bond_for_each_slave(bond
, slave
, i
) {
1664 if (!memcmp(slave
->dev
->dev_addr
, bond_dev
->dev_addr
, ETH_ALEN
)) {
1671 alb_swap_mac_addr(bond
, swap_slave
, bond
->curr_active_slave
);
1673 alb_set_slave_mac_addr(bond
->curr_active_slave
, bond_dev
->dev_addr
,
1674 bond
->alb_info
.rlb_enabled
);
1676 alb_send_learning_packets(bond
->curr_active_slave
, bond_dev
->dev_addr
);
1677 if (bond
->alb_info
.rlb_enabled
) {
1678 /* inform clients mac address has changed */
1679 rlb_req_update_slave_clients(bond
, bond
->curr_active_slave
);
1686 void bond_alb_clear_vlan(struct bonding
*bond
, unsigned short vlan_id
)
1688 if (bond
->alb_info
.current_alb_vlan
&&
1689 (bond
->alb_info
.current_alb_vlan
->vlan_id
== vlan_id
)) {
1690 bond
->alb_info
.current_alb_vlan
= NULL
;
1693 if (bond
->alb_info
.rlb_enabled
) {
1694 rlb_clear_vlan(bond
, vlan_id
);