Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jikos/trivial
[linux-2.6/linux-2.6-openrd.git] / drivers / net / bonding / bond_alb.c
blob00ab51ef3129c835523bc2aefebbb076f7a89aa9
1 /*
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
12 * for more details.
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 #include <linux/skbuff.h>
24 #include <linux/netdevice.h>
25 #include <linux/etherdevice.h>
26 #include <linux/pkt_sched.h>
27 #include <linux/spinlock.h>
28 #include <linux/slab.h>
29 #include <linux/timer.h>
30 #include <linux/ip.h>
31 #include <linux/ipv6.h>
32 #include <linux/if_arp.h>
33 #include <linux/if_ether.h>
34 #include <linux/if_bonding.h>
35 #include <linux/if_vlan.h>
36 #include <linux/in.h>
37 #include <net/ipx.h>
38 #include <net/arp.h>
39 #include <net/ipv6.h>
40 #include <asm/byteorder.h>
41 #include "bonding.h"
42 #include "bond_alb.h"
45 #define ALB_TIMER_TICKS_PER_SEC 10 /* should be a divisor of HZ */
46 #define BOND_TLB_REBALANCE_INTERVAL 10 /* In seconds, periodic re-balancing.
47 * Used for division - never set
48 * to zero !!!
50 #define BOND_ALB_LP_INTERVAL 1 /* In seconds, periodic send of
51 * learning packets to the switch
54 #define BOND_TLB_REBALANCE_TICKS (BOND_TLB_REBALANCE_INTERVAL \
55 * ALB_TIMER_TICKS_PER_SEC)
57 #define BOND_ALB_LP_TICKS (BOND_ALB_LP_INTERVAL \
58 * ALB_TIMER_TICKS_PER_SEC)
60 #define TLB_HASH_TABLE_SIZE 256 /* The size of the clients hash table.
61 * Note that this value MUST NOT be smaller
62 * because the key hash table is BYTE wide !
66 #define TLB_NULL_INDEX 0xffffffff
67 #define MAX_LP_BURST 3
69 /* rlb defs */
70 #define RLB_HASH_TABLE_SIZE 256
71 #define RLB_NULL_INDEX 0xffffffff
72 #define RLB_UPDATE_DELAY 2*ALB_TIMER_TICKS_PER_SEC /* 2 seconds */
73 #define RLB_ARP_BURST_SIZE 2
74 #define RLB_UPDATE_RETRY 3 /* 3-ticks - must be smaller than the rlb
75 * rebalance interval (5 min).
77 /* RLB_PROMISC_TIMEOUT = 10 sec equals the time that the current slave is
78 * promiscuous after failover
80 #define RLB_PROMISC_TIMEOUT 10*ALB_TIMER_TICKS_PER_SEC
82 #ifndef __long_aligned
83 #define __long_aligned __attribute__((aligned((sizeof(long)))))
84 #endif
85 static const u8 mac_bcast[ETH_ALEN] __long_aligned = {
86 0xff, 0xff, 0xff, 0xff, 0xff, 0xff
88 static const u8 mac_v6_allmcast[ETH_ALEN] __long_aligned = {
89 0x33, 0x33, 0x00, 0x00, 0x00, 0x01
91 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
93 #pragma pack(1)
94 struct learning_pkt {
95 u8 mac_dst[ETH_ALEN];
96 u8 mac_src[ETH_ALEN];
97 __be16 type;
98 u8 padding[ETH_ZLEN - ETH_HLEN];
101 struct arp_pkt {
102 __be16 hw_addr_space;
103 __be16 prot_addr_space;
104 u8 hw_addr_len;
105 u8 prot_addr_len;
106 __be16 op_code;
107 u8 mac_src[ETH_ALEN]; /* sender hardware address */
108 __be32 ip_src; /* sender IP address */
109 u8 mac_dst[ETH_ALEN]; /* target hardware address */
110 __be32 ip_dst; /* target IP address */
112 #pragma pack()
114 static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb)
116 return (struct arp_pkt *)skb_network_header(skb);
119 /* Forward declaration */
120 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[]);
122 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
124 int i;
125 u8 hash = 0;
127 for (i = 0; i < hash_size; i++) {
128 hash ^= hash_start[i];
131 return hash;
134 /*********************** tlb specific functions ***************************/
136 static inline void _lock_tx_hashtbl(struct bonding *bond)
138 spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
141 static inline void _unlock_tx_hashtbl(struct bonding *bond)
143 spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
146 /* Caller must hold tx_hashtbl lock */
147 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
149 if (save_load) {
150 entry->load_history = 1 + entry->tx_bytes /
151 BOND_TLB_REBALANCE_INTERVAL;
152 entry->tx_bytes = 0;
155 entry->tx_slave = NULL;
156 entry->next = TLB_NULL_INDEX;
157 entry->prev = TLB_NULL_INDEX;
160 static inline void tlb_init_slave(struct slave *slave)
162 SLAVE_TLB_INFO(slave).load = 0;
163 SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
166 /* Caller must hold bond lock for read */
167 static void tlb_clear_slave(struct bonding *bond, struct slave *slave, int save_load)
169 struct tlb_client_info *tx_hash_table;
170 u32 index;
172 _lock_tx_hashtbl(bond);
174 /* clear slave from tx_hashtbl */
175 tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
177 /* skip this if we've already freed the tx hash table */
178 if (tx_hash_table) {
179 index = SLAVE_TLB_INFO(slave).head;
180 while (index != TLB_NULL_INDEX) {
181 u32 next_index = tx_hash_table[index].next;
182 tlb_init_table_entry(&tx_hash_table[index], save_load);
183 index = next_index;
187 tlb_init_slave(slave);
189 _unlock_tx_hashtbl(bond);
192 /* Must be called before starting the monitor timer */
193 static int tlb_initialize(struct bonding *bond)
195 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
196 int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
197 struct tlb_client_info *new_hashtbl;
198 int i;
200 spin_lock_init(&(bond_info->tx_hashtbl_lock));
202 new_hashtbl = kzalloc(size, GFP_KERNEL);
203 if (!new_hashtbl) {
204 pr_err(DRV_NAME
205 ": %s: Error: Failed to allocate TLB hash table\n",
206 bond->dev->name);
207 return -1;
209 _lock_tx_hashtbl(bond);
211 bond_info->tx_hashtbl = new_hashtbl;
213 for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
214 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 1);
217 _unlock_tx_hashtbl(bond);
219 return 0;
222 /* Must be called only after all slaves have been released */
223 static void tlb_deinitialize(struct bonding *bond)
225 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
227 _lock_tx_hashtbl(bond);
229 kfree(bond_info->tx_hashtbl);
230 bond_info->tx_hashtbl = NULL;
232 _unlock_tx_hashtbl(bond);
235 /* Caller must hold bond lock for read */
236 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
238 struct slave *slave, *least_loaded;
239 s64 max_gap;
240 int i, found = 0;
242 /* Find the first enabled slave */
243 bond_for_each_slave(bond, slave, i) {
244 if (SLAVE_IS_OK(slave)) {
245 found = 1;
246 break;
250 if (!found) {
251 return NULL;
254 least_loaded = slave;
255 max_gap = (s64)(slave->speed << 20) - /* Convert to Megabit per sec */
256 (s64)(SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
258 /* Find the slave with the largest gap */
259 bond_for_each_slave_from(bond, slave, i, least_loaded) {
260 if (SLAVE_IS_OK(slave)) {
261 s64 gap = (s64)(slave->speed << 20) -
262 (s64)(SLAVE_TLB_INFO(slave).load << 3);
263 if (max_gap < gap) {
264 least_loaded = slave;
265 max_gap = gap;
270 return least_loaded;
273 /* Caller must hold bond lock for read */
274 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index, u32 skb_len)
276 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
277 struct tlb_client_info *hash_table;
278 struct slave *assigned_slave;
280 _lock_tx_hashtbl(bond);
282 hash_table = bond_info->tx_hashtbl;
283 assigned_slave = hash_table[hash_index].tx_slave;
284 if (!assigned_slave) {
285 assigned_slave = tlb_get_least_loaded_slave(bond);
287 if (assigned_slave) {
288 struct tlb_slave_info *slave_info =
289 &(SLAVE_TLB_INFO(assigned_slave));
290 u32 next_index = slave_info->head;
292 hash_table[hash_index].tx_slave = assigned_slave;
293 hash_table[hash_index].next = next_index;
294 hash_table[hash_index].prev = TLB_NULL_INDEX;
296 if (next_index != TLB_NULL_INDEX) {
297 hash_table[next_index].prev = hash_index;
300 slave_info->head = hash_index;
301 slave_info->load +=
302 hash_table[hash_index].load_history;
306 if (assigned_slave) {
307 hash_table[hash_index].tx_bytes += skb_len;
310 _unlock_tx_hashtbl(bond);
312 return assigned_slave;
315 /*********************** rlb specific functions ***************************/
316 static inline void _lock_rx_hashtbl(struct bonding *bond)
318 spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
321 static inline void _unlock_rx_hashtbl(struct bonding *bond)
323 spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
326 /* when an ARP REPLY is received from a client update its info
327 * in the rx_hashtbl
329 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
331 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
332 struct rlb_client_info *client_info;
333 u32 hash_index;
335 _lock_rx_hashtbl(bond);
337 hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
338 client_info = &(bond_info->rx_hashtbl[hash_index]);
340 if ((client_info->assigned) &&
341 (client_info->ip_src == arp->ip_dst) &&
342 (client_info->ip_dst == arp->ip_src)) {
343 /* update the clients MAC address */
344 memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
345 client_info->ntt = 1;
346 bond_info->rx_ntt = 1;
349 _unlock_rx_hashtbl(bond);
352 static int rlb_arp_recv(struct sk_buff *skb, struct net_device *bond_dev, struct packet_type *ptype, struct net_device *orig_dev)
354 struct bonding *bond;
355 struct arp_pkt *arp = (struct arp_pkt *)skb->data;
356 int res = NET_RX_DROP;
358 while (bond_dev->priv_flags & IFF_802_1Q_VLAN)
359 bond_dev = vlan_dev_real_dev(bond_dev);
361 if (!(bond_dev->priv_flags & IFF_BONDING) ||
362 !(bond_dev->flags & IFF_MASTER))
363 goto out;
365 if (!arp) {
366 pr_debug("Packet has no ARP data\n");
367 goto out;
370 if (skb->len < sizeof(struct arp_pkt)) {
371 pr_debug("Packet is too small to be an ARP\n");
372 goto out;
375 if (arp->op_code == htons(ARPOP_REPLY)) {
376 /* update rx hash table for this ARP */
377 bond = netdev_priv(bond_dev);
378 rlb_update_entry_from_arp(bond, arp);
379 pr_debug("Server received an ARP Reply from client\n");
382 res = NET_RX_SUCCESS;
384 out:
385 dev_kfree_skb(skb);
387 return res;
390 /* Caller must hold bond lock for read */
391 static struct slave *rlb_next_rx_slave(struct bonding *bond)
393 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
394 struct slave *rx_slave, *slave, *start_at;
395 int i = 0;
397 if (bond_info->next_rx_slave) {
398 start_at = bond_info->next_rx_slave;
399 } else {
400 start_at = bond->first_slave;
403 rx_slave = NULL;
405 bond_for_each_slave_from(bond, slave, i, start_at) {
406 if (SLAVE_IS_OK(slave)) {
407 if (!rx_slave) {
408 rx_slave = slave;
409 } else if (slave->speed > rx_slave->speed) {
410 rx_slave = slave;
415 if (rx_slave) {
416 bond_info->next_rx_slave = rx_slave->next;
419 return rx_slave;
422 /* teach the switch the mac of a disabled slave
423 * on the primary for fault tolerance
425 * Caller must hold bond->curr_slave_lock for write or bond lock for write
427 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
429 if (!bond->curr_active_slave) {
430 return;
433 if (!bond->alb_info.primary_is_promisc) {
434 if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1))
435 bond->alb_info.primary_is_promisc = 1;
436 else
437 bond->alb_info.primary_is_promisc = 0;
440 bond->alb_info.rlb_promisc_timeout_counter = 0;
442 alb_send_learning_packets(bond->curr_active_slave, addr);
445 /* slave being removed should not be active at this point
447 * Caller must hold bond lock for read
449 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
451 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
452 struct rlb_client_info *rx_hash_table;
453 u32 index, next_index;
455 /* clear slave from rx_hashtbl */
456 _lock_rx_hashtbl(bond);
458 rx_hash_table = bond_info->rx_hashtbl;
459 index = bond_info->rx_hashtbl_head;
460 for (; index != RLB_NULL_INDEX; index = next_index) {
461 next_index = rx_hash_table[index].next;
462 if (rx_hash_table[index].slave == slave) {
463 struct slave *assigned_slave = rlb_next_rx_slave(bond);
465 if (assigned_slave) {
466 rx_hash_table[index].slave = assigned_slave;
467 if (compare_ether_addr_64bits(rx_hash_table[index].mac_dst,
468 mac_bcast)) {
469 bond_info->rx_hashtbl[index].ntt = 1;
470 bond_info->rx_ntt = 1;
471 /* A slave has been removed from the
472 * table because it is either disabled
473 * or being released. We must retry the
474 * update to avoid clients from not
475 * being updated & disconnecting when
476 * there is stress
478 bond_info->rlb_update_retry_counter =
479 RLB_UPDATE_RETRY;
481 } else { /* there is no active slave */
482 rx_hash_table[index].slave = NULL;
487 _unlock_rx_hashtbl(bond);
489 write_lock_bh(&bond->curr_slave_lock);
491 if (slave != bond->curr_active_slave) {
492 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
495 write_unlock_bh(&bond->curr_slave_lock);
498 static void rlb_update_client(struct rlb_client_info *client_info)
500 int i;
502 if (!client_info->slave) {
503 return;
506 for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
507 struct sk_buff *skb;
509 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
510 client_info->ip_dst,
511 client_info->slave->dev,
512 client_info->ip_src,
513 client_info->mac_dst,
514 client_info->slave->dev->dev_addr,
515 client_info->mac_dst);
516 if (!skb) {
517 pr_err(DRV_NAME
518 ": %s: Error: failed to create an ARP packet\n",
519 client_info->slave->dev->master->name);
520 continue;
523 skb->dev = client_info->slave->dev;
525 if (client_info->tag) {
526 skb = vlan_put_tag(skb, client_info->vlan_id);
527 if (!skb) {
528 pr_err(DRV_NAME
529 ": %s: Error: failed to insert VLAN tag\n",
530 client_info->slave->dev->master->name);
531 continue;
535 arp_xmit(skb);
539 /* sends ARP REPLIES that update the clients that need updating */
540 static void rlb_update_rx_clients(struct bonding *bond)
542 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
543 struct rlb_client_info *client_info;
544 u32 hash_index;
546 _lock_rx_hashtbl(bond);
548 hash_index = bond_info->rx_hashtbl_head;
549 for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
550 client_info = &(bond_info->rx_hashtbl[hash_index]);
551 if (client_info->ntt) {
552 rlb_update_client(client_info);
553 if (bond_info->rlb_update_retry_counter == 0) {
554 client_info->ntt = 0;
559 /* do not update the entries again until this counter is zero so that
560 * not to confuse the clients.
562 bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
564 _unlock_rx_hashtbl(bond);
567 /* The slave was assigned a new mac address - update the clients */
568 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
570 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
571 struct rlb_client_info *client_info;
572 int ntt = 0;
573 u32 hash_index;
575 _lock_rx_hashtbl(bond);
577 hash_index = bond_info->rx_hashtbl_head;
578 for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
579 client_info = &(bond_info->rx_hashtbl[hash_index]);
581 if ((client_info->slave == slave) &&
582 compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
583 client_info->ntt = 1;
584 ntt = 1;
588 // update the team's flag only after the whole iteration
589 if (ntt) {
590 bond_info->rx_ntt = 1;
591 //fasten the change
592 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
595 _unlock_rx_hashtbl(bond);
598 /* mark all clients using src_ip to be updated */
599 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
601 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
602 struct rlb_client_info *client_info;
603 u32 hash_index;
605 _lock_rx_hashtbl(bond);
607 hash_index = bond_info->rx_hashtbl_head;
608 for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
609 client_info = &(bond_info->rx_hashtbl[hash_index]);
611 if (!client_info->slave) {
612 pr_err(DRV_NAME
613 ": %s: Error: found a client with no channel in "
614 "the client's hash table\n",
615 bond->dev->name);
616 continue;
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 compare_ether_addr_64bits(client_info->slave->dev->dev_addr,
624 bond->dev->dev_addr) &&
625 compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
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 static 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 = arp_pkt(skb);
639 struct slave *assigned_slave;
640 struct rlb_client_info *client_info;
641 u32 hash_index = 0;
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 (compare_ether_addr_64bits(arp->mac_dst, mac_bcast)) {
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;
662 } else {
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 (compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
688 client_info->ntt = 1;
689 bond->alb_info.rx_ntt = 1;
690 } else {
691 client_info->ntt = 0;
694 if (!list_empty(&bond->vlan_list)) {
695 if (!vlan_get_tag(skb, &client_info->vlan_id))
696 client_info->tag = 1;
699 if (!client_info->assigned) {
700 u32 prev_tbl_head = bond_info->rx_hashtbl_head;
701 bond_info->rx_hashtbl_head = hash_index;
702 client_info->next = prev_tbl_head;
703 if (prev_tbl_head != RLB_NULL_INDEX) {
704 bond_info->rx_hashtbl[prev_tbl_head].prev =
705 hash_index;
707 client_info->assigned = 1;
711 _unlock_rx_hashtbl(bond);
713 return assigned_slave;
716 /* chooses (and returns) transmit channel for arp reply
717 * does not choose channel for other arp types since they are
718 * sent on the curr_active_slave
720 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
722 struct arp_pkt *arp = arp_pkt(skb);
723 struct slave *tx_slave = NULL;
725 if (arp->op_code == htons(ARPOP_REPLY)) {
726 /* the arp must be sent on the selected
727 * rx channel
729 tx_slave = rlb_choose_channel(skb, bond);
730 if (tx_slave) {
731 memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
733 pr_debug("Server sent ARP Reply packet\n");
734 } else if (arp->op_code == htons(ARPOP_REQUEST)) {
735 /* Create an entry in the rx_hashtbl for this client as a
736 * place holder.
737 * When the arp reply is received the entry will be updated
738 * with the correct unicast address of the client.
740 rlb_choose_channel(skb, bond);
742 /* The ARP relpy packets must be delayed so that
743 * they can cancel out the influence of the ARP request.
745 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
747 /* arp requests are broadcast and are sent on the primary
748 * the arp request will collapse all clients on the subnet to
749 * the primary slave. We must register these clients to be
750 * updated with their assigned mac.
752 rlb_req_update_subnet_clients(bond, arp->ip_src);
753 pr_debug("Server sent ARP Request packet\n");
756 return tx_slave;
759 /* Caller must hold bond lock for read */
760 static void rlb_rebalance(struct bonding *bond)
762 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
763 struct slave *assigned_slave;
764 struct rlb_client_info *client_info;
765 int ntt;
766 u32 hash_index;
768 _lock_rx_hashtbl(bond);
770 ntt = 0;
771 hash_index = bond_info->rx_hashtbl_head;
772 for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
773 client_info = &(bond_info->rx_hashtbl[hash_index]);
774 assigned_slave = rlb_next_rx_slave(bond);
775 if (assigned_slave && (client_info->slave != assigned_slave)) {
776 client_info->slave = assigned_slave;
777 client_info->ntt = 1;
778 ntt = 1;
782 /* update the team's flag only after the whole iteration */
783 if (ntt) {
784 bond_info->rx_ntt = 1;
786 _unlock_rx_hashtbl(bond);
789 /* Caller must hold rx_hashtbl lock */
790 static void rlb_init_table_entry(struct rlb_client_info *entry)
792 memset(entry, 0, sizeof(struct rlb_client_info));
793 entry->next = RLB_NULL_INDEX;
794 entry->prev = RLB_NULL_INDEX;
797 static int rlb_initialize(struct bonding *bond)
799 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
800 struct packet_type *pk_type = &(BOND_ALB_INFO(bond).rlb_pkt_type);
801 struct rlb_client_info *new_hashtbl;
802 int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
803 int i;
805 spin_lock_init(&(bond_info->rx_hashtbl_lock));
807 new_hashtbl = kmalloc(size, GFP_KERNEL);
808 if (!new_hashtbl) {
809 pr_err(DRV_NAME
810 ": %s: Error: Failed to allocate RLB hash table\n",
811 bond->dev->name);
812 return -1;
814 _lock_rx_hashtbl(bond);
816 bond_info->rx_hashtbl = new_hashtbl;
818 bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
820 for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
821 rlb_init_table_entry(bond_info->rx_hashtbl + i);
824 _unlock_rx_hashtbl(bond);
826 /*initialize packet type*/
827 pk_type->type = cpu_to_be16(ETH_P_ARP);
828 pk_type->dev = NULL;
829 pk_type->func = rlb_arp_recv;
831 /* register to receive ARPs */
832 dev_add_pack(pk_type);
834 return 0;
837 static void rlb_deinitialize(struct bonding *bond)
839 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
841 dev_remove_pack(&(bond_info->rlb_pkt_type));
843 _lock_rx_hashtbl(bond);
845 kfree(bond_info->rx_hashtbl);
846 bond_info->rx_hashtbl = NULL;
847 bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
849 _unlock_rx_hashtbl(bond);
852 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
854 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
855 u32 curr_index;
857 _lock_rx_hashtbl(bond);
859 curr_index = bond_info->rx_hashtbl_head;
860 while (curr_index != RLB_NULL_INDEX) {
861 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
862 u32 next_index = bond_info->rx_hashtbl[curr_index].next;
863 u32 prev_index = bond_info->rx_hashtbl[curr_index].prev;
865 if (curr->tag && (curr->vlan_id == vlan_id)) {
866 if (curr_index == bond_info->rx_hashtbl_head) {
867 bond_info->rx_hashtbl_head = next_index;
869 if (prev_index != RLB_NULL_INDEX) {
870 bond_info->rx_hashtbl[prev_index].next = next_index;
872 if (next_index != RLB_NULL_INDEX) {
873 bond_info->rx_hashtbl[next_index].prev = prev_index;
876 rlb_init_table_entry(curr);
879 curr_index = next_index;
882 _unlock_rx_hashtbl(bond);
885 /*********************** tlb/rlb shared functions *********************/
887 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
889 struct bonding *bond = bond_get_bond_by_slave(slave);
890 struct learning_pkt pkt;
891 int size = sizeof(struct learning_pkt);
892 int i;
894 memset(&pkt, 0, size);
895 memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
896 memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
897 pkt.type = cpu_to_be16(ETH_P_LOOP);
899 for (i = 0; i < MAX_LP_BURST; i++) {
900 struct sk_buff *skb;
901 char *data;
903 skb = dev_alloc_skb(size);
904 if (!skb) {
905 return;
908 data = skb_put(skb, size);
909 memcpy(data, &pkt, size);
911 skb_reset_mac_header(skb);
912 skb->network_header = skb->mac_header + ETH_HLEN;
913 skb->protocol = pkt.type;
914 skb->priority = TC_PRIO_CONTROL;
915 skb->dev = slave->dev;
917 if (!list_empty(&bond->vlan_list)) {
918 struct vlan_entry *vlan;
920 vlan = bond_next_vlan(bond,
921 bond->alb_info.current_alb_vlan);
923 bond->alb_info.current_alb_vlan = vlan;
924 if (!vlan) {
925 kfree_skb(skb);
926 continue;
929 skb = vlan_put_tag(skb, vlan->vlan_id);
930 if (!skb) {
931 pr_err(DRV_NAME
932 ": %s: Error: failed to insert VLAN tag\n",
933 bond->dev->name);
934 continue;
938 dev_queue_xmit(skb);
942 /* hw is a boolean parameter that determines whether we should try and
943 * set the hw address of the device as well as the hw address of the
944 * net_device
946 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[], int hw)
948 struct net_device *dev = slave->dev;
949 struct sockaddr s_addr;
951 if (!hw) {
952 memcpy(dev->dev_addr, addr, dev->addr_len);
953 return 0;
956 /* for rlb each slave must have a unique hw mac addresses so that */
957 /* each slave will receive packets destined to a different mac */
958 memcpy(s_addr.sa_data, addr, dev->addr_len);
959 s_addr.sa_family = dev->type;
960 if (dev_set_mac_address(dev, &s_addr)) {
961 pr_err(DRV_NAME
962 ": %s: Error: dev_set_mac_address of dev %s failed! ALB "
963 "mode requires that the base driver support setting "
964 "the hw address also when the network device's "
965 "interface is open\n",
966 dev->master->name, dev->name);
967 return -EOPNOTSUPP;
969 return 0;
973 * Swap MAC addresses between two slaves.
975 * Called with RTNL held, and no other locks.
979 static void alb_swap_mac_addr(struct bonding *bond, struct slave *slave1, struct slave *slave2)
981 u8 tmp_mac_addr[ETH_ALEN];
983 memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
984 alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr, bond->alb_info.rlb_enabled);
985 alb_set_slave_mac_addr(slave2, tmp_mac_addr, bond->alb_info.rlb_enabled);
990 * Send learning packets after MAC address swap.
992 * Called with RTNL and no other locks
994 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
995 struct slave *slave2)
997 int slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
998 struct slave *disabled_slave = NULL;
1000 ASSERT_RTNL();
1002 /* fasten the change in the switch */
1003 if (SLAVE_IS_OK(slave1)) {
1004 alb_send_learning_packets(slave1, slave1->dev->dev_addr);
1005 if (bond->alb_info.rlb_enabled) {
1006 /* inform the clients that the mac address
1007 * has changed
1009 rlb_req_update_slave_clients(bond, slave1);
1011 } else {
1012 disabled_slave = slave1;
1015 if (SLAVE_IS_OK(slave2)) {
1016 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
1017 if (bond->alb_info.rlb_enabled) {
1018 /* inform the clients that the mac address
1019 * has changed
1021 rlb_req_update_slave_clients(bond, slave2);
1023 } else {
1024 disabled_slave = slave2;
1027 if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1028 /* A disabled slave was assigned an active mac addr */
1029 rlb_teach_disabled_mac_on_primary(bond,
1030 disabled_slave->dev->dev_addr);
1035 * alb_change_hw_addr_on_detach
1036 * @bond: bonding we're working on
1037 * @slave: the slave that was just detached
1039 * We assume that @slave was already detached from the slave list.
1041 * If @slave's permanent hw address is different both from its current
1042 * address and from @bond's address, then somewhere in the bond there's
1043 * a slave that has @slave's permanet address as its current address.
1044 * We'll make sure that that slave no longer uses @slave's permanent address.
1046 * Caller must hold RTNL and no other locks
1048 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1050 int perm_curr_diff;
1051 int perm_bond_diff;
1053 perm_curr_diff = compare_ether_addr_64bits(slave->perm_hwaddr,
1054 slave->dev->dev_addr);
1055 perm_bond_diff = compare_ether_addr_64bits(slave->perm_hwaddr,
1056 bond->dev->dev_addr);
1058 if (perm_curr_diff && perm_bond_diff) {
1059 struct slave *tmp_slave;
1060 int i, found = 0;
1062 bond_for_each_slave(bond, tmp_slave, i) {
1063 if (!compare_ether_addr_64bits(slave->perm_hwaddr,
1064 tmp_slave->dev->dev_addr)) {
1065 found = 1;
1066 break;
1070 if (found) {
1071 /* locking: needs RTNL and nothing else */
1072 alb_swap_mac_addr(bond, slave, tmp_slave);
1073 alb_fasten_mac_swap(bond, slave, tmp_slave);
1079 * alb_handle_addr_collision_on_attach
1080 * @bond: bonding we're working on
1081 * @slave: the slave that was just attached
1083 * checks uniqueness of slave's mac address and handles the case the
1084 * new slave uses the bonds mac address.
1086 * If the permanent hw address of @slave is @bond's hw address, we need to
1087 * find a different hw address to give @slave, that isn't in use by any other
1088 * slave in the bond. This address must be, of course, one of the premanent
1089 * addresses of the other slaves.
1091 * We go over the slave list, and for each slave there we compare its
1092 * permanent hw address with the current address of all the other slaves.
1093 * If no match was found, then we've found a slave with a permanent address
1094 * that isn't used by any other slave in the bond, so we can assign it to
1095 * @slave.
1097 * assumption: this function is called before @slave is attached to the
1098 * bond slave list.
1100 * caller must hold the bond lock for write since the mac addresses are compared
1101 * and may be swapped.
1103 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1105 struct slave *tmp_slave1, *tmp_slave2, *free_mac_slave;
1106 struct slave *has_bond_addr = bond->curr_active_slave;
1107 int i, j, found = 0;
1109 if (bond->slave_cnt == 0) {
1110 /* this is the first slave */
1111 return 0;
1114 /* if slave's mac address differs from bond's mac address
1115 * check uniqueness of slave's mac address against the other
1116 * slaves in the bond.
1118 if (compare_ether_addr_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1119 bond_for_each_slave(bond, tmp_slave1, i) {
1120 if (!compare_ether_addr_64bits(tmp_slave1->dev->dev_addr,
1121 slave->dev->dev_addr)) {
1122 found = 1;
1123 break;
1127 if (!found)
1128 return 0;
1130 /* Try setting slave mac to bond address and fall-through
1131 to code handling that situation below... */
1132 alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1133 bond->alb_info.rlb_enabled);
1136 /* The slave's address is equal to the address of the bond.
1137 * Search for a spare address in the bond for this slave.
1139 free_mac_slave = NULL;
1141 bond_for_each_slave(bond, tmp_slave1, i) {
1142 found = 0;
1143 bond_for_each_slave(bond, tmp_slave2, j) {
1144 if (!compare_ether_addr_64bits(tmp_slave1->perm_hwaddr,
1145 tmp_slave2->dev->dev_addr)) {
1146 found = 1;
1147 break;
1151 if (!found) {
1152 /* no slave has tmp_slave1's perm addr
1153 * as its curr addr
1155 free_mac_slave = tmp_slave1;
1156 break;
1159 if (!has_bond_addr) {
1160 if (!compare_ether_addr_64bits(tmp_slave1->dev->dev_addr,
1161 bond->dev->dev_addr)) {
1163 has_bond_addr = tmp_slave1;
1168 if (free_mac_slave) {
1169 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1170 bond->alb_info.rlb_enabled);
1172 pr_warning(DRV_NAME
1173 ": %s: Warning: the hw address of slave %s is "
1174 "in use by the bond; giving it the hw address "
1175 "of %s\n",
1176 bond->dev->name, slave->dev->name,
1177 free_mac_slave->dev->name);
1179 } else if (has_bond_addr) {
1180 pr_err(DRV_NAME
1181 ": %s: Error: the hw address of slave %s is in use by the "
1182 "bond; couldn't find a slave with a free hw address to "
1183 "give it (this should not have happened)\n",
1184 bond->dev->name, slave->dev->name);
1185 return -EFAULT;
1188 return 0;
1192 * alb_set_mac_address
1193 * @bond:
1194 * @addr:
1196 * In TLB mode all slaves are configured to the bond's hw address, but set
1197 * their dev_addr field to different addresses (based on their permanent hw
1198 * addresses).
1200 * For each slave, this function sets the interface to the new address and then
1201 * changes its dev_addr field to its previous value.
1203 * Unwinding assumes bond's mac address has not yet changed.
1205 static int alb_set_mac_address(struct bonding *bond, void *addr)
1207 struct sockaddr sa;
1208 struct slave *slave, *stop_at;
1209 char tmp_addr[ETH_ALEN];
1210 int res;
1211 int i;
1213 if (bond->alb_info.rlb_enabled) {
1214 return 0;
1217 bond_for_each_slave(bond, slave, i) {
1218 /* save net_device's current hw address */
1219 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1221 res = dev_set_mac_address(slave->dev, addr);
1223 /* restore net_device's hw address */
1224 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1226 if (res)
1227 goto unwind;
1230 return 0;
1232 unwind:
1233 memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1234 sa.sa_family = bond->dev->type;
1236 /* unwind from head to the slave that failed */
1237 stop_at = slave;
1238 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
1239 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1240 dev_set_mac_address(slave->dev, &sa);
1241 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1244 return res;
1247 /************************ exported alb funcions ************************/
1249 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1251 int res;
1253 res = tlb_initialize(bond);
1254 if (res) {
1255 return res;
1258 if (rlb_enabled) {
1259 bond->alb_info.rlb_enabled = 1;
1260 /* initialize rlb */
1261 res = rlb_initialize(bond);
1262 if (res) {
1263 tlb_deinitialize(bond);
1264 return res;
1266 } else {
1267 bond->alb_info.rlb_enabled = 0;
1270 return 0;
1273 void bond_alb_deinitialize(struct bonding *bond)
1275 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1277 tlb_deinitialize(bond);
1279 if (bond_info->rlb_enabled) {
1280 rlb_deinitialize(bond);
1284 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1286 struct bonding *bond = netdev_priv(bond_dev);
1287 struct ethhdr *eth_data;
1288 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1289 struct slave *tx_slave = NULL;
1290 static const __be32 ip_bcast = htonl(0xffffffff);
1291 int hash_size = 0;
1292 int do_tx_balance = 1;
1293 u32 hash_index = 0;
1294 const u8 *hash_start = NULL;
1295 int res = 1;
1296 struct ipv6hdr *ip6hdr;
1298 skb_reset_mac_header(skb);
1299 eth_data = eth_hdr(skb);
1301 /* make sure that the curr_active_slave and the slaves list do
1302 * not change during tx
1304 read_lock(&bond->lock);
1305 read_lock(&bond->curr_slave_lock);
1307 if (!BOND_IS_OK(bond)) {
1308 goto out;
1311 switch (ntohs(skb->protocol)) {
1312 case ETH_P_IP: {
1313 const struct iphdr *iph = ip_hdr(skb);
1315 if (!compare_ether_addr_64bits(eth_data->h_dest, mac_bcast) ||
1316 (iph->daddr == ip_bcast) ||
1317 (iph->protocol == IPPROTO_IGMP)) {
1318 do_tx_balance = 0;
1319 break;
1321 hash_start = (char *)&(iph->daddr);
1322 hash_size = sizeof(iph->daddr);
1324 break;
1325 case ETH_P_IPV6:
1326 /* IPv6 doesn't really use broadcast mac address, but leave
1327 * that here just in case.
1329 if (!compare_ether_addr_64bits(eth_data->h_dest, mac_bcast)) {
1330 do_tx_balance = 0;
1331 break;
1334 /* IPv6 uses all-nodes multicast as an equivalent to
1335 * broadcasts in IPv4.
1337 if (!compare_ether_addr_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1338 do_tx_balance = 0;
1339 break;
1342 /* Additianally, DAD probes should not be tx-balanced as that
1343 * will lead to false positives for duplicate addresses and
1344 * prevent address configuration from working.
1346 ip6hdr = ipv6_hdr(skb);
1347 if (ipv6_addr_any(&ip6hdr->saddr)) {
1348 do_tx_balance = 0;
1349 break;
1352 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1353 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1354 break;
1355 case ETH_P_IPX:
1356 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1357 /* something is wrong with this packet */
1358 do_tx_balance = 0;
1359 break;
1362 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1363 /* The only protocol worth balancing in
1364 * this family since it has an "ARP" like
1365 * mechanism
1367 do_tx_balance = 0;
1368 break;
1371 hash_start = (char*)eth_data->h_dest;
1372 hash_size = ETH_ALEN;
1373 break;
1374 case ETH_P_ARP:
1375 do_tx_balance = 0;
1376 if (bond_info->rlb_enabled) {
1377 tx_slave = rlb_arp_xmit(skb, bond);
1379 break;
1380 default:
1381 do_tx_balance = 0;
1382 break;
1385 if (do_tx_balance) {
1386 hash_index = _simple_hash(hash_start, hash_size);
1387 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1390 if (!tx_slave) {
1391 /* unbalanced or unassigned, send through primary */
1392 tx_slave = bond->curr_active_slave;
1393 bond_info->unbalanced_load += skb->len;
1396 if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1397 if (tx_slave != bond->curr_active_slave) {
1398 memcpy(eth_data->h_source,
1399 tx_slave->dev->dev_addr,
1400 ETH_ALEN);
1403 res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1404 } else {
1405 if (tx_slave) {
1406 tlb_clear_slave(bond, tx_slave, 0);
1410 out:
1411 if (res) {
1412 /* no suitable interface, frame not sent */
1413 dev_kfree_skb(skb);
1415 read_unlock(&bond->curr_slave_lock);
1416 read_unlock(&bond->lock);
1417 return NETDEV_TX_OK;
1420 void bond_alb_monitor(struct work_struct *work)
1422 struct bonding *bond = container_of(work, struct bonding,
1423 alb_work.work);
1424 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1425 struct slave *slave;
1426 int i;
1428 read_lock(&bond->lock);
1430 if (bond->kill_timers) {
1431 goto out;
1434 if (bond->slave_cnt == 0) {
1435 bond_info->tx_rebalance_counter = 0;
1436 bond_info->lp_counter = 0;
1437 goto re_arm;
1440 bond_info->tx_rebalance_counter++;
1441 bond_info->lp_counter++;
1443 /* send learning packets */
1444 if (bond_info->lp_counter >= BOND_ALB_LP_TICKS) {
1445 /* change of curr_active_slave involves swapping of mac addresses.
1446 * in order to avoid this swapping from happening while
1447 * sending the learning packets, the curr_slave_lock must be held for
1448 * read.
1450 read_lock(&bond->curr_slave_lock);
1452 bond_for_each_slave(bond, slave, i) {
1453 alb_send_learning_packets(slave, slave->dev->dev_addr);
1456 read_unlock(&bond->curr_slave_lock);
1458 bond_info->lp_counter = 0;
1461 /* rebalance tx traffic */
1462 if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1464 read_lock(&bond->curr_slave_lock);
1466 bond_for_each_slave(bond, slave, i) {
1467 tlb_clear_slave(bond, slave, 1);
1468 if (slave == bond->curr_active_slave) {
1469 SLAVE_TLB_INFO(slave).load =
1470 bond_info->unbalanced_load /
1471 BOND_TLB_REBALANCE_INTERVAL;
1472 bond_info->unbalanced_load = 0;
1476 read_unlock(&bond->curr_slave_lock);
1478 bond_info->tx_rebalance_counter = 0;
1481 /* handle rlb stuff */
1482 if (bond_info->rlb_enabled) {
1483 if (bond_info->primary_is_promisc &&
1484 (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1487 * dev_set_promiscuity requires rtnl and
1488 * nothing else.
1490 read_unlock(&bond->lock);
1491 rtnl_lock();
1493 bond_info->rlb_promisc_timeout_counter = 0;
1495 /* If the primary was set to promiscuous mode
1496 * because a slave was disabled then
1497 * it can now leave promiscuous mode.
1499 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1500 bond_info->primary_is_promisc = 0;
1502 rtnl_unlock();
1503 read_lock(&bond->lock);
1506 if (bond_info->rlb_rebalance) {
1507 bond_info->rlb_rebalance = 0;
1508 rlb_rebalance(bond);
1511 /* check if clients need updating */
1512 if (bond_info->rx_ntt) {
1513 if (bond_info->rlb_update_delay_counter) {
1514 --bond_info->rlb_update_delay_counter;
1515 } else {
1516 rlb_update_rx_clients(bond);
1517 if (bond_info->rlb_update_retry_counter) {
1518 --bond_info->rlb_update_retry_counter;
1519 } else {
1520 bond_info->rx_ntt = 0;
1526 re_arm:
1527 queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1528 out:
1529 read_unlock(&bond->lock);
1532 /* assumption: called before the slave is attached to the bond
1533 * and not locked by the bond lock
1535 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1537 int res;
1539 res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1540 bond->alb_info.rlb_enabled);
1541 if (res) {
1542 return res;
1545 /* caller must hold the bond lock for write since the mac addresses
1546 * are compared and may be swapped.
1548 read_lock(&bond->lock);
1550 res = alb_handle_addr_collision_on_attach(bond, slave);
1552 read_unlock(&bond->lock);
1554 if (res) {
1555 return res;
1558 tlb_init_slave(slave);
1560 /* order a rebalance ASAP */
1561 bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1563 if (bond->alb_info.rlb_enabled) {
1564 bond->alb_info.rlb_rebalance = 1;
1567 return 0;
1571 * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1572 * if necessary.
1574 * Caller must hold RTNL and no other locks
1576 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1578 if (bond->slave_cnt > 1) {
1579 alb_change_hw_addr_on_detach(bond, slave);
1582 tlb_clear_slave(bond, slave, 0);
1584 if (bond->alb_info.rlb_enabled) {
1585 bond->alb_info.next_rx_slave = NULL;
1586 rlb_clear_slave(bond, slave);
1590 /* Caller must hold bond lock for read */
1591 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1593 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1595 if (link == BOND_LINK_DOWN) {
1596 tlb_clear_slave(bond, slave, 0);
1597 if (bond->alb_info.rlb_enabled) {
1598 rlb_clear_slave(bond, slave);
1600 } else if (link == BOND_LINK_UP) {
1601 /* order a rebalance ASAP */
1602 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1603 if (bond->alb_info.rlb_enabled) {
1604 bond->alb_info.rlb_rebalance = 1;
1605 /* If the updelay module parameter is smaller than the
1606 * forwarding delay of the switch the rebalance will
1607 * not work because the rebalance arp replies will
1608 * not be forwarded to the clients..
1615 * bond_alb_handle_active_change - assign new curr_active_slave
1616 * @bond: our bonding struct
1617 * @new_slave: new slave to assign
1619 * Set the bond->curr_active_slave to @new_slave and handle
1620 * mac address swapping and promiscuity changes as needed.
1622 * If new_slave is NULL, caller must hold curr_slave_lock or
1623 * bond->lock for write.
1625 * If new_slave is not NULL, caller must hold RTNL, bond->lock for
1626 * read and curr_slave_lock for write. Processing here may sleep, so
1627 * no other locks may be held.
1629 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1630 __releases(&bond->curr_slave_lock)
1631 __releases(&bond->lock)
1632 __acquires(&bond->lock)
1633 __acquires(&bond->curr_slave_lock)
1635 struct slave *swap_slave;
1636 int i;
1638 if (bond->curr_active_slave == new_slave) {
1639 return;
1642 if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1643 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1644 bond->alb_info.primary_is_promisc = 0;
1645 bond->alb_info.rlb_promisc_timeout_counter = 0;
1648 swap_slave = bond->curr_active_slave;
1649 bond->curr_active_slave = new_slave;
1651 if (!new_slave || (bond->slave_cnt == 0)) {
1652 return;
1655 /* set the new curr_active_slave to the bonds mac address
1656 * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1658 if (!swap_slave) {
1659 struct slave *tmp_slave;
1660 /* find slave that is holding the bond's mac address */
1661 bond_for_each_slave(bond, tmp_slave, i) {
1662 if (!compare_ether_addr_64bits(tmp_slave->dev->dev_addr,
1663 bond->dev->dev_addr)) {
1664 swap_slave = tmp_slave;
1665 break;
1671 * Arrange for swap_slave and new_slave to temporarily be
1672 * ignored so we can mess with their MAC addresses without
1673 * fear of interference from transmit activity.
1675 if (swap_slave) {
1676 tlb_clear_slave(bond, swap_slave, 1);
1678 tlb_clear_slave(bond, new_slave, 1);
1680 write_unlock_bh(&bond->curr_slave_lock);
1681 read_unlock(&bond->lock);
1683 ASSERT_RTNL();
1685 /* curr_active_slave must be set before calling alb_swap_mac_addr */
1686 if (swap_slave) {
1687 /* swap mac address */
1688 alb_swap_mac_addr(bond, swap_slave, new_slave);
1689 } else {
1690 /* set the new_slave to the bond mac address */
1691 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1692 bond->alb_info.rlb_enabled);
1695 if (swap_slave) {
1696 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1697 read_lock(&bond->lock);
1698 } else {
1699 read_lock(&bond->lock);
1700 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1703 write_lock_bh(&bond->curr_slave_lock);
1707 * Called with RTNL
1709 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1710 __acquires(&bond->lock)
1711 __releases(&bond->lock)
1713 struct bonding *bond = netdev_priv(bond_dev);
1714 struct sockaddr *sa = addr;
1715 struct slave *slave, *swap_slave;
1716 int res;
1717 int i;
1719 if (!is_valid_ether_addr(sa->sa_data)) {
1720 return -EADDRNOTAVAIL;
1723 res = alb_set_mac_address(bond, addr);
1724 if (res) {
1725 return res;
1728 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1730 /* If there is no curr_active_slave there is nothing else to do.
1731 * Otherwise we'll need to pass the new address to it and handle
1732 * duplications.
1734 if (!bond->curr_active_slave) {
1735 return 0;
1738 swap_slave = NULL;
1740 bond_for_each_slave(bond, slave, i) {
1741 if (!compare_ether_addr_64bits(slave->dev->dev_addr,
1742 bond_dev->dev_addr)) {
1743 swap_slave = slave;
1744 break;
1748 if (swap_slave) {
1749 alb_swap_mac_addr(bond, swap_slave, bond->curr_active_slave);
1750 alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave);
1751 } else {
1752 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr,
1753 bond->alb_info.rlb_enabled);
1755 read_lock(&bond->lock);
1756 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1757 if (bond->alb_info.rlb_enabled) {
1758 /* inform clients mac address has changed */
1759 rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1761 read_unlock(&bond->lock);
1764 return 0;
1767 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1769 if (bond->alb_info.current_alb_vlan &&
1770 (bond->alb_info.current_alb_vlan->vlan_id == vlan_id)) {
1771 bond->alb_info.current_alb_vlan = NULL;
1774 if (bond->alb_info.rlb_enabled) {
1775 rlb_clear_vlan(bond, vlan_id);