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 BONDING_DEBUG 1
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>
47 #define ALB_TIMER_TICKS_PER_SEC 10 /* should be a divisor of HZ */
48 #define BOND_TLB_REBALANCE_INTERVAL 10 /* In seconds, periodic re-balancing.
49 * Used for division - never set
52 #define BOND_ALB_LP_INTERVAL 1 /* In seconds, periodic send of
53 * learning packets to the switch
56 #define BOND_TLB_REBALANCE_TICKS (BOND_TLB_REBALANCE_INTERVAL \
57 * ALB_TIMER_TICKS_PER_SEC)
59 #define BOND_ALB_LP_TICKS (BOND_ALB_LP_INTERVAL \
60 * ALB_TIMER_TICKS_PER_SEC)
62 #define TLB_HASH_TABLE_SIZE 256 /* The size of the clients hash table.
63 * Note that this value MUST NOT be smaller
64 * because the key hash table is BYTE wide !
68 #define TLB_NULL_INDEX 0xffffffff
69 #define MAX_LP_BURST 3
72 #define RLB_HASH_TABLE_SIZE 256
73 #define RLB_NULL_INDEX 0xffffffff
74 #define RLB_UPDATE_DELAY 2*ALB_TIMER_TICKS_PER_SEC /* 2 seconds */
75 #define RLB_ARP_BURST_SIZE 2
76 #define RLB_UPDATE_RETRY 3 /* 3-ticks - must be smaller than the rlb
77 * rebalance interval (5 min).
79 /* RLB_PROMISC_TIMEOUT = 10 sec equals the time that the current slave is
80 * promiscuous after failover
82 #define RLB_PROMISC_TIMEOUT 10*ALB_TIMER_TICKS_PER_SEC
84 static const u8 mac_bcast
[ETH_ALEN
] = {0xff,0xff,0xff,0xff,0xff,0xff};
85 static const u8 mac_v6_allmcast
[ETH_ALEN
] = {0x33,0x33,0x00,0x00,0x00,0x01};
86 static const int alb_delta_in_ticks
= HZ
/ ALB_TIMER_TICKS_PER_SEC
;
93 u8 padding
[ETH_ZLEN
- ETH_HLEN
];
98 __be16 prot_addr_space
;
102 u8 mac_src
[ETH_ALEN
]; /* sender hardware address */
103 __be32 ip_src
; /* sender IP address */
104 u8 mac_dst
[ETH_ALEN
]; /* target hardware address */
105 __be32 ip_dst
; /* target IP address */
109 static inline struct arp_pkt
*arp_pkt(const struct sk_buff
*skb
)
111 return (struct arp_pkt
*)skb_network_header(skb
);
114 /* Forward declaration */
115 static void alb_send_learning_packets(struct slave
*slave
, u8 mac_addr
[]);
117 static inline u8
_simple_hash(const u8
*hash_start
, int hash_size
)
122 for (i
= 0; i
< hash_size
; i
++) {
123 hash
^= hash_start
[i
];
129 /*********************** tlb specific functions ***************************/
131 static inline void _lock_tx_hashtbl(struct bonding
*bond
)
133 spin_lock_bh(&(BOND_ALB_INFO(bond
).tx_hashtbl_lock
));
136 static inline void _unlock_tx_hashtbl(struct bonding
*bond
)
138 spin_unlock_bh(&(BOND_ALB_INFO(bond
).tx_hashtbl_lock
));
141 /* Caller must hold tx_hashtbl lock */
142 static inline void tlb_init_table_entry(struct tlb_client_info
*entry
, int save_load
)
145 entry
->load_history
= 1 + entry
->tx_bytes
/
146 BOND_TLB_REBALANCE_INTERVAL
;
150 entry
->tx_slave
= NULL
;
151 entry
->next
= TLB_NULL_INDEX
;
152 entry
->prev
= TLB_NULL_INDEX
;
155 static inline void tlb_init_slave(struct slave
*slave
)
157 SLAVE_TLB_INFO(slave
).load
= 0;
158 SLAVE_TLB_INFO(slave
).head
= TLB_NULL_INDEX
;
161 /* Caller must hold bond lock for read */
162 static void tlb_clear_slave(struct bonding
*bond
, struct slave
*slave
, int save_load
)
164 struct tlb_client_info
*tx_hash_table
;
167 _lock_tx_hashtbl(bond
);
169 /* clear slave from tx_hashtbl */
170 tx_hash_table
= BOND_ALB_INFO(bond
).tx_hashtbl
;
172 index
= SLAVE_TLB_INFO(slave
).head
;
173 while (index
!= TLB_NULL_INDEX
) {
174 u32 next_index
= tx_hash_table
[index
].next
;
175 tlb_init_table_entry(&tx_hash_table
[index
], save_load
);
179 tlb_init_slave(slave
);
181 _unlock_tx_hashtbl(bond
);
184 /* Must be called before starting the monitor timer */
185 static int tlb_initialize(struct bonding
*bond
)
187 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
188 int size
= TLB_HASH_TABLE_SIZE
* sizeof(struct tlb_client_info
);
189 struct tlb_client_info
*new_hashtbl
;
192 spin_lock_init(&(bond_info
->tx_hashtbl_lock
));
194 new_hashtbl
= kzalloc(size
, GFP_KERNEL
);
196 printk(KERN_ERR DRV_NAME
197 ": %s: Error: Failed to allocate TLB hash table\n",
201 _lock_tx_hashtbl(bond
);
203 bond_info
->tx_hashtbl
= new_hashtbl
;
205 for (i
= 0; i
< TLB_HASH_TABLE_SIZE
; i
++) {
206 tlb_init_table_entry(&bond_info
->tx_hashtbl
[i
], 1);
209 _unlock_tx_hashtbl(bond
);
214 /* Must be called only after all slaves have been released */
215 static void tlb_deinitialize(struct bonding
*bond
)
217 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
219 _lock_tx_hashtbl(bond
);
221 kfree(bond_info
->tx_hashtbl
);
222 bond_info
->tx_hashtbl
= NULL
;
224 _unlock_tx_hashtbl(bond
);
227 /* Caller must hold bond lock for read */
228 static struct slave
*tlb_get_least_loaded_slave(struct bonding
*bond
)
230 struct slave
*slave
, *least_loaded
;
234 /* Find the first enabled slave */
235 bond_for_each_slave(bond
, slave
, i
) {
236 if (SLAVE_IS_OK(slave
)) {
246 least_loaded
= slave
;
247 max_gap
= (s64
)(slave
->speed
<< 20) - /* Convert to Megabit per sec */
248 (s64
)(SLAVE_TLB_INFO(slave
).load
<< 3); /* Bytes to bits */
250 /* Find the slave with the largest gap */
251 bond_for_each_slave_from(bond
, slave
, i
, least_loaded
) {
252 if (SLAVE_IS_OK(slave
)) {
253 s64 gap
= (s64
)(slave
->speed
<< 20) -
254 (s64
)(SLAVE_TLB_INFO(slave
).load
<< 3);
256 least_loaded
= slave
;
265 /* Caller must hold bond lock for read */
266 static struct slave
*tlb_choose_channel(struct bonding
*bond
, u32 hash_index
, u32 skb_len
)
268 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
269 struct tlb_client_info
*hash_table
;
270 struct slave
*assigned_slave
;
272 _lock_tx_hashtbl(bond
);
274 hash_table
= bond_info
->tx_hashtbl
;
275 assigned_slave
= hash_table
[hash_index
].tx_slave
;
276 if (!assigned_slave
) {
277 assigned_slave
= tlb_get_least_loaded_slave(bond
);
279 if (assigned_slave
) {
280 struct tlb_slave_info
*slave_info
=
281 &(SLAVE_TLB_INFO(assigned_slave
));
282 u32 next_index
= slave_info
->head
;
284 hash_table
[hash_index
].tx_slave
= assigned_slave
;
285 hash_table
[hash_index
].next
= next_index
;
286 hash_table
[hash_index
].prev
= TLB_NULL_INDEX
;
288 if (next_index
!= TLB_NULL_INDEX
) {
289 hash_table
[next_index
].prev
= hash_index
;
292 slave_info
->head
= hash_index
;
294 hash_table
[hash_index
].load_history
;
298 if (assigned_slave
) {
299 hash_table
[hash_index
].tx_bytes
+= skb_len
;
302 _unlock_tx_hashtbl(bond
);
304 return assigned_slave
;
307 /*********************** rlb specific functions ***************************/
308 static inline void _lock_rx_hashtbl(struct bonding
*bond
)
310 spin_lock_bh(&(BOND_ALB_INFO(bond
).rx_hashtbl_lock
));
313 static inline void _unlock_rx_hashtbl(struct bonding
*bond
)
315 spin_unlock_bh(&(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(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 /* update the clients MAC address */
336 memcpy(client_info
->mac_dst
, arp
->mac_src
, ETH_ALEN
);
337 client_info
->ntt
= 1;
338 bond_info
->rx_ntt
= 1;
341 _unlock_rx_hashtbl(bond
);
344 static int rlb_arp_recv(struct sk_buff
*skb
, struct net_device
*bond_dev
, struct packet_type
*ptype
, struct net_device
*orig_dev
)
346 struct bonding
*bond
= bond_dev
->priv
;
347 struct arp_pkt
*arp
= (struct arp_pkt
*)skb
->data
;
348 int res
= NET_RX_DROP
;
350 if (dev_net(bond_dev
) != &init_net
)
353 if (!(bond_dev
->flags
& IFF_MASTER
))
357 dprintk("Packet has no ARP data\n");
361 if (skb
->len
< sizeof(struct arp_pkt
)) {
362 dprintk("Packet is too small to be an ARP\n");
366 if (arp
->op_code
== htons(ARPOP_REPLY
)) {
367 /* update rx hash table for this ARP */
368 rlb_update_entry_from_arp(bond
, arp
);
369 dprintk("Server received an ARP Reply from client\n");
372 res
= NET_RX_SUCCESS
;
380 /* Caller must hold bond lock for read */
381 static struct slave
*rlb_next_rx_slave(struct bonding
*bond
)
383 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
384 struct slave
*rx_slave
, *slave
, *start_at
;
387 if (bond_info
->next_rx_slave
) {
388 start_at
= bond_info
->next_rx_slave
;
390 start_at
= bond
->first_slave
;
395 bond_for_each_slave_from(bond
, slave
, i
, start_at
) {
396 if (SLAVE_IS_OK(slave
)) {
399 } else if (slave
->speed
> rx_slave
->speed
) {
406 bond_info
->next_rx_slave
= rx_slave
->next
;
412 /* teach the switch the mac of a disabled slave
413 * on the primary for fault tolerance
415 * Caller must hold bond->curr_slave_lock for write or bond lock for write
417 static void rlb_teach_disabled_mac_on_primary(struct bonding
*bond
, u8 addr
[])
419 if (!bond
->curr_active_slave
) {
423 if (!bond
->alb_info
.primary_is_promisc
) {
424 if (!dev_set_promiscuity(bond
->curr_active_slave
->dev
, 1))
425 bond
->alb_info
.primary_is_promisc
= 1;
427 bond
->alb_info
.primary_is_promisc
= 0;
430 bond
->alb_info
.rlb_promisc_timeout_counter
= 0;
432 alb_send_learning_packets(bond
->curr_active_slave
, addr
);
435 /* slave being removed should not be active at this point
437 * Caller must hold bond lock for read
439 static void rlb_clear_slave(struct bonding
*bond
, struct slave
*slave
)
441 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
442 struct rlb_client_info
*rx_hash_table
;
443 u32 index
, next_index
;
445 /* clear slave from rx_hashtbl */
446 _lock_rx_hashtbl(bond
);
448 rx_hash_table
= bond_info
->rx_hashtbl
;
449 index
= bond_info
->rx_hashtbl_head
;
450 for (; index
!= RLB_NULL_INDEX
; index
= next_index
) {
451 next_index
= rx_hash_table
[index
].next
;
452 if (rx_hash_table
[index
].slave
== slave
) {
453 struct slave
*assigned_slave
= rlb_next_rx_slave(bond
);
455 if (assigned_slave
) {
456 rx_hash_table
[index
].slave
= assigned_slave
;
457 if (memcmp(rx_hash_table
[index
].mac_dst
,
458 mac_bcast
, ETH_ALEN
)) {
459 bond_info
->rx_hashtbl
[index
].ntt
= 1;
460 bond_info
->rx_ntt
= 1;
461 /* A slave has been removed from the
462 * table because it is either disabled
463 * or being released. We must retry the
464 * update to avoid clients from not
465 * being updated & disconnecting when
468 bond_info
->rlb_update_retry_counter
=
471 } else { /* there is no active slave */
472 rx_hash_table
[index
].slave
= NULL
;
477 _unlock_rx_hashtbl(bond
);
479 write_lock_bh(&bond
->curr_slave_lock
);
481 if (slave
!= bond
->curr_active_slave
) {
482 rlb_teach_disabled_mac_on_primary(bond
, slave
->dev
->dev_addr
);
485 write_unlock_bh(&bond
->curr_slave_lock
);
488 static void rlb_update_client(struct rlb_client_info
*client_info
)
492 if (!client_info
->slave
) {
496 for (i
= 0; i
< RLB_ARP_BURST_SIZE
; i
++) {
499 skb
= arp_create(ARPOP_REPLY
, ETH_P_ARP
,
501 client_info
->slave
->dev
,
503 client_info
->mac_dst
,
504 client_info
->slave
->dev
->dev_addr
,
505 client_info
->mac_dst
);
507 printk(KERN_ERR DRV_NAME
508 ": %s: Error: failed to create an ARP packet\n",
509 client_info
->slave
->dev
->master
->name
);
513 skb
->dev
= client_info
->slave
->dev
;
515 if (client_info
->tag
) {
516 skb
= vlan_put_tag(skb
, client_info
->vlan_id
);
518 printk(KERN_ERR DRV_NAME
519 ": %s: Error: failed to insert VLAN tag\n",
520 client_info
->slave
->dev
->master
->name
);
529 /* sends ARP REPLIES that update the clients that need updating */
530 static void rlb_update_rx_clients(struct bonding
*bond
)
532 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
533 struct rlb_client_info
*client_info
;
536 _lock_rx_hashtbl(bond
);
538 hash_index
= bond_info
->rx_hashtbl_head
;
539 for (; hash_index
!= RLB_NULL_INDEX
; hash_index
= client_info
->next
) {
540 client_info
= &(bond_info
->rx_hashtbl
[hash_index
]);
541 if (client_info
->ntt
) {
542 rlb_update_client(client_info
);
543 if (bond_info
->rlb_update_retry_counter
== 0) {
544 client_info
->ntt
= 0;
549 /* do not update the entries again untill this counter is zero so that
550 * not to confuse the clients.
552 bond_info
->rlb_update_delay_counter
= RLB_UPDATE_DELAY
;
554 _unlock_rx_hashtbl(bond
);
557 /* The slave was assigned a new mac address - update the clients */
558 static void rlb_req_update_slave_clients(struct bonding
*bond
, struct slave
*slave
)
560 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
561 struct rlb_client_info
*client_info
;
565 _lock_rx_hashtbl(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
== slave
) &&
572 memcmp(client_info
->mac_dst
, mac_bcast
, ETH_ALEN
)) {
573 client_info
->ntt
= 1;
578 // update the team's flag only after the whole iteration
580 bond_info
->rx_ntt
= 1;
582 bond_info
->rlb_update_retry_counter
= RLB_UPDATE_RETRY
;
585 _unlock_rx_hashtbl(bond
);
588 /* mark all clients using src_ip to be updated */
589 static void rlb_req_update_subnet_clients(struct bonding
*bond
, __be32 src_ip
)
591 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
592 struct rlb_client_info
*client_info
;
595 _lock_rx_hashtbl(bond
);
597 hash_index
= bond_info
->rx_hashtbl_head
;
598 for (; hash_index
!= RLB_NULL_INDEX
; hash_index
= client_info
->next
) {
599 client_info
= &(bond_info
->rx_hashtbl
[hash_index
]);
601 if (!client_info
->slave
) {
602 printk(KERN_ERR DRV_NAME
603 ": %s: Error: found a client with no channel in "
604 "the client's hash table\n",
608 /*update all clients using this src_ip, that are not assigned
609 * to the team's address (curr_active_slave) and have a known
610 * unicast mac address.
612 if ((client_info
->ip_src
== src_ip
) &&
613 memcmp(client_info
->slave
->dev
->dev_addr
,
614 bond
->dev
->dev_addr
, ETH_ALEN
) &&
615 memcmp(client_info
->mac_dst
, mac_bcast
, ETH_ALEN
)) {
616 client_info
->ntt
= 1;
617 bond_info
->rx_ntt
= 1;
621 _unlock_rx_hashtbl(bond
);
624 /* Caller must hold both bond and ptr locks for read */
625 static struct slave
*rlb_choose_channel(struct sk_buff
*skb
, struct bonding
*bond
)
627 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
628 struct arp_pkt
*arp
= arp_pkt(skb
);
629 struct slave
*assigned_slave
;
630 struct rlb_client_info
*client_info
;
633 _lock_rx_hashtbl(bond
);
635 hash_index
= _simple_hash((u8
*)&arp
->ip_dst
, sizeof(arp
->ip_src
));
636 client_info
= &(bond_info
->rx_hashtbl
[hash_index
]);
638 if (client_info
->assigned
) {
639 if ((client_info
->ip_src
== arp
->ip_src
) &&
640 (client_info
->ip_dst
== arp
->ip_dst
)) {
641 /* the entry is already assigned to this client */
642 if (memcmp(arp
->mac_dst
, mac_bcast
, ETH_ALEN
)) {
643 /* update mac address from arp */
644 memcpy(client_info
->mac_dst
, arp
->mac_dst
, ETH_ALEN
);
647 assigned_slave
= client_info
->slave
;
648 if (assigned_slave
) {
649 _unlock_rx_hashtbl(bond
);
650 return assigned_slave
;
653 /* the entry is already assigned to some other client,
654 * move the old client to primary (curr_active_slave) so
655 * that the new client can be assigned to this entry.
657 if (bond
->curr_active_slave
&&
658 client_info
->slave
!= bond
->curr_active_slave
) {
659 client_info
->slave
= bond
->curr_active_slave
;
660 rlb_update_client(client_info
);
664 /* assign a new slave */
665 assigned_slave
= rlb_next_rx_slave(bond
);
667 if (assigned_slave
) {
668 client_info
->ip_src
= arp
->ip_src
;
669 client_info
->ip_dst
= arp
->ip_dst
;
670 /* arp->mac_dst is broadcast for arp reqeusts.
671 * will be updated with clients actual unicast mac address
672 * upon receiving an arp reply.
674 memcpy(client_info
->mac_dst
, arp
->mac_dst
, ETH_ALEN
);
675 client_info
->slave
= assigned_slave
;
677 if (memcmp(client_info
->mac_dst
, mac_bcast
, ETH_ALEN
)) {
678 client_info
->ntt
= 1;
679 bond
->alb_info
.rx_ntt
= 1;
681 client_info
->ntt
= 0;
684 if (!list_empty(&bond
->vlan_list
)) {
685 if (!vlan_get_tag(skb
, &client_info
->vlan_id
))
686 client_info
->tag
= 1;
689 if (!client_info
->assigned
) {
690 u32 prev_tbl_head
= bond_info
->rx_hashtbl_head
;
691 bond_info
->rx_hashtbl_head
= hash_index
;
692 client_info
->next
= prev_tbl_head
;
693 if (prev_tbl_head
!= RLB_NULL_INDEX
) {
694 bond_info
->rx_hashtbl
[prev_tbl_head
].prev
=
697 client_info
->assigned
= 1;
701 _unlock_rx_hashtbl(bond
);
703 return assigned_slave
;
706 /* chooses (and returns) transmit channel for arp reply
707 * does not choose channel for other arp types since they are
708 * sent on the curr_active_slave
710 static struct slave
*rlb_arp_xmit(struct sk_buff
*skb
, struct bonding
*bond
)
712 struct arp_pkt
*arp
= arp_pkt(skb
);
713 struct slave
*tx_slave
= NULL
;
715 if (arp
->op_code
== htons(ARPOP_REPLY
)) {
716 /* the arp must be sent on the selected
719 tx_slave
= rlb_choose_channel(skb
, bond
);
721 memcpy(arp
->mac_src
,tx_slave
->dev
->dev_addr
, ETH_ALEN
);
723 dprintk("Server sent ARP Reply packet\n");
724 } else if (arp
->op_code
== htons(ARPOP_REQUEST
)) {
725 /* Create an entry in the rx_hashtbl for this client as a
727 * When the arp reply is received the entry will be updated
728 * with the correct unicast address of the client.
730 rlb_choose_channel(skb
, bond
);
732 /* The ARP relpy packets must be delayed so that
733 * they can cancel out the influence of the ARP request.
735 bond
->alb_info
.rlb_update_delay_counter
= RLB_UPDATE_DELAY
;
737 /* arp requests are broadcast and are sent on the primary
738 * the arp request will collapse all clients on the subnet to
739 * the primary slave. We must register these clients to be
740 * updated with their assigned mac.
742 rlb_req_update_subnet_clients(bond
, arp
->ip_src
);
743 dprintk("Server sent ARP Request packet\n");
749 /* Caller must hold bond lock for read */
750 static void rlb_rebalance(struct bonding
*bond
)
752 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
753 struct slave
*assigned_slave
;
754 struct rlb_client_info
*client_info
;
758 _lock_rx_hashtbl(bond
);
761 hash_index
= bond_info
->rx_hashtbl_head
;
762 for (; hash_index
!= RLB_NULL_INDEX
; hash_index
= client_info
->next
) {
763 client_info
= &(bond_info
->rx_hashtbl
[hash_index
]);
764 assigned_slave
= rlb_next_rx_slave(bond
);
765 if (assigned_slave
&& (client_info
->slave
!= assigned_slave
)) {
766 client_info
->slave
= assigned_slave
;
767 client_info
->ntt
= 1;
772 /* update the team's flag only after the whole iteration */
774 bond_info
->rx_ntt
= 1;
776 _unlock_rx_hashtbl(bond
);
779 /* Caller must hold rx_hashtbl lock */
780 static void rlb_init_table_entry(struct rlb_client_info
*entry
)
782 memset(entry
, 0, sizeof(struct rlb_client_info
));
783 entry
->next
= RLB_NULL_INDEX
;
784 entry
->prev
= RLB_NULL_INDEX
;
787 static int rlb_initialize(struct bonding
*bond
)
789 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
790 struct packet_type
*pk_type
= &(BOND_ALB_INFO(bond
).rlb_pkt_type
);
791 struct rlb_client_info
*new_hashtbl
;
792 int size
= RLB_HASH_TABLE_SIZE
* sizeof(struct rlb_client_info
);
795 spin_lock_init(&(bond_info
->rx_hashtbl_lock
));
797 new_hashtbl
= kmalloc(size
, GFP_KERNEL
);
799 printk(KERN_ERR DRV_NAME
800 ": %s: Error: Failed to allocate RLB hash table\n",
804 _lock_rx_hashtbl(bond
);
806 bond_info
->rx_hashtbl
= new_hashtbl
;
808 bond_info
->rx_hashtbl_head
= RLB_NULL_INDEX
;
810 for (i
= 0; i
< RLB_HASH_TABLE_SIZE
; i
++) {
811 rlb_init_table_entry(bond_info
->rx_hashtbl
+ i
);
814 _unlock_rx_hashtbl(bond
);
816 /*initialize packet type*/
817 pk_type
->type
= __constant_htons(ETH_P_ARP
);
818 pk_type
->dev
= bond
->dev
;
819 pk_type
->func
= rlb_arp_recv
;
821 /* register to receive ARPs */
822 dev_add_pack(pk_type
);
827 static void rlb_deinitialize(struct bonding
*bond
)
829 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
831 dev_remove_pack(&(bond_info
->rlb_pkt_type
));
833 _lock_rx_hashtbl(bond
);
835 kfree(bond_info
->rx_hashtbl
);
836 bond_info
->rx_hashtbl
= NULL
;
837 bond_info
->rx_hashtbl_head
= RLB_NULL_INDEX
;
839 _unlock_rx_hashtbl(bond
);
842 static void rlb_clear_vlan(struct bonding
*bond
, unsigned short vlan_id
)
844 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
847 _lock_rx_hashtbl(bond
);
849 curr_index
= bond_info
->rx_hashtbl_head
;
850 while (curr_index
!= RLB_NULL_INDEX
) {
851 struct rlb_client_info
*curr
= &(bond_info
->rx_hashtbl
[curr_index
]);
852 u32 next_index
= bond_info
->rx_hashtbl
[curr_index
].next
;
853 u32 prev_index
= bond_info
->rx_hashtbl
[curr_index
].prev
;
855 if (curr
->tag
&& (curr
->vlan_id
== vlan_id
)) {
856 if (curr_index
== bond_info
->rx_hashtbl_head
) {
857 bond_info
->rx_hashtbl_head
= next_index
;
859 if (prev_index
!= RLB_NULL_INDEX
) {
860 bond_info
->rx_hashtbl
[prev_index
].next
= next_index
;
862 if (next_index
!= RLB_NULL_INDEX
) {
863 bond_info
->rx_hashtbl
[next_index
].prev
= prev_index
;
866 rlb_init_table_entry(curr
);
869 curr_index
= next_index
;
872 _unlock_rx_hashtbl(bond
);
875 /*********************** tlb/rlb shared functions *********************/
877 static void alb_send_learning_packets(struct slave
*slave
, u8 mac_addr
[])
879 struct bonding
*bond
= bond_get_bond_by_slave(slave
);
880 struct learning_pkt pkt
;
881 int size
= sizeof(struct learning_pkt
);
884 memset(&pkt
, 0, size
);
885 memcpy(pkt
.mac_dst
, mac_addr
, ETH_ALEN
);
886 memcpy(pkt
.mac_src
, mac_addr
, ETH_ALEN
);
887 pkt
.type
= __constant_htons(ETH_P_LOOP
);
889 for (i
= 0; i
< MAX_LP_BURST
; i
++) {
893 skb
= dev_alloc_skb(size
);
898 data
= skb_put(skb
, size
);
899 memcpy(data
, &pkt
, size
);
901 skb_reset_mac_header(skb
);
902 skb
->network_header
= skb
->mac_header
+ ETH_HLEN
;
903 skb
->protocol
= pkt
.type
;
904 skb
->priority
= TC_PRIO_CONTROL
;
905 skb
->dev
= slave
->dev
;
907 if (!list_empty(&bond
->vlan_list
)) {
908 struct vlan_entry
*vlan
;
910 vlan
= bond_next_vlan(bond
,
911 bond
->alb_info
.current_alb_vlan
);
913 bond
->alb_info
.current_alb_vlan
= vlan
;
919 skb
= vlan_put_tag(skb
, vlan
->vlan_id
);
921 printk(KERN_ERR DRV_NAME
922 ": %s: Error: failed to insert VLAN tag\n",
932 /* hw is a boolean parameter that determines whether we should try and
933 * set the hw address of the device as well as the hw address of the
936 static int alb_set_slave_mac_addr(struct slave
*slave
, u8 addr
[], int hw
)
938 struct net_device
*dev
= slave
->dev
;
939 struct sockaddr s_addr
;
942 memcpy(dev
->dev_addr
, addr
, dev
->addr_len
);
946 /* for rlb each slave must have a unique hw mac addresses so that */
947 /* each slave will receive packets destined to a different mac */
948 memcpy(s_addr
.sa_data
, addr
, dev
->addr_len
);
949 s_addr
.sa_family
= dev
->type
;
950 if (dev_set_mac_address(dev
, &s_addr
)) {
951 printk(KERN_ERR DRV_NAME
952 ": %s: Error: dev_set_mac_address of dev %s failed! ALB "
953 "mode requires that the base driver support setting "
954 "the hw address also when the network device's "
955 "interface is open\n",
956 dev
->master
->name
, dev
->name
);
963 * Swap MAC addresses between two slaves.
965 * Called with RTNL held, and no other locks.
969 static void alb_swap_mac_addr(struct bonding
*bond
, struct slave
*slave1
, struct slave
*slave2
)
971 u8 tmp_mac_addr
[ETH_ALEN
];
973 memcpy(tmp_mac_addr
, slave1
->dev
->dev_addr
, ETH_ALEN
);
974 alb_set_slave_mac_addr(slave1
, slave2
->dev
->dev_addr
, bond
->alb_info
.rlb_enabled
);
975 alb_set_slave_mac_addr(slave2
, tmp_mac_addr
, bond
->alb_info
.rlb_enabled
);
980 * Send learning packets after MAC address swap.
982 * Called with RTNL and no other locks
984 static void alb_fasten_mac_swap(struct bonding
*bond
, struct slave
*slave1
,
985 struct slave
*slave2
)
987 int slaves_state_differ
= (SLAVE_IS_OK(slave1
) != SLAVE_IS_OK(slave2
));
988 struct slave
*disabled_slave
= NULL
;
992 /* fasten the change in the switch */
993 if (SLAVE_IS_OK(slave1
)) {
994 alb_send_learning_packets(slave1
, slave1
->dev
->dev_addr
);
995 if (bond
->alb_info
.rlb_enabled
) {
996 /* inform the clients that the mac address
999 rlb_req_update_slave_clients(bond
, slave1
);
1002 disabled_slave
= slave1
;
1005 if (SLAVE_IS_OK(slave2
)) {
1006 alb_send_learning_packets(slave2
, slave2
->dev
->dev_addr
);
1007 if (bond
->alb_info
.rlb_enabled
) {
1008 /* inform the clients that the mac address
1011 rlb_req_update_slave_clients(bond
, slave2
);
1014 disabled_slave
= slave2
;
1017 if (bond
->alb_info
.rlb_enabled
&& slaves_state_differ
) {
1018 /* A disabled slave was assigned an active mac addr */
1019 rlb_teach_disabled_mac_on_primary(bond
,
1020 disabled_slave
->dev
->dev_addr
);
1025 * alb_change_hw_addr_on_detach
1026 * @bond: bonding we're working on
1027 * @slave: the slave that was just detached
1029 * We assume that @slave was already detached from the slave list.
1031 * If @slave's permanent hw address is different both from its current
1032 * address and from @bond's address, then somewhere in the bond there's
1033 * a slave that has @slave's permanet address as its current address.
1034 * We'll make sure that that slave no longer uses @slave's permanent address.
1036 * Caller must hold RTNL and no other locks
1038 static void alb_change_hw_addr_on_detach(struct bonding
*bond
, struct slave
*slave
)
1043 perm_curr_diff
= memcmp(slave
->perm_hwaddr
,
1044 slave
->dev
->dev_addr
,
1046 perm_bond_diff
= memcmp(slave
->perm_hwaddr
,
1047 bond
->dev
->dev_addr
,
1050 if (perm_curr_diff
&& perm_bond_diff
) {
1051 struct slave
*tmp_slave
;
1054 bond_for_each_slave(bond
, tmp_slave
, i
) {
1055 if (!memcmp(slave
->perm_hwaddr
,
1056 tmp_slave
->dev
->dev_addr
,
1064 /* locking: needs RTNL and nothing else */
1065 alb_swap_mac_addr(bond
, slave
, tmp_slave
);
1066 alb_fasten_mac_swap(bond
, slave
, tmp_slave
);
1072 * alb_handle_addr_collision_on_attach
1073 * @bond: bonding we're working on
1074 * @slave: the slave that was just attached
1076 * checks uniqueness of slave's mac address and handles the case the
1077 * new slave uses the bonds mac address.
1079 * If the permanent hw address of @slave is @bond's hw address, we need to
1080 * find a different hw address to give @slave, that isn't in use by any other
1081 * slave in the bond. This address must be, of course, one of the premanent
1082 * addresses of the other slaves.
1084 * We go over the slave list, and for each slave there we compare its
1085 * permanent hw address with the current address of all the other slaves.
1086 * If no match was found, then we've found a slave with a permanent address
1087 * that isn't used by any other slave in the bond, so we can assign it to
1090 * assumption: this function is called before @slave is attached to the
1093 * caller must hold the bond lock for write since the mac addresses are compared
1094 * and may be swapped.
1096 static int alb_handle_addr_collision_on_attach(struct bonding
*bond
, struct slave
*slave
)
1098 struct slave
*tmp_slave1
, *tmp_slave2
, *free_mac_slave
;
1099 struct slave
*has_bond_addr
= bond
->curr_active_slave
;
1100 int i
, j
, found
= 0;
1102 if (bond
->slave_cnt
== 0) {
1103 /* this is the first slave */
1107 /* if slave's mac address differs from bond's mac address
1108 * check uniqueness of slave's mac address against the other
1109 * slaves in the bond.
1111 if (memcmp(slave
->perm_hwaddr
, bond
->dev
->dev_addr
, ETH_ALEN
)) {
1112 bond_for_each_slave(bond
, tmp_slave1
, i
) {
1113 if (!memcmp(tmp_slave1
->dev
->dev_addr
, slave
->dev
->dev_addr
,
1123 /* Try setting slave mac to bond address and fall-through
1124 to code handling that situation below... */
1125 alb_set_slave_mac_addr(slave
, bond
->dev
->dev_addr
,
1126 bond
->alb_info
.rlb_enabled
);
1129 /* The slave's address is equal to the address of the bond.
1130 * Search for a spare address in the bond for this slave.
1132 free_mac_slave
= NULL
;
1134 bond_for_each_slave(bond
, tmp_slave1
, i
) {
1136 bond_for_each_slave(bond
, tmp_slave2
, j
) {
1137 if (!memcmp(tmp_slave1
->perm_hwaddr
,
1138 tmp_slave2
->dev
->dev_addr
,
1146 /* no slave has tmp_slave1's perm addr
1149 free_mac_slave
= tmp_slave1
;
1153 if (!has_bond_addr
) {
1154 if (!memcmp(tmp_slave1
->dev
->dev_addr
,
1155 bond
->dev
->dev_addr
,
1158 has_bond_addr
= tmp_slave1
;
1163 if (free_mac_slave
) {
1164 alb_set_slave_mac_addr(slave
, free_mac_slave
->perm_hwaddr
,
1165 bond
->alb_info
.rlb_enabled
);
1167 printk(KERN_WARNING DRV_NAME
1168 ": %s: Warning: the hw address of slave %s is in use by "
1169 "the bond; giving it the hw address of %s\n",
1170 bond
->dev
->name
, slave
->dev
->name
, free_mac_slave
->dev
->name
);
1172 } else if (has_bond_addr
) {
1173 printk(KERN_ERR DRV_NAME
1174 ": %s: Error: the hw address of slave %s is in use by the "
1175 "bond; couldn't find a slave with a free hw address to "
1176 "give it (this should not have happened)\n",
1177 bond
->dev
->name
, slave
->dev
->name
);
1185 * alb_set_mac_address
1189 * In TLB mode all slaves are configured to the bond's hw address, but set
1190 * their dev_addr field to different addresses (based on their permanent hw
1193 * For each slave, this function sets the interface to the new address and then
1194 * changes its dev_addr field to its previous value.
1196 * Unwinding assumes bond's mac address has not yet changed.
1198 static int alb_set_mac_address(struct bonding
*bond
, void *addr
)
1201 struct slave
*slave
, *stop_at
;
1202 char tmp_addr
[ETH_ALEN
];
1206 if (bond
->alb_info
.rlb_enabled
) {
1210 bond_for_each_slave(bond
, slave
, i
) {
1211 if (slave
->dev
->set_mac_address
== NULL
) {
1216 /* save net_device's current hw address */
1217 memcpy(tmp_addr
, slave
->dev
->dev_addr
, ETH_ALEN
);
1219 res
= dev_set_mac_address(slave
->dev
, addr
);
1221 /* restore net_device's hw address */
1222 memcpy(slave
->dev
->dev_addr
, tmp_addr
, ETH_ALEN
);
1232 memcpy(sa
.sa_data
, bond
->dev
->dev_addr
, bond
->dev
->addr_len
);
1233 sa
.sa_family
= bond
->dev
->type
;
1235 /* unwind from head to the slave that failed */
1237 bond_for_each_slave_from_to(bond
, slave
, i
, bond
->first_slave
, stop_at
) {
1238 memcpy(tmp_addr
, slave
->dev
->dev_addr
, ETH_ALEN
);
1239 dev_set_mac_address(slave
->dev
, &sa
);
1240 memcpy(slave
->dev
->dev_addr
, tmp_addr
, ETH_ALEN
);
1246 /************************ exported alb funcions ************************/
1248 int bond_alb_initialize(struct bonding
*bond
, int rlb_enabled
)
1252 res
= tlb_initialize(bond
);
1258 bond
->alb_info
.rlb_enabled
= 1;
1259 /* initialize rlb */
1260 res
= rlb_initialize(bond
);
1262 tlb_deinitialize(bond
);
1266 bond
->alb_info
.rlb_enabled
= 0;
1272 void bond_alb_deinitialize(struct bonding
*bond
)
1274 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
1276 tlb_deinitialize(bond
);
1278 if (bond_info
->rlb_enabled
) {
1279 rlb_deinitialize(bond
);
1283 int bond_alb_xmit(struct sk_buff
*skb
, struct net_device
*bond_dev
)
1285 struct bonding
*bond
= bond_dev
->priv
;
1286 struct ethhdr
*eth_data
;
1287 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
1288 struct slave
*tx_slave
= NULL
;
1289 static const __be32 ip_bcast
= htonl(0xffffffff);
1291 int do_tx_balance
= 1;
1293 const u8
*hash_start
= NULL
;
1295 struct ipv6hdr
*ip6hdr
;
1297 skb_reset_mac_header(skb
);
1298 eth_data
= eth_hdr(skb
);
1300 /* make sure that the curr_active_slave and the slaves list do
1301 * not change during tx
1303 read_lock(&bond
->lock
);
1304 read_lock(&bond
->curr_slave_lock
);
1306 if (!BOND_IS_OK(bond
)) {
1310 switch (ntohs(skb
->protocol
)) {
1312 const struct iphdr
*iph
= ip_hdr(skb
);
1314 if ((memcmp(eth_data
->h_dest
, mac_bcast
, ETH_ALEN
) == 0) ||
1315 (iph
->daddr
== ip_bcast
) ||
1316 (iph
->protocol
== IPPROTO_IGMP
)) {
1320 hash_start
= (char *)&(iph
->daddr
);
1321 hash_size
= sizeof(iph
->daddr
);
1325 /* IPv6 doesn't really use broadcast mac address, but leave
1326 * that here just in case.
1328 if (memcmp(eth_data
->h_dest
, mac_bcast
, ETH_ALEN
) == 0) {
1333 /* IPv6 uses all-nodes multicast as an equivalent to
1334 * broadcasts in IPv4.
1336 if (memcmp(eth_data
->h_dest
, mac_v6_allmcast
, ETH_ALEN
) == 0) {
1341 /* Additianally, DAD probes should not be tx-balanced as that
1342 * will lead to false positives for duplicate addresses and
1343 * prevent address configuration from working.
1345 ip6hdr
= ipv6_hdr(skb
);
1346 if (ipv6_addr_any(&ip6hdr
->saddr
)) {
1351 hash_start
= (char *)&(ipv6_hdr(skb
)->daddr
);
1352 hash_size
= sizeof(ipv6_hdr(skb
)->daddr
);
1355 if (ipx_hdr(skb
)->ipx_checksum
!= IPX_NO_CHECKSUM
) {
1356 /* something is wrong with this packet */
1361 if (ipx_hdr(skb
)->ipx_type
!= IPX_TYPE_NCP
) {
1362 /* The only protocol worth balancing in
1363 * this family since it has an "ARP" like
1370 hash_start
= (char*)eth_data
->h_dest
;
1371 hash_size
= ETH_ALEN
;
1375 if (bond_info
->rlb_enabled
) {
1376 tx_slave
= rlb_arp_xmit(skb
, bond
);
1384 if (do_tx_balance
) {
1385 hash_index
= _simple_hash(hash_start
, hash_size
);
1386 tx_slave
= tlb_choose_channel(bond
, hash_index
, skb
->len
);
1390 /* unbalanced or unassigned, send through primary */
1391 tx_slave
= bond
->curr_active_slave
;
1392 bond_info
->unbalanced_load
+= skb
->len
;
1395 if (tx_slave
&& SLAVE_IS_OK(tx_slave
)) {
1396 if (tx_slave
!= bond
->curr_active_slave
) {
1397 memcpy(eth_data
->h_source
,
1398 tx_slave
->dev
->dev_addr
,
1402 res
= bond_dev_queue_xmit(bond
, skb
, tx_slave
->dev
);
1405 tlb_clear_slave(bond
, tx_slave
, 0);
1411 /* no suitable interface, frame not sent */
1414 read_unlock(&bond
->curr_slave_lock
);
1415 read_unlock(&bond
->lock
);
1419 void bond_alb_monitor(struct work_struct
*work
)
1421 struct bonding
*bond
= container_of(work
, struct bonding
,
1423 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
1424 struct slave
*slave
;
1427 read_lock(&bond
->lock
);
1429 if (bond
->kill_timers
) {
1433 if (bond
->slave_cnt
== 0) {
1434 bond_info
->tx_rebalance_counter
= 0;
1435 bond_info
->lp_counter
= 0;
1439 bond_info
->tx_rebalance_counter
++;
1440 bond_info
->lp_counter
++;
1442 /* send learning packets */
1443 if (bond_info
->lp_counter
>= BOND_ALB_LP_TICKS
) {
1444 /* change of curr_active_slave involves swapping of mac addresses.
1445 * in order to avoid this swapping from happening while
1446 * sending the learning packets, the curr_slave_lock must be held for
1449 read_lock(&bond
->curr_slave_lock
);
1451 bond_for_each_slave(bond
, slave
, i
) {
1452 alb_send_learning_packets(slave
, slave
->dev
->dev_addr
);
1455 read_unlock(&bond
->curr_slave_lock
);
1457 bond_info
->lp_counter
= 0;
1460 /* rebalance tx traffic */
1461 if (bond_info
->tx_rebalance_counter
>= BOND_TLB_REBALANCE_TICKS
) {
1463 read_lock(&bond
->curr_slave_lock
);
1465 bond_for_each_slave(bond
, slave
, i
) {
1466 tlb_clear_slave(bond
, slave
, 1);
1467 if (slave
== bond
->curr_active_slave
) {
1468 SLAVE_TLB_INFO(slave
).load
=
1469 bond_info
->unbalanced_load
/
1470 BOND_TLB_REBALANCE_INTERVAL
;
1471 bond_info
->unbalanced_load
= 0;
1475 read_unlock(&bond
->curr_slave_lock
);
1477 bond_info
->tx_rebalance_counter
= 0;
1480 /* handle rlb stuff */
1481 if (bond_info
->rlb_enabled
) {
1482 if (bond_info
->primary_is_promisc
&&
1483 (++bond_info
->rlb_promisc_timeout_counter
>= RLB_PROMISC_TIMEOUT
)) {
1486 * dev_set_promiscuity requires rtnl and
1489 read_unlock(&bond
->lock
);
1492 bond_info
->rlb_promisc_timeout_counter
= 0;
1494 /* If the primary was set to promiscuous mode
1495 * because a slave was disabled then
1496 * it can now leave promiscuous mode.
1498 dev_set_promiscuity(bond
->curr_active_slave
->dev
, -1);
1499 bond_info
->primary_is_promisc
= 0;
1502 read_lock(&bond
->lock
);
1505 if (bond_info
->rlb_rebalance
) {
1506 bond_info
->rlb_rebalance
= 0;
1507 rlb_rebalance(bond
);
1510 /* check if clients need updating */
1511 if (bond_info
->rx_ntt
) {
1512 if (bond_info
->rlb_update_delay_counter
) {
1513 --bond_info
->rlb_update_delay_counter
;
1515 rlb_update_rx_clients(bond
);
1516 if (bond_info
->rlb_update_retry_counter
) {
1517 --bond_info
->rlb_update_retry_counter
;
1519 bond_info
->rx_ntt
= 0;
1526 queue_delayed_work(bond
->wq
, &bond
->alb_work
, alb_delta_in_ticks
);
1528 read_unlock(&bond
->lock
);
1531 /* assumption: called before the slave is attached to the bond
1532 * and not locked by the bond lock
1534 int bond_alb_init_slave(struct bonding
*bond
, struct slave
*slave
)
1538 res
= alb_set_slave_mac_addr(slave
, slave
->perm_hwaddr
,
1539 bond
->alb_info
.rlb_enabled
);
1544 /* caller must hold the bond lock for write since the mac addresses
1545 * are compared and may be swapped.
1547 read_lock(&bond
->lock
);
1549 res
= alb_handle_addr_collision_on_attach(bond
, slave
);
1551 read_unlock(&bond
->lock
);
1557 tlb_init_slave(slave
);
1559 /* order a rebalance ASAP */
1560 bond
->alb_info
.tx_rebalance_counter
= BOND_TLB_REBALANCE_TICKS
;
1562 if (bond
->alb_info
.rlb_enabled
) {
1563 bond
->alb_info
.rlb_rebalance
= 1;
1570 * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1573 * Caller must hold RTNL and no other locks
1575 void bond_alb_deinit_slave(struct bonding
*bond
, struct slave
*slave
)
1577 if (bond
->slave_cnt
> 1) {
1578 alb_change_hw_addr_on_detach(bond
, slave
);
1581 tlb_clear_slave(bond
, slave
, 0);
1583 if (bond
->alb_info
.rlb_enabled
) {
1584 bond
->alb_info
.next_rx_slave
= NULL
;
1585 rlb_clear_slave(bond
, slave
);
1589 /* Caller must hold bond lock for read */
1590 void bond_alb_handle_link_change(struct bonding
*bond
, struct slave
*slave
, char link
)
1592 struct alb_bond_info
*bond_info
= &(BOND_ALB_INFO(bond
));
1594 if (link
== BOND_LINK_DOWN
) {
1595 tlb_clear_slave(bond
, slave
, 0);
1596 if (bond
->alb_info
.rlb_enabled
) {
1597 rlb_clear_slave(bond
, slave
);
1599 } else if (link
== BOND_LINK_UP
) {
1600 /* order a rebalance ASAP */
1601 bond_info
->tx_rebalance_counter
= BOND_TLB_REBALANCE_TICKS
;
1602 if (bond
->alb_info
.rlb_enabled
) {
1603 bond
->alb_info
.rlb_rebalance
= 1;
1604 /* If the updelay module parameter is smaller than the
1605 * forwarding delay of the switch the rebalance will
1606 * not work because the rebalance arp replies will
1607 * not be forwarded to the clients..
1614 * bond_alb_handle_active_change - assign new curr_active_slave
1615 * @bond: our bonding struct
1616 * @new_slave: new slave to assign
1618 * Set the bond->curr_active_slave to @new_slave and handle
1619 * mac address swapping and promiscuity changes as needed.
1621 * If new_slave is NULL, caller must hold curr_slave_lock or
1622 * bond->lock for write.
1624 * If new_slave is not NULL, caller must hold RTNL, bond->lock for
1625 * read and curr_slave_lock for write. Processing here may sleep, so
1626 * no other locks may be held.
1628 void bond_alb_handle_active_change(struct bonding
*bond
, struct slave
*new_slave
)
1630 struct slave
*swap_slave
;
1633 if (bond
->curr_active_slave
== new_slave
) {
1637 if (bond
->curr_active_slave
&& bond
->alb_info
.primary_is_promisc
) {
1638 dev_set_promiscuity(bond
->curr_active_slave
->dev
, -1);
1639 bond
->alb_info
.primary_is_promisc
= 0;
1640 bond
->alb_info
.rlb_promisc_timeout_counter
= 0;
1643 swap_slave
= bond
->curr_active_slave
;
1644 bond
->curr_active_slave
= new_slave
;
1646 if (!new_slave
|| (bond
->slave_cnt
== 0)) {
1650 /* set the new curr_active_slave to the bonds mac address
1651 * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1654 struct slave
*tmp_slave
;
1655 /* find slave that is holding the bond's mac address */
1656 bond_for_each_slave(bond
, tmp_slave
, i
) {
1657 if (!memcmp(tmp_slave
->dev
->dev_addr
,
1658 bond
->dev
->dev_addr
, ETH_ALEN
)) {
1659 swap_slave
= tmp_slave
;
1666 * Arrange for swap_slave and new_slave to temporarily be
1667 * ignored so we can mess with their MAC addresses without
1668 * fear of interference from transmit activity.
1671 tlb_clear_slave(bond
, swap_slave
, 1);
1673 tlb_clear_slave(bond
, new_slave
, 1);
1675 write_unlock_bh(&bond
->curr_slave_lock
);
1676 read_unlock(&bond
->lock
);
1680 /* curr_active_slave must be set before calling alb_swap_mac_addr */
1682 /* swap mac address */
1683 alb_swap_mac_addr(bond
, swap_slave
, new_slave
);
1685 /* set the new_slave to the bond mac address */
1686 alb_set_slave_mac_addr(new_slave
, bond
->dev
->dev_addr
,
1687 bond
->alb_info
.rlb_enabled
);
1691 alb_fasten_mac_swap(bond
, swap_slave
, new_slave
);
1692 read_lock(&bond
->lock
);
1694 read_lock(&bond
->lock
);
1695 alb_send_learning_packets(new_slave
, bond
->dev
->dev_addr
);
1698 write_lock_bh(&bond
->curr_slave_lock
);
1704 int bond_alb_set_mac_address(struct net_device
*bond_dev
, void *addr
)
1706 struct bonding
*bond
= bond_dev
->priv
;
1707 struct sockaddr
*sa
= addr
;
1708 struct slave
*slave
, *swap_slave
;
1712 if (!is_valid_ether_addr(sa
->sa_data
)) {
1713 return -EADDRNOTAVAIL
;
1716 res
= alb_set_mac_address(bond
, addr
);
1721 memcpy(bond_dev
->dev_addr
, sa
->sa_data
, bond_dev
->addr_len
);
1723 /* If there is no curr_active_slave there is nothing else to do.
1724 * Otherwise we'll need to pass the new address to it and handle
1727 if (!bond
->curr_active_slave
) {
1733 bond_for_each_slave(bond
, slave
, i
) {
1734 if (!memcmp(slave
->dev
->dev_addr
, bond_dev
->dev_addr
, ETH_ALEN
)) {
1740 write_unlock_bh(&bond
->curr_slave_lock
);
1741 read_unlock(&bond
->lock
);
1744 alb_swap_mac_addr(bond
, swap_slave
, bond
->curr_active_slave
);
1745 alb_fasten_mac_swap(bond
, swap_slave
, bond
->curr_active_slave
);
1747 alb_set_slave_mac_addr(bond
->curr_active_slave
, bond_dev
->dev_addr
,
1748 bond
->alb_info
.rlb_enabled
);
1750 alb_send_learning_packets(bond
->curr_active_slave
, bond_dev
->dev_addr
);
1751 if (bond
->alb_info
.rlb_enabled
) {
1752 /* inform clients mac address has changed */
1753 rlb_req_update_slave_clients(bond
, bond
->curr_active_slave
);
1757 read_lock(&bond
->lock
);
1758 write_lock_bh(&bond
->curr_slave_lock
);
1763 void bond_alb_clear_vlan(struct bonding
*bond
, unsigned short vlan_id
)
1765 if (bond
->alb_info
.current_alb_vlan
&&
1766 (bond
->alb_info
.current_alb_vlan
->vlan_id
== vlan_id
)) {
1767 bond
->alb_info
.current_alb_vlan
= NULL
;
1770 if (bond
->alb_info
.rlb_enabled
) {
1771 rlb_clear_vlan(bond
, vlan_id
);