Linux 3.4.102
[linux/fpc-iii.git] / drivers / net / bonding / bond_alb.c
blobb6ed7e9bbb80cda409efa6c02d1ab549084cc40b
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
25 #include <linux/skbuff.h>
26 #include <linux/netdevice.h>
27 #include <linux/etherdevice.h>
28 #include <linux/pkt_sched.h>
29 #include <linux/spinlock.h>
30 #include <linux/slab.h>
31 #include <linux/timer.h>
32 #include <linux/ip.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>
38 #include <linux/in.h>
39 #include <net/ipx.h>
40 #include <net/arp.h>
41 #include <net/ipv6.h>
42 #include <asm/byteorder.h>
43 #include "bonding.h"
44 #include "bond_alb.h"
48 #ifndef __long_aligned
49 #define __long_aligned __attribute__((aligned((sizeof(long)))))
50 #endif
51 static const u8 mac_bcast[ETH_ALEN] __long_aligned = {
52 0xff, 0xff, 0xff, 0xff, 0xff, 0xff
54 static const u8 mac_v6_allmcast[ETH_ALEN] __long_aligned = {
55 0x33, 0x33, 0x00, 0x00, 0x00, 0x01
57 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
59 #pragma pack(1)
60 struct learning_pkt {
61 u8 mac_dst[ETH_ALEN];
62 u8 mac_src[ETH_ALEN];
63 __be16 type;
64 u8 padding[ETH_ZLEN - ETH_HLEN];
67 struct arp_pkt {
68 __be16 hw_addr_space;
69 __be16 prot_addr_space;
70 u8 hw_addr_len;
71 u8 prot_addr_len;
72 __be16 op_code;
73 u8 mac_src[ETH_ALEN]; /* sender hardware address */
74 __be32 ip_src; /* sender IP address */
75 u8 mac_dst[ETH_ALEN]; /* target hardware address */
76 __be32 ip_dst; /* target IP address */
78 #pragma pack()
80 static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb)
82 return (struct arp_pkt *)skb_network_header(skb);
85 /* Forward declaration */
86 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[]);
88 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
90 int i;
91 u8 hash = 0;
93 for (i = 0; i < hash_size; i++) {
94 hash ^= hash_start[i];
97 return hash;
100 /*********************** tlb specific functions ***************************/
102 static inline void _lock_tx_hashtbl_bh(struct bonding *bond)
104 spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
107 static inline void _unlock_tx_hashtbl_bh(struct bonding *bond)
109 spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
112 static inline void _lock_tx_hashtbl(struct bonding *bond)
114 spin_lock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
117 static inline void _unlock_tx_hashtbl(struct bonding *bond)
119 spin_unlock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
122 /* Caller must hold tx_hashtbl lock */
123 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
125 if (save_load) {
126 entry->load_history = 1 + entry->tx_bytes /
127 BOND_TLB_REBALANCE_INTERVAL;
128 entry->tx_bytes = 0;
131 entry->tx_slave = NULL;
132 entry->next = TLB_NULL_INDEX;
133 entry->prev = TLB_NULL_INDEX;
136 static inline void tlb_init_slave(struct slave *slave)
138 SLAVE_TLB_INFO(slave).load = 0;
139 SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
142 /* Caller must hold bond lock for read, BH disabled */
143 static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
144 int save_load)
146 struct tlb_client_info *tx_hash_table;
147 u32 index;
149 /* clear slave from tx_hashtbl */
150 tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
152 /* skip this if we've already freed the tx hash table */
153 if (tx_hash_table) {
154 index = SLAVE_TLB_INFO(slave).head;
155 while (index != TLB_NULL_INDEX) {
156 u32 next_index = tx_hash_table[index].next;
157 tlb_init_table_entry(&tx_hash_table[index], save_load);
158 index = next_index;
162 tlb_init_slave(slave);
165 /* Caller must hold bond lock for read */
166 static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
167 int save_load)
169 _lock_tx_hashtbl_bh(bond);
170 __tlb_clear_slave(bond, slave, save_load);
171 _unlock_tx_hashtbl_bh(bond);
174 /* Must be called before starting the monitor timer */
175 static int tlb_initialize(struct bonding *bond)
177 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
178 int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
179 struct tlb_client_info *new_hashtbl;
180 int i;
182 new_hashtbl = kzalloc(size, GFP_KERNEL);
183 if (!new_hashtbl)
184 return -1;
186 _lock_tx_hashtbl_bh(bond);
188 bond_info->tx_hashtbl = new_hashtbl;
190 for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
191 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
194 _unlock_tx_hashtbl_bh(bond);
196 return 0;
199 /* Must be called only after all slaves have been released */
200 static void tlb_deinitialize(struct bonding *bond)
202 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
204 _lock_tx_hashtbl_bh(bond);
206 kfree(bond_info->tx_hashtbl);
207 bond_info->tx_hashtbl = NULL;
209 _unlock_tx_hashtbl_bh(bond);
212 static long long compute_gap(struct slave *slave)
214 return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
215 (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
218 /* Caller must hold bond lock for read */
219 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
221 struct slave *slave, *least_loaded;
222 long long max_gap;
223 int i;
225 least_loaded = NULL;
226 max_gap = LLONG_MIN;
228 /* Find the slave with the largest gap */
229 bond_for_each_slave(bond, slave, i) {
230 if (SLAVE_IS_OK(slave)) {
231 long long gap = compute_gap(slave);
233 if (max_gap < gap) {
234 least_loaded = slave;
235 max_gap = gap;
240 return least_loaded;
243 static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
244 u32 skb_len)
246 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
247 struct tlb_client_info *hash_table;
248 struct slave *assigned_slave;
250 hash_table = bond_info->tx_hashtbl;
251 assigned_slave = hash_table[hash_index].tx_slave;
252 if (!assigned_slave) {
253 assigned_slave = tlb_get_least_loaded_slave(bond);
255 if (assigned_slave) {
256 struct tlb_slave_info *slave_info =
257 &(SLAVE_TLB_INFO(assigned_slave));
258 u32 next_index = slave_info->head;
260 hash_table[hash_index].tx_slave = assigned_slave;
261 hash_table[hash_index].next = next_index;
262 hash_table[hash_index].prev = TLB_NULL_INDEX;
264 if (next_index != TLB_NULL_INDEX) {
265 hash_table[next_index].prev = hash_index;
268 slave_info->head = hash_index;
269 slave_info->load +=
270 hash_table[hash_index].load_history;
274 if (assigned_slave) {
275 hash_table[hash_index].tx_bytes += skb_len;
278 return assigned_slave;
281 /* Caller must hold bond lock for read */
282 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
283 u32 skb_len)
285 struct slave *tx_slave;
287 * We don't need to disable softirq here, becase
288 * tlb_choose_channel() is only called by bond_alb_xmit()
289 * which already has softirq disabled.
291 _lock_tx_hashtbl(bond);
292 tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
293 _unlock_tx_hashtbl(bond);
294 return tx_slave;
297 /*********************** rlb specific functions ***************************/
298 static inline void _lock_rx_hashtbl_bh(struct bonding *bond)
300 spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
303 static inline void _unlock_rx_hashtbl_bh(struct bonding *bond)
305 spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
308 static inline void _lock_rx_hashtbl(struct bonding *bond)
310 spin_lock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
313 static inline void _unlock_rx_hashtbl(struct bonding *bond)
315 spin_unlock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
318 /* when an ARP REPLY is received from a client update its info
319 * in the rx_hashtbl
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;
325 u32 hash_index;
327 _lock_rx_hashtbl_bh(bond);
329 hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
330 client_info = &(bond_info->rx_hashtbl[hash_index]);
332 if ((client_info->assigned) &&
333 (client_info->ip_src == arp->ip_dst) &&
334 (client_info->ip_dst == arp->ip_src) &&
335 (compare_ether_addr_64bits(client_info->mac_dst, arp->mac_src))) {
336 /* update the clients MAC address */
337 memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
338 client_info->ntt = 1;
339 bond_info->rx_ntt = 1;
342 _unlock_rx_hashtbl_bh(bond);
345 static int rlb_arp_recv(struct sk_buff *skb, struct bonding *bond,
346 struct slave *slave)
348 struct arp_pkt *arp;
350 if (skb->protocol != cpu_to_be16(ETH_P_ARP))
351 goto out;
353 arp = (struct arp_pkt *) skb->data;
354 if (!arp) {
355 pr_debug("Packet has no ARP data\n");
356 goto out;
359 if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
360 goto out;
362 if (skb->len < sizeof(struct arp_pkt)) {
363 pr_debug("Packet is too small to be an ARP\n");
364 goto out;
367 if (arp->op_code == htons(ARPOP_REPLY)) {
368 /* update rx hash table for this ARP */
369 rlb_update_entry_from_arp(bond, arp);
370 pr_debug("Server received an ARP Reply from client\n");
372 out:
373 return RX_HANDLER_ANOTHER;
376 /* Caller must hold bond lock for read */
377 static struct slave *rlb_next_rx_slave(struct bonding *bond)
379 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
380 struct slave *rx_slave, *slave, *start_at;
381 int i = 0;
383 if (bond_info->next_rx_slave) {
384 start_at = bond_info->next_rx_slave;
385 } else {
386 start_at = bond->first_slave;
389 rx_slave = NULL;
391 bond_for_each_slave_from(bond, slave, i, start_at) {
392 if (SLAVE_IS_OK(slave)) {
393 if (!rx_slave) {
394 rx_slave = slave;
395 } else if (slave->speed > rx_slave->speed) {
396 rx_slave = slave;
401 if (rx_slave) {
402 bond_info->next_rx_slave = rx_slave->next;
405 return rx_slave;
408 /* teach the switch the mac of a disabled slave
409 * on the primary for fault tolerance
411 * Caller must hold bond->curr_slave_lock for write or bond lock for write
413 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
415 if (!bond->curr_active_slave) {
416 return;
419 if (!bond->alb_info.primary_is_promisc) {
420 if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1))
421 bond->alb_info.primary_is_promisc = 1;
422 else
423 bond->alb_info.primary_is_promisc = 0;
426 bond->alb_info.rlb_promisc_timeout_counter = 0;
428 alb_send_learning_packets(bond->curr_active_slave, addr);
431 /* slave being removed should not be active at this point
433 * Caller must hold bond lock for read
435 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
437 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
438 struct rlb_client_info *rx_hash_table;
439 u32 index, next_index;
441 /* clear slave from rx_hashtbl */
442 _lock_rx_hashtbl_bh(bond);
444 rx_hash_table = bond_info->rx_hashtbl;
445 index = bond_info->rx_hashtbl_head;
446 for (; index != RLB_NULL_INDEX; index = next_index) {
447 next_index = rx_hash_table[index].next;
448 if (rx_hash_table[index].slave == slave) {
449 struct slave *assigned_slave = rlb_next_rx_slave(bond);
451 if (assigned_slave) {
452 rx_hash_table[index].slave = assigned_slave;
453 if (compare_ether_addr_64bits(rx_hash_table[index].mac_dst,
454 mac_bcast)) {
455 bond_info->rx_hashtbl[index].ntt = 1;
456 bond_info->rx_ntt = 1;
457 /* A slave has been removed from the
458 * table because it is either disabled
459 * or being released. We must retry the
460 * update to avoid clients from not
461 * being updated & disconnecting when
462 * there is stress
464 bond_info->rlb_update_retry_counter =
465 RLB_UPDATE_RETRY;
467 } else { /* there is no active slave */
468 rx_hash_table[index].slave = NULL;
473 _unlock_rx_hashtbl_bh(bond);
475 write_lock_bh(&bond->curr_slave_lock);
477 if (slave != bond->curr_active_slave) {
478 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
481 write_unlock_bh(&bond->curr_slave_lock);
484 static void rlb_update_client(struct rlb_client_info *client_info)
486 int i;
488 if (!client_info->slave) {
489 return;
492 for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
493 struct sk_buff *skb;
495 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
496 client_info->ip_dst,
497 client_info->slave->dev,
498 client_info->ip_src,
499 client_info->mac_dst,
500 client_info->slave->dev->dev_addr,
501 client_info->mac_dst);
502 if (!skb) {
503 pr_err("%s: Error: failed to create an ARP packet\n",
504 client_info->slave->dev->master->name);
505 continue;
508 skb->dev = client_info->slave->dev;
510 if (client_info->tag) {
511 skb = vlan_put_tag(skb, client_info->vlan_id);
512 if (!skb) {
513 pr_err("%s: Error: failed to insert VLAN tag\n",
514 client_info->slave->dev->master->name);
515 continue;
519 arp_xmit(skb);
523 /* sends ARP REPLIES that update the clients that need updating */
524 static void rlb_update_rx_clients(struct bonding *bond)
526 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
527 struct rlb_client_info *client_info;
528 u32 hash_index;
530 _lock_rx_hashtbl_bh(bond);
532 hash_index = bond_info->rx_hashtbl_head;
533 for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
534 client_info = &(bond_info->rx_hashtbl[hash_index]);
535 if (client_info->ntt) {
536 rlb_update_client(client_info);
537 if (bond_info->rlb_update_retry_counter == 0) {
538 client_info->ntt = 0;
543 /* do not update the entries again until this counter is zero so that
544 * not to confuse the clients.
546 bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
548 _unlock_rx_hashtbl_bh(bond);
551 /* The slave was assigned a new mac address - update the clients */
552 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
554 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
555 struct rlb_client_info *client_info;
556 int ntt = 0;
557 u32 hash_index;
559 _lock_rx_hashtbl_bh(bond);
561 hash_index = bond_info->rx_hashtbl_head;
562 for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
563 client_info = &(bond_info->rx_hashtbl[hash_index]);
565 if ((client_info->slave == slave) &&
566 compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
567 client_info->ntt = 1;
568 ntt = 1;
572 // update the team's flag only after the whole iteration
573 if (ntt) {
574 bond_info->rx_ntt = 1;
575 //fasten the change
576 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
579 _unlock_rx_hashtbl_bh(bond);
582 /* mark all clients using src_ip to be updated */
583 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
585 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
586 struct rlb_client_info *client_info;
587 u32 hash_index;
589 _lock_rx_hashtbl(bond);
591 hash_index = bond_info->rx_hashtbl_head;
592 for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
593 client_info = &(bond_info->rx_hashtbl[hash_index]);
595 if (!client_info->slave) {
596 pr_err("%s: Error: found a client with no channel in the client's hash table\n",
597 bond->dev->name);
598 continue;
600 /*update all clients using this src_ip, that are not assigned
601 * to the team's address (curr_active_slave) and have a known
602 * unicast mac address.
604 if ((client_info->ip_src == src_ip) &&
605 compare_ether_addr_64bits(client_info->slave->dev->dev_addr,
606 bond->dev->dev_addr) &&
607 compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
608 client_info->ntt = 1;
609 bond_info->rx_ntt = 1;
613 _unlock_rx_hashtbl(bond);
616 /* Caller must hold both bond and ptr locks for read */
617 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
619 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
620 struct arp_pkt *arp = arp_pkt(skb);
621 struct slave *assigned_slave;
622 struct rlb_client_info *client_info;
623 u32 hash_index = 0;
625 _lock_rx_hashtbl(bond);
627 hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
628 client_info = &(bond_info->rx_hashtbl[hash_index]);
630 if (client_info->assigned) {
631 if ((client_info->ip_src == arp->ip_src) &&
632 (client_info->ip_dst == arp->ip_dst)) {
633 /* the entry is already assigned to this client */
634 if (compare_ether_addr_64bits(arp->mac_dst, mac_bcast)) {
635 /* update mac address from arp */
636 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
639 assigned_slave = client_info->slave;
640 if (assigned_slave) {
641 _unlock_rx_hashtbl(bond);
642 return assigned_slave;
644 } else {
645 /* the entry is already assigned to some other client,
646 * move the old client to primary (curr_active_slave) so
647 * that the new client can be assigned to this entry.
649 if (bond->curr_active_slave &&
650 client_info->slave != bond->curr_active_slave) {
651 client_info->slave = bond->curr_active_slave;
652 rlb_update_client(client_info);
656 /* assign a new slave */
657 assigned_slave = rlb_next_rx_slave(bond);
659 if (assigned_slave) {
660 client_info->ip_src = arp->ip_src;
661 client_info->ip_dst = arp->ip_dst;
662 /* arp->mac_dst is broadcast for arp reqeusts.
663 * will be updated with clients actual unicast mac address
664 * upon receiving an arp reply.
666 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
667 client_info->slave = assigned_slave;
669 if (compare_ether_addr_64bits(client_info->mac_dst, mac_bcast)) {
670 client_info->ntt = 1;
671 bond->alb_info.rx_ntt = 1;
672 } else {
673 client_info->ntt = 0;
676 if (bond_vlan_used(bond)) {
677 if (!vlan_get_tag(skb, &client_info->vlan_id))
678 client_info->tag = 1;
681 if (!client_info->assigned) {
682 u32 prev_tbl_head = bond_info->rx_hashtbl_head;
683 bond_info->rx_hashtbl_head = hash_index;
684 client_info->next = prev_tbl_head;
685 if (prev_tbl_head != RLB_NULL_INDEX) {
686 bond_info->rx_hashtbl[prev_tbl_head].prev =
687 hash_index;
689 client_info->assigned = 1;
693 _unlock_rx_hashtbl(bond);
695 return assigned_slave;
698 /* chooses (and returns) transmit channel for arp reply
699 * does not choose channel for other arp types since they are
700 * sent on the curr_active_slave
702 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
704 struct arp_pkt *arp = arp_pkt(skb);
705 struct slave *tx_slave = NULL;
707 /* Don't modify or load balance ARPs that do not originate locally
708 * (e.g.,arrive via a bridge).
710 if (!bond_slave_has_mac(bond, arp->mac_src))
711 return NULL;
713 if (arp->op_code == htons(ARPOP_REPLY)) {
714 /* the arp must be sent on the selected
715 * rx channel
717 tx_slave = rlb_choose_channel(skb, bond);
718 if (tx_slave) {
719 memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
721 pr_debug("Server sent ARP Reply packet\n");
722 } else if (arp->op_code == htons(ARPOP_REQUEST)) {
723 /* Create an entry in the rx_hashtbl for this client as a
724 * place holder.
725 * When the arp reply is received the entry will be updated
726 * with the correct unicast address of the client.
728 rlb_choose_channel(skb, bond);
730 /* The ARP reply packets must be delayed so that
731 * they can cancel out the influence of the ARP request.
733 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
735 /* arp requests are broadcast and are sent on the primary
736 * the arp request will collapse all clients on the subnet to
737 * the primary slave. We must register these clients to be
738 * updated with their assigned mac.
740 rlb_req_update_subnet_clients(bond, arp->ip_src);
741 pr_debug("Server sent ARP Request packet\n");
744 return tx_slave;
747 /* Caller must hold bond lock for read */
748 static void rlb_rebalance(struct bonding *bond)
750 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
751 struct slave *assigned_slave;
752 struct rlb_client_info *client_info;
753 int ntt;
754 u32 hash_index;
756 _lock_rx_hashtbl_bh(bond);
758 ntt = 0;
759 hash_index = bond_info->rx_hashtbl_head;
760 for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
761 client_info = &(bond_info->rx_hashtbl[hash_index]);
762 assigned_slave = rlb_next_rx_slave(bond);
763 if (assigned_slave && (client_info->slave != assigned_slave)) {
764 client_info->slave = assigned_slave;
765 client_info->ntt = 1;
766 ntt = 1;
770 /* update the team's flag only after the whole iteration */
771 if (ntt) {
772 bond_info->rx_ntt = 1;
774 _unlock_rx_hashtbl_bh(bond);
777 /* Caller must hold rx_hashtbl lock */
778 static void rlb_init_table_entry(struct rlb_client_info *entry)
780 memset(entry, 0, sizeof(struct rlb_client_info));
781 entry->next = RLB_NULL_INDEX;
782 entry->prev = RLB_NULL_INDEX;
785 static int rlb_initialize(struct bonding *bond)
787 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
788 struct rlb_client_info *new_hashtbl;
789 int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
790 int i;
792 new_hashtbl = kmalloc(size, GFP_KERNEL);
793 if (!new_hashtbl)
794 return -1;
796 _lock_rx_hashtbl_bh(bond);
798 bond_info->rx_hashtbl = new_hashtbl;
800 bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
802 for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
803 rlb_init_table_entry(bond_info->rx_hashtbl + i);
806 _unlock_rx_hashtbl_bh(bond);
808 /* register to receive ARPs */
809 bond->recv_probe = rlb_arp_recv;
811 return 0;
814 static void rlb_deinitialize(struct bonding *bond)
816 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
818 _lock_rx_hashtbl_bh(bond);
820 kfree(bond_info->rx_hashtbl);
821 bond_info->rx_hashtbl = NULL;
822 bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
824 _unlock_rx_hashtbl_bh(bond);
827 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
829 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
830 u32 curr_index;
832 _lock_rx_hashtbl_bh(bond);
834 curr_index = bond_info->rx_hashtbl_head;
835 while (curr_index != RLB_NULL_INDEX) {
836 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
837 u32 next_index = bond_info->rx_hashtbl[curr_index].next;
838 u32 prev_index = bond_info->rx_hashtbl[curr_index].prev;
840 if (curr->tag && (curr->vlan_id == vlan_id)) {
841 if (curr_index == bond_info->rx_hashtbl_head) {
842 bond_info->rx_hashtbl_head = next_index;
844 if (prev_index != RLB_NULL_INDEX) {
845 bond_info->rx_hashtbl[prev_index].next = next_index;
847 if (next_index != RLB_NULL_INDEX) {
848 bond_info->rx_hashtbl[next_index].prev = prev_index;
851 rlb_init_table_entry(curr);
854 curr_index = next_index;
857 _unlock_rx_hashtbl_bh(bond);
860 /*********************** tlb/rlb shared functions *********************/
862 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
864 struct bonding *bond = bond_get_bond_by_slave(slave);
865 struct learning_pkt pkt;
866 int size = sizeof(struct learning_pkt);
867 int i;
869 memset(&pkt, 0, size);
870 memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
871 memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
872 pkt.type = cpu_to_be16(ETH_P_LOOP);
874 for (i = 0; i < MAX_LP_BURST; i++) {
875 struct sk_buff *skb;
876 char *data;
878 skb = dev_alloc_skb(size);
879 if (!skb) {
880 return;
883 data = skb_put(skb, size);
884 memcpy(data, &pkt, size);
886 skb_reset_mac_header(skb);
887 skb->network_header = skb->mac_header + ETH_HLEN;
888 skb->protocol = pkt.type;
889 skb->priority = TC_PRIO_CONTROL;
890 skb->dev = slave->dev;
892 if (bond_vlan_used(bond)) {
893 struct vlan_entry *vlan;
895 vlan = bond_next_vlan(bond,
896 bond->alb_info.current_alb_vlan);
898 bond->alb_info.current_alb_vlan = vlan;
899 if (!vlan) {
900 kfree_skb(skb);
901 continue;
904 skb = vlan_put_tag(skb, vlan->vlan_id);
905 if (!skb) {
906 pr_err("%s: Error: failed to insert VLAN tag\n",
907 bond->dev->name);
908 continue;
912 dev_queue_xmit(skb);
916 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[])
918 struct net_device *dev = slave->dev;
919 struct sockaddr s_addr;
921 if (slave->bond->params.mode == BOND_MODE_TLB) {
922 memcpy(dev->dev_addr, addr, dev->addr_len);
923 return 0;
926 /* for rlb each slave must have a unique hw mac addresses so that */
927 /* each slave will receive packets destined to a different mac */
928 memcpy(s_addr.sa_data, addr, dev->addr_len);
929 s_addr.sa_family = dev->type;
930 if (dev_set_mac_address(dev, &s_addr)) {
931 pr_err("%s: Error: dev_set_mac_address of dev %s failed!\n"
932 "ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
933 dev->master->name, dev->name);
934 return -EOPNOTSUPP;
936 return 0;
940 * Swap MAC addresses between two slaves.
942 * Called with RTNL held, and no other locks.
946 static void alb_swap_mac_addr(struct bonding *bond, struct slave *slave1, struct slave *slave2)
948 u8 tmp_mac_addr[ETH_ALEN];
950 memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
951 alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr);
952 alb_set_slave_mac_addr(slave2, tmp_mac_addr);
957 * Send learning packets after MAC address swap.
959 * Called with RTNL and no other locks
961 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
962 struct slave *slave2)
964 int slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
965 struct slave *disabled_slave = NULL;
967 ASSERT_RTNL();
969 /* fasten the change in the switch */
970 if (SLAVE_IS_OK(slave1)) {
971 alb_send_learning_packets(slave1, slave1->dev->dev_addr);
972 if (bond->alb_info.rlb_enabled) {
973 /* inform the clients that the mac address
974 * has changed
976 rlb_req_update_slave_clients(bond, slave1);
978 } else {
979 disabled_slave = slave1;
982 if (SLAVE_IS_OK(slave2)) {
983 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
984 if (bond->alb_info.rlb_enabled) {
985 /* inform the clients that the mac address
986 * has changed
988 rlb_req_update_slave_clients(bond, slave2);
990 } else {
991 disabled_slave = slave2;
994 if (bond->alb_info.rlb_enabled && slaves_state_differ) {
995 /* A disabled slave was assigned an active mac addr */
996 rlb_teach_disabled_mac_on_primary(bond,
997 disabled_slave->dev->dev_addr);
1002 * alb_change_hw_addr_on_detach
1003 * @bond: bonding we're working on
1004 * @slave: the slave that was just detached
1006 * We assume that @slave was already detached from the slave list.
1008 * If @slave's permanent hw address is different both from its current
1009 * address and from @bond's address, then somewhere in the bond there's
1010 * a slave that has @slave's permanet address as its current address.
1011 * We'll make sure that that slave no longer uses @slave's permanent address.
1013 * Caller must hold RTNL and no other locks
1015 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1017 int perm_curr_diff;
1018 int perm_bond_diff;
1020 perm_curr_diff = compare_ether_addr_64bits(slave->perm_hwaddr,
1021 slave->dev->dev_addr);
1022 perm_bond_diff = compare_ether_addr_64bits(slave->perm_hwaddr,
1023 bond->dev->dev_addr);
1025 if (perm_curr_diff && perm_bond_diff) {
1026 struct slave *tmp_slave;
1027 int i, found = 0;
1029 bond_for_each_slave(bond, tmp_slave, i) {
1030 if (!compare_ether_addr_64bits(slave->perm_hwaddr,
1031 tmp_slave->dev->dev_addr)) {
1032 found = 1;
1033 break;
1037 if (found) {
1038 /* locking: needs RTNL and nothing else */
1039 alb_swap_mac_addr(bond, slave, tmp_slave);
1040 alb_fasten_mac_swap(bond, slave, tmp_slave);
1046 * alb_handle_addr_collision_on_attach
1047 * @bond: bonding we're working on
1048 * @slave: the slave that was just attached
1050 * checks uniqueness of slave's mac address and handles the case the
1051 * new slave uses the bonds mac address.
1053 * If the permanent hw address of @slave is @bond's hw address, we need to
1054 * find a different hw address to give @slave, that isn't in use by any other
1055 * slave in the bond. This address must be, of course, one of the permanent
1056 * addresses of the other slaves.
1058 * We go over the slave list, and for each slave there we compare its
1059 * permanent hw address with the current address of all the other slaves.
1060 * If no match was found, then we've found a slave with a permanent address
1061 * that isn't used by any other slave in the bond, so we can assign it to
1062 * @slave.
1064 * assumption: this function is called before @slave is attached to the
1065 * bond slave list.
1067 * caller must hold the bond lock for write since the mac addresses are compared
1068 * and may be swapped.
1070 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1072 struct slave *tmp_slave1, *tmp_slave2, *free_mac_slave;
1073 struct slave *has_bond_addr = bond->curr_active_slave;
1074 int i, j, found = 0;
1076 if (bond->slave_cnt == 0) {
1077 /* this is the first slave */
1078 return 0;
1081 /* if slave's mac address differs from bond's mac address
1082 * check uniqueness of slave's mac address against the other
1083 * slaves in the bond.
1085 if (compare_ether_addr_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1086 bond_for_each_slave(bond, tmp_slave1, i) {
1087 if (!compare_ether_addr_64bits(tmp_slave1->dev->dev_addr,
1088 slave->dev->dev_addr)) {
1089 found = 1;
1090 break;
1094 if (!found)
1095 return 0;
1097 /* Try setting slave mac to bond address and fall-through
1098 to code handling that situation below... */
1099 alb_set_slave_mac_addr(slave, bond->dev->dev_addr);
1102 /* The slave's address is equal to the address of the bond.
1103 * Search for a spare address in the bond for this slave.
1105 free_mac_slave = NULL;
1107 bond_for_each_slave(bond, tmp_slave1, i) {
1108 found = 0;
1109 bond_for_each_slave(bond, tmp_slave2, j) {
1110 if (!compare_ether_addr_64bits(tmp_slave1->perm_hwaddr,
1111 tmp_slave2->dev->dev_addr)) {
1112 found = 1;
1113 break;
1117 if (!found) {
1118 /* no slave has tmp_slave1's perm addr
1119 * as its curr addr
1121 free_mac_slave = tmp_slave1;
1122 break;
1125 if (!has_bond_addr) {
1126 if (!compare_ether_addr_64bits(tmp_slave1->dev->dev_addr,
1127 bond->dev->dev_addr)) {
1129 has_bond_addr = tmp_slave1;
1134 if (free_mac_slave) {
1135 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr);
1137 pr_warning("%s: Warning: the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1138 bond->dev->name, slave->dev->name,
1139 free_mac_slave->dev->name);
1141 } else if (has_bond_addr) {
1142 pr_err("%s: Error: the hw address of slave %s is in use by the bond; couldn't find a slave with a free hw address to give it (this should not have happened)\n",
1143 bond->dev->name, slave->dev->name);
1144 return -EFAULT;
1147 return 0;
1151 * alb_set_mac_address
1152 * @bond:
1153 * @addr:
1155 * In TLB mode all slaves are configured to the bond's hw address, but set
1156 * their dev_addr field to different addresses (based on their permanent hw
1157 * addresses).
1159 * For each slave, this function sets the interface to the new address and then
1160 * changes its dev_addr field to its previous value.
1162 * Unwinding assumes bond's mac address has not yet changed.
1164 static int alb_set_mac_address(struct bonding *bond, void *addr)
1166 struct sockaddr sa;
1167 struct slave *slave, *stop_at;
1168 char tmp_addr[ETH_ALEN];
1169 int res;
1170 int i;
1172 if (bond->alb_info.rlb_enabled) {
1173 return 0;
1176 bond_for_each_slave(bond, slave, i) {
1177 /* save net_device's current hw address */
1178 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1180 res = dev_set_mac_address(slave->dev, addr);
1182 /* restore net_device's hw address */
1183 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1185 if (res)
1186 goto unwind;
1189 return 0;
1191 unwind:
1192 memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1193 sa.sa_family = bond->dev->type;
1195 /* unwind from head to the slave that failed */
1196 stop_at = slave;
1197 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
1198 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1199 dev_set_mac_address(slave->dev, &sa);
1200 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1203 return res;
1206 /************************ exported alb funcions ************************/
1208 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1210 int res;
1212 res = tlb_initialize(bond);
1213 if (res) {
1214 return res;
1217 if (rlb_enabled) {
1218 bond->alb_info.rlb_enabled = 1;
1219 /* initialize rlb */
1220 res = rlb_initialize(bond);
1221 if (res) {
1222 tlb_deinitialize(bond);
1223 return res;
1225 } else {
1226 bond->alb_info.rlb_enabled = 0;
1229 return 0;
1232 void bond_alb_deinitialize(struct bonding *bond)
1234 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1236 tlb_deinitialize(bond);
1238 if (bond_info->rlb_enabled) {
1239 rlb_deinitialize(bond);
1243 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1245 struct bonding *bond = netdev_priv(bond_dev);
1246 struct ethhdr *eth_data;
1247 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1248 struct slave *tx_slave = NULL;
1249 static const __be32 ip_bcast = htonl(0xffffffff);
1250 int hash_size = 0;
1251 int do_tx_balance = 1;
1252 u32 hash_index = 0;
1253 const u8 *hash_start = NULL;
1254 int res = 1;
1255 struct ipv6hdr *ip6hdr;
1257 skb_reset_mac_header(skb);
1258 eth_data = eth_hdr(skb);
1260 /* make sure that the curr_active_slave do not change during tx
1262 read_lock(&bond->curr_slave_lock);
1264 switch (ntohs(skb->protocol)) {
1265 case ETH_P_IP: {
1266 const struct iphdr *iph = ip_hdr(skb);
1268 if (!compare_ether_addr_64bits(eth_data->h_dest, mac_bcast) ||
1269 (iph->daddr == ip_bcast) ||
1270 (iph->protocol == IPPROTO_IGMP)) {
1271 do_tx_balance = 0;
1272 break;
1274 hash_start = (char *)&(iph->daddr);
1275 hash_size = sizeof(iph->daddr);
1277 break;
1278 case ETH_P_IPV6:
1279 /* IPv6 doesn't really use broadcast mac address, but leave
1280 * that here just in case.
1282 if (!compare_ether_addr_64bits(eth_data->h_dest, mac_bcast)) {
1283 do_tx_balance = 0;
1284 break;
1287 /* IPv6 uses all-nodes multicast as an equivalent to
1288 * broadcasts in IPv4.
1290 if (!compare_ether_addr_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1291 do_tx_balance = 0;
1292 break;
1295 /* Additianally, DAD probes should not be tx-balanced as that
1296 * will lead to false positives for duplicate addresses and
1297 * prevent address configuration from working.
1299 ip6hdr = ipv6_hdr(skb);
1300 if (ipv6_addr_any(&ip6hdr->saddr)) {
1301 do_tx_balance = 0;
1302 break;
1305 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1306 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1307 break;
1308 case ETH_P_IPX:
1309 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1310 /* something is wrong with this packet */
1311 do_tx_balance = 0;
1312 break;
1315 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1316 /* The only protocol worth balancing in
1317 * this family since it has an "ARP" like
1318 * mechanism
1320 do_tx_balance = 0;
1321 break;
1324 hash_start = (char*)eth_data->h_dest;
1325 hash_size = ETH_ALEN;
1326 break;
1327 case ETH_P_ARP:
1328 do_tx_balance = 0;
1329 if (bond_info->rlb_enabled) {
1330 tx_slave = rlb_arp_xmit(skb, bond);
1332 break;
1333 default:
1334 do_tx_balance = 0;
1335 break;
1338 if (do_tx_balance) {
1339 hash_index = _simple_hash(hash_start, hash_size);
1340 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1343 if (!tx_slave) {
1344 /* unbalanced or unassigned, send through primary */
1345 tx_slave = bond->curr_active_slave;
1346 bond_info->unbalanced_load += skb->len;
1349 if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1350 if (tx_slave != bond->curr_active_slave) {
1351 memcpy(eth_data->h_source,
1352 tx_slave->dev->dev_addr,
1353 ETH_ALEN);
1356 res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1357 } else {
1358 if (tx_slave) {
1359 _lock_tx_hashtbl(bond);
1360 __tlb_clear_slave(bond, tx_slave, 0);
1361 _unlock_tx_hashtbl(bond);
1365 if (res) {
1366 /* no suitable interface, frame not sent */
1367 dev_kfree_skb(skb);
1369 read_unlock(&bond->curr_slave_lock);
1371 return NETDEV_TX_OK;
1374 void bond_alb_monitor(struct work_struct *work)
1376 struct bonding *bond = container_of(work, struct bonding,
1377 alb_work.work);
1378 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1379 struct slave *slave;
1380 int i;
1382 read_lock(&bond->lock);
1384 if (bond->slave_cnt == 0) {
1385 bond_info->tx_rebalance_counter = 0;
1386 bond_info->lp_counter = 0;
1387 goto re_arm;
1390 bond_info->tx_rebalance_counter++;
1391 bond_info->lp_counter++;
1393 /* send learning packets */
1394 if (bond_info->lp_counter >= BOND_ALB_LP_TICKS) {
1395 /* change of curr_active_slave involves swapping of mac addresses.
1396 * in order to avoid this swapping from happening while
1397 * sending the learning packets, the curr_slave_lock must be held for
1398 * read.
1400 read_lock(&bond->curr_slave_lock);
1402 bond_for_each_slave(bond, slave, i) {
1403 alb_send_learning_packets(slave, slave->dev->dev_addr);
1406 read_unlock(&bond->curr_slave_lock);
1408 bond_info->lp_counter = 0;
1411 /* rebalance tx traffic */
1412 if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1414 read_lock(&bond->curr_slave_lock);
1416 bond_for_each_slave(bond, slave, i) {
1417 tlb_clear_slave(bond, slave, 1);
1418 if (slave == bond->curr_active_slave) {
1419 SLAVE_TLB_INFO(slave).load =
1420 bond_info->unbalanced_load /
1421 BOND_TLB_REBALANCE_INTERVAL;
1422 bond_info->unbalanced_load = 0;
1426 read_unlock(&bond->curr_slave_lock);
1428 bond_info->tx_rebalance_counter = 0;
1431 /* handle rlb stuff */
1432 if (bond_info->rlb_enabled) {
1433 if (bond_info->primary_is_promisc &&
1434 (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1437 * dev_set_promiscuity requires rtnl and
1438 * nothing else. Avoid race with bond_close.
1440 read_unlock(&bond->lock);
1441 if (!rtnl_trylock()) {
1442 read_lock(&bond->lock);
1443 goto re_arm;
1446 bond_info->rlb_promisc_timeout_counter = 0;
1448 /* If the primary was set to promiscuous mode
1449 * because a slave was disabled then
1450 * it can now leave promiscuous mode.
1452 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1453 bond_info->primary_is_promisc = 0;
1455 rtnl_unlock();
1456 read_lock(&bond->lock);
1459 if (bond_info->rlb_rebalance) {
1460 bond_info->rlb_rebalance = 0;
1461 rlb_rebalance(bond);
1464 /* check if clients need updating */
1465 if (bond_info->rx_ntt) {
1466 if (bond_info->rlb_update_delay_counter) {
1467 --bond_info->rlb_update_delay_counter;
1468 } else {
1469 rlb_update_rx_clients(bond);
1470 if (bond_info->rlb_update_retry_counter) {
1471 --bond_info->rlb_update_retry_counter;
1472 } else {
1473 bond_info->rx_ntt = 0;
1479 re_arm:
1480 queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1482 read_unlock(&bond->lock);
1485 /* assumption: called before the slave is attached to the bond
1486 * and not locked by the bond lock
1488 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1490 int res;
1492 res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr);
1493 if (res) {
1494 return res;
1497 /* caller must hold the bond lock for write since the mac addresses
1498 * are compared and may be swapped.
1500 read_lock(&bond->lock);
1502 res = alb_handle_addr_collision_on_attach(bond, slave);
1504 read_unlock(&bond->lock);
1506 if (res) {
1507 return res;
1510 tlb_init_slave(slave);
1512 /* order a rebalance ASAP */
1513 bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1515 if (bond->alb_info.rlb_enabled) {
1516 bond->alb_info.rlb_rebalance = 1;
1519 return 0;
1523 * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1524 * if necessary.
1526 * Caller must hold RTNL and no other locks
1528 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1530 if (bond->slave_cnt > 1) {
1531 alb_change_hw_addr_on_detach(bond, slave);
1534 tlb_clear_slave(bond, slave, 0);
1536 if (bond->alb_info.rlb_enabled) {
1537 bond->alb_info.next_rx_slave = NULL;
1538 rlb_clear_slave(bond, slave);
1542 /* Caller must hold bond lock for read */
1543 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1545 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1547 if (link == BOND_LINK_DOWN) {
1548 tlb_clear_slave(bond, slave, 0);
1549 if (bond->alb_info.rlb_enabled) {
1550 rlb_clear_slave(bond, slave);
1552 } else if (link == BOND_LINK_UP) {
1553 /* order a rebalance ASAP */
1554 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1555 if (bond->alb_info.rlb_enabled) {
1556 bond->alb_info.rlb_rebalance = 1;
1557 /* If the updelay module parameter is smaller than the
1558 * forwarding delay of the switch the rebalance will
1559 * not work because the rebalance arp replies will
1560 * not be forwarded to the clients..
1567 * bond_alb_handle_active_change - assign new curr_active_slave
1568 * @bond: our bonding struct
1569 * @new_slave: new slave to assign
1571 * Set the bond->curr_active_slave to @new_slave and handle
1572 * mac address swapping and promiscuity changes as needed.
1574 * If new_slave is NULL, caller must hold curr_slave_lock or
1575 * bond->lock for write.
1577 * If new_slave is not NULL, caller must hold RTNL, bond->lock for
1578 * read and curr_slave_lock for write. Processing here may sleep, so
1579 * no other locks may be held.
1581 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1582 __releases(&bond->curr_slave_lock)
1583 __releases(&bond->lock)
1584 __acquires(&bond->lock)
1585 __acquires(&bond->curr_slave_lock)
1587 struct slave *swap_slave;
1588 int i;
1590 if (bond->curr_active_slave == new_slave) {
1591 return;
1594 if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1595 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1596 bond->alb_info.primary_is_promisc = 0;
1597 bond->alb_info.rlb_promisc_timeout_counter = 0;
1600 swap_slave = bond->curr_active_slave;
1601 bond->curr_active_slave = new_slave;
1603 if (!new_slave || (bond->slave_cnt == 0)) {
1604 return;
1607 /* set the new curr_active_slave to the bonds mac address
1608 * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1610 if (!swap_slave) {
1611 struct slave *tmp_slave;
1612 /* find slave that is holding the bond's mac address */
1613 bond_for_each_slave(bond, tmp_slave, i) {
1614 if (!compare_ether_addr_64bits(tmp_slave->dev->dev_addr,
1615 bond->dev->dev_addr)) {
1616 swap_slave = tmp_slave;
1617 break;
1623 * Arrange for swap_slave and new_slave to temporarily be
1624 * ignored so we can mess with their MAC addresses without
1625 * fear of interference from transmit activity.
1627 if (swap_slave) {
1628 tlb_clear_slave(bond, swap_slave, 1);
1630 tlb_clear_slave(bond, new_slave, 1);
1632 write_unlock_bh(&bond->curr_slave_lock);
1633 read_unlock(&bond->lock);
1635 ASSERT_RTNL();
1637 /* curr_active_slave must be set before calling alb_swap_mac_addr */
1638 if (swap_slave) {
1639 /* swap mac address */
1640 alb_swap_mac_addr(bond, swap_slave, new_slave);
1641 } else {
1642 /* set the new_slave to the bond mac address */
1643 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr);
1646 if (swap_slave) {
1647 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1648 read_lock(&bond->lock);
1649 } else {
1650 read_lock(&bond->lock);
1651 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1654 write_lock_bh(&bond->curr_slave_lock);
1658 * Called with RTNL
1660 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1661 __acquires(&bond->lock)
1662 __releases(&bond->lock)
1664 struct bonding *bond = netdev_priv(bond_dev);
1665 struct sockaddr *sa = addr;
1666 struct slave *slave, *swap_slave;
1667 int res;
1668 int i;
1670 if (!is_valid_ether_addr(sa->sa_data)) {
1671 return -EADDRNOTAVAIL;
1674 res = alb_set_mac_address(bond, addr);
1675 if (res) {
1676 return res;
1679 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1681 /* If there is no curr_active_slave there is nothing else to do.
1682 * Otherwise we'll need to pass the new address to it and handle
1683 * duplications.
1685 if (!bond->curr_active_slave) {
1686 return 0;
1689 swap_slave = NULL;
1691 bond_for_each_slave(bond, slave, i) {
1692 if (!compare_ether_addr_64bits(slave->dev->dev_addr,
1693 bond_dev->dev_addr)) {
1694 swap_slave = slave;
1695 break;
1699 if (swap_slave) {
1700 alb_swap_mac_addr(bond, swap_slave, bond->curr_active_slave);
1701 alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave);
1702 } else {
1703 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr);
1705 read_lock(&bond->lock);
1706 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1707 if (bond->alb_info.rlb_enabled) {
1708 /* inform clients mac address has changed */
1709 rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1711 read_unlock(&bond->lock);
1714 return 0;
1717 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1719 if (bond->alb_info.current_alb_vlan &&
1720 (bond->alb_info.current_alb_vlan->vlan_id == vlan_id)) {
1721 bond->alb_info.current_alb_vlan = NULL;
1724 if (bond->alb_info.rlb_enabled) {
1725 rlb_clear_vlan(bond, vlan_id);