mfd: wm8350-i2c: Make sure the i2c regmap functions are compiled
[linux/fpc-iii.git] / drivers / net / bonding / bond_alb.c
blob175f266ce82ed0d09f82d6a18d6006cd75336ef3
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[]);
87 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp);
88 static void rlb_src_unlink(struct bonding *bond, u32 index);
89 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash,
90 u32 ip_dst_hash);
92 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
94 int i;
95 u8 hash = 0;
97 for (i = 0; i < hash_size; i++) {
98 hash ^= hash_start[i];
101 return hash;
104 /*********************** tlb specific functions ***************************/
106 static inline void _lock_tx_hashtbl_bh(struct bonding *bond)
108 spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
111 static inline void _unlock_tx_hashtbl_bh(struct bonding *bond)
113 spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
116 static inline void _lock_tx_hashtbl(struct bonding *bond)
118 spin_lock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
121 static inline void _unlock_tx_hashtbl(struct bonding *bond)
123 spin_unlock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
126 /* Caller must hold tx_hashtbl lock */
127 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
129 if (save_load) {
130 entry->load_history = 1 + entry->tx_bytes /
131 BOND_TLB_REBALANCE_INTERVAL;
132 entry->tx_bytes = 0;
135 entry->tx_slave = NULL;
136 entry->next = TLB_NULL_INDEX;
137 entry->prev = TLB_NULL_INDEX;
140 static inline void tlb_init_slave(struct slave *slave)
142 SLAVE_TLB_INFO(slave).load = 0;
143 SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
146 /* Caller must hold bond lock for read, BH disabled */
147 static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
148 int save_load)
150 struct tlb_client_info *tx_hash_table;
151 u32 index;
153 /* clear slave from tx_hashtbl */
154 tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
156 /* skip this if we've already freed the tx hash table */
157 if (tx_hash_table) {
158 index = SLAVE_TLB_INFO(slave).head;
159 while (index != TLB_NULL_INDEX) {
160 u32 next_index = tx_hash_table[index].next;
161 tlb_init_table_entry(&tx_hash_table[index], save_load);
162 index = next_index;
166 tlb_init_slave(slave);
169 /* Caller must hold bond lock for read */
170 static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
171 int save_load)
173 _lock_tx_hashtbl_bh(bond);
174 __tlb_clear_slave(bond, slave, save_load);
175 _unlock_tx_hashtbl_bh(bond);
178 /* Must be called before starting the monitor timer */
179 static int tlb_initialize(struct bonding *bond)
181 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
182 int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
183 struct tlb_client_info *new_hashtbl;
184 int i;
186 new_hashtbl = kzalloc(size, GFP_KERNEL);
187 if (!new_hashtbl)
188 return -1;
190 _lock_tx_hashtbl_bh(bond);
192 bond_info->tx_hashtbl = new_hashtbl;
194 for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
195 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
198 _unlock_tx_hashtbl_bh(bond);
200 return 0;
203 /* Must be called only after all slaves have been released */
204 static void tlb_deinitialize(struct bonding *bond)
206 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
208 _lock_tx_hashtbl_bh(bond);
210 kfree(bond_info->tx_hashtbl);
211 bond_info->tx_hashtbl = NULL;
213 _unlock_tx_hashtbl_bh(bond);
216 static long long compute_gap(struct slave *slave)
218 return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
219 (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
222 /* Caller must hold bond lock for read */
223 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
225 struct slave *slave, *least_loaded;
226 long long max_gap;
228 least_loaded = NULL;
229 max_gap = LLONG_MIN;
231 /* Find the slave with the largest gap */
232 bond_for_each_slave(bond, slave) {
233 if (SLAVE_IS_OK(slave)) {
234 long long gap = compute_gap(slave);
236 if (max_gap < gap) {
237 least_loaded = slave;
238 max_gap = gap;
243 return least_loaded;
246 static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
247 u32 skb_len)
249 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
250 struct tlb_client_info *hash_table;
251 struct slave *assigned_slave;
253 hash_table = bond_info->tx_hashtbl;
254 assigned_slave = hash_table[hash_index].tx_slave;
255 if (!assigned_slave) {
256 assigned_slave = tlb_get_least_loaded_slave(bond);
258 if (assigned_slave) {
259 struct tlb_slave_info *slave_info =
260 &(SLAVE_TLB_INFO(assigned_slave));
261 u32 next_index = slave_info->head;
263 hash_table[hash_index].tx_slave = assigned_slave;
264 hash_table[hash_index].next = next_index;
265 hash_table[hash_index].prev = TLB_NULL_INDEX;
267 if (next_index != TLB_NULL_INDEX) {
268 hash_table[next_index].prev = hash_index;
271 slave_info->head = hash_index;
272 slave_info->load +=
273 hash_table[hash_index].load_history;
277 if (assigned_slave) {
278 hash_table[hash_index].tx_bytes += skb_len;
281 return assigned_slave;
284 /* Caller must hold bond lock for read */
285 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
286 u32 skb_len)
288 struct slave *tx_slave;
290 * We don't need to disable softirq here, becase
291 * tlb_choose_channel() is only called by bond_alb_xmit()
292 * which already has softirq disabled.
294 _lock_tx_hashtbl(bond);
295 tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
296 _unlock_tx_hashtbl(bond);
297 return tx_slave;
300 /*********************** rlb specific functions ***************************/
301 static inline void _lock_rx_hashtbl_bh(struct bonding *bond)
303 spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
306 static inline void _unlock_rx_hashtbl_bh(struct bonding *bond)
308 spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
311 static inline void _lock_rx_hashtbl(struct bonding *bond)
313 spin_lock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
316 static inline void _unlock_rx_hashtbl(struct bonding *bond)
318 spin_unlock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
321 /* when an ARP REPLY is received from a client update its info
322 * in the rx_hashtbl
324 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
326 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
327 struct rlb_client_info *client_info;
328 u32 hash_index;
330 _lock_rx_hashtbl_bh(bond);
332 hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
333 client_info = &(bond_info->rx_hashtbl[hash_index]);
335 if ((client_info->assigned) &&
336 (client_info->ip_src == arp->ip_dst) &&
337 (client_info->ip_dst == arp->ip_src) &&
338 (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) {
339 /* update the clients MAC address */
340 memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
341 client_info->ntt = 1;
342 bond_info->rx_ntt = 1;
345 _unlock_rx_hashtbl_bh(bond);
348 static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
349 struct slave *slave)
351 struct arp_pkt *arp, _arp;
353 if (skb->protocol != cpu_to_be16(ETH_P_ARP))
354 goto out;
356 arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp);
357 if (!arp)
358 goto out;
360 /* We received an ARP from arp->ip_src.
361 * We might have used this IP address previously (on the bonding host
362 * itself or on a system that is bridged together with the bond).
363 * However, if arp->mac_src is different than what is stored in
364 * rx_hashtbl, some other host is now using the IP and we must prevent
365 * sending out client updates with this IP address and the old MAC
366 * address.
367 * Clean up all hash table entries that have this address as ip_src but
368 * have a different mac_src.
370 rlb_purge_src_ip(bond, arp);
372 if (arp->op_code == htons(ARPOP_REPLY)) {
373 /* update rx hash table for this ARP */
374 rlb_update_entry_from_arp(bond, arp);
375 pr_debug("Server received an ARP Reply from client\n");
377 out:
378 return RX_HANDLER_ANOTHER;
381 /* Caller must hold bond lock for read */
382 static struct slave *rlb_next_rx_slave(struct bonding *bond)
384 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
385 struct slave *rx_slave, *slave, *start_at;
386 int i = 0;
388 if (bond_info->next_rx_slave)
389 start_at = bond_info->next_rx_slave;
390 else
391 start_at = bond_first_slave(bond);
393 rx_slave = NULL;
395 bond_for_each_slave_from(bond, slave, i, start_at) {
396 if (SLAVE_IS_OK(slave)) {
397 if (!rx_slave) {
398 rx_slave = slave;
399 } else if (slave->speed > rx_slave->speed) {
400 rx_slave = slave;
405 if (rx_slave) {
406 slave = bond_next_slave(bond, rx_slave);
407 bond_info->next_rx_slave = slave;
410 return rx_slave;
413 /* teach the switch the mac of a disabled slave
414 * on the primary for fault tolerance
416 * Caller must hold bond->curr_slave_lock for write or bond lock for write
418 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
420 if (!bond->curr_active_slave) {
421 return;
424 if (!bond->alb_info.primary_is_promisc) {
425 if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1))
426 bond->alb_info.primary_is_promisc = 1;
427 else
428 bond->alb_info.primary_is_promisc = 0;
431 bond->alb_info.rlb_promisc_timeout_counter = 0;
433 alb_send_learning_packets(bond->curr_active_slave, addr);
436 /* slave being removed should not be active at this point
438 * Caller must hold bond lock for read
440 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
442 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
443 struct rlb_client_info *rx_hash_table;
444 u32 index, next_index;
446 /* clear slave from rx_hashtbl */
447 _lock_rx_hashtbl_bh(bond);
449 rx_hash_table = bond_info->rx_hashtbl;
450 index = bond_info->rx_hashtbl_used_head;
451 for (; index != RLB_NULL_INDEX; index = next_index) {
452 next_index = rx_hash_table[index].used_next;
453 if (rx_hash_table[index].slave == slave) {
454 struct slave *assigned_slave = rlb_next_rx_slave(bond);
456 if (assigned_slave) {
457 rx_hash_table[index].slave = assigned_slave;
458 if (!ether_addr_equal_64bits(rx_hash_table[index].mac_dst,
459 mac_bcast)) {
460 bond_info->rx_hashtbl[index].ntt = 1;
461 bond_info->rx_ntt = 1;
462 /* A slave has been removed from the
463 * table because it is either disabled
464 * or being released. We must retry the
465 * update to avoid clients from not
466 * being updated & disconnecting when
467 * there is stress
469 bond_info->rlb_update_retry_counter =
470 RLB_UPDATE_RETRY;
472 } else { /* there is no active slave */
473 rx_hash_table[index].slave = NULL;
478 _unlock_rx_hashtbl_bh(bond);
480 write_lock_bh(&bond->curr_slave_lock);
482 if (slave != bond->curr_active_slave) {
483 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
486 write_unlock_bh(&bond->curr_slave_lock);
489 static void rlb_update_client(struct rlb_client_info *client_info)
491 int i;
493 if (!client_info->slave) {
494 return;
497 for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
498 struct sk_buff *skb;
500 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
501 client_info->ip_dst,
502 client_info->slave->dev,
503 client_info->ip_src,
504 client_info->mac_dst,
505 client_info->slave->dev->dev_addr,
506 client_info->mac_dst);
507 if (!skb) {
508 pr_err("%s: Error: failed to create an ARP packet\n",
509 client_info->slave->bond->dev->name);
510 continue;
513 skb->dev = client_info->slave->dev;
515 if (client_info->vlan_id) {
516 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), client_info->vlan_id);
517 if (!skb) {
518 pr_err("%s: Error: failed to insert VLAN tag\n",
519 client_info->slave->bond->dev->name);
520 continue;
524 arp_xmit(skb);
528 /* sends ARP REPLIES that update the clients that need updating */
529 static void rlb_update_rx_clients(struct bonding *bond)
531 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
532 struct rlb_client_info *client_info;
533 u32 hash_index;
535 _lock_rx_hashtbl_bh(bond);
537 hash_index = bond_info->rx_hashtbl_used_head;
538 for (; hash_index != RLB_NULL_INDEX;
539 hash_index = client_info->used_next) {
540 client_info = &(bond_info->rx_hashtbl[hash_index]);
541 if (client_info->ntt) {
542 rlb_update_client(client_info);
543 if (bond_info->rlb_update_retry_counter == 0) {
544 client_info->ntt = 0;
549 /* do not update the entries again until this counter is zero so that
550 * not to confuse the clients.
552 bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
554 _unlock_rx_hashtbl_bh(bond);
557 /* The slave was assigned a new mac address - update the clients */
558 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
560 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
561 struct rlb_client_info *client_info;
562 int ntt = 0;
563 u32 hash_index;
565 _lock_rx_hashtbl_bh(bond);
567 hash_index = bond_info->rx_hashtbl_used_head;
568 for (; hash_index != RLB_NULL_INDEX;
569 hash_index = client_info->used_next) {
570 client_info = &(bond_info->rx_hashtbl[hash_index]);
572 if ((client_info->slave == slave) &&
573 !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
574 client_info->ntt = 1;
575 ntt = 1;
579 // update the team's flag only after the whole iteration
580 if (ntt) {
581 bond_info->rx_ntt = 1;
582 //fasten the change
583 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
586 _unlock_rx_hashtbl_bh(bond);
589 /* mark all clients using src_ip to be updated */
590 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
592 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
593 struct rlb_client_info *client_info;
594 u32 hash_index;
596 _lock_rx_hashtbl(bond);
598 hash_index = bond_info->rx_hashtbl_used_head;
599 for (; hash_index != RLB_NULL_INDEX;
600 hash_index = client_info->used_next) {
601 client_info = &(bond_info->rx_hashtbl[hash_index]);
603 if (!client_info->slave) {
604 pr_err("%s: Error: found a client with no channel in the client's hash table\n",
605 bond->dev->name);
606 continue;
608 /*update all clients using this src_ip, that are not assigned
609 * to the team's address (curr_active_slave) and have a known
610 * unicast mac address.
612 if ((client_info->ip_src == src_ip) &&
613 !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
614 bond->dev->dev_addr) &&
615 !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
616 client_info->ntt = 1;
617 bond_info->rx_ntt = 1;
621 _unlock_rx_hashtbl(bond);
624 /* Caller must hold both bond and ptr locks for read */
625 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
627 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
628 struct arp_pkt *arp = arp_pkt(skb);
629 struct slave *assigned_slave;
630 struct rlb_client_info *client_info;
631 u32 hash_index = 0;
633 _lock_rx_hashtbl(bond);
635 hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
636 client_info = &(bond_info->rx_hashtbl[hash_index]);
638 if (client_info->assigned) {
639 if ((client_info->ip_src == arp->ip_src) &&
640 (client_info->ip_dst == arp->ip_dst)) {
641 /* the entry is already assigned to this client */
642 if (!ether_addr_equal_64bits(arp->mac_dst, mac_bcast)) {
643 /* update mac address from arp */
644 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
646 memcpy(client_info->mac_src, arp->mac_src, ETH_ALEN);
648 assigned_slave = client_info->slave;
649 if (assigned_slave) {
650 _unlock_rx_hashtbl(bond);
651 return assigned_slave;
653 } else {
654 /* the entry is already assigned to some other client,
655 * move the old client to primary (curr_active_slave) so
656 * that the new client can be assigned to this entry.
658 if (bond->curr_active_slave &&
659 client_info->slave != bond->curr_active_slave) {
660 client_info->slave = bond->curr_active_slave;
661 rlb_update_client(client_info);
665 /* assign a new slave */
666 assigned_slave = rlb_next_rx_slave(bond);
668 if (assigned_slave) {
669 if (!(client_info->assigned &&
670 client_info->ip_src == arp->ip_src)) {
671 /* ip_src is going to be updated,
672 * fix the src hash list
674 u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
675 sizeof(arp->ip_src));
676 rlb_src_unlink(bond, hash_index);
677 rlb_src_link(bond, hash_src, hash_index);
680 client_info->ip_src = arp->ip_src;
681 client_info->ip_dst = arp->ip_dst;
682 /* arp->mac_dst is broadcast for arp reqeusts.
683 * will be updated with clients actual unicast mac address
684 * upon receiving an arp reply.
686 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
687 memcpy(client_info->mac_src, arp->mac_src, ETH_ALEN);
688 client_info->slave = assigned_slave;
690 if (!ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
691 client_info->ntt = 1;
692 bond->alb_info.rx_ntt = 1;
693 } else {
694 client_info->ntt = 0;
697 if (vlan_get_tag(skb, &client_info->vlan_id))
698 client_info->vlan_id = 0;
700 if (!client_info->assigned) {
701 u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
702 bond_info->rx_hashtbl_used_head = hash_index;
703 client_info->used_next = prev_tbl_head;
704 if (prev_tbl_head != RLB_NULL_INDEX) {
705 bond_info->rx_hashtbl[prev_tbl_head].used_prev =
706 hash_index;
708 client_info->assigned = 1;
712 _unlock_rx_hashtbl(bond);
714 return assigned_slave;
717 /* chooses (and returns) transmit channel for arp reply
718 * does not choose channel for other arp types since they are
719 * sent on the curr_active_slave
721 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
723 struct arp_pkt *arp = arp_pkt(skb);
724 struct slave *tx_slave = NULL;
726 /* Don't modify or load balance ARPs that do not originate locally
727 * (e.g.,arrive via a bridge).
729 if (!bond_slave_has_mac(bond, arp->mac_src))
730 return NULL;
732 if (arp->op_code == htons(ARPOP_REPLY)) {
733 /* the arp must be sent on the selected
734 * rx channel
736 tx_slave = rlb_choose_channel(skb, bond);
737 if (tx_slave) {
738 memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
740 pr_debug("Server sent ARP Reply packet\n");
741 } else if (arp->op_code == htons(ARPOP_REQUEST)) {
742 /* Create an entry in the rx_hashtbl for this client as a
743 * place holder.
744 * When the arp reply is received the entry will be updated
745 * with the correct unicast address of the client.
747 rlb_choose_channel(skb, bond);
749 /* The ARP reply packets must be delayed so that
750 * they can cancel out the influence of the ARP request.
752 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
754 /* arp requests are broadcast and are sent on the primary
755 * the arp request will collapse all clients on the subnet to
756 * the primary slave. We must register these clients to be
757 * updated with their assigned mac.
759 rlb_req_update_subnet_clients(bond, arp->ip_src);
760 pr_debug("Server sent ARP Request packet\n");
763 return tx_slave;
766 /* Caller must hold bond lock for read */
767 static void rlb_rebalance(struct bonding *bond)
769 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
770 struct slave *assigned_slave;
771 struct rlb_client_info *client_info;
772 int ntt;
773 u32 hash_index;
775 _lock_rx_hashtbl_bh(bond);
777 ntt = 0;
778 hash_index = bond_info->rx_hashtbl_used_head;
779 for (; hash_index != RLB_NULL_INDEX;
780 hash_index = client_info->used_next) {
781 client_info = &(bond_info->rx_hashtbl[hash_index]);
782 assigned_slave = rlb_next_rx_slave(bond);
783 if (assigned_slave && (client_info->slave != assigned_slave)) {
784 client_info->slave = assigned_slave;
785 client_info->ntt = 1;
786 ntt = 1;
790 /* update the team's flag only after the whole iteration */
791 if (ntt) {
792 bond_info->rx_ntt = 1;
794 _unlock_rx_hashtbl_bh(bond);
797 /* Caller must hold rx_hashtbl lock */
798 static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
800 entry->used_next = RLB_NULL_INDEX;
801 entry->used_prev = RLB_NULL_INDEX;
802 entry->assigned = 0;
803 entry->slave = NULL;
804 entry->vlan_id = 0;
806 static void rlb_init_table_entry_src(struct rlb_client_info *entry)
808 entry->src_first = RLB_NULL_INDEX;
809 entry->src_prev = RLB_NULL_INDEX;
810 entry->src_next = RLB_NULL_INDEX;
813 static void rlb_init_table_entry(struct rlb_client_info *entry)
815 memset(entry, 0, sizeof(struct rlb_client_info));
816 rlb_init_table_entry_dst(entry);
817 rlb_init_table_entry_src(entry);
820 static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
822 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
823 u32 next_index = bond_info->rx_hashtbl[index].used_next;
824 u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
826 if (index == bond_info->rx_hashtbl_used_head)
827 bond_info->rx_hashtbl_used_head = next_index;
828 if (prev_index != RLB_NULL_INDEX)
829 bond_info->rx_hashtbl[prev_index].used_next = next_index;
830 if (next_index != RLB_NULL_INDEX)
831 bond_info->rx_hashtbl[next_index].used_prev = prev_index;
834 /* unlink a rlb hash table entry from the src list */
835 static void rlb_src_unlink(struct bonding *bond, u32 index)
837 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
838 u32 next_index = bond_info->rx_hashtbl[index].src_next;
839 u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
841 bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
842 bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
844 if (next_index != RLB_NULL_INDEX)
845 bond_info->rx_hashtbl[next_index].src_prev = prev_index;
847 if (prev_index == RLB_NULL_INDEX)
848 return;
850 /* is prev_index pointing to the head of this list? */
851 if (bond_info->rx_hashtbl[prev_index].src_first == index)
852 bond_info->rx_hashtbl[prev_index].src_first = next_index;
853 else
854 bond_info->rx_hashtbl[prev_index].src_next = next_index;
858 static void rlb_delete_table_entry(struct bonding *bond, u32 index)
860 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
861 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
863 rlb_delete_table_entry_dst(bond, index);
864 rlb_init_table_entry_dst(entry);
866 rlb_src_unlink(bond, index);
869 /* add the rx_hashtbl[ip_dst_hash] entry to the list
870 * of entries with identical ip_src_hash
872 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
874 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
875 u32 next;
877 bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
878 next = bond_info->rx_hashtbl[ip_src_hash].src_first;
879 bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
880 if (next != RLB_NULL_INDEX)
881 bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
882 bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
885 /* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does
886 * not match arp->mac_src */
887 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
889 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
890 u32 ip_src_hash = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
891 u32 index;
893 _lock_rx_hashtbl_bh(bond);
895 index = bond_info->rx_hashtbl[ip_src_hash].src_first;
896 while (index != RLB_NULL_INDEX) {
897 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
898 u32 next_index = entry->src_next;
899 if (entry->ip_src == arp->ip_src &&
900 !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
901 rlb_delete_table_entry(bond, index);
902 index = next_index;
904 _unlock_rx_hashtbl_bh(bond);
907 static int rlb_initialize(struct bonding *bond)
909 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
910 struct rlb_client_info *new_hashtbl;
911 int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
912 int i;
914 new_hashtbl = kmalloc(size, GFP_KERNEL);
915 if (!new_hashtbl)
916 return -1;
918 _lock_rx_hashtbl_bh(bond);
920 bond_info->rx_hashtbl = new_hashtbl;
922 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
924 for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
925 rlb_init_table_entry(bond_info->rx_hashtbl + i);
928 _unlock_rx_hashtbl_bh(bond);
930 /* register to receive ARPs */
931 bond->recv_probe = rlb_arp_recv;
933 return 0;
936 static void rlb_deinitialize(struct bonding *bond)
938 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
940 _lock_rx_hashtbl_bh(bond);
942 kfree(bond_info->rx_hashtbl);
943 bond_info->rx_hashtbl = NULL;
944 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
946 _unlock_rx_hashtbl_bh(bond);
949 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
951 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
952 u32 curr_index;
954 _lock_rx_hashtbl_bh(bond);
956 curr_index = bond_info->rx_hashtbl_used_head;
957 while (curr_index != RLB_NULL_INDEX) {
958 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
959 u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
961 if (curr->vlan_id == vlan_id)
962 rlb_delete_table_entry(bond, curr_index);
964 curr_index = next_index;
967 _unlock_rx_hashtbl_bh(bond);
970 /*********************** tlb/rlb shared functions *********************/
972 static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
973 u16 vid)
975 struct learning_pkt pkt;
976 struct sk_buff *skb;
977 int size = sizeof(struct learning_pkt);
978 char *data;
980 memset(&pkt, 0, size);
981 memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
982 memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
983 pkt.type = cpu_to_be16(ETH_P_LOOP);
985 skb = dev_alloc_skb(size);
986 if (!skb)
987 return;
989 data = skb_put(skb, size);
990 memcpy(data, &pkt, size);
992 skb_reset_mac_header(skb);
993 skb->network_header = skb->mac_header + ETH_HLEN;
994 skb->protocol = pkt.type;
995 skb->priority = TC_PRIO_CONTROL;
996 skb->dev = slave->dev;
998 if (vid) {
999 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vid);
1000 if (!skb) {
1001 pr_err("%s: Error: failed to insert VLAN tag\n",
1002 slave->bond->dev->name);
1003 return;
1007 dev_queue_xmit(skb);
1011 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
1013 struct bonding *bond = bond_get_bond_by_slave(slave);
1014 struct net_device *upper;
1015 struct list_head *iter;
1017 /* send untagged */
1018 alb_send_lp_vid(slave, mac_addr, 0);
1020 /* loop through vlans and send one packet for each */
1021 rcu_read_lock();
1022 netdev_for_each_upper_dev_rcu(bond->dev, upper, iter) {
1023 if (upper->priv_flags & IFF_802_1Q_VLAN)
1024 alb_send_lp_vid(slave, mac_addr,
1025 vlan_dev_vlan_id(upper));
1027 rcu_read_unlock();
1030 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[])
1032 struct net_device *dev = slave->dev;
1033 struct sockaddr s_addr;
1035 if (slave->bond->params.mode == BOND_MODE_TLB) {
1036 memcpy(dev->dev_addr, addr, dev->addr_len);
1037 return 0;
1040 /* for rlb each slave must have a unique hw mac addresses so that */
1041 /* each slave will receive packets destined to a different mac */
1042 memcpy(s_addr.sa_data, addr, dev->addr_len);
1043 s_addr.sa_family = dev->type;
1044 if (dev_set_mac_address(dev, &s_addr)) {
1045 pr_err("%s: Error: dev_set_mac_address of dev %s failed!\n"
1046 "ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
1047 slave->bond->dev->name, dev->name);
1048 return -EOPNOTSUPP;
1050 return 0;
1054 * Swap MAC addresses between two slaves.
1056 * Called with RTNL held, and no other locks.
1060 static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1062 u8 tmp_mac_addr[ETH_ALEN];
1064 memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
1065 alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr);
1066 alb_set_slave_mac_addr(slave2, tmp_mac_addr);
1071 * Send learning packets after MAC address swap.
1073 * Called with RTNL and no other locks
1075 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1076 struct slave *slave2)
1078 int slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
1079 struct slave *disabled_slave = NULL;
1081 ASSERT_RTNL();
1083 /* fasten the change in the switch */
1084 if (SLAVE_IS_OK(slave1)) {
1085 alb_send_learning_packets(slave1, slave1->dev->dev_addr);
1086 if (bond->alb_info.rlb_enabled) {
1087 /* inform the clients that the mac address
1088 * has changed
1090 rlb_req_update_slave_clients(bond, slave1);
1092 } else {
1093 disabled_slave = slave1;
1096 if (SLAVE_IS_OK(slave2)) {
1097 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
1098 if (bond->alb_info.rlb_enabled) {
1099 /* inform the clients that the mac address
1100 * has changed
1102 rlb_req_update_slave_clients(bond, slave2);
1104 } else {
1105 disabled_slave = slave2;
1108 if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1109 /* A disabled slave was assigned an active mac addr */
1110 rlb_teach_disabled_mac_on_primary(bond,
1111 disabled_slave->dev->dev_addr);
1116 * alb_change_hw_addr_on_detach
1117 * @bond: bonding we're working on
1118 * @slave: the slave that was just detached
1120 * We assume that @slave was already detached from the slave list.
1122 * If @slave's permanent hw address is different both from its current
1123 * address and from @bond's address, then somewhere in the bond there's
1124 * a slave that has @slave's permanet address as its current address.
1125 * We'll make sure that that slave no longer uses @slave's permanent address.
1127 * Caller must hold RTNL and no other locks
1129 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1131 int perm_curr_diff;
1132 int perm_bond_diff;
1133 struct slave *found_slave;
1135 perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1136 slave->dev->dev_addr);
1137 perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1138 bond->dev->dev_addr);
1140 if (perm_curr_diff && perm_bond_diff) {
1141 found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1143 if (found_slave) {
1144 /* locking: needs RTNL and nothing else */
1145 alb_swap_mac_addr(slave, found_slave);
1146 alb_fasten_mac_swap(bond, slave, found_slave);
1152 * alb_handle_addr_collision_on_attach
1153 * @bond: bonding we're working on
1154 * @slave: the slave that was just attached
1156 * checks uniqueness of slave's mac address and handles the case the
1157 * new slave uses the bonds mac address.
1159 * If the permanent hw address of @slave is @bond's hw address, we need to
1160 * find a different hw address to give @slave, that isn't in use by any other
1161 * slave in the bond. This address must be, of course, one of the permanent
1162 * addresses of the other slaves.
1164 * We go over the slave list, and for each slave there we compare its
1165 * permanent hw address with the current address of all the other slaves.
1166 * If no match was found, then we've found a slave with a permanent address
1167 * that isn't used by any other slave in the bond, so we can assign it to
1168 * @slave.
1170 * assumption: this function is called before @slave is attached to the
1171 * bond slave list.
1173 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1175 struct slave *tmp_slave1, *free_mac_slave = NULL;
1176 struct slave *has_bond_addr = bond->curr_active_slave;
1178 if (list_empty(&bond->slave_list)) {
1179 /* this is the first slave */
1180 return 0;
1183 /* if slave's mac address differs from bond's mac address
1184 * check uniqueness of slave's mac address against the other
1185 * slaves in the bond.
1187 if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1188 if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1189 return 0;
1191 /* Try setting slave mac to bond address and fall-through
1192 to code handling that situation below... */
1193 alb_set_slave_mac_addr(slave, bond->dev->dev_addr);
1196 /* The slave's address is equal to the address of the bond.
1197 * Search for a spare address in the bond for this slave.
1199 bond_for_each_slave(bond, tmp_slave1) {
1200 if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1201 /* no slave has tmp_slave1's perm addr
1202 * as its curr addr
1204 free_mac_slave = tmp_slave1;
1205 break;
1208 if (!has_bond_addr) {
1209 if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1210 bond->dev->dev_addr)) {
1212 has_bond_addr = tmp_slave1;
1217 if (free_mac_slave) {
1218 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr);
1220 pr_warning("%s: Warning: the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1221 bond->dev->name, slave->dev->name,
1222 free_mac_slave->dev->name);
1224 } else if (has_bond_addr) {
1225 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",
1226 bond->dev->name, slave->dev->name);
1227 return -EFAULT;
1230 return 0;
1234 * alb_set_mac_address
1235 * @bond:
1236 * @addr:
1238 * In TLB mode all slaves are configured to the bond's hw address, but set
1239 * their dev_addr field to different addresses (based on their permanent hw
1240 * addresses).
1242 * For each slave, this function sets the interface to the new address and then
1243 * changes its dev_addr field to its previous value.
1245 * Unwinding assumes bond's mac address has not yet changed.
1247 static int alb_set_mac_address(struct bonding *bond, void *addr)
1249 char tmp_addr[ETH_ALEN];
1250 struct slave *slave;
1251 struct sockaddr sa;
1252 int res;
1254 if (bond->alb_info.rlb_enabled)
1255 return 0;
1257 bond_for_each_slave(bond, slave) {
1258 /* save net_device's current hw address */
1259 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1261 res = dev_set_mac_address(slave->dev, addr);
1263 /* restore net_device's hw address */
1264 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1266 if (res)
1267 goto unwind;
1270 return 0;
1272 unwind:
1273 memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1274 sa.sa_family = bond->dev->type;
1276 /* unwind from head to the slave that failed */
1277 bond_for_each_slave_continue_reverse(bond, slave) {
1278 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1279 dev_set_mac_address(slave->dev, &sa);
1280 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1283 return res;
1286 /************************ exported alb funcions ************************/
1288 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1290 int res;
1292 res = tlb_initialize(bond);
1293 if (res) {
1294 return res;
1297 if (rlb_enabled) {
1298 bond->alb_info.rlb_enabled = 1;
1299 /* initialize rlb */
1300 res = rlb_initialize(bond);
1301 if (res) {
1302 tlb_deinitialize(bond);
1303 return res;
1305 } else {
1306 bond->alb_info.rlb_enabled = 0;
1309 return 0;
1312 void bond_alb_deinitialize(struct bonding *bond)
1314 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1316 tlb_deinitialize(bond);
1318 if (bond_info->rlb_enabled) {
1319 rlb_deinitialize(bond);
1323 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1325 struct bonding *bond = netdev_priv(bond_dev);
1326 struct ethhdr *eth_data;
1327 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1328 struct slave *tx_slave = NULL;
1329 static const __be32 ip_bcast = htonl(0xffffffff);
1330 int hash_size = 0;
1331 int do_tx_balance = 1;
1332 u32 hash_index = 0;
1333 const u8 *hash_start = NULL;
1334 int res = 1;
1335 struct ipv6hdr *ip6hdr;
1337 skb_reset_mac_header(skb);
1338 eth_data = eth_hdr(skb);
1340 /* make sure that the curr_active_slave do not change during tx
1342 read_lock(&bond->lock);
1343 read_lock(&bond->curr_slave_lock);
1345 switch (ntohs(skb->protocol)) {
1346 case ETH_P_IP: {
1347 const struct iphdr *iph = ip_hdr(skb);
1349 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast) ||
1350 (iph->daddr == ip_bcast) ||
1351 (iph->protocol == IPPROTO_IGMP)) {
1352 do_tx_balance = 0;
1353 break;
1355 hash_start = (char *)&(iph->daddr);
1356 hash_size = sizeof(iph->daddr);
1358 break;
1359 case ETH_P_IPV6:
1360 /* IPv6 doesn't really use broadcast mac address, but leave
1361 * that here just in case.
1363 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast)) {
1364 do_tx_balance = 0;
1365 break;
1368 /* IPv6 uses all-nodes multicast as an equivalent to
1369 * broadcasts in IPv4.
1371 if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1372 do_tx_balance = 0;
1373 break;
1376 /* Additianally, DAD probes should not be tx-balanced as that
1377 * will lead to false positives for duplicate addresses and
1378 * prevent address configuration from working.
1380 ip6hdr = ipv6_hdr(skb);
1381 if (ipv6_addr_any(&ip6hdr->saddr)) {
1382 do_tx_balance = 0;
1383 break;
1386 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1387 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1388 break;
1389 case ETH_P_IPX:
1390 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1391 /* something is wrong with this packet */
1392 do_tx_balance = 0;
1393 break;
1396 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1397 /* The only protocol worth balancing in
1398 * this family since it has an "ARP" like
1399 * mechanism
1401 do_tx_balance = 0;
1402 break;
1405 hash_start = (char*)eth_data->h_dest;
1406 hash_size = ETH_ALEN;
1407 break;
1408 case ETH_P_ARP:
1409 do_tx_balance = 0;
1410 if (bond_info->rlb_enabled) {
1411 tx_slave = rlb_arp_xmit(skb, bond);
1413 break;
1414 default:
1415 do_tx_balance = 0;
1416 break;
1419 if (do_tx_balance) {
1420 hash_index = _simple_hash(hash_start, hash_size);
1421 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1424 if (!tx_slave) {
1425 /* unbalanced or unassigned, send through primary */
1426 tx_slave = bond->curr_active_slave;
1427 bond_info->unbalanced_load += skb->len;
1430 if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1431 if (tx_slave != bond->curr_active_slave) {
1432 memcpy(eth_data->h_source,
1433 tx_slave->dev->dev_addr,
1434 ETH_ALEN);
1437 res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1438 } else {
1439 if (tx_slave) {
1440 _lock_tx_hashtbl(bond);
1441 __tlb_clear_slave(bond, tx_slave, 0);
1442 _unlock_tx_hashtbl(bond);
1446 read_unlock(&bond->curr_slave_lock);
1447 read_unlock(&bond->lock);
1448 if (res) {
1449 /* no suitable interface, frame not sent */
1450 dev_kfree_skb_any(skb);
1453 return NETDEV_TX_OK;
1456 void bond_alb_monitor(struct work_struct *work)
1458 struct bonding *bond = container_of(work, struct bonding,
1459 alb_work.work);
1460 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1461 struct slave *slave;
1463 read_lock(&bond->lock);
1465 if (list_empty(&bond->slave_list)) {
1466 bond_info->tx_rebalance_counter = 0;
1467 bond_info->lp_counter = 0;
1468 goto re_arm;
1471 bond_info->tx_rebalance_counter++;
1472 bond_info->lp_counter++;
1474 /* send learning packets */
1475 if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1476 /* change of curr_active_slave involves swapping of mac addresses.
1477 * in order to avoid this swapping from happening while
1478 * sending the learning packets, the curr_slave_lock must be held for
1479 * read.
1481 read_lock(&bond->curr_slave_lock);
1483 bond_for_each_slave(bond, slave)
1484 alb_send_learning_packets(slave, slave->dev->dev_addr);
1486 read_unlock(&bond->curr_slave_lock);
1488 bond_info->lp_counter = 0;
1491 /* rebalance tx traffic */
1492 if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1494 read_lock(&bond->curr_slave_lock);
1496 bond_for_each_slave(bond, slave) {
1497 tlb_clear_slave(bond, slave, 1);
1498 if (slave == bond->curr_active_slave) {
1499 SLAVE_TLB_INFO(slave).load =
1500 bond_info->unbalanced_load /
1501 BOND_TLB_REBALANCE_INTERVAL;
1502 bond_info->unbalanced_load = 0;
1506 read_unlock(&bond->curr_slave_lock);
1508 bond_info->tx_rebalance_counter = 0;
1511 /* handle rlb stuff */
1512 if (bond_info->rlb_enabled) {
1513 if (bond_info->primary_is_promisc &&
1514 (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1517 * dev_set_promiscuity requires rtnl and
1518 * nothing else. Avoid race with bond_close.
1520 read_unlock(&bond->lock);
1521 if (!rtnl_trylock()) {
1522 read_lock(&bond->lock);
1523 goto re_arm;
1526 bond_info->rlb_promisc_timeout_counter = 0;
1528 /* If the primary was set to promiscuous mode
1529 * because a slave was disabled then
1530 * it can now leave promiscuous mode.
1532 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1533 bond_info->primary_is_promisc = 0;
1535 rtnl_unlock();
1536 read_lock(&bond->lock);
1539 if (bond_info->rlb_rebalance) {
1540 bond_info->rlb_rebalance = 0;
1541 rlb_rebalance(bond);
1544 /* check if clients need updating */
1545 if (bond_info->rx_ntt) {
1546 if (bond_info->rlb_update_delay_counter) {
1547 --bond_info->rlb_update_delay_counter;
1548 } else {
1549 rlb_update_rx_clients(bond);
1550 if (bond_info->rlb_update_retry_counter) {
1551 --bond_info->rlb_update_retry_counter;
1552 } else {
1553 bond_info->rx_ntt = 0;
1559 re_arm:
1560 queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1562 read_unlock(&bond->lock);
1565 /* assumption: called before the slave is attached to the bond
1566 * and not locked by the bond lock
1568 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1570 int res;
1572 res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr);
1573 if (res) {
1574 return res;
1577 res = alb_handle_addr_collision_on_attach(bond, slave);
1578 if (res) {
1579 return res;
1582 tlb_init_slave(slave);
1584 /* order a rebalance ASAP */
1585 bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1587 if (bond->alb_info.rlb_enabled) {
1588 bond->alb_info.rlb_rebalance = 1;
1591 return 0;
1595 * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1596 * if necessary.
1598 * Caller must hold RTNL and no other locks
1600 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1602 if (!list_empty(&bond->slave_list))
1603 alb_change_hw_addr_on_detach(bond, slave);
1605 tlb_clear_slave(bond, slave, 0);
1607 if (bond->alb_info.rlb_enabled) {
1608 bond->alb_info.next_rx_slave = NULL;
1609 rlb_clear_slave(bond, slave);
1613 /* Caller must hold bond lock for read */
1614 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1616 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1618 if (link == BOND_LINK_DOWN) {
1619 tlb_clear_slave(bond, slave, 0);
1620 if (bond->alb_info.rlb_enabled) {
1621 rlb_clear_slave(bond, slave);
1623 } else if (link == BOND_LINK_UP) {
1624 /* order a rebalance ASAP */
1625 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1626 if (bond->alb_info.rlb_enabled) {
1627 bond->alb_info.rlb_rebalance = 1;
1628 /* If the updelay module parameter is smaller than the
1629 * forwarding delay of the switch the rebalance will
1630 * not work because the rebalance arp replies will
1631 * not be forwarded to the clients..
1638 * bond_alb_handle_active_change - assign new curr_active_slave
1639 * @bond: our bonding struct
1640 * @new_slave: new slave to assign
1642 * Set the bond->curr_active_slave to @new_slave and handle
1643 * mac address swapping and promiscuity changes as needed.
1645 * If new_slave is NULL, caller must hold curr_slave_lock or
1646 * bond->lock for write.
1648 * If new_slave is not NULL, caller must hold RTNL, bond->lock for
1649 * read and curr_slave_lock for write. Processing here may sleep, so
1650 * no other locks may be held.
1652 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1653 __releases(&bond->curr_slave_lock)
1654 __releases(&bond->lock)
1655 __acquires(&bond->lock)
1656 __acquires(&bond->curr_slave_lock)
1658 struct slave *swap_slave;
1660 if (bond->curr_active_slave == new_slave)
1661 return;
1663 if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1664 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1665 bond->alb_info.primary_is_promisc = 0;
1666 bond->alb_info.rlb_promisc_timeout_counter = 0;
1669 swap_slave = bond->curr_active_slave;
1670 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1672 if (!new_slave || list_empty(&bond->slave_list))
1673 return;
1675 /* set the new curr_active_slave to the bonds mac address
1676 * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1678 if (!swap_slave)
1679 swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1682 * Arrange for swap_slave and new_slave to temporarily be
1683 * ignored so we can mess with their MAC addresses without
1684 * fear of interference from transmit activity.
1686 if (swap_slave)
1687 tlb_clear_slave(bond, swap_slave, 1);
1688 tlb_clear_slave(bond, new_slave, 1);
1690 write_unlock_bh(&bond->curr_slave_lock);
1691 read_unlock(&bond->lock);
1693 ASSERT_RTNL();
1695 /* curr_active_slave must be set before calling alb_swap_mac_addr */
1696 if (swap_slave) {
1697 /* swap mac address */
1698 alb_swap_mac_addr(swap_slave, new_slave);
1699 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1700 read_lock(&bond->lock);
1701 } else {
1702 /* set the new_slave to the bond mac address */
1703 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr);
1704 read_lock(&bond->lock);
1705 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1708 write_lock_bh(&bond->curr_slave_lock);
1712 * Called with RTNL
1714 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1715 __acquires(&bond->lock)
1716 __releases(&bond->lock)
1718 struct bonding *bond = netdev_priv(bond_dev);
1719 struct sockaddr *sa = addr;
1720 struct slave *swap_slave;
1721 int res;
1723 if (!is_valid_ether_addr(sa->sa_data)) {
1724 return -EADDRNOTAVAIL;
1727 res = alb_set_mac_address(bond, addr);
1728 if (res) {
1729 return res;
1732 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1734 /* If there is no curr_active_slave there is nothing else to do.
1735 * Otherwise we'll need to pass the new address to it and handle
1736 * duplications.
1738 if (!bond->curr_active_slave) {
1739 return 0;
1742 swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1744 if (swap_slave) {
1745 alb_swap_mac_addr(swap_slave, bond->curr_active_slave);
1746 alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave);
1747 } else {
1748 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr);
1750 read_lock(&bond->lock);
1751 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1752 if (bond->alb_info.rlb_enabled) {
1753 /* inform clients mac address has changed */
1754 rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1756 read_unlock(&bond->lock);
1759 return 0;
1762 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1764 if (bond->alb_info.rlb_enabled) {
1765 rlb_clear_vlan(bond, vlan_id);