blob: d3c6801f101e5e89d093d88d390f406209494dcc [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved.
3 *
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.
8 *
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.
13 *
14 * You should have received a copy of the GNU General Public License along
Jeff Kirsheradf8d3f2013-12-06 06:28:47 -080015 * with this program; if not, see <http://www.gnu.org/licenses/>.
Linus Torvalds1da177e2005-04-16 15:20:36 -070016 *
17 * The full GNU General Public License is included in this distribution in the
18 * file called LICENSE.
19 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070020 */
21
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/skbuff.h>
23#include <linux/netdevice.h>
24#include <linux/etherdevice.h>
25#include <linux/pkt_sched.h>
26#include <linux/spinlock.h>
27#include <linux/slab.h>
28#include <linux/timer.h>
29#include <linux/ip.h>
30#include <linux/ipv6.h>
31#include <linux/if_arp.h>
32#include <linux/if_ether.h>
33#include <linux/if_bonding.h>
34#include <linux/if_vlan.h>
35#include <linux/in.h>
36#include <net/ipx.h>
37#include <net/arp.h>
Vlad Yasevich2d1ea192008-08-28 15:38:41 -040038#include <net/ipv6.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <asm/byteorder.h>
40#include "bonding.h"
41#include "bond_alb.h"
42
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
Eric Dumazet885a1362009-09-01 06:31:18 +000045#ifndef __long_aligned
46#define __long_aligned __attribute__((aligned((sizeof(long)))))
47#endif
48static const u8 mac_bcast[ETH_ALEN] __long_aligned = {
49 0xff, 0xff, 0xff, 0xff, 0xff, 0xff
50};
51static const u8 mac_v6_allmcast[ETH_ALEN] __long_aligned = {
52 0x33, 0x33, 0x00, 0x00, 0x00, 0x01
53};
Linus Torvalds1da177e2005-04-16 15:20:36 -070054static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
55
56#pragma pack(1)
57struct learning_pkt {
58 u8 mac_dst[ETH_ALEN];
59 u8 mac_src[ETH_ALEN];
Al Virod3bb52b2007-08-22 20:06:58 -040060 __be16 type;
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 u8 padding[ETH_ZLEN - ETH_HLEN];
62};
63
64struct arp_pkt {
Al Virod3bb52b2007-08-22 20:06:58 -040065 __be16 hw_addr_space;
66 __be16 prot_addr_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 u8 hw_addr_len;
68 u8 prot_addr_len;
Al Virod3bb52b2007-08-22 20:06:58 -040069 __be16 op_code;
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 u8 mac_src[ETH_ALEN]; /* sender hardware address */
Al Virod3bb52b2007-08-22 20:06:58 -040071 __be32 ip_src; /* sender IP address */
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 u8 mac_dst[ETH_ALEN]; /* target hardware address */
Al Virod3bb52b2007-08-22 20:06:58 -040073 __be32 ip_dst; /* target IP address */
Linus Torvalds1da177e2005-04-16 15:20:36 -070074};
75#pragma pack()
76
Arnaldo Carvalho de Meloa16aeb32007-03-10 16:07:19 -030077static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb)
78{
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -070079 return (struct arp_pkt *)skb_network_header(skb);
Arnaldo Carvalho de Meloa16aeb32007-03-10 16:07:19 -030080}
81
Linus Torvalds1da177e2005-04-16 15:20:36 -070082/* Forward declaration */
Vlad Yasevichd0c21d42014-05-21 13:19:48 -040083static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
84 bool strict_match);
Jiri Bohace53665c2012-11-28 04:42:14 +000085static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp);
86static void rlb_src_unlink(struct bonding *bond, u32 index);
87static void rlb_src_link(struct bonding *bond, u32 ip_src_hash,
88 u32 ip_dst_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -070090static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -070091{
92 int i;
93 u8 hash = 0;
94
Wang Yufenfdb89d72014-02-14 17:15:14 +080095 for (i = 0; i < hash_size; i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 hash ^= hash_start[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
98 return hash;
99}
100
101/*********************** tlb specific functions ***************************/
102
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000103static inline void _lock_tx_hashtbl_bh(struct bonding *bond)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104{
Jay Vosburgh6603a6f2007-10-17 17:37:50 -0700105 spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106}
107
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000108static inline void _unlock_tx_hashtbl_bh(struct bonding *bond)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109{
Jay Vosburgh6603a6f2007-10-17 17:37:50 -0700110 spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111}
112
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000113static inline void _lock_tx_hashtbl(struct bonding *bond)
114{
115 spin_lock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
116}
117
118static inline void _unlock_tx_hashtbl(struct bonding *bond)
119{
120 spin_unlock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
121}
122
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123/* Caller must hold tx_hashtbl lock */
124static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
125{
126 if (save_load) {
127 entry->load_history = 1 + entry->tx_bytes /
128 BOND_TLB_REBALANCE_INTERVAL;
129 entry->tx_bytes = 0;
130 }
131
132 entry->tx_slave = NULL;
133 entry->next = TLB_NULL_INDEX;
134 entry->prev = TLB_NULL_INDEX;
135}
136
137static inline void tlb_init_slave(struct slave *slave)
138{
139 SLAVE_TLB_INFO(slave).load = 0;
140 SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
141}
142
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000143/* Caller must hold bond lock for read, BH disabled */
144static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
145 int save_load)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146{
147 struct tlb_client_info *tx_hash_table;
148 u32 index;
149
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 /* clear slave from tx_hashtbl */
151 tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
152
Andy Gospodarekce39a802008-10-30 17:41:16 -0700153 /* skip this if we've already freed the tx hash table */
154 if (tx_hash_table) {
155 index = SLAVE_TLB_INFO(slave).head;
156 while (index != TLB_NULL_INDEX) {
157 u32 next_index = tx_hash_table[index].next;
158 tlb_init_table_entry(&tx_hash_table[index], save_load);
159 index = next_index;
160 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 }
162
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 tlb_init_slave(slave);
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000164}
Jay Vosburgh5af47b22006-01-09 12:14:00 -0800165
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000166/* Caller must hold bond lock for read */
167static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
168 int save_load)
169{
170 _lock_tx_hashtbl_bh(bond);
171 __tlb_clear_slave(bond, slave, save_load);
172 _unlock_tx_hashtbl_bh(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173}
174
175/* Must be called before starting the monitor timer */
176static int tlb_initialize(struct bonding *bond)
177{
178 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
179 int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
Mitch Williams0d206a32005-11-09 10:35:30 -0800180 struct tlb_client_info *new_hashtbl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 int i;
182
Joe Jin243cb4e2007-02-06 14:16:40 -0800183 new_hashtbl = kzalloc(size, GFP_KERNEL);
Joe Perchese404dec2012-01-29 12:56:23 +0000184 if (!new_hashtbl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 return -1;
Joe Perchese404dec2012-01-29 12:56:23 +0000186
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000187 _lock_tx_hashtbl_bh(bond);
Mitch Williams0d206a32005-11-09 10:35:30 -0800188
189 bond_info->tx_hashtbl = new_hashtbl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
Wang Yufenfdb89d72014-02-14 17:15:14 +0800191 for (i = 0; i < TLB_HASH_TABLE_SIZE; i++)
Peter Pan(潘卫平)38dbaf02011-04-08 03:40:19 +0000192 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000194 _unlock_tx_hashtbl_bh(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
196 return 0;
197}
198
199/* Must be called only after all slaves have been released */
200static void tlb_deinitialize(struct bonding *bond)
201{
202 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
Mahesh Bandewar6b794c12014-07-16 11:10:36 -0700203 struct tlb_up_slave *arr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000205 _lock_tx_hashtbl_bh(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206
207 kfree(bond_info->tx_hashtbl);
208 bond_info->tx_hashtbl = NULL;
209
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000210 _unlock_tx_hashtbl_bh(bond);
Mahesh Bandewar6b794c12014-07-16 11:10:36 -0700211
212 arr = rtnl_dereference(bond_info->slave_arr);
213 if (arr)
214 kfree_rcu(arr, rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215}
216
Jiri Pirko097811b2010-05-19 03:26:39 +0000217static long long compute_gap(struct slave *slave)
218{
219 return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
220 (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
221}
222
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223/* Caller must hold bond lock for read */
224static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
225{
226 struct slave *slave, *least_loaded;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +0200227 struct list_head *iter;
Jiri Pirko097811b2010-05-19 03:26:39 +0000228 long long max_gap;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
Jiri Pirko097811b2010-05-19 03:26:39 +0000230 least_loaded = NULL;
231 max_gap = LLONG_MIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
233 /* Find the slave with the largest gap */
dingtianhong28c71922013-10-15 16:28:39 +0800234 bond_for_each_slave_rcu(bond, slave, iter) {
Veaceslav Falico8557cd72014-05-15 21:39:59 +0200235 if (bond_slave_can_tx(slave)) {
Jiri Pirko097811b2010-05-19 03:26:39 +0000236 long long gap = compute_gap(slave);
237
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 if (max_gap < gap) {
239 least_loaded = slave;
240 max_gap = gap;
241 }
242 }
243 }
244
245 return least_loaded;
246}
247
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000248static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
249 u32 skb_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250{
251 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
252 struct tlb_client_info *hash_table;
253 struct slave *assigned_slave;
254
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 hash_table = bond_info->tx_hashtbl;
256 assigned_slave = hash_table[hash_index].tx_slave;
257 if (!assigned_slave) {
258 assigned_slave = tlb_get_least_loaded_slave(bond);
259
260 if (assigned_slave) {
261 struct tlb_slave_info *slave_info =
262 &(SLAVE_TLB_INFO(assigned_slave));
263 u32 next_index = slave_info->head;
264
265 hash_table[hash_index].tx_slave = assigned_slave;
266 hash_table[hash_index].next = next_index;
267 hash_table[hash_index].prev = TLB_NULL_INDEX;
268
Wang Yufenfdb89d72014-02-14 17:15:14 +0800269 if (next_index != TLB_NULL_INDEX)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 hash_table[next_index].prev = hash_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271
272 slave_info->head = hash_index;
273 slave_info->load +=
274 hash_table[hash_index].load_history;
275 }
276 }
277
Wang Yufenfdb89d72014-02-14 17:15:14 +0800278 if (assigned_slave)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 hash_table[hash_index].tx_bytes += skb_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 return assigned_slave;
282}
283
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000284/* Caller must hold bond lock for read */
285static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
286 u32 skb_len)
287{
288 struct slave *tx_slave;
289 /*
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.
293 */
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;
298}
299
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300/*********************** rlb specific functions ***************************/
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000301static inline void _lock_rx_hashtbl_bh(struct bonding *bond)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302{
Jay Vosburgh6603a6f2007-10-17 17:37:50 -0700303 spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304}
305
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000306static inline void _unlock_rx_hashtbl_bh(struct bonding *bond)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307{
Jay Vosburgh6603a6f2007-10-17 17:37:50 -0700308 spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309}
310
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000311static inline void _lock_rx_hashtbl(struct bonding *bond)
312{
313 spin_lock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
314}
315
316static inline void _unlock_rx_hashtbl(struct bonding *bond)
317{
318 spin_unlock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
319}
320
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321/* when an ARP REPLY is received from a client update its info
322 * in the rx_hashtbl
323 */
324static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
325{
326 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
327 struct rlb_client_info *client_info;
328 u32 hash_index;
329
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000330 _lock_rx_hashtbl_bh(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331
Wang Yufenb85b6fb2014-02-14 17:15:12 +0800332 hash_index = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 client_info = &(bond_info->rx_hashtbl[hash_index]);
334
335 if ((client_info->assigned) &&
336 (client_info->ip_src == arp->ip_dst) &&
Flavio Leitner42d782a2010-06-29 08:24:39 +0000337 (client_info->ip_dst == arp->ip_src) &&
Joe Perchesa6700db2012-05-09 17:04:04 +0000338 (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 /* update the clients MAC address */
Joe Perchesada0f862014-02-15 16:02:17 -0800340 ether_addr_copy(client_info->mac_dst, arp->mac_src);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 client_info->ntt = 1;
342 bond_info->rx_ntt = 1;
343 }
344
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000345 _unlock_rx_hashtbl_bh(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346}
347
Eric Dumazetde063b72012-06-11 19:23:07 +0000348static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
349 struct slave *slave)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350{
Eric Dumazetde063b72012-06-11 19:23:07 +0000351 struct arp_pkt *arp, _arp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352
Jiri Pirko3aba8912011-04-19 03:48:16 +0000353 if (skb->protocol != cpu_to_be16(ETH_P_ARP))
David S. Millerb99215c2012-05-13 15:45:13 -0400354 goto out;
Jay Vosburgh6146b1a2008-11-04 17:51:15 -0800355
Eric Dumazetde063b72012-06-11 19:23:07 +0000356 arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp);
357 if (!arp)
David S. Millerb99215c2012-05-13 15:45:13 -0400358 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359
Jiri Bohace53665c2012-11-28 04:42:14 +0000360 /* 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.
369 */
370 rlb_purge_src_ip(bond, arp);
371
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 if (arp->op_code == htons(ARPOP_REPLY)) {
373 /* update rx hash table for this ARP */
374 rlb_update_entry_from_arp(bond, arp);
Veaceslav Falico0a111a02014-07-15 19:36:02 +0200375 netdev_dbg(bond->dev, "Server received an ARP Reply from client\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 }
David S. Millerb99215c2012-05-13 15:45:13 -0400377out:
378 return RX_HANDLER_ANOTHER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379}
380
381/* Caller must hold bond lock for read */
382static struct slave *rlb_next_rx_slave(struct bonding *bond)
383{
384 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
Veaceslav Falico6475ae42013-09-25 09:20:17 +0200385 struct slave *before = NULL, *rx_slave = NULL, *slave;
386 struct list_head *iter;
387 bool found = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
Veaceslav Falico6475ae42013-09-25 09:20:17 +0200389 bond_for_each_slave(bond, slave, iter) {
Veaceslav Falico8557cd72014-05-15 21:39:59 +0200390 if (!bond_slave_can_tx(slave))
Veaceslav Falico6475ae42013-09-25 09:20:17 +0200391 continue;
392 if (!found) {
393 if (!before || before->speed < slave->speed)
394 before = slave;
395 } else {
396 if (!rx_slave || rx_slave->speed < slave->speed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 rx_slave = slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 }
Veaceslav Falico6475ae42013-09-25 09:20:17 +0200399 if (slave == bond_info->rx_slave)
400 found = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 }
Veaceslav Falico6475ae42013-09-25 09:20:17 +0200402 /* we didn't find anything after the current or we have something
403 * better before and up to the current slave
404 */
405 if (!rx_slave || (before && rx_slave->speed < before->speed))
406 rx_slave = before;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407
Veaceslav Falico6475ae42013-09-25 09:20:17 +0200408 if (rx_slave)
409 bond_info->rx_slave = rx_slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410
411 return rx_slave;
412}
413
dingtianhong28c71922013-10-15 16:28:39 +0800414/* Caller must hold rcu_read_lock() for read */
415static struct slave *__rlb_next_rx_slave(struct bonding *bond)
416{
417 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
418 struct slave *before = NULL, *rx_slave = NULL, *slave;
419 struct list_head *iter;
420 bool found = false;
421
422 bond_for_each_slave_rcu(bond, slave, iter) {
Veaceslav Falico8557cd72014-05-15 21:39:59 +0200423 if (!bond_slave_can_tx(slave))
dingtianhong28c71922013-10-15 16:28:39 +0800424 continue;
425 if (!found) {
426 if (!before || before->speed < slave->speed)
427 before = slave;
428 } else {
429 if (!rx_slave || rx_slave->speed < slave->speed)
430 rx_slave = slave;
431 }
432 if (slave == bond_info->rx_slave)
433 found = true;
434 }
435 /* we didn't find anything after the current or we have something
436 * better before and up to the current slave
437 */
438 if (!rx_slave || (before && rx_slave->speed < before->speed))
439 rx_slave = before;
440
441 if (rx_slave)
442 bond_info->rx_slave = rx_slave;
443
444 return rx_slave;
445}
446
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447/* teach the switch the mac of a disabled slave
448 * on the primary for fault tolerance
449 *
450 * Caller must hold bond->curr_slave_lock for write or bond lock for write
451 */
452static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
453{
Eric Dumazet4740d632014-07-15 06:56:55 -0700454 struct slave *curr_active = bond_deref_active_protected(bond);
455
456 if (!curr_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458
459 if (!bond->alb_info.primary_is_promisc) {
Eric Dumazet4740d632014-07-15 06:56:55 -0700460 if (!dev_set_promiscuity(curr_active->dev, 1))
Wang Chen7e1a1ac2008-07-14 20:51:36 -0700461 bond->alb_info.primary_is_promisc = 1;
462 else
463 bond->alb_info.primary_is_promisc = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 }
465
466 bond->alb_info.rlb_promisc_timeout_counter = 0;
467
Eric Dumazet4740d632014-07-15 06:56:55 -0700468 alb_send_learning_packets(curr_active, addr, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469}
470
471/* slave being removed should not be active at this point
472 *
dingtianhongb2e7ace2013-12-13 10:19:32 +0800473 * Caller must hold rtnl.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 */
475static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
476{
477 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
478 struct rlb_client_info *rx_hash_table;
479 u32 index, next_index;
480
481 /* clear slave from rx_hashtbl */
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000482 _lock_rx_hashtbl_bh(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
484 rx_hash_table = bond_info->rx_hashtbl;
Jiri Bohace53665c2012-11-28 04:42:14 +0000485 index = bond_info->rx_hashtbl_used_head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 for (; index != RLB_NULL_INDEX; index = next_index) {
Jiri Bohace53665c2012-11-28 04:42:14 +0000487 next_index = rx_hash_table[index].used_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 if (rx_hash_table[index].slave == slave) {
489 struct slave *assigned_slave = rlb_next_rx_slave(bond);
490
491 if (assigned_slave) {
492 rx_hash_table[index].slave = assigned_slave;
Joe Perchesa6700db2012-05-09 17:04:04 +0000493 if (!ether_addr_equal_64bits(rx_hash_table[index].mac_dst,
494 mac_bcast)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 bond_info->rx_hashtbl[index].ntt = 1;
496 bond_info->rx_ntt = 1;
497 /* A slave has been removed from the
498 * table because it is either disabled
499 * or being released. We must retry the
500 * update to avoid clients from not
501 * being updated & disconnecting when
502 * there is stress
503 */
504 bond_info->rlb_update_retry_counter =
505 RLB_UPDATE_RETRY;
506 }
507 } else { /* there is no active slave */
508 rx_hash_table[index].slave = NULL;
509 }
510 }
511 }
512
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000513 _unlock_rx_hashtbl_bh(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514
Jay Vosburgh6603a6f2007-10-17 17:37:50 -0700515 write_lock_bh(&bond->curr_slave_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516
Eric Dumazet4740d632014-07-15 06:56:55 -0700517 if (slave != bond_deref_active_protected(bond))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
Jay Vosburgh6603a6f2007-10-17 17:37:50 -0700520 write_unlock_bh(&bond->curr_slave_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521}
522
523static void rlb_update_client(struct rlb_client_info *client_info)
524{
525 int i;
526
Wang Yufen35d75ee2014-02-14 17:15:15 +0800527 if (!client_info->slave)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529
530 for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
531 struct sk_buff *skb;
532
533 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
534 client_info->ip_dst,
535 client_info->slave->dev,
536 client_info->ip_src,
537 client_info->mac_dst,
538 client_info->slave->dev->dev_addr,
539 client_info->mac_dst);
540 if (!skb) {
Veaceslav Falico0a111a02014-07-15 19:36:02 +0200541 netdev_err(client_info->slave->bond->dev,
542 "failed to create an ARP packet\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 continue;
544 }
545
546 skb->dev = client_info->slave->dev;
547
Veaceslav Falicod3ab3ff2013-08-29 23:38:57 +0200548 if (client_info->vlan_id) {
Patrick McHardy86a9bad2013-04-19 02:04:30 +0000549 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), client_info->vlan_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 if (!skb) {
Veaceslav Falico0a111a02014-07-15 19:36:02 +0200551 netdev_err(client_info->slave->bond->dev,
552 "failed to insert VLAN tag\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 continue;
554 }
555 }
556
557 arp_xmit(skb);
558 }
559}
560
561/* sends ARP REPLIES that update the clients that need updating */
562static void rlb_update_rx_clients(struct bonding *bond)
563{
564 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
565 struct rlb_client_info *client_info;
566 u32 hash_index;
567
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000568 _lock_rx_hashtbl_bh(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
Jiri Bohace53665c2012-11-28 04:42:14 +0000570 hash_index = bond_info->rx_hashtbl_used_head;
571 for (; hash_index != RLB_NULL_INDEX;
572 hash_index = client_info->used_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 client_info = &(bond_info->rx_hashtbl[hash_index]);
574 if (client_info->ntt) {
575 rlb_update_client(client_info);
Wang Yufen35d75ee2014-02-14 17:15:15 +0800576 if (bond_info->rlb_update_retry_counter == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 client_info->ntt = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 }
579 }
580
Thadeu Lima de Souza Cascardo94e2bd62009-10-16 15:20:49 +0200581 /* do not update the entries again until this counter is zero so that
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 * not to confuse the clients.
583 */
584 bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
585
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000586 _unlock_rx_hashtbl_bh(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587}
588
589/* The slave was assigned a new mac address - update the clients */
590static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
591{
592 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
593 struct rlb_client_info *client_info;
594 int ntt = 0;
595 u32 hash_index;
596
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000597 _lock_rx_hashtbl_bh(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
Jiri Bohace53665c2012-11-28 04:42:14 +0000599 hash_index = bond_info->rx_hashtbl_used_head;
600 for (; hash_index != RLB_NULL_INDEX;
601 hash_index = client_info->used_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 client_info = &(bond_info->rx_hashtbl[hash_index]);
603
604 if ((client_info->slave == slave) &&
Joe Perchesa6700db2012-05-09 17:04:04 +0000605 !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 client_info->ntt = 1;
607 ntt = 1;
608 }
609 }
610
Wang Yufen4708a1b2014-02-14 17:15:13 +0800611 /* update the team's flag only after the whole iteration */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 if (ntt) {
613 bond_info->rx_ntt = 1;
Wang Yufen4708a1b2014-02-14 17:15:13 +0800614 /* fasten the change */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
616 }
617
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000618 _unlock_rx_hashtbl_bh(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619}
620
621/* mark all clients using src_ip to be updated */
Al Virod3bb52b2007-08-22 20:06:58 -0400622static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623{
624 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
625 struct rlb_client_info *client_info;
626 u32 hash_index;
627
628 _lock_rx_hashtbl(bond);
629
Jiri Bohace53665c2012-11-28 04:42:14 +0000630 hash_index = bond_info->rx_hashtbl_used_head;
631 for (; hash_index != RLB_NULL_INDEX;
632 hash_index = client_info->used_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 client_info = &(bond_info->rx_hashtbl[hash_index]);
634
635 if (!client_info->slave) {
Veaceslav Falico0a111a02014-07-15 19:36:02 +0200636 netdev_err(bond->dev, "found a client with no channel in the client's hash table\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 continue;
638 }
639 /*update all clients using this src_ip, that are not assigned
640 * to the team's address (curr_active_slave) and have a known
641 * unicast mac address.
642 */
643 if ((client_info->ip_src == src_ip) &&
Joe Perchesa6700db2012-05-09 17:04:04 +0000644 !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
645 bond->dev->dev_addr) &&
646 !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 client_info->ntt = 1;
648 bond_info->rx_ntt = 1;
649 }
650 }
651
652 _unlock_rx_hashtbl(bond);
653}
654
655/* Caller must hold both bond and ptr locks for read */
656static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
657{
658 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
Arnaldo Carvalho de Meloa16aeb32007-03-10 16:07:19 -0300659 struct arp_pkt *arp = arp_pkt(skb);
dingtianhong28c71922013-10-15 16:28:39 +0800660 struct slave *assigned_slave, *curr_active_slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 struct rlb_client_info *client_info;
662 u32 hash_index = 0;
663
664 _lock_rx_hashtbl(bond);
665
dingtianhong28c71922013-10-15 16:28:39 +0800666 curr_active_slave = rcu_dereference(bond->curr_active_slave);
667
Amerigo Wange364a342011-02-27 23:34:28 +0000668 hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 client_info = &(bond_info->rx_hashtbl[hash_index]);
670
671 if (client_info->assigned) {
672 if ((client_info->ip_src == arp->ip_src) &&
673 (client_info->ip_dst == arp->ip_dst)) {
674 /* the entry is already assigned to this client */
Joe Perchesa6700db2012-05-09 17:04:04 +0000675 if (!ether_addr_equal_64bits(arp->mac_dst, mac_bcast)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 /* update mac address from arp */
Joe Perchesada0f862014-02-15 16:02:17 -0800677 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 }
Joe Perchesada0f862014-02-15 16:02:17 -0800679 ether_addr_copy(client_info->mac_src, arp->mac_src);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680
681 assigned_slave = client_info->slave;
682 if (assigned_slave) {
683 _unlock_rx_hashtbl(bond);
684 return assigned_slave;
685 }
686 } else {
687 /* the entry is already assigned to some other client,
688 * move the old client to primary (curr_active_slave) so
689 * that the new client can be assigned to this entry.
690 */
Eric Dumazet4740d632014-07-15 06:56:55 -0700691 if (curr_active_slave &&
dingtianhong28c71922013-10-15 16:28:39 +0800692 client_info->slave != curr_active_slave) {
693 client_info->slave = curr_active_slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 rlb_update_client(client_info);
695 }
696 }
697 }
698 /* assign a new slave */
dingtianhong28c71922013-10-15 16:28:39 +0800699 assigned_slave = __rlb_next_rx_slave(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
701 if (assigned_slave) {
Jiri Bohace53665c2012-11-28 04:42:14 +0000702 if (!(client_info->assigned &&
703 client_info->ip_src == arp->ip_src)) {
704 /* ip_src is going to be updated,
705 * fix the src hash list
706 */
707 u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
708 sizeof(arp->ip_src));
709 rlb_src_unlink(bond, hash_index);
710 rlb_src_link(bond, hash_src, hash_index);
711 }
712
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 client_info->ip_src = arp->ip_src;
714 client_info->ip_dst = arp->ip_dst;
715 /* arp->mac_dst is broadcast for arp reqeusts.
716 * will be updated with clients actual unicast mac address
717 * upon receiving an arp reply.
718 */
Joe Perchesada0f862014-02-15 16:02:17 -0800719 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
720 ether_addr_copy(client_info->mac_src, arp->mac_src);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 client_info->slave = assigned_slave;
722
Joe Perchesa6700db2012-05-09 17:04:04 +0000723 if (!ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 client_info->ntt = 1;
725 bond->alb_info.rx_ntt = 1;
726 } else {
727 client_info->ntt = 0;
728 }
729
dingtianhongfb00bc22014-03-12 17:31:59 +0800730 if (vlan_get_tag(skb, &client_info->vlan_id))
Veaceslav Falicod3ab3ff2013-08-29 23:38:57 +0200731 client_info->vlan_id = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
733 if (!client_info->assigned) {
Jiri Bohace53665c2012-11-28 04:42:14 +0000734 u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
735 bond_info->rx_hashtbl_used_head = hash_index;
736 client_info->used_next = prev_tbl_head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 if (prev_tbl_head != RLB_NULL_INDEX) {
Jiri Bohace53665c2012-11-28 04:42:14 +0000738 bond_info->rx_hashtbl[prev_tbl_head].used_prev =
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 hash_index;
740 }
741 client_info->assigned = 1;
742 }
743 }
744
745 _unlock_rx_hashtbl(bond);
746
747 return assigned_slave;
748}
749
750/* chooses (and returns) transmit channel for arp reply
751 * does not choose channel for other arp types since they are
752 * sent on the curr_active_slave
753 */
754static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
755{
Arnaldo Carvalho de Meloa16aeb32007-03-10 16:07:19 -0300756 struct arp_pkt *arp = arp_pkt(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 struct slave *tx_slave = NULL;
758
zheng.li567b8712012-11-27 23:57:04 +0000759 /* Don't modify or load balance ARPs that do not originate locally
760 * (e.g.,arrive via a bridge).
761 */
Vlad Yasevich14af9962014-06-04 16:23:38 -0400762 if (!bond_slave_has_mac_rx(bond, arp->mac_src))
zheng.li567b8712012-11-27 23:57:04 +0000763 return NULL;
764
Brian Haleyf14c4e42008-09-02 10:08:08 -0400765 if (arp->op_code == htons(ARPOP_REPLY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 /* the arp must be sent on the selected
767 * rx channel
768 */
769 tx_slave = rlb_choose_channel(skb, bond);
Wang Yufen35d75ee2014-02-14 17:15:15 +0800770 if (tx_slave)
Joe Perchesada0f862014-02-15 16:02:17 -0800771 ether_addr_copy(arp->mac_src, tx_slave->dev->dev_addr);
Veaceslav Falico0a111a02014-07-15 19:36:02 +0200772 netdev_dbg(bond->dev, "Server sent ARP Reply packet\n");
Brian Haleyf14c4e42008-09-02 10:08:08 -0400773 } else if (arp->op_code == htons(ARPOP_REQUEST)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 /* Create an entry in the rx_hashtbl for this client as a
775 * place holder.
776 * When the arp reply is received the entry will be updated
777 * with the correct unicast address of the client.
778 */
779 rlb_choose_channel(skb, bond);
780
Peter Pan(潘卫平)77c8e2c2011-04-11 00:16:32 +0000781 /* The ARP reply packets must be delayed so that
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 * they can cancel out the influence of the ARP request.
783 */
784 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
785
786 /* arp requests are broadcast and are sent on the primary
787 * the arp request will collapse all clients on the subnet to
788 * the primary slave. We must register these clients to be
789 * updated with their assigned mac.
790 */
791 rlb_req_update_subnet_clients(bond, arp->ip_src);
Veaceslav Falico0a111a02014-07-15 19:36:02 +0200792 netdev_dbg(bond->dev, "Server sent ARP Request packet\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 }
794
795 return tx_slave;
796}
797
798/* Caller must hold bond lock for read */
799static void rlb_rebalance(struct bonding *bond)
800{
801 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
802 struct slave *assigned_slave;
803 struct rlb_client_info *client_info;
804 int ntt;
805 u32 hash_index;
806
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000807 _lock_rx_hashtbl_bh(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808
809 ntt = 0;
Jiri Bohace53665c2012-11-28 04:42:14 +0000810 hash_index = bond_info->rx_hashtbl_used_head;
811 for (; hash_index != RLB_NULL_INDEX;
812 hash_index = client_info->used_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 client_info = &(bond_info->rx_hashtbl[hash_index]);
dingtianhong733ab632013-12-13 10:19:45 +0800814 assigned_slave = __rlb_next_rx_slave(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 if (assigned_slave && (client_info->slave != assigned_slave)) {
816 client_info->slave = assigned_slave;
817 client_info->ntt = 1;
818 ntt = 1;
819 }
820 }
821
822 /* update the team's flag only after the whole iteration */
Wang Yufen35d75ee2014-02-14 17:15:15 +0800823 if (ntt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 bond_info->rx_ntt = 1;
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000825 _unlock_rx_hashtbl_bh(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826}
827
828/* Caller must hold rx_hashtbl lock */
Jiri Bohace53665c2012-11-28 04:42:14 +0000829static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
830{
831 entry->used_next = RLB_NULL_INDEX;
832 entry->used_prev = RLB_NULL_INDEX;
833 entry->assigned = 0;
834 entry->slave = NULL;
Veaceslav Falicod3ab3ff2013-08-29 23:38:57 +0200835 entry->vlan_id = 0;
Jiri Bohace53665c2012-11-28 04:42:14 +0000836}
837static void rlb_init_table_entry_src(struct rlb_client_info *entry)
838{
839 entry->src_first = RLB_NULL_INDEX;
840 entry->src_prev = RLB_NULL_INDEX;
841 entry->src_next = RLB_NULL_INDEX;
842}
843
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844static void rlb_init_table_entry(struct rlb_client_info *entry)
845{
846 memset(entry, 0, sizeof(struct rlb_client_info));
Jiri Bohace53665c2012-11-28 04:42:14 +0000847 rlb_init_table_entry_dst(entry);
848 rlb_init_table_entry_src(entry);
849}
850
851static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
852{
853 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
854 u32 next_index = bond_info->rx_hashtbl[index].used_next;
855 u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
856
857 if (index == bond_info->rx_hashtbl_used_head)
858 bond_info->rx_hashtbl_used_head = next_index;
859 if (prev_index != RLB_NULL_INDEX)
860 bond_info->rx_hashtbl[prev_index].used_next = next_index;
861 if (next_index != RLB_NULL_INDEX)
862 bond_info->rx_hashtbl[next_index].used_prev = prev_index;
863}
864
865/* unlink a rlb hash table entry from the src list */
866static void rlb_src_unlink(struct bonding *bond, u32 index)
867{
868 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
869 u32 next_index = bond_info->rx_hashtbl[index].src_next;
870 u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
871
872 bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
873 bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
874
875 if (next_index != RLB_NULL_INDEX)
876 bond_info->rx_hashtbl[next_index].src_prev = prev_index;
877
878 if (prev_index == RLB_NULL_INDEX)
879 return;
880
881 /* is prev_index pointing to the head of this list? */
882 if (bond_info->rx_hashtbl[prev_index].src_first == index)
883 bond_info->rx_hashtbl[prev_index].src_first = next_index;
884 else
885 bond_info->rx_hashtbl[prev_index].src_next = next_index;
886
887}
888
889static void rlb_delete_table_entry(struct bonding *bond, u32 index)
890{
891 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
892 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
893
894 rlb_delete_table_entry_dst(bond, index);
895 rlb_init_table_entry_dst(entry);
896
897 rlb_src_unlink(bond, index);
898}
899
900/* add the rx_hashtbl[ip_dst_hash] entry to the list
901 * of entries with identical ip_src_hash
902 */
903static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
904{
905 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
906 u32 next;
907
908 bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
909 next = bond_info->rx_hashtbl[ip_src_hash].src_first;
910 bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
911 if (next != RLB_NULL_INDEX)
912 bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
913 bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
914}
915
916/* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does
917 * not match arp->mac_src */
918static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
919{
920 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
Wang Yufenb85b6fb2014-02-14 17:15:12 +0800921 u32 ip_src_hash = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
Jiri Bohace53665c2012-11-28 04:42:14 +0000922 u32 index;
923
924 _lock_rx_hashtbl_bh(bond);
925
926 index = bond_info->rx_hashtbl[ip_src_hash].src_first;
927 while (index != RLB_NULL_INDEX) {
928 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
929 u32 next_index = entry->src_next;
930 if (entry->ip_src == arp->ip_src &&
931 !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
932 rlb_delete_table_entry(bond, index);
933 index = next_index;
934 }
935 _unlock_rx_hashtbl_bh(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936}
937
938static int rlb_initialize(struct bonding *bond)
939{
940 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
Mitch Williams0d206a32005-11-09 10:35:30 -0800941 struct rlb_client_info *new_hashtbl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
943 int i;
944
Mitch Williams0d206a32005-11-09 10:35:30 -0800945 new_hashtbl = kmalloc(size, GFP_KERNEL);
Joe Perchese404dec2012-01-29 12:56:23 +0000946 if (!new_hashtbl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 return -1;
Joe Perchese404dec2012-01-29 12:56:23 +0000948
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000949 _lock_rx_hashtbl_bh(bond);
Mitch Williams0d206a32005-11-09 10:35:30 -0800950
951 bond_info->rx_hashtbl = new_hashtbl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952
Jiri Bohace53665c2012-11-28 04:42:14 +0000953 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954
Wang Yufen35d75ee2014-02-14 17:15:15 +0800955 for (i = 0; i < RLB_HASH_TABLE_SIZE; i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 rlb_init_table_entry(bond_info->rx_hashtbl + i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000958 _unlock_rx_hashtbl_bh(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 /* register to receive ARPs */
Jiri Pirko3aba8912011-04-19 03:48:16 +0000961 bond->recv_probe = rlb_arp_recv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962
963 return 0;
964}
965
966static void rlb_deinitialize(struct bonding *bond)
967{
968 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
969
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000970 _lock_rx_hashtbl_bh(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
972 kfree(bond_info->rx_hashtbl);
973 bond_info->rx_hashtbl = NULL;
Jiri Bohace53665c2012-11-28 04:42:14 +0000974 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000976 _unlock_rx_hashtbl_bh(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977}
978
979static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
980{
981 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
982 u32 curr_index;
983
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000984 _lock_rx_hashtbl_bh(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985
Jiri Bohace53665c2012-11-28 04:42:14 +0000986 curr_index = bond_info->rx_hashtbl_used_head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 while (curr_index != RLB_NULL_INDEX) {
988 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
Jiri Bohace53665c2012-11-28 04:42:14 +0000989 u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990
Veaceslav Falicod3ab3ff2013-08-29 23:38:57 +0200991 if (curr->vlan_id == vlan_id)
Jiri Bohace53665c2012-11-28 04:42:14 +0000992 rlb_delete_table_entry(bond, curr_index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
994 curr_index = next_index;
995 }
996
Maxim Uvarovf515e6b2012-01-09 12:01:37 +0000997 _unlock_rx_hashtbl_bh(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998}
999
1000/*********************** tlb/rlb shared functions *********************/
1001
Veaceslav Falico7aa64982013-08-28 23:25:13 +02001002static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
Vlad Yasevichd6b694c2014-05-21 11:24:39 -04001003 __be16 vlan_proto, u16 vid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 struct learning_pkt pkt;
Veaceslav Falico7aa64982013-08-28 23:25:13 +02001006 struct sk_buff *skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 int size = sizeof(struct learning_pkt);
Veaceslav Falico7aa64982013-08-28 23:25:13 +02001008 char *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
1010 memset(&pkt, 0, size);
Joe Perchesada0f862014-02-15 16:02:17 -08001011 ether_addr_copy(pkt.mac_dst, mac_addr);
1012 ether_addr_copy(pkt.mac_src, mac_addr);
Veaceslav Falico96a09222014-03-13 12:41:58 +01001013 pkt.type = cpu_to_be16(ETH_P_LOOPBACK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014
Veaceslav Falico7aa64982013-08-28 23:25:13 +02001015 skb = dev_alloc_skb(size);
1016 if (!skb)
1017 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018
Veaceslav Falico7aa64982013-08-28 23:25:13 +02001019 data = skb_put(skb, size);
1020 memcpy(data, &pkt, size);
1021
1022 skb_reset_mac_header(skb);
1023 skb->network_header = skb->mac_header + ETH_HLEN;
1024 skb->protocol = pkt.type;
1025 skb->priority = TC_PRIO_CONTROL;
1026 skb->dev = slave->dev;
1027
1028 if (vid) {
Vlad Yasevichd6b694c2014-05-21 11:24:39 -04001029 skb = vlan_put_tag(skb, vlan_proto, vid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 if (!skb) {
Veaceslav Falico0a111a02014-07-15 19:36:02 +02001031 netdev_err(slave->bond->dev, "failed to insert VLAN tag\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 return;
1033 }
Veaceslav Falico7aa64982013-08-28 23:25:13 +02001034 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035
Veaceslav Falico7aa64982013-08-28 23:25:13 +02001036 dev_queue_xmit(skb);
1037}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038
Vlad Yasevichd0c21d42014-05-21 13:19:48 -04001039static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
1040 bool strict_match)
Veaceslav Falico7aa64982013-08-28 23:25:13 +02001041{
1042 struct bonding *bond = bond_get_bond_by_slave(slave);
Veaceslav Falico5bf94b82013-08-28 23:25:14 +02001043 struct net_device *upper;
1044 struct list_head *iter;
Vlad Yasevich14af9962014-06-04 16:23:38 -04001045 struct bond_vlan_tag tags[BOND_MAX_VLAN_ENCAP];
Veaceslav Falico7aa64982013-08-28 23:25:13 +02001046
Veaceslav Falico5bf94b82013-08-28 23:25:14 +02001047 /* send untagged */
Vlad Yasevichd6b694c2014-05-21 11:24:39 -04001048 alb_send_lp_vid(slave, mac_addr, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049
Vlad Yasevich14af9962014-06-04 16:23:38 -04001050 /* loop through all devices and see if we need to send a packet
1051 * for that device.
1052 */
Veaceslav Falico5bf94b82013-08-28 23:25:14 +02001053 rcu_read_lock();
Veaceslav Falico2f268f12013-09-25 09:20:07 +02001054 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
Vlad Yasevichd0c21d42014-05-21 13:19:48 -04001055 if (is_vlan_dev(upper) && vlan_get_encap_level(upper) == 0) {
1056 if (strict_match &&
1057 ether_addr_equal_64bits(mac_addr,
1058 upper->dev_addr)) {
1059 alb_send_lp_vid(slave, mac_addr,
1060 vlan_dev_vlan_proto(upper),
1061 vlan_dev_vlan_id(upper));
1062 } else if (!strict_match) {
1063 alb_send_lp_vid(slave, upper->dev_addr,
1064 vlan_dev_vlan_proto(upper),
1065 vlan_dev_vlan_id(upper));
1066 }
1067 }
Vlad Yasevich14af9962014-06-04 16:23:38 -04001068
1069 /* If this is a macvlan device, then only send updates
1070 * when strict_match is turned off.
1071 */
1072 if (netif_is_macvlan(upper) && !strict_match) {
1073 memset(tags, 0, sizeof(tags));
1074 bond_verify_device_path(bond->dev, upper, tags);
1075 alb_send_lp_vid(slave, upper->dev_addr,
1076 tags[0].vlan_proto, tags[0].vlan_id);
1077 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 }
Veaceslav Falico5bf94b82013-08-28 23:25:14 +02001079 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080}
1081
Jiri Bohacb9245512012-01-18 12:24:54 +00001082static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[])
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083{
1084 struct net_device *dev = slave->dev;
1085 struct sockaddr s_addr;
1086
Veaceslav Falico01844092014-05-15 21:39:55 +02001087 if (BOND_MODE(slave->bond) == BOND_MODE_TLB) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 memcpy(dev->dev_addr, addr, dev->addr_len);
1089 return 0;
1090 }
1091
1092 /* for rlb each slave must have a unique hw mac addresses so that */
1093 /* each slave will receive packets destined to a different mac */
1094 memcpy(s_addr.sa_data, addr, dev->addr_len);
1095 s_addr.sa_family = dev->type;
1096 if (dev_set_mac_address(dev, &s_addr)) {
Veaceslav Falico0a111a02014-07-15 19:36:02 +02001097 netdev_err(slave->bond->dev, "dev_set_mac_address of dev %s failed! ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
1098 dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 return -EOPNOTSUPP;
1100 }
1101 return 0;
1102}
1103
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001104/*
1105 * Swap MAC addresses between two slaves.
1106 *
1107 * Called with RTNL held, and no other locks.
1108 *
1109 */
1110
Veaceslav Falico43547ea2013-05-27 23:14:51 +00001111static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 u8 tmp_mac_addr[ETH_ALEN];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114
Joe Perchesada0f862014-02-15 16:02:17 -08001115 ether_addr_copy(tmp_mac_addr, slave1->dev->dev_addr);
Jiri Bohacb9245512012-01-18 12:24:54 +00001116 alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr);
1117 alb_set_slave_mac_addr(slave2, tmp_mac_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001119}
1120
1121/*
1122 * Send learning packets after MAC address swap.
1123 *
Jay Vosburgh25433312008-01-17 16:24:59 -08001124 * Called with RTNL and no other locks
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001125 */
1126static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1127 struct slave *slave2)
1128{
Veaceslav Falico8557cd72014-05-15 21:39:59 +02001129 int slaves_state_differ = (bond_slave_can_tx(slave1) != bond_slave_can_tx(slave2));
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001130 struct slave *disabled_slave = NULL;
1131
Jay Vosburgh25433312008-01-17 16:24:59 -08001132 ASSERT_RTNL();
1133
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 /* fasten the change in the switch */
Veaceslav Falico8557cd72014-05-15 21:39:59 +02001135 if (bond_slave_can_tx(slave1)) {
Vlad Yasevichd0c21d42014-05-21 13:19:48 -04001136 alb_send_learning_packets(slave1, slave1->dev->dev_addr, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 if (bond->alb_info.rlb_enabled) {
1138 /* inform the clients that the mac address
1139 * has changed
1140 */
1141 rlb_req_update_slave_clients(bond, slave1);
1142 }
1143 } else {
1144 disabled_slave = slave1;
1145 }
1146
Veaceslav Falico8557cd72014-05-15 21:39:59 +02001147 if (bond_slave_can_tx(slave2)) {
Vlad Yasevichd0c21d42014-05-21 13:19:48 -04001148 alb_send_learning_packets(slave2, slave2->dev->dev_addr, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 if (bond->alb_info.rlb_enabled) {
1150 /* inform the clients that the mac address
1151 * has changed
1152 */
1153 rlb_req_update_slave_clients(bond, slave2);
1154 }
1155 } else {
1156 disabled_slave = slave2;
1157 }
1158
1159 if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1160 /* A disabled slave was assigned an active mac addr */
1161 rlb_teach_disabled_mac_on_primary(bond,
1162 disabled_slave->dev->dev_addr);
1163 }
1164}
1165
1166/**
1167 * alb_change_hw_addr_on_detach
1168 * @bond: bonding we're working on
1169 * @slave: the slave that was just detached
1170 *
1171 * We assume that @slave was already detached from the slave list.
1172 *
1173 * If @slave's permanent hw address is different both from its current
1174 * address and from @bond's address, then somewhere in the bond there's
1175 * a slave that has @slave's permanet address as its current address.
1176 * We'll make sure that that slave no longer uses @slave's permanent address.
1177 *
Jay Vosburgh25433312008-01-17 16:24:59 -08001178 * Caller must hold RTNL and no other locks
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 */
1180static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1181{
1182 int perm_curr_diff;
1183 int perm_bond_diff;
Veaceslav Falicob88ec382013-06-18 13:44:52 +02001184 struct slave *found_slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
Joe Perchesa6700db2012-05-09 17:04:04 +00001186 perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1187 slave->dev->dev_addr);
1188 perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1189 bond->dev->dev_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190
1191 if (perm_curr_diff && perm_bond_diff) {
Veaceslav Falicob88ec382013-06-18 13:44:52 +02001192 found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
Veaceslav Falicob88ec382013-06-18 13:44:52 +02001194 if (found_slave) {
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001195 /* locking: needs RTNL and nothing else */
Veaceslav Falicob88ec382013-06-18 13:44:52 +02001196 alb_swap_mac_addr(slave, found_slave);
1197 alb_fasten_mac_swap(bond, slave, found_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 }
1199 }
1200}
1201
1202/**
1203 * alb_handle_addr_collision_on_attach
1204 * @bond: bonding we're working on
1205 * @slave: the slave that was just attached
1206 *
1207 * checks uniqueness of slave's mac address and handles the case the
1208 * new slave uses the bonds mac address.
1209 *
1210 * If the permanent hw address of @slave is @bond's hw address, we need to
1211 * find a different hw address to give @slave, that isn't in use by any other
Peter Pan(潘卫平)77c8e2c2011-04-11 00:16:32 +00001212 * slave in the bond. This address must be, of course, one of the permanent
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 * addresses of the other slaves.
1214 *
1215 * We go over the slave list, and for each slave there we compare its
1216 * permanent hw address with the current address of all the other slaves.
1217 * If no match was found, then we've found a slave with a permanent address
1218 * that isn't used by any other slave in the bond, so we can assign it to
1219 * @slave.
1220 *
1221 * assumption: this function is called before @slave is attached to the
Veaceslav Falicocedb7432013-06-17 19:30:35 +02001222 * bond slave list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 */
1224static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1225{
Eric Dumazet4740d632014-07-15 06:56:55 -07001226 struct slave *has_bond_addr = rcu_access_pointer(bond->curr_active_slave);
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001227 struct slave *tmp_slave1, *free_mac_slave = NULL;
1228 struct list_head *iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001230 if (!bond_has_slaves(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 /* this is the first slave */
1232 return 0;
1233 }
1234
1235 /* if slave's mac address differs from bond's mac address
1236 * check uniqueness of slave's mac address against the other
1237 * slaves in the bond.
1238 */
Joe Perchesa6700db2012-05-09 17:04:04 +00001239 if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
Veaceslav Falicocedb7432013-06-17 19:30:35 +02001240 if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
John W. Linville6b38aef2005-07-28 15:00:15 -04001241 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242
John W. Linville6b38aef2005-07-28 15:00:15 -04001243 /* Try setting slave mac to bond address and fall-through
1244 to code handling that situation below... */
Jiri Bohacb9245512012-01-18 12:24:54 +00001245 alb_set_slave_mac_addr(slave, bond->dev->dev_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 }
1247
1248 /* The slave's address is equal to the address of the bond.
1249 * Search for a spare address in the bond for this slave.
1250 */
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001251 bond_for_each_slave(bond, tmp_slave1, iter) {
Veaceslav Falicocedb7432013-06-17 19:30:35 +02001252 if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 /* no slave has tmp_slave1's perm addr
1254 * as its curr addr
1255 */
1256 free_mac_slave = tmp_slave1;
1257 break;
1258 }
1259
1260 if (!has_bond_addr) {
Joe Perchesa6700db2012-05-09 17:04:04 +00001261 if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1262 bond->dev->dev_addr)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263
1264 has_bond_addr = tmp_slave1;
1265 }
1266 }
1267 }
1268
1269 if (free_mac_slave) {
Jiri Bohacb9245512012-01-18 12:24:54 +00001270 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271
Veaceslav Falico0a111a02014-07-15 19:36:02 +02001272 netdev_warn(bond->dev, "the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1273 slave->dev->name, free_mac_slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274
1275 } else if (has_bond_addr) {
Veaceslav Falico0a111a02014-07-15 19:36:02 +02001276 netdev_err(bond->dev, "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",
1277 slave->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 return -EFAULT;
1279 }
1280
1281 return 0;
1282}
1283
1284/**
1285 * alb_set_mac_address
1286 * @bond:
1287 * @addr:
1288 *
1289 * In TLB mode all slaves are configured to the bond's hw address, but set
1290 * their dev_addr field to different addresses (based on their permanent hw
1291 * addresses).
1292 *
1293 * For each slave, this function sets the interface to the new address and then
1294 * changes its dev_addr field to its previous value.
1295 *
1296 * Unwinding assumes bond's mac address has not yet changed.
1297 */
1298static int alb_set_mac_address(struct bonding *bond, void *addr)
1299{
Veaceslav Falico81f23b12013-09-25 09:20:13 +02001300 struct slave *slave, *rollback_slave;
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001301 struct list_head *iter;
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02001302 struct sockaddr sa;
Veaceslav Falico81f23b12013-09-25 09:20:13 +02001303 char tmp_addr[ETH_ALEN];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02001306 if (bond->alb_info.rlb_enabled)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001309 bond_for_each_slave(bond, slave, iter) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 /* save net_device's current hw address */
Joe Perchesada0f862014-02-15 16:02:17 -08001311 ether_addr_copy(tmp_addr, slave->dev->dev_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312
1313 res = dev_set_mac_address(slave->dev, addr);
1314
1315 /* restore net_device's hw address */
Joe Perchesada0f862014-02-15 16:02:17 -08001316 ether_addr_copy(slave->dev->dev_addr, tmp_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317
Stephen Hemmingereb7cc592008-11-19 21:56:05 -08001318 if (res)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 goto unwind;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 }
1321
1322 return 0;
1323
1324unwind:
1325 memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1326 sa.sa_family = bond->dev->type;
1327
1328 /* unwind from head to the slave that failed */
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001329 bond_for_each_slave(bond, rollback_slave, iter) {
Veaceslav Falico81f23b12013-09-25 09:20:13 +02001330 if (rollback_slave == slave)
1331 break;
Joe Perchesada0f862014-02-15 16:02:17 -08001332 ether_addr_copy(tmp_addr, rollback_slave->dev->dev_addr);
Veaceslav Falico81f23b12013-09-25 09:20:13 +02001333 dev_set_mac_address(rollback_slave->dev, &sa);
Joe Perchesada0f862014-02-15 16:02:17 -08001334 ether_addr_copy(rollback_slave->dev->dev_addr, tmp_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 }
1336
1337 return res;
1338}
1339
1340/************************ exported alb funcions ************************/
1341
1342int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1343{
1344 int res;
1345
1346 res = tlb_initialize(bond);
Wang Yufen35d75ee2014-02-14 17:15:15 +08001347 if (res)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349
1350 if (rlb_enabled) {
1351 bond->alb_info.rlb_enabled = 1;
1352 /* initialize rlb */
1353 res = rlb_initialize(bond);
1354 if (res) {
1355 tlb_deinitialize(bond);
1356 return res;
1357 }
Mitch Williamsb76850a2005-11-09 10:35:35 -08001358 } else {
1359 bond->alb_info.rlb_enabled = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 }
1361
1362 return 0;
1363}
1364
1365void bond_alb_deinitialize(struct bonding *bond)
1366{
1367 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1368
1369 tlb_deinitialize(bond);
1370
Wang Yufendda0fd52014-02-14 17:15:16 +08001371 if (bond_info->rlb_enabled)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 rlb_deinitialize(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373}
1374
Mahesh Bandewar9a49aba2014-04-22 16:30:18 -07001375static int bond_do_alb_xmit(struct sk_buff *skb, struct bonding *bond,
1376 struct slave *tx_slave)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
Mahesh Bandewar9a49aba2014-04-22 16:30:18 -07001379 struct ethhdr *eth_data = eth_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380
1381 if (!tx_slave) {
1382 /* unbalanced or unassigned, send through primary */
dingtianhong28c71922013-10-15 16:28:39 +08001383 tx_slave = rcu_dereference(bond->curr_active_slave);
Mahesh Bandeware9f0fb82014-04-22 16:30:22 -07001384 if (bond->params.tlb_dynamic_lb)
1385 bond_info->unbalanced_load += skb->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 }
1387
Veaceslav Falico8557cd72014-05-15 21:39:59 +02001388 if (tx_slave && bond_slave_can_tx(tx_slave)) {
dingtianhong28c71922013-10-15 16:28:39 +08001389 if (tx_slave != rcu_dereference(bond->curr_active_slave)) {
Joe Perches2a7c1832014-02-18 09:42:45 -08001390 ether_addr_copy(eth_data->h_source,
1391 tx_slave->dev->dev_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 }
1393
dingtianhong4d4ac1b2014-01-02 09:13:02 +08001394 bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1395 goto out;
Joe Perches157550fb2014-02-18 09:42:46 -08001396 }
1397
Mahesh Bandeware9f0fb82014-04-22 16:30:22 -07001398 if (tx_slave && bond->params.tlb_dynamic_lb) {
Joe Perches157550fb2014-02-18 09:42:46 -08001399 _lock_tx_hashtbl(bond);
1400 __tlb_clear_slave(bond, tx_slave, 0);
1401 _unlock_tx_hashtbl(bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 }
1403
dingtianhong4d4ac1b2014-01-02 09:13:02 +08001404 /* no suitable interface, frame not sent */
Eric W. Biederman2bb77ab2014-03-11 14:16:58 -07001405 dev_kfree_skb_any(skb);
dingtianhong4d4ac1b2014-01-02 09:13:02 +08001406out:
Patrick McHardyec634fe2009-07-05 19:23:38 -07001407 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408}
1409
Mahesh Bandewar6b794c12014-07-16 11:10:36 -07001410static int bond_tlb_update_slave_arr(struct bonding *bond,
1411 struct slave *skipslave)
1412{
1413 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1414 struct slave *tx_slave;
1415 struct list_head *iter;
1416 struct tlb_up_slave *new_arr, *old_arr;
1417
1418 new_arr = kzalloc(offsetof(struct tlb_up_slave, arr[bond->slave_cnt]),
1419 GFP_ATOMIC);
1420 if (!new_arr)
1421 return -ENOMEM;
1422
1423 bond_for_each_slave(bond, tx_slave, iter) {
1424 if (!bond_slave_can_tx(tx_slave))
1425 continue;
1426 if (skipslave == tx_slave)
1427 continue;
1428 new_arr->arr[new_arr->count++] = tx_slave;
1429 }
1430
1431 old_arr = rtnl_dereference(bond_info->slave_arr);
1432 rcu_assign_pointer(bond_info->slave_arr, new_arr);
1433 if (old_arr)
1434 kfree_rcu(old_arr, rcu);
1435
1436 return 0;
1437}
1438
Mahesh Bandewarf05b42e2014-04-22 16:30:20 -07001439int bond_tlb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1440{
1441 struct bonding *bond = netdev_priv(bond_dev);
Mahesh Bandewar6b794c12014-07-16 11:10:36 -07001442 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
Mahesh Bandewarf05b42e2014-04-22 16:30:20 -07001443 struct ethhdr *eth_data;
1444 struct slave *tx_slave = NULL;
1445 u32 hash_index;
1446
1447 skb_reset_mac_header(skb);
1448 eth_data = eth_hdr(skb);
1449
1450 /* Do not TX balance any multicast or broadcast */
1451 if (!is_multicast_ether_addr(eth_data->h_dest)) {
1452 switch (skb->protocol) {
1453 case htons(ETH_P_IP):
1454 case htons(ETH_P_IPX):
1455 /* In case of IPX, it will falback to L2 hash */
1456 case htons(ETH_P_IPV6):
1457 hash_index = bond_xmit_hash(bond, skb);
Mahesh Bandeware9f0fb82014-04-22 16:30:22 -07001458 if (bond->params.tlb_dynamic_lb) {
1459 tx_slave = tlb_choose_channel(bond,
1460 hash_index & 0xFF,
1461 skb->len);
1462 } else {
Mahesh Bandewar6b794c12014-07-16 11:10:36 -07001463 struct tlb_up_slave *slaves;
Mahesh Bandeware9f0fb82014-04-22 16:30:22 -07001464
Mahesh Bandewar6b794c12014-07-16 11:10:36 -07001465 slaves = rcu_dereference(bond_info->slave_arr);
1466 if (slaves && slaves->count)
1467 tx_slave = slaves->arr[hash_index %
1468 slaves->count];
Mahesh Bandeware9f0fb82014-04-22 16:30:22 -07001469 }
Mahesh Bandewarf05b42e2014-04-22 16:30:20 -07001470 break;
1471 }
1472 }
Mahesh Bandewarf05b42e2014-04-22 16:30:20 -07001473 return bond_do_alb_xmit(skb, bond, tx_slave);
1474}
1475
Mahesh Bandewar9a49aba2014-04-22 16:30:18 -07001476int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1477{
1478 struct bonding *bond = netdev_priv(bond_dev);
1479 struct ethhdr *eth_data;
1480 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1481 struct slave *tx_slave = NULL;
1482 static const __be32 ip_bcast = htonl(0xffffffff);
1483 int hash_size = 0;
1484 bool do_tx_balance = true;
1485 u32 hash_index = 0;
1486 const u8 *hash_start = NULL;
1487 struct ipv6hdr *ip6hdr;
1488
1489 skb_reset_mac_header(skb);
1490 eth_data = eth_hdr(skb);
1491
1492 switch (ntohs(skb->protocol)) {
1493 case ETH_P_IP: {
1494 const struct iphdr *iph = ip_hdr(skb);
1495
1496 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast) ||
1497 (iph->daddr == ip_bcast) ||
1498 (iph->protocol == IPPROTO_IGMP)) {
1499 do_tx_balance = false;
1500 break;
1501 }
1502 hash_start = (char *)&(iph->daddr);
1503 hash_size = sizeof(iph->daddr);
1504 }
1505 break;
1506 case ETH_P_IPV6:
1507 /* IPv6 doesn't really use broadcast mac address, but leave
1508 * that here just in case.
1509 */
1510 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast)) {
1511 do_tx_balance = false;
1512 break;
1513 }
1514
1515 /* IPv6 uses all-nodes multicast as an equivalent to
1516 * broadcasts in IPv4.
1517 */
1518 if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1519 do_tx_balance = false;
1520 break;
1521 }
1522
1523 /* Additianally, DAD probes should not be tx-balanced as that
1524 * will lead to false positives for duplicate addresses and
1525 * prevent address configuration from working.
1526 */
1527 ip6hdr = ipv6_hdr(skb);
1528 if (ipv6_addr_any(&ip6hdr->saddr)) {
1529 do_tx_balance = false;
1530 break;
1531 }
1532
1533 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1534 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1535 break;
1536 case ETH_P_IPX:
1537 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1538 /* something is wrong with this packet */
1539 do_tx_balance = false;
1540 break;
1541 }
1542
1543 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1544 /* The only protocol worth balancing in
1545 * this family since it has an "ARP" like
1546 * mechanism
1547 */
1548 do_tx_balance = false;
1549 break;
1550 }
1551
1552 hash_start = (char *)eth_data->h_dest;
1553 hash_size = ETH_ALEN;
1554 break;
1555 case ETH_P_ARP:
1556 do_tx_balance = false;
1557 if (bond_info->rlb_enabled)
1558 tx_slave = rlb_arp_xmit(skb, bond);
1559 break;
1560 default:
1561 do_tx_balance = false;
1562 break;
1563 }
1564
1565 if (do_tx_balance) {
1566 hash_index = _simple_hash(hash_start, hash_size);
1567 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1568 }
1569
1570 return bond_do_alb_xmit(skb, bond, tx_slave);
1571}
1572
Jay Vosburgh1b76b312007-10-17 17:37:45 -07001573void bond_alb_monitor(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574{
Jay Vosburgh1b76b312007-10-17 17:37:45 -07001575 struct bonding *bond = container_of(work, struct bonding,
1576 alb_work.work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
Veaceslav Falico9caff1e72013-09-25 09:20:14 +02001578 struct list_head *iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 struct slave *slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001581 if (!bond_has_slaves(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 bond_info->tx_rebalance_counter = 0;
1583 bond_info->lp_counter = 0;
1584 goto re_arm;
1585 }
1586
dingtianhong733ab632013-12-13 10:19:45 +08001587 rcu_read_lock();
1588
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 bond_info->tx_rebalance_counter++;
1590 bond_info->lp_counter++;
1591
1592 /* send learning packets */
Neil Horman7eacd032013-09-13 11:05:33 -04001593 if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
Vlad Yasevichd0c21d42014-05-21 13:19:48 -04001594 bool strict_match;
1595
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 /* change of curr_active_slave involves swapping of mac addresses.
1597 * in order to avoid this swapping from happening while
1598 * sending the learning packets, the curr_slave_lock must be held for
1599 * read.
1600 */
1601 read_lock(&bond->curr_slave_lock);
1602
Vlad Yasevichd0c21d42014-05-21 13:19:48 -04001603 bond_for_each_slave_rcu(bond, slave, iter) {
1604 /* If updating current_active, use all currently
1605 * user mac addreses (!strict_match). Otherwise, only
1606 * use mac of the slave device.
Vlad Yasevich14af9962014-06-04 16:23:38 -04001607 * In RLB mode, we always use strict matches.
Vlad Yasevichd0c21d42014-05-21 13:19:48 -04001608 */
Eric Dumazet4740d632014-07-15 06:56:55 -07001609 strict_match = (slave != rcu_access_pointer(bond->curr_active_slave) ||
Vlad Yasevich14af9962014-06-04 16:23:38 -04001610 bond_info->rlb_enabled);
Vlad Yasevichd0c21d42014-05-21 13:19:48 -04001611 alb_send_learning_packets(slave, slave->dev->dev_addr,
1612 strict_match);
1613 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614
1615 read_unlock(&bond->curr_slave_lock);
1616
1617 bond_info->lp_counter = 0;
1618 }
1619
1620 /* rebalance tx traffic */
1621 if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1622
1623 read_lock(&bond->curr_slave_lock);
1624
dingtianhong733ab632013-12-13 10:19:45 +08001625 bond_for_each_slave_rcu(bond, slave, iter) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 tlb_clear_slave(bond, slave, 1);
Eric Dumazet4740d632014-07-15 06:56:55 -07001627 if (slave == rcu_access_pointer(bond->curr_active_slave)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 SLAVE_TLB_INFO(slave).load =
1629 bond_info->unbalanced_load /
1630 BOND_TLB_REBALANCE_INTERVAL;
1631 bond_info->unbalanced_load = 0;
1632 }
1633 }
1634
1635 read_unlock(&bond->curr_slave_lock);
1636
1637 bond_info->tx_rebalance_counter = 0;
1638 }
1639
1640 /* handle rlb stuff */
1641 if (bond_info->rlb_enabled) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 if (bond_info->primary_is_promisc &&
1643 (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1644
David S. Miller1f2cd842013-10-28 00:11:22 -04001645 /*
1646 * dev_set_promiscuity requires rtnl and
1647 * nothing else. Avoid race with bond_close.
1648 */
dingtianhong733ab632013-12-13 10:19:45 +08001649 rcu_read_unlock();
1650 if (!rtnl_trylock())
David S. Miller1f2cd842013-10-28 00:11:22 -04001651 goto re_arm;
David S. Miller1f2cd842013-10-28 00:11:22 -04001652
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 bond_info->rlb_promisc_timeout_counter = 0;
1654
1655 /* If the primary was set to promiscuous mode
1656 * because a slave was disabled then
1657 * it can now leave promiscuous mode.
1658 */
Eric Dumazet4740d632014-07-15 06:56:55 -07001659 dev_set_promiscuity(rtnl_dereference(bond->curr_active_slave)->dev,
1660 -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 bond_info->primary_is_promisc = 0;
David S. Miller1f2cd842013-10-28 00:11:22 -04001662
1663 rtnl_unlock();
dingtianhong733ab632013-12-13 10:19:45 +08001664 rcu_read_lock();
Jay Vosburghd0e81b72007-10-17 17:37:51 -07001665 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666
1667 if (bond_info->rlb_rebalance) {
1668 bond_info->rlb_rebalance = 0;
1669 rlb_rebalance(bond);
1670 }
1671
1672 /* check if clients need updating */
1673 if (bond_info->rx_ntt) {
1674 if (bond_info->rlb_update_delay_counter) {
1675 --bond_info->rlb_update_delay_counter;
1676 } else {
1677 rlb_update_rx_clients(bond);
Wang Yufendda0fd52014-02-14 17:15:16 +08001678 if (bond_info->rlb_update_retry_counter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 --bond_info->rlb_update_retry_counter;
Wang Yufendda0fd52014-02-14 17:15:16 +08001680 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 bond_info->rx_ntt = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 }
1683 }
1684 }
dingtianhong733ab632013-12-13 10:19:45 +08001685 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686re_arm:
Jay Vosburghe6d265e2011-10-28 15:42:50 +00001687 queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688}
1689
1690/* assumption: called before the slave is attached to the bond
1691 * and not locked by the bond lock
1692 */
1693int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1694{
1695 int res;
1696
Jiri Bohacb9245512012-01-18 12:24:54 +00001697 res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr);
Wang Yufendda0fd52014-02-14 17:15:16 +08001698 if (res)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 res = alb_handle_addr_collision_on_attach(bond, slave);
Wang Yufendda0fd52014-02-14 17:15:16 +08001702 if (res)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704
1705 tlb_init_slave(slave);
1706
1707 /* order a rebalance ASAP */
1708 bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1709
Wang Yufendda0fd52014-02-14 17:15:16 +08001710 if (bond->alb_info.rlb_enabled)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 bond->alb_info.rlb_rebalance = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712
1713 return 0;
1714}
1715
Jay Vosburgh25433312008-01-17 16:24:59 -08001716/*
1717 * Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1718 * if necessary.
1719 *
1720 * Caller must hold RTNL and no other locks
1721 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1723{
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001724 if (bond_has_slaves(bond))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 alb_change_hw_addr_on_detach(bond, slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726
1727 tlb_clear_slave(bond, slave, 0);
1728
1729 if (bond->alb_info.rlb_enabled) {
Veaceslav Falico6475ae42013-09-25 09:20:17 +02001730 bond->alb_info.rx_slave = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 rlb_clear_slave(bond, slave);
1732 }
Mahesh Bandewar6b794c12014-07-16 11:10:36 -07001733
1734 if (bond_is_nondyn_tlb(bond))
1735 if (bond_tlb_update_slave_arr(bond, slave))
1736 pr_err("Failed to build slave-array for TLB mode.\n");
1737
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738}
1739
1740/* Caller must hold bond lock for read */
1741void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1742{
1743 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1744
1745 if (link == BOND_LINK_DOWN) {
1746 tlb_clear_slave(bond, slave, 0);
Wang Yufen73ac0cd2014-02-14 17:15:17 +08001747 if (bond->alb_info.rlb_enabled)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 rlb_clear_slave(bond, slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 } else if (link == BOND_LINK_UP) {
1750 /* order a rebalance ASAP */
1751 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1752 if (bond->alb_info.rlb_enabled) {
1753 bond->alb_info.rlb_rebalance = 1;
1754 /* If the updelay module parameter is smaller than the
1755 * forwarding delay of the switch the rebalance will
1756 * not work because the rebalance arp replies will
1757 * not be forwarded to the clients..
1758 */
1759 }
1760 }
Mahesh Bandewar6b794c12014-07-16 11:10:36 -07001761
1762 if (bond_is_nondyn_tlb(bond)) {
1763 if (bond_tlb_update_slave_arr(bond, NULL))
1764 pr_err("Failed to build slave-array for TLB mode.\n");
1765 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766}
1767
1768/**
1769 * bond_alb_handle_active_change - assign new curr_active_slave
1770 * @bond: our bonding struct
1771 * @new_slave: new slave to assign
1772 *
1773 * Set the bond->curr_active_slave to @new_slave and handle
1774 * mac address swapping and promiscuity changes as needed.
1775 *
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001776 * If new_slave is NULL, caller must hold curr_slave_lock or
1777 * bond->lock for write.
1778 *
dingtianhongb2e7ace2013-12-13 10:19:32 +08001779 * If new_slave is not NULL, caller must hold RTNL, curr_slave_lock
1780 * for write. Processing here may sleep, so no other locks may be held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 */
1782void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
Hannes Eder1f78d9f2009-02-14 11:15:33 +00001783 __releases(&bond->curr_slave_lock)
Hannes Eder1f78d9f2009-02-14 11:15:33 +00001784 __acquires(&bond->curr_slave_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785{
1786 struct slave *swap_slave;
Eric Dumazet4740d632014-07-15 06:56:55 -07001787 struct slave *curr_active;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788
Eric Dumazet4740d632014-07-15 06:56:55 -07001789 curr_active = rcu_dereference_protected(bond->curr_active_slave,
1790 !new_slave ||
1791 lockdep_is_held(&bond->curr_slave_lock));
1792 if (curr_active == new_slave)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794
Eric Dumazet4740d632014-07-15 06:56:55 -07001795 if (curr_active && bond->alb_info.primary_is_promisc) {
1796 dev_set_promiscuity(curr_active->dev, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 bond->alb_info.primary_is_promisc = 0;
1798 bond->alb_info.rlb_promisc_timeout_counter = 0;
1799 }
1800
Eric Dumazet4740d632014-07-15 06:56:55 -07001801 swap_slave = curr_active;
nikolay@redhat.com278b2082013-08-01 16:54:51 +02001802 rcu_assign_pointer(bond->curr_active_slave, new_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803
Veaceslav Falico0965a1f2013-09-25 09:20:21 +02001804 if (!new_slave || !bond_has_slaves(bond))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806
1807 /* set the new curr_active_slave to the bonds mac address
1808 * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1809 */
Veaceslav Falicob88ec382013-06-18 13:44:52 +02001810 if (!swap_slave)
1811 swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001813 /*
1814 * Arrange for swap_slave and new_slave to temporarily be
1815 * ignored so we can mess with their MAC addresses without
1816 * fear of interference from transmit activity.
1817 */
nikolay@redhat.comdec1e902013-08-01 16:54:47 +02001818 if (swap_slave)
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001819 tlb_clear_slave(bond, swap_slave, 1);
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001820 tlb_clear_slave(bond, new_slave, 1);
1821
1822 write_unlock_bh(&bond->curr_slave_lock);
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001823
Jay Vosburghe0138a62008-01-17 16:24:58 -08001824 ASSERT_RTNL();
1825
Veaceslav Falico4996b902013-10-07 09:17:20 +02001826 /* in TLB mode, the slave might flip down/up with the old dev_addr,
1827 * and thus filter bond->dev_addr's packets, so force bond's mac
1828 */
Veaceslav Falico01844092014-05-15 21:39:55 +02001829 if (BOND_MODE(bond) == BOND_MODE_TLB) {
Veaceslav Falico4996b902013-10-07 09:17:20 +02001830 struct sockaddr sa;
1831 u8 tmp_addr[ETH_ALEN];
1832
Joe Perchesada0f862014-02-15 16:02:17 -08001833 ether_addr_copy(tmp_addr, new_slave->dev->dev_addr);
Veaceslav Falico4996b902013-10-07 09:17:20 +02001834
1835 memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1836 sa.sa_family = bond->dev->type;
1837 /* we don't care if it can't change its mac, best effort */
1838 dev_set_mac_address(new_slave->dev, &sa);
1839
Joe Perchesada0f862014-02-15 16:02:17 -08001840 ether_addr_copy(new_slave->dev->dev_addr, tmp_addr);
Veaceslav Falico4996b902013-10-07 09:17:20 +02001841 }
1842
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 /* curr_active_slave must be set before calling alb_swap_mac_addr */
1844 if (swap_slave) {
1845 /* swap mac address */
Veaceslav Falico43547ea2013-05-27 23:14:51 +00001846 alb_swap_mac_addr(swap_slave, new_slave);
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001847 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1848 } else {
Veaceslav Falicob88ec382013-06-18 13:44:52 +02001849 /* set the new_slave to the bond mac address */
1850 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr);
Vlad Yasevichd0c21d42014-05-21 13:19:48 -04001851 alb_send_learning_packets(new_slave, bond->dev->dev_addr,
1852 false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 }
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001854
1855 write_lock_bh(&bond->curr_slave_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856}
1857
Jay Vosburgh059fe7a2007-10-17 17:37:49 -07001858/*
1859 * Called with RTNL
1860 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
Hannes Eder1f78d9f2009-02-14 11:15:33 +00001862 __acquires(&bond->lock)
Jay Vosburgh815bcc22009-05-04 09:03:37 +00001863 __releases(&bond->lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864{
Wang Chen454d7c92008-11-12 23:37:49 -08001865 struct bonding *bond = netdev_priv(bond_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 struct sockaddr *sa = addr;
Eric Dumazet4740d632014-07-15 06:56:55 -07001867 struct slave *curr_active;
Veaceslav Falicob88ec382013-06-18 13:44:52 +02001868 struct slave *swap_slave;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870
Wang Yufen73ac0cd2014-02-14 17:15:17 +08001871 if (!is_valid_ether_addr(sa->sa_data))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 return -EADDRNOTAVAIL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873
1874 res = alb_set_mac_address(bond, addr);
Wang Yufen73ac0cd2014-02-14 17:15:17 +08001875 if (res)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877
1878 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1879
1880 /* If there is no curr_active_slave there is nothing else to do.
1881 * Otherwise we'll need to pass the new address to it and handle
1882 * duplications.
1883 */
Eric Dumazet4740d632014-07-15 06:56:55 -07001884 curr_active = rtnl_dereference(bond->curr_active_slave);
1885 if (!curr_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887
Veaceslav Falicob88ec382013-06-18 13:44:52 +02001888 swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889
1890 if (swap_slave) {
Eric Dumazet4740d632014-07-15 06:56:55 -07001891 alb_swap_mac_addr(swap_slave, curr_active);
1892 alb_fasten_mac_swap(bond, swap_slave, curr_active);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 } else {
Eric Dumazet4740d632014-07-15 06:56:55 -07001894 alb_set_slave_mac_addr(curr_active, bond_dev->dev_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895
Jay Vosburgh815bcc22009-05-04 09:03:37 +00001896 read_lock(&bond->lock);
Eric Dumazet4740d632014-07-15 06:56:55 -07001897 alb_send_learning_packets(curr_active,
Vlad Yasevichd0c21d42014-05-21 13:19:48 -04001898 bond_dev->dev_addr, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 if (bond->alb_info.rlb_enabled) {
1900 /* inform clients mac address has changed */
Eric Dumazet4740d632014-07-15 06:56:55 -07001901 rlb_req_update_slave_clients(bond, curr_active);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 }
Jay Vosburgh815bcc22009-05-04 09:03:37 +00001903 read_unlock(&bond->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 }
1905
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 return 0;
1907}
1908
1909void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1910{
Wang Yufen73ac0cd2014-02-14 17:15:17 +08001911 if (bond->alb_info.rlb_enabled)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 rlb_clear_vlan(bond, vlan_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913}
1914