Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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 | |
15 | * with this program; if not, write to the Free Software Foundation, Inc., | |
16 | * 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | |
17 | * | |
18 | * The full GNU General Public License is included in this distribution in the | |
19 | * file called LICENSE. | |
20 | * | |
1da177e4 LT |
21 | */ |
22 | ||
a4aee5c8 JP |
23 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
24 | ||
1da177e4 LT |
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> | |
2d1ea19d | 41 | #include <net/ipv6.h> |
1da177e4 LT |
42 | #include <asm/byteorder.h> |
43 | #include "bonding.h" | |
44 | #include "bond_alb.h" | |
45 | ||
46 | ||
1da177e4 | 47 | |
885a136c ED |
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 | |
53 | }; | |
54 | static const u8 mac_v6_allmcast[ETH_ALEN] __long_aligned = { | |
55 | 0x33, 0x33, 0x00, 0x00, 0x00, 0x01 | |
56 | }; | |
1da177e4 LT |
57 | static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC; |
58 | ||
59 | #pragma pack(1) | |
60 | struct learning_pkt { | |
61 | u8 mac_dst[ETH_ALEN]; | |
62 | u8 mac_src[ETH_ALEN]; | |
d3bb52b0 | 63 | __be16 type; |
1da177e4 LT |
64 | u8 padding[ETH_ZLEN - ETH_HLEN]; |
65 | }; | |
66 | ||
67 | struct arp_pkt { | |
d3bb52b0 AV |
68 | __be16 hw_addr_space; |
69 | __be16 prot_addr_space; | |
1da177e4 LT |
70 | u8 hw_addr_len; |
71 | u8 prot_addr_len; | |
d3bb52b0 | 72 | __be16 op_code; |
1da177e4 | 73 | u8 mac_src[ETH_ALEN]; /* sender hardware address */ |
d3bb52b0 | 74 | __be32 ip_src; /* sender IP address */ |
1da177e4 | 75 | u8 mac_dst[ETH_ALEN]; /* target hardware address */ |
d3bb52b0 | 76 | __be32 ip_dst; /* target IP address */ |
1da177e4 LT |
77 | }; |
78 | #pragma pack() | |
79 | ||
a16aeb36 ACM |
80 | static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb) |
81 | { | |
d56f90a7 | 82 | return (struct arp_pkt *)skb_network_header(skb); |
a16aeb36 ACM |
83 | } |
84 | ||
1da177e4 LT |
85 | /* Forward declaration */ |
86 | static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[]); | |
e53665c6 JB |
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); | |
1da177e4 | 91 | |
eddc9ec5 | 92 | static inline u8 _simple_hash(const u8 *hash_start, int hash_size) |
1da177e4 LT |
93 | { |
94 | int i; | |
95 | u8 hash = 0; | |
96 | ||
97 | for (i = 0; i < hash_size; i++) { | |
98 | hash ^= hash_start[i]; | |
99 | } | |
100 | ||
101 | return hash; | |
102 | } | |
103 | ||
104 | /*********************** tlb specific functions ***************************/ | |
105 | ||
f515e6b7 | 106 | static inline void _lock_tx_hashtbl_bh(struct bonding *bond) |
1da177e4 | 107 | { |
6603a6f2 | 108 | spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock)); |
1da177e4 LT |
109 | } |
110 | ||
f515e6b7 | 111 | static inline void _unlock_tx_hashtbl_bh(struct bonding *bond) |
1da177e4 | 112 | { |
6603a6f2 | 113 | spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock)); |
1da177e4 LT |
114 | } |
115 | ||
f515e6b7 MU |
116 | static inline void _lock_tx_hashtbl(struct bonding *bond) |
117 | { | |
118 | spin_lock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock)); | |
119 | } | |
120 | ||
121 | static inline void _unlock_tx_hashtbl(struct bonding *bond) | |
122 | { | |
123 | spin_unlock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock)); | |
124 | } | |
125 | ||
1da177e4 LT |
126 | /* Caller must hold tx_hashtbl lock */ |
127 | static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load) | |
128 | { | |
129 | if (save_load) { | |
130 | entry->load_history = 1 + entry->tx_bytes / | |
131 | BOND_TLB_REBALANCE_INTERVAL; | |
132 | entry->tx_bytes = 0; | |
133 | } | |
134 | ||
135 | entry->tx_slave = NULL; | |
136 | entry->next = TLB_NULL_INDEX; | |
137 | entry->prev = TLB_NULL_INDEX; | |
138 | } | |
139 | ||
140 | static inline void tlb_init_slave(struct slave *slave) | |
141 | { | |
142 | SLAVE_TLB_INFO(slave).load = 0; | |
143 | SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX; | |
144 | } | |
145 | ||
f515e6b7 MU |
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) | |
1da177e4 LT |
149 | { |
150 | struct tlb_client_info *tx_hash_table; | |
151 | u32 index; | |
152 | ||
1da177e4 LT |
153 | /* clear slave from tx_hashtbl */ |
154 | tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl; | |
155 | ||
ce39a800 AG |
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; | |
163 | } | |
1da177e4 LT |
164 | } |
165 | ||
1da177e4 | 166 | tlb_init_slave(slave); |
f515e6b7 | 167 | } |
5af47b2f | 168 | |
f515e6b7 MU |
169 | /* Caller must hold bond lock for read */ |
170 | static void tlb_clear_slave(struct bonding *bond, struct slave *slave, | |
171 | int save_load) | |
172 | { | |
173 | _lock_tx_hashtbl_bh(bond); | |
174 | __tlb_clear_slave(bond, slave, save_load); | |
175 | _unlock_tx_hashtbl_bh(bond); | |
1da177e4 LT |
176 | } |
177 | ||
178 | /* Must be called before starting the monitor timer */ | |
179 | static int tlb_initialize(struct bonding *bond) | |
180 | { | |
181 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
182 | int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info); | |
0d206a3a | 183 | struct tlb_client_info *new_hashtbl; |
1da177e4 LT |
184 | int i; |
185 | ||
243cb4e5 | 186 | new_hashtbl = kzalloc(size, GFP_KERNEL); |
e404decb | 187 | if (!new_hashtbl) |
1da177e4 | 188 | return -1; |
e404decb | 189 | |
f515e6b7 | 190 | _lock_tx_hashtbl_bh(bond); |
0d206a3a MW |
191 | |
192 | bond_info->tx_hashtbl = new_hashtbl; | |
1da177e4 | 193 | |
1da177e4 | 194 | for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) { |
38dbaf0a | 195 | tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0); |
1da177e4 LT |
196 | } |
197 | ||
f515e6b7 | 198 | _unlock_tx_hashtbl_bh(bond); |
1da177e4 LT |
199 | |
200 | return 0; | |
201 | } | |
202 | ||
203 | /* Must be called only after all slaves have been released */ | |
204 | static void tlb_deinitialize(struct bonding *bond) | |
205 | { | |
206 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
207 | ||
f515e6b7 | 208 | _lock_tx_hashtbl_bh(bond); |
1da177e4 LT |
209 | |
210 | kfree(bond_info->tx_hashtbl); | |
211 | bond_info->tx_hashtbl = NULL; | |
212 | ||
f515e6b7 | 213 | _unlock_tx_hashtbl_bh(bond); |
1da177e4 LT |
214 | } |
215 | ||
097811bb JP |
216 | static long long compute_gap(struct slave *slave) |
217 | { | |
218 | return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */ | |
219 | (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */ | |
220 | } | |
221 | ||
1da177e4 LT |
222 | /* Caller must hold bond lock for read */ |
223 | static struct slave *tlb_get_least_loaded_slave(struct bonding *bond) | |
224 | { | |
225 | struct slave *slave, *least_loaded; | |
097811bb JP |
226 | long long max_gap; |
227 | int i; | |
1da177e4 | 228 | |
097811bb JP |
229 | least_loaded = NULL; |
230 | max_gap = LLONG_MIN; | |
1da177e4 LT |
231 | |
232 | /* Find the slave with the largest gap */ | |
097811bb | 233 | bond_for_each_slave(bond, slave, i) { |
1da177e4 | 234 | if (SLAVE_IS_OK(slave)) { |
097811bb JP |
235 | long long gap = compute_gap(slave); |
236 | ||
1da177e4 LT |
237 | if (max_gap < gap) { |
238 | least_loaded = slave; | |
239 | max_gap = gap; | |
240 | } | |
241 | } | |
242 | } | |
243 | ||
244 | return least_loaded; | |
245 | } | |
246 | ||
f515e6b7 MU |
247 | static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index, |
248 | u32 skb_len) | |
1da177e4 LT |
249 | { |
250 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
251 | struct tlb_client_info *hash_table; | |
252 | struct slave *assigned_slave; | |
253 | ||
1da177e4 LT |
254 | hash_table = bond_info->tx_hashtbl; |
255 | assigned_slave = hash_table[hash_index].tx_slave; | |
256 | if (!assigned_slave) { | |
257 | assigned_slave = tlb_get_least_loaded_slave(bond); | |
258 | ||
259 | if (assigned_slave) { | |
260 | struct tlb_slave_info *slave_info = | |
261 | &(SLAVE_TLB_INFO(assigned_slave)); | |
262 | u32 next_index = slave_info->head; | |
263 | ||
264 | hash_table[hash_index].tx_slave = assigned_slave; | |
265 | hash_table[hash_index].next = next_index; | |
266 | hash_table[hash_index].prev = TLB_NULL_INDEX; | |
267 | ||
268 | if (next_index != TLB_NULL_INDEX) { | |
269 | hash_table[next_index].prev = hash_index; | |
270 | } | |
271 | ||
272 | slave_info->head = hash_index; | |
273 | slave_info->load += | |
274 | hash_table[hash_index].load_history; | |
275 | } | |
276 | } | |
277 | ||
278 | if (assigned_slave) { | |
279 | hash_table[hash_index].tx_bytes += skb_len; | |
280 | } | |
281 | ||
1da177e4 LT |
282 | return assigned_slave; |
283 | } | |
284 | ||
f515e6b7 MU |
285 | /* Caller must hold bond lock for read */ |
286 | static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index, | |
287 | u32 skb_len) | |
288 | { | |
289 | struct slave *tx_slave; | |
290 | /* | |
291 | * We don't need to disable softirq here, becase | |
292 | * tlb_choose_channel() is only called by bond_alb_xmit() | |
293 | * which already has softirq disabled. | |
294 | */ | |
295 | _lock_tx_hashtbl(bond); | |
296 | tx_slave = __tlb_choose_channel(bond, hash_index, skb_len); | |
297 | _unlock_tx_hashtbl(bond); | |
298 | return tx_slave; | |
299 | } | |
300 | ||
1da177e4 | 301 | /*********************** rlb specific functions ***************************/ |
f515e6b7 | 302 | static inline void _lock_rx_hashtbl_bh(struct bonding *bond) |
1da177e4 | 303 | { |
6603a6f2 | 304 | spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock)); |
1da177e4 LT |
305 | } |
306 | ||
f515e6b7 | 307 | static inline void _unlock_rx_hashtbl_bh(struct bonding *bond) |
1da177e4 | 308 | { |
6603a6f2 | 309 | spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock)); |
1da177e4 LT |
310 | } |
311 | ||
f515e6b7 MU |
312 | static inline void _lock_rx_hashtbl(struct bonding *bond) |
313 | { | |
314 | spin_lock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock)); | |
315 | } | |
316 | ||
317 | static inline void _unlock_rx_hashtbl(struct bonding *bond) | |
318 | { | |
319 | spin_unlock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock)); | |
320 | } | |
321 | ||
1da177e4 LT |
322 | /* when an ARP REPLY is received from a client update its info |
323 | * in the rx_hashtbl | |
324 | */ | |
325 | static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp) | |
326 | { | |
327 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
328 | struct rlb_client_info *client_info; | |
329 | u32 hash_index; | |
330 | ||
f515e6b7 | 331 | _lock_rx_hashtbl_bh(bond); |
1da177e4 LT |
332 | |
333 | hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src)); | |
334 | client_info = &(bond_info->rx_hashtbl[hash_index]); | |
335 | ||
336 | if ((client_info->assigned) && | |
337 | (client_info->ip_src == arp->ip_dst) && | |
42d782ac | 338 | (client_info->ip_dst == arp->ip_src) && |
a6700db1 | 339 | (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) { |
1da177e4 LT |
340 | /* update the clients MAC address */ |
341 | memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN); | |
342 | client_info->ntt = 1; | |
343 | bond_info->rx_ntt = 1; | |
344 | } | |
345 | ||
f515e6b7 | 346 | _unlock_rx_hashtbl_bh(bond); |
1da177e4 LT |
347 | } |
348 | ||
de063b70 ED |
349 | static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond, |
350 | struct slave *slave) | |
1da177e4 | 351 | { |
de063b70 | 352 | struct arp_pkt *arp, _arp; |
1da177e4 | 353 | |
3aba891d | 354 | if (skb->protocol != cpu_to_be16(ETH_P_ARP)) |
b99215cd | 355 | goto out; |
1da177e4 | 356 | |
de063b70 ED |
357 | arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp); |
358 | if (!arp) | |
b99215cd | 359 | goto out; |
1da177e4 | 360 | |
e53665c6 JB |
361 | /* We received an ARP from arp->ip_src. |
362 | * We might have used this IP address previously (on the bonding host | |
363 | * itself or on a system that is bridged together with the bond). | |
364 | * However, if arp->mac_src is different than what is stored in | |
365 | * rx_hashtbl, some other host is now using the IP and we must prevent | |
366 | * sending out client updates with this IP address and the old MAC | |
367 | * address. | |
368 | * Clean up all hash table entries that have this address as ip_src but | |
369 | * have a different mac_src. | |
370 | */ | |
371 | rlb_purge_src_ip(bond, arp); | |
372 | ||
1da177e4 LT |
373 | if (arp->op_code == htons(ARPOP_REPLY)) { |
374 | /* update rx hash table for this ARP */ | |
375 | rlb_update_entry_from_arp(bond, arp); | |
5a03cdb7 | 376 | pr_debug("Server received an ARP Reply from client\n"); |
1da177e4 | 377 | } |
b99215cd DM |
378 | out: |
379 | return RX_HANDLER_ANOTHER; | |
1da177e4 LT |
380 | } |
381 | ||
382 | /* Caller must hold bond lock for read */ | |
383 | static struct slave *rlb_next_rx_slave(struct bonding *bond) | |
384 | { | |
385 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
386 | struct slave *rx_slave, *slave, *start_at; | |
387 | int i = 0; | |
388 | ||
389 | if (bond_info->next_rx_slave) { | |
390 | start_at = bond_info->next_rx_slave; | |
391 | } else { | |
392 | start_at = bond->first_slave; | |
393 | } | |
394 | ||
395 | rx_slave = NULL; | |
396 | ||
397 | bond_for_each_slave_from(bond, slave, i, start_at) { | |
398 | if (SLAVE_IS_OK(slave)) { | |
399 | if (!rx_slave) { | |
400 | rx_slave = slave; | |
401 | } else if (slave->speed > rx_slave->speed) { | |
402 | rx_slave = slave; | |
403 | } | |
404 | } | |
405 | } | |
406 | ||
407 | if (rx_slave) { | |
408 | bond_info->next_rx_slave = rx_slave->next; | |
409 | } | |
410 | ||
411 | return rx_slave; | |
412 | } | |
413 | ||
414 | /* teach the switch the mac of a disabled slave | |
415 | * on the primary for fault tolerance | |
416 | * | |
417 | * Caller must hold bond->curr_slave_lock for write or bond lock for write | |
418 | */ | |
419 | static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[]) | |
420 | { | |
421 | if (!bond->curr_active_slave) { | |
422 | return; | |
423 | } | |
424 | ||
425 | if (!bond->alb_info.primary_is_promisc) { | |
7e1a1ac1 WC |
426 | if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1)) |
427 | bond->alb_info.primary_is_promisc = 1; | |
428 | else | |
429 | bond->alb_info.primary_is_promisc = 0; | |
1da177e4 LT |
430 | } |
431 | ||
432 | bond->alb_info.rlb_promisc_timeout_counter = 0; | |
433 | ||
434 | alb_send_learning_packets(bond->curr_active_slave, addr); | |
435 | } | |
436 | ||
437 | /* slave being removed should not be active at this point | |
438 | * | |
439 | * Caller must hold bond lock for read | |
440 | */ | |
441 | static void rlb_clear_slave(struct bonding *bond, struct slave *slave) | |
442 | { | |
443 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
444 | struct rlb_client_info *rx_hash_table; | |
445 | u32 index, next_index; | |
446 | ||
447 | /* clear slave from rx_hashtbl */ | |
f515e6b7 | 448 | _lock_rx_hashtbl_bh(bond); |
1da177e4 LT |
449 | |
450 | rx_hash_table = bond_info->rx_hashtbl; | |
e53665c6 | 451 | index = bond_info->rx_hashtbl_used_head; |
1da177e4 | 452 | for (; index != RLB_NULL_INDEX; index = next_index) { |
e53665c6 | 453 | next_index = rx_hash_table[index].used_next; |
1da177e4 LT |
454 | if (rx_hash_table[index].slave == slave) { |
455 | struct slave *assigned_slave = rlb_next_rx_slave(bond); | |
456 | ||
457 | if (assigned_slave) { | |
458 | rx_hash_table[index].slave = assigned_slave; | |
a6700db1 JP |
459 | if (!ether_addr_equal_64bits(rx_hash_table[index].mac_dst, |
460 | mac_bcast)) { | |
1da177e4 LT |
461 | bond_info->rx_hashtbl[index].ntt = 1; |
462 | bond_info->rx_ntt = 1; | |
463 | /* A slave has been removed from the | |
464 | * table because it is either disabled | |
465 | * or being released. We must retry the | |
466 | * update to avoid clients from not | |
467 | * being updated & disconnecting when | |
468 | * there is stress | |
469 | */ | |
470 | bond_info->rlb_update_retry_counter = | |
471 | RLB_UPDATE_RETRY; | |
472 | } | |
473 | } else { /* there is no active slave */ | |
474 | rx_hash_table[index].slave = NULL; | |
475 | } | |
476 | } | |
477 | } | |
478 | ||
f515e6b7 | 479 | _unlock_rx_hashtbl_bh(bond); |
1da177e4 | 480 | |
6603a6f2 | 481 | write_lock_bh(&bond->curr_slave_lock); |
1da177e4 LT |
482 | |
483 | if (slave != bond->curr_active_slave) { | |
484 | rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr); | |
485 | } | |
486 | ||
6603a6f2 | 487 | write_unlock_bh(&bond->curr_slave_lock); |
1da177e4 LT |
488 | } |
489 | ||
490 | static void rlb_update_client(struct rlb_client_info *client_info) | |
491 | { | |
492 | int i; | |
493 | ||
494 | if (!client_info->slave) { | |
495 | return; | |
496 | } | |
497 | ||
498 | for (i = 0; i < RLB_ARP_BURST_SIZE; i++) { | |
499 | struct sk_buff *skb; | |
500 | ||
501 | skb = arp_create(ARPOP_REPLY, ETH_P_ARP, | |
502 | client_info->ip_dst, | |
503 | client_info->slave->dev, | |
504 | client_info->ip_src, | |
505 | client_info->mac_dst, | |
506 | client_info->slave->dev->dev_addr, | |
507 | client_info->mac_dst); | |
508 | if (!skb) { | |
a4aee5c8 | 509 | pr_err("%s: Error: failed to create an ARP packet\n", |
4e0952c7 | 510 | client_info->slave->dev->master->name); |
1da177e4 LT |
511 | continue; |
512 | } | |
513 | ||
514 | skb->dev = client_info->slave->dev; | |
515 | ||
516 | if (client_info->tag) { | |
517 | skb = vlan_put_tag(skb, client_info->vlan_id); | |
518 | if (!skb) { | |
a4aee5c8 | 519 | pr_err("%s: Error: failed to insert VLAN tag\n", |
4e0952c7 | 520 | client_info->slave->dev->master->name); |
1da177e4 LT |
521 | continue; |
522 | } | |
523 | } | |
524 | ||
525 | arp_xmit(skb); | |
526 | } | |
527 | } | |
528 | ||
529 | /* sends ARP REPLIES that update the clients that need updating */ | |
530 | static void rlb_update_rx_clients(struct bonding *bond) | |
531 | { | |
532 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
533 | struct rlb_client_info *client_info; | |
534 | u32 hash_index; | |
535 | ||
f515e6b7 | 536 | _lock_rx_hashtbl_bh(bond); |
1da177e4 | 537 | |
e53665c6 JB |
538 | hash_index = bond_info->rx_hashtbl_used_head; |
539 | for (; hash_index != RLB_NULL_INDEX; | |
540 | hash_index = client_info->used_next) { | |
1da177e4 LT |
541 | client_info = &(bond_info->rx_hashtbl[hash_index]); |
542 | if (client_info->ntt) { | |
543 | rlb_update_client(client_info); | |
544 | if (bond_info->rlb_update_retry_counter == 0) { | |
545 | client_info->ntt = 0; | |
546 | } | |
547 | } | |
548 | } | |
549 | ||
94e2bd68 | 550 | /* do not update the entries again until this counter is zero so that |
1da177e4 LT |
551 | * not to confuse the clients. |
552 | */ | |
553 | bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY; | |
554 | ||
f515e6b7 | 555 | _unlock_rx_hashtbl_bh(bond); |
1da177e4 LT |
556 | } |
557 | ||
558 | /* The slave was assigned a new mac address - update the clients */ | |
559 | static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave) | |
560 | { | |
561 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
562 | struct rlb_client_info *client_info; | |
563 | int ntt = 0; | |
564 | u32 hash_index; | |
565 | ||
f515e6b7 | 566 | _lock_rx_hashtbl_bh(bond); |
1da177e4 | 567 | |
e53665c6 JB |
568 | hash_index = bond_info->rx_hashtbl_used_head; |
569 | for (; hash_index != RLB_NULL_INDEX; | |
570 | hash_index = client_info->used_next) { | |
1da177e4 LT |
571 | client_info = &(bond_info->rx_hashtbl[hash_index]); |
572 | ||
573 | if ((client_info->slave == slave) && | |
a6700db1 | 574 | !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) { |
1da177e4 LT |
575 | client_info->ntt = 1; |
576 | ntt = 1; | |
577 | } | |
578 | } | |
579 | ||
580 | // update the team's flag only after the whole iteration | |
581 | if (ntt) { | |
582 | bond_info->rx_ntt = 1; | |
583 | //fasten the change | |
584 | bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY; | |
585 | } | |
586 | ||
f515e6b7 | 587 | _unlock_rx_hashtbl_bh(bond); |
1da177e4 LT |
588 | } |
589 | ||
590 | /* mark all clients using src_ip to be updated */ | |
d3bb52b0 | 591 | static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip) |
1da177e4 LT |
592 | { |
593 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
594 | struct rlb_client_info *client_info; | |
595 | u32 hash_index; | |
596 | ||
597 | _lock_rx_hashtbl(bond); | |
598 | ||
e53665c6 JB |
599 | hash_index = bond_info->rx_hashtbl_used_head; |
600 | for (; hash_index != RLB_NULL_INDEX; | |
601 | hash_index = client_info->used_next) { | |
1da177e4 LT |
602 | client_info = &(bond_info->rx_hashtbl[hash_index]); |
603 | ||
604 | if (!client_info->slave) { | |
a4aee5c8 | 605 | pr_err("%s: Error: found a client with no channel in the client's hash table\n", |
4e0952c7 | 606 | bond->dev->name); |
1da177e4 LT |
607 | continue; |
608 | } | |
609 | /*update all clients using this src_ip, that are not assigned | |
610 | * to the team's address (curr_active_slave) and have a known | |
611 | * unicast mac address. | |
612 | */ | |
613 | if ((client_info->ip_src == src_ip) && | |
a6700db1 JP |
614 | !ether_addr_equal_64bits(client_info->slave->dev->dev_addr, |
615 | bond->dev->dev_addr) && | |
616 | !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) { | |
1da177e4 LT |
617 | client_info->ntt = 1; |
618 | bond_info->rx_ntt = 1; | |
619 | } | |
620 | } | |
621 | ||
622 | _unlock_rx_hashtbl(bond); | |
623 | } | |
624 | ||
625 | /* Caller must hold both bond and ptr locks for read */ | |
626 | static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond) | |
627 | { | |
628 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
a16aeb36 | 629 | struct arp_pkt *arp = arp_pkt(skb); |
1da177e4 LT |
630 | struct slave *assigned_slave; |
631 | struct rlb_client_info *client_info; | |
632 | u32 hash_index = 0; | |
633 | ||
634 | _lock_rx_hashtbl(bond); | |
635 | ||
e364a341 | 636 | hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst)); |
1da177e4 LT |
637 | client_info = &(bond_info->rx_hashtbl[hash_index]); |
638 | ||
639 | if (client_info->assigned) { | |
640 | if ((client_info->ip_src == arp->ip_src) && | |
641 | (client_info->ip_dst == arp->ip_dst)) { | |
642 | /* the entry is already assigned to this client */ | |
a6700db1 | 643 | if (!ether_addr_equal_64bits(arp->mac_dst, mac_bcast)) { |
1da177e4 LT |
644 | /* update mac address from arp */ |
645 | memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN); | |
646 | } | |
e53665c6 | 647 | memcpy(client_info->mac_src, arp->mac_src, ETH_ALEN); |
1da177e4 LT |
648 | |
649 | assigned_slave = client_info->slave; | |
650 | if (assigned_slave) { | |
651 | _unlock_rx_hashtbl(bond); | |
652 | return assigned_slave; | |
653 | } | |
654 | } else { | |
655 | /* the entry is already assigned to some other client, | |
656 | * move the old client to primary (curr_active_slave) so | |
657 | * that the new client can be assigned to this entry. | |
658 | */ | |
659 | if (bond->curr_active_slave && | |
660 | client_info->slave != bond->curr_active_slave) { | |
661 | client_info->slave = bond->curr_active_slave; | |
662 | rlb_update_client(client_info); | |
663 | } | |
664 | } | |
665 | } | |
666 | /* assign a new slave */ | |
667 | assigned_slave = rlb_next_rx_slave(bond); | |
668 | ||
669 | if (assigned_slave) { | |
e53665c6 JB |
670 | if (!(client_info->assigned && |
671 | client_info->ip_src == arp->ip_src)) { | |
672 | /* ip_src is going to be updated, | |
673 | * fix the src hash list | |
674 | */ | |
675 | u32 hash_src = _simple_hash((u8 *)&arp->ip_src, | |
676 | sizeof(arp->ip_src)); | |
677 | rlb_src_unlink(bond, hash_index); | |
678 | rlb_src_link(bond, hash_src, hash_index); | |
679 | } | |
680 | ||
1da177e4 LT |
681 | client_info->ip_src = arp->ip_src; |
682 | client_info->ip_dst = arp->ip_dst; | |
683 | /* arp->mac_dst is broadcast for arp reqeusts. | |
684 | * will be updated with clients actual unicast mac address | |
685 | * upon receiving an arp reply. | |
686 | */ | |
687 | memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN); | |
e53665c6 | 688 | memcpy(client_info->mac_src, arp->mac_src, ETH_ALEN); |
1da177e4 LT |
689 | client_info->slave = assigned_slave; |
690 | ||
a6700db1 | 691 | if (!ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) { |
1da177e4 LT |
692 | client_info->ntt = 1; |
693 | bond->alb_info.rx_ntt = 1; | |
694 | } else { | |
695 | client_info->ntt = 0; | |
696 | } | |
697 | ||
cc0e4070 | 698 | if (bond_vlan_used(bond)) { |
966bc6f4 | 699 | if (!vlan_get_tag(skb, &client_info->vlan_id)) |
1da177e4 | 700 | client_info->tag = 1; |
1da177e4 LT |
701 | } |
702 | ||
703 | if (!client_info->assigned) { | |
e53665c6 JB |
704 | u32 prev_tbl_head = bond_info->rx_hashtbl_used_head; |
705 | bond_info->rx_hashtbl_used_head = hash_index; | |
706 | client_info->used_next = prev_tbl_head; | |
1da177e4 | 707 | if (prev_tbl_head != RLB_NULL_INDEX) { |
e53665c6 | 708 | bond_info->rx_hashtbl[prev_tbl_head].used_prev = |
1da177e4 LT |
709 | hash_index; |
710 | } | |
711 | client_info->assigned = 1; | |
712 | } | |
713 | } | |
714 | ||
715 | _unlock_rx_hashtbl(bond); | |
716 | ||
717 | return assigned_slave; | |
718 | } | |
719 | ||
720 | /* chooses (and returns) transmit channel for arp reply | |
721 | * does not choose channel for other arp types since they are | |
722 | * sent on the curr_active_slave | |
723 | */ | |
724 | static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond) | |
725 | { | |
a16aeb36 | 726 | struct arp_pkt *arp = arp_pkt(skb); |
1da177e4 LT |
727 | struct slave *tx_slave = NULL; |
728 | ||
567b871e | 729 | /* Don't modify or load balance ARPs that do not originate locally |
730 | * (e.g.,arrive via a bridge). | |
731 | */ | |
732 | if (!bond_slave_has_mac(bond, arp->mac_src)) | |
733 | return NULL; | |
734 | ||
f14c4e4e | 735 | if (arp->op_code == htons(ARPOP_REPLY)) { |
1da177e4 LT |
736 | /* the arp must be sent on the selected |
737 | * rx channel | |
738 | */ | |
739 | tx_slave = rlb_choose_channel(skb, bond); | |
740 | if (tx_slave) { | |
741 | memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN); | |
742 | } | |
5a03cdb7 | 743 | pr_debug("Server sent ARP Reply packet\n"); |
f14c4e4e | 744 | } else if (arp->op_code == htons(ARPOP_REQUEST)) { |
1da177e4 LT |
745 | /* Create an entry in the rx_hashtbl for this client as a |
746 | * place holder. | |
747 | * When the arp reply is received the entry will be updated | |
748 | * with the correct unicast address of the client. | |
749 | */ | |
750 | rlb_choose_channel(skb, bond); | |
751 | ||
77c8e2c0 | 752 | /* The ARP reply packets must be delayed so that |
1da177e4 LT |
753 | * they can cancel out the influence of the ARP request. |
754 | */ | |
755 | bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY; | |
756 | ||
757 | /* arp requests are broadcast and are sent on the primary | |
758 | * the arp request will collapse all clients on the subnet to | |
759 | * the primary slave. We must register these clients to be | |
760 | * updated with their assigned mac. | |
761 | */ | |
762 | rlb_req_update_subnet_clients(bond, arp->ip_src); | |
5a03cdb7 | 763 | pr_debug("Server sent ARP Request packet\n"); |
1da177e4 LT |
764 | } |
765 | ||
766 | return tx_slave; | |
767 | } | |
768 | ||
769 | /* Caller must hold bond lock for read */ | |
770 | static void rlb_rebalance(struct bonding *bond) | |
771 | { | |
772 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
773 | struct slave *assigned_slave; | |
774 | struct rlb_client_info *client_info; | |
775 | int ntt; | |
776 | u32 hash_index; | |
777 | ||
f515e6b7 | 778 | _lock_rx_hashtbl_bh(bond); |
1da177e4 LT |
779 | |
780 | ntt = 0; | |
e53665c6 JB |
781 | hash_index = bond_info->rx_hashtbl_used_head; |
782 | for (; hash_index != RLB_NULL_INDEX; | |
783 | hash_index = client_info->used_next) { | |
1da177e4 LT |
784 | client_info = &(bond_info->rx_hashtbl[hash_index]); |
785 | assigned_slave = rlb_next_rx_slave(bond); | |
786 | if (assigned_slave && (client_info->slave != assigned_slave)) { | |
787 | client_info->slave = assigned_slave; | |
788 | client_info->ntt = 1; | |
789 | ntt = 1; | |
790 | } | |
791 | } | |
792 | ||
793 | /* update the team's flag only after the whole iteration */ | |
794 | if (ntt) { | |
795 | bond_info->rx_ntt = 1; | |
796 | } | |
f515e6b7 | 797 | _unlock_rx_hashtbl_bh(bond); |
1da177e4 LT |
798 | } |
799 | ||
800 | /* Caller must hold rx_hashtbl lock */ | |
e53665c6 JB |
801 | static void rlb_init_table_entry_dst(struct rlb_client_info *entry) |
802 | { | |
803 | entry->used_next = RLB_NULL_INDEX; | |
804 | entry->used_prev = RLB_NULL_INDEX; | |
805 | entry->assigned = 0; | |
806 | entry->slave = NULL; | |
807 | entry->tag = 0; | |
808 | } | |
809 | static void rlb_init_table_entry_src(struct rlb_client_info *entry) | |
810 | { | |
811 | entry->src_first = RLB_NULL_INDEX; | |
812 | entry->src_prev = RLB_NULL_INDEX; | |
813 | entry->src_next = RLB_NULL_INDEX; | |
814 | } | |
815 | ||
1da177e4 LT |
816 | static void rlb_init_table_entry(struct rlb_client_info *entry) |
817 | { | |
818 | memset(entry, 0, sizeof(struct rlb_client_info)); | |
e53665c6 JB |
819 | rlb_init_table_entry_dst(entry); |
820 | rlb_init_table_entry_src(entry); | |
821 | } | |
822 | ||
823 | static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index) | |
824 | { | |
825 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
826 | u32 next_index = bond_info->rx_hashtbl[index].used_next; | |
827 | u32 prev_index = bond_info->rx_hashtbl[index].used_prev; | |
828 | ||
829 | if (index == bond_info->rx_hashtbl_used_head) | |
830 | bond_info->rx_hashtbl_used_head = next_index; | |
831 | if (prev_index != RLB_NULL_INDEX) | |
832 | bond_info->rx_hashtbl[prev_index].used_next = next_index; | |
833 | if (next_index != RLB_NULL_INDEX) | |
834 | bond_info->rx_hashtbl[next_index].used_prev = prev_index; | |
835 | } | |
836 | ||
837 | /* unlink a rlb hash table entry from the src list */ | |
838 | static void rlb_src_unlink(struct bonding *bond, u32 index) | |
839 | { | |
840 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
841 | u32 next_index = bond_info->rx_hashtbl[index].src_next; | |
842 | u32 prev_index = bond_info->rx_hashtbl[index].src_prev; | |
843 | ||
844 | bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX; | |
845 | bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX; | |
846 | ||
847 | if (next_index != RLB_NULL_INDEX) | |
848 | bond_info->rx_hashtbl[next_index].src_prev = prev_index; | |
849 | ||
850 | if (prev_index == RLB_NULL_INDEX) | |
851 | return; | |
852 | ||
853 | /* is prev_index pointing to the head of this list? */ | |
854 | if (bond_info->rx_hashtbl[prev_index].src_first == index) | |
855 | bond_info->rx_hashtbl[prev_index].src_first = next_index; | |
856 | else | |
857 | bond_info->rx_hashtbl[prev_index].src_next = next_index; | |
858 | ||
859 | } | |
860 | ||
861 | static void rlb_delete_table_entry(struct bonding *bond, u32 index) | |
862 | { | |
863 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
864 | struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]); | |
865 | ||
866 | rlb_delete_table_entry_dst(bond, index); | |
867 | rlb_init_table_entry_dst(entry); | |
868 | ||
869 | rlb_src_unlink(bond, index); | |
870 | } | |
871 | ||
872 | /* add the rx_hashtbl[ip_dst_hash] entry to the list | |
873 | * of entries with identical ip_src_hash | |
874 | */ | |
875 | static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash) | |
876 | { | |
877 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
878 | u32 next; | |
879 | ||
880 | bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash; | |
881 | next = bond_info->rx_hashtbl[ip_src_hash].src_first; | |
882 | bond_info->rx_hashtbl[ip_dst_hash].src_next = next; | |
883 | if (next != RLB_NULL_INDEX) | |
884 | bond_info->rx_hashtbl[next].src_prev = ip_dst_hash; | |
885 | bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash; | |
886 | } | |
887 | ||
888 | /* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does | |
889 | * not match arp->mac_src */ | |
890 | static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp) | |
891 | { | |
892 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
893 | u32 ip_src_hash = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src)); | |
894 | u32 index; | |
895 | ||
896 | _lock_rx_hashtbl_bh(bond); | |
897 | ||
898 | index = bond_info->rx_hashtbl[ip_src_hash].src_first; | |
899 | while (index != RLB_NULL_INDEX) { | |
900 | struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]); | |
901 | u32 next_index = entry->src_next; | |
902 | if (entry->ip_src == arp->ip_src && | |
903 | !ether_addr_equal_64bits(arp->mac_src, entry->mac_src)) | |
904 | rlb_delete_table_entry(bond, index); | |
905 | index = next_index; | |
906 | } | |
907 | _unlock_rx_hashtbl_bh(bond); | |
1da177e4 LT |
908 | } |
909 | ||
910 | static int rlb_initialize(struct bonding *bond) | |
911 | { | |
912 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
0d206a3a | 913 | struct rlb_client_info *new_hashtbl; |
1da177e4 LT |
914 | int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info); |
915 | int i; | |
916 | ||
0d206a3a | 917 | new_hashtbl = kmalloc(size, GFP_KERNEL); |
e404decb | 918 | if (!new_hashtbl) |
1da177e4 | 919 | return -1; |
e404decb | 920 | |
f515e6b7 | 921 | _lock_rx_hashtbl_bh(bond); |
0d206a3a MW |
922 | |
923 | bond_info->rx_hashtbl = new_hashtbl; | |
1da177e4 | 924 | |
e53665c6 | 925 | bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX; |
1da177e4 LT |
926 | |
927 | for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) { | |
928 | rlb_init_table_entry(bond_info->rx_hashtbl + i); | |
929 | } | |
930 | ||
f515e6b7 | 931 | _unlock_rx_hashtbl_bh(bond); |
1da177e4 | 932 | |
1da177e4 | 933 | /* register to receive ARPs */ |
3aba891d | 934 | bond->recv_probe = rlb_arp_recv; |
1da177e4 LT |
935 | |
936 | return 0; | |
937 | } | |
938 | ||
939 | static void rlb_deinitialize(struct bonding *bond) | |
940 | { | |
941 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
942 | ||
f515e6b7 | 943 | _lock_rx_hashtbl_bh(bond); |
1da177e4 LT |
944 | |
945 | kfree(bond_info->rx_hashtbl); | |
946 | bond_info->rx_hashtbl = NULL; | |
e53665c6 | 947 | bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX; |
1da177e4 | 948 | |
f515e6b7 | 949 | _unlock_rx_hashtbl_bh(bond); |
1da177e4 LT |
950 | } |
951 | ||
952 | static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id) | |
953 | { | |
954 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
955 | u32 curr_index; | |
956 | ||
f515e6b7 | 957 | _lock_rx_hashtbl_bh(bond); |
1da177e4 | 958 | |
e53665c6 | 959 | curr_index = bond_info->rx_hashtbl_used_head; |
1da177e4 LT |
960 | while (curr_index != RLB_NULL_INDEX) { |
961 | struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]); | |
e53665c6 | 962 | u32 next_index = bond_info->rx_hashtbl[curr_index].used_next; |
1da177e4 | 963 | |
e53665c6 JB |
964 | if (curr->tag && (curr->vlan_id == vlan_id)) |
965 | rlb_delete_table_entry(bond, curr_index); | |
1da177e4 LT |
966 | |
967 | curr_index = next_index; | |
968 | } | |
969 | ||
f515e6b7 | 970 | _unlock_rx_hashtbl_bh(bond); |
1da177e4 LT |
971 | } |
972 | ||
973 | /*********************** tlb/rlb shared functions *********************/ | |
974 | ||
975 | static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[]) | |
976 | { | |
977 | struct bonding *bond = bond_get_bond_by_slave(slave); | |
978 | struct learning_pkt pkt; | |
979 | int size = sizeof(struct learning_pkt); | |
980 | int i; | |
981 | ||
982 | memset(&pkt, 0, size); | |
983 | memcpy(pkt.mac_dst, mac_addr, ETH_ALEN); | |
984 | memcpy(pkt.mac_src, mac_addr, ETH_ALEN); | |
09640e63 | 985 | pkt.type = cpu_to_be16(ETH_P_LOOP); |
1da177e4 LT |
986 | |
987 | for (i = 0; i < MAX_LP_BURST; i++) { | |
988 | struct sk_buff *skb; | |
989 | char *data; | |
990 | ||
991 | skb = dev_alloc_skb(size); | |
992 | if (!skb) { | |
993 | return; | |
994 | } | |
995 | ||
996 | data = skb_put(skb, size); | |
997 | memcpy(data, &pkt, size); | |
998 | ||
459a98ed | 999 | skb_reset_mac_header(skb); |
b0e380b1 | 1000 | skb->network_header = skb->mac_header + ETH_HLEN; |
1da177e4 LT |
1001 | skb->protocol = pkt.type; |
1002 | skb->priority = TC_PRIO_CONTROL; | |
1003 | skb->dev = slave->dev; | |
1004 | ||
cc0e4070 | 1005 | if (bond_vlan_used(bond)) { |
1da177e4 LT |
1006 | struct vlan_entry *vlan; |
1007 | ||
1008 | vlan = bond_next_vlan(bond, | |
1009 | bond->alb_info.current_alb_vlan); | |
1010 | ||
1011 | bond->alb_info.current_alb_vlan = vlan; | |
1012 | if (!vlan) { | |
1013 | kfree_skb(skb); | |
1014 | continue; | |
1015 | } | |
1016 | ||
1017 | skb = vlan_put_tag(skb, vlan->vlan_id); | |
1018 | if (!skb) { | |
a4aee5c8 | 1019 | pr_err("%s: Error: failed to insert VLAN tag\n", |
4e0952c7 | 1020 | bond->dev->name); |
1da177e4 LT |
1021 | continue; |
1022 | } | |
1023 | } | |
1024 | ||
1025 | dev_queue_xmit(skb); | |
1026 | } | |
1027 | } | |
1028 | ||
b924551b | 1029 | static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[]) |
1da177e4 LT |
1030 | { |
1031 | struct net_device *dev = slave->dev; | |
1032 | struct sockaddr s_addr; | |
1033 | ||
b924551b | 1034 | if (slave->bond->params.mode == BOND_MODE_TLB) { |
1da177e4 LT |
1035 | memcpy(dev->dev_addr, addr, dev->addr_len); |
1036 | return 0; | |
1037 | } | |
1038 | ||
1039 | /* for rlb each slave must have a unique hw mac addresses so that */ | |
1040 | /* each slave will receive packets destined to a different mac */ | |
1041 | memcpy(s_addr.sa_data, addr, dev->addr_len); | |
1042 | s_addr.sa_family = dev->type; | |
1043 | if (dev_set_mac_address(dev, &s_addr)) { | |
a4aee5c8 JP |
1044 | pr_err("%s: Error: dev_set_mac_address of dev %s failed!\n" |
1045 | "ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n", | |
4e0952c7 | 1046 | dev->master->name, dev->name); |
1da177e4 LT |
1047 | return -EOPNOTSUPP; |
1048 | } | |
1049 | return 0; | |
1050 | } | |
1051 | ||
059fe7a5 JV |
1052 | /* |
1053 | * Swap MAC addresses between two slaves. | |
1054 | * | |
1055 | * Called with RTNL held, and no other locks. | |
1056 | * | |
1057 | */ | |
1058 | ||
1da177e4 LT |
1059 | static void alb_swap_mac_addr(struct bonding *bond, struct slave *slave1, struct slave *slave2) |
1060 | { | |
1da177e4 | 1061 | u8 tmp_mac_addr[ETH_ALEN]; |
1da177e4 LT |
1062 | |
1063 | memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN); | |
b924551b JB |
1064 | alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr); |
1065 | alb_set_slave_mac_addr(slave2, tmp_mac_addr); | |
1da177e4 | 1066 | |
059fe7a5 JV |
1067 | } |
1068 | ||
1069 | /* | |
1070 | * Send learning packets after MAC address swap. | |
1071 | * | |
2543331d | 1072 | * Called with RTNL and no other locks |
059fe7a5 JV |
1073 | */ |
1074 | static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1, | |
1075 | struct slave *slave2) | |
1076 | { | |
1077 | int slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2)); | |
1078 | struct slave *disabled_slave = NULL; | |
1079 | ||
2543331d JV |
1080 | ASSERT_RTNL(); |
1081 | ||
1da177e4 LT |
1082 | /* fasten the change in the switch */ |
1083 | if (SLAVE_IS_OK(slave1)) { | |
1084 | alb_send_learning_packets(slave1, slave1->dev->dev_addr); | |
1085 | if (bond->alb_info.rlb_enabled) { | |
1086 | /* inform the clients that the mac address | |
1087 | * has changed | |
1088 | */ | |
1089 | rlb_req_update_slave_clients(bond, slave1); | |
1090 | } | |
1091 | } else { | |
1092 | disabled_slave = slave1; | |
1093 | } | |
1094 | ||
1095 | if (SLAVE_IS_OK(slave2)) { | |
1096 | alb_send_learning_packets(slave2, slave2->dev->dev_addr); | |
1097 | if (bond->alb_info.rlb_enabled) { | |
1098 | /* inform the clients that the mac address | |
1099 | * has changed | |
1100 | */ | |
1101 | rlb_req_update_slave_clients(bond, slave2); | |
1102 | } | |
1103 | } else { | |
1104 | disabled_slave = slave2; | |
1105 | } | |
1106 | ||
1107 | if (bond->alb_info.rlb_enabled && slaves_state_differ) { | |
1108 | /* A disabled slave was assigned an active mac addr */ | |
1109 | rlb_teach_disabled_mac_on_primary(bond, | |
1110 | disabled_slave->dev->dev_addr); | |
1111 | } | |
1112 | } | |
1113 | ||
1114 | /** | |
1115 | * alb_change_hw_addr_on_detach | |
1116 | * @bond: bonding we're working on | |
1117 | * @slave: the slave that was just detached | |
1118 | * | |
1119 | * We assume that @slave was already detached from the slave list. | |
1120 | * | |
1121 | * If @slave's permanent hw address is different both from its current | |
1122 | * address and from @bond's address, then somewhere in the bond there's | |
1123 | * a slave that has @slave's permanet address as its current address. | |
1124 | * We'll make sure that that slave no longer uses @slave's permanent address. | |
1125 | * | |
2543331d | 1126 | * Caller must hold RTNL and no other locks |
1da177e4 LT |
1127 | */ |
1128 | static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave) | |
1129 | { | |
1130 | int perm_curr_diff; | |
1131 | int perm_bond_diff; | |
1132 | ||
a6700db1 JP |
1133 | perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr, |
1134 | slave->dev->dev_addr); | |
1135 | perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr, | |
1136 | bond->dev->dev_addr); | |
1da177e4 LT |
1137 | |
1138 | if (perm_curr_diff && perm_bond_diff) { | |
1139 | struct slave *tmp_slave; | |
1140 | int i, found = 0; | |
1141 | ||
1142 | bond_for_each_slave(bond, tmp_slave, i) { | |
a6700db1 JP |
1143 | if (ether_addr_equal_64bits(slave->perm_hwaddr, |
1144 | tmp_slave->dev->dev_addr)) { | |
1da177e4 LT |
1145 | found = 1; |
1146 | break; | |
1147 | } | |
1148 | } | |
1149 | ||
1150 | if (found) { | |
059fe7a5 | 1151 | /* locking: needs RTNL and nothing else */ |
1da177e4 | 1152 | alb_swap_mac_addr(bond, slave, tmp_slave); |
059fe7a5 | 1153 | alb_fasten_mac_swap(bond, slave, tmp_slave); |
1da177e4 LT |
1154 | } |
1155 | } | |
1156 | } | |
1157 | ||
1158 | /** | |
1159 | * alb_handle_addr_collision_on_attach | |
1160 | * @bond: bonding we're working on | |
1161 | * @slave: the slave that was just attached | |
1162 | * | |
1163 | * checks uniqueness of slave's mac address and handles the case the | |
1164 | * new slave uses the bonds mac address. | |
1165 | * | |
1166 | * If the permanent hw address of @slave is @bond's hw address, we need to | |
1167 | * find a different hw address to give @slave, that isn't in use by any other | |
77c8e2c0 | 1168 | * slave in the bond. This address must be, of course, one of the permanent |
1da177e4 LT |
1169 | * addresses of the other slaves. |
1170 | * | |
1171 | * We go over the slave list, and for each slave there we compare its | |
1172 | * permanent hw address with the current address of all the other slaves. | |
1173 | * If no match was found, then we've found a slave with a permanent address | |
1174 | * that isn't used by any other slave in the bond, so we can assign it to | |
1175 | * @slave. | |
1176 | * | |
1177 | * assumption: this function is called before @slave is attached to the | |
1178 | * bond slave list. | |
1179 | * | |
1180 | * caller must hold the bond lock for write since the mac addresses are compared | |
1181 | * and may be swapped. | |
1182 | */ | |
1183 | static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave) | |
1184 | { | |
1185 | struct slave *tmp_slave1, *tmp_slave2, *free_mac_slave; | |
1186 | struct slave *has_bond_addr = bond->curr_active_slave; | |
1187 | int i, j, found = 0; | |
1188 | ||
1189 | if (bond->slave_cnt == 0) { | |
1190 | /* this is the first slave */ | |
1191 | return 0; | |
1192 | } | |
1193 | ||
1194 | /* if slave's mac address differs from bond's mac address | |
1195 | * check uniqueness of slave's mac address against the other | |
1196 | * slaves in the bond. | |
1197 | */ | |
a6700db1 | 1198 | if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) { |
1da177e4 | 1199 | bond_for_each_slave(bond, tmp_slave1, i) { |
a6700db1 JP |
1200 | if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr, |
1201 | slave->dev->dev_addr)) { | |
1da177e4 LT |
1202 | found = 1; |
1203 | break; | |
1204 | } | |
1205 | } | |
1206 | ||
6b38aefe JL |
1207 | if (!found) |
1208 | return 0; | |
1da177e4 | 1209 | |
6b38aefe JL |
1210 | /* Try setting slave mac to bond address and fall-through |
1211 | to code handling that situation below... */ | |
b924551b | 1212 | alb_set_slave_mac_addr(slave, bond->dev->dev_addr); |
1da177e4 LT |
1213 | } |
1214 | ||
1215 | /* The slave's address is equal to the address of the bond. | |
1216 | * Search for a spare address in the bond for this slave. | |
1217 | */ | |
1218 | free_mac_slave = NULL; | |
1219 | ||
1220 | bond_for_each_slave(bond, tmp_slave1, i) { | |
1221 | found = 0; | |
1222 | bond_for_each_slave(bond, tmp_slave2, j) { | |
a6700db1 JP |
1223 | if (ether_addr_equal_64bits(tmp_slave1->perm_hwaddr, |
1224 | tmp_slave2->dev->dev_addr)) { | |
1da177e4 LT |
1225 | found = 1; |
1226 | break; | |
1227 | } | |
1228 | } | |
1229 | ||
1230 | if (!found) { | |
1231 | /* no slave has tmp_slave1's perm addr | |
1232 | * as its curr addr | |
1233 | */ | |
1234 | free_mac_slave = tmp_slave1; | |
1235 | break; | |
1236 | } | |
1237 | ||
1238 | if (!has_bond_addr) { | |
a6700db1 JP |
1239 | if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr, |
1240 | bond->dev->dev_addr)) { | |
1da177e4 LT |
1241 | |
1242 | has_bond_addr = tmp_slave1; | |
1243 | } | |
1244 | } | |
1245 | } | |
1246 | ||
1247 | if (free_mac_slave) { | |
b924551b | 1248 | alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr); |
1da177e4 | 1249 | |
a4aee5c8 | 1250 | pr_warning("%s: Warning: the hw address of slave %s is in use by the bond; giving it the hw address of %s\n", |
e5e2a8fd JP |
1251 | bond->dev->name, slave->dev->name, |
1252 | free_mac_slave->dev->name); | |
1da177e4 LT |
1253 | |
1254 | } else if (has_bond_addr) { | |
a4aee5c8 | 1255 | 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", |
4e0952c7 | 1256 | bond->dev->name, slave->dev->name); |
1da177e4 LT |
1257 | return -EFAULT; |
1258 | } | |
1259 | ||
1260 | return 0; | |
1261 | } | |
1262 | ||
1263 | /** | |
1264 | * alb_set_mac_address | |
1265 | * @bond: | |
1266 | * @addr: | |
1267 | * | |
1268 | * In TLB mode all slaves are configured to the bond's hw address, but set | |
1269 | * their dev_addr field to different addresses (based on their permanent hw | |
1270 | * addresses). | |
1271 | * | |
1272 | * For each slave, this function sets the interface to the new address and then | |
1273 | * changes its dev_addr field to its previous value. | |
1274 | * | |
1275 | * Unwinding assumes bond's mac address has not yet changed. | |
1276 | */ | |
1277 | static int alb_set_mac_address(struct bonding *bond, void *addr) | |
1278 | { | |
1279 | struct sockaddr sa; | |
1280 | struct slave *slave, *stop_at; | |
1281 | char tmp_addr[ETH_ALEN]; | |
1282 | int res; | |
1283 | int i; | |
1284 | ||
1285 | if (bond->alb_info.rlb_enabled) { | |
1286 | return 0; | |
1287 | } | |
1288 | ||
1289 | bond_for_each_slave(bond, slave, i) { | |
1da177e4 LT |
1290 | /* save net_device's current hw address */ |
1291 | memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN); | |
1292 | ||
1293 | res = dev_set_mac_address(slave->dev, addr); | |
1294 | ||
1295 | /* restore net_device's hw address */ | |
1296 | memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN); | |
1297 | ||
eb7cc59a | 1298 | if (res) |
1da177e4 | 1299 | goto unwind; |
1da177e4 LT |
1300 | } |
1301 | ||
1302 | return 0; | |
1303 | ||
1304 | unwind: | |
1305 | memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len); | |
1306 | sa.sa_family = bond->dev->type; | |
1307 | ||
1308 | /* unwind from head to the slave that failed */ | |
1309 | stop_at = slave; | |
1310 | bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) { | |
1311 | memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN); | |
1312 | dev_set_mac_address(slave->dev, &sa); | |
1313 | memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN); | |
1314 | } | |
1315 | ||
1316 | return res; | |
1317 | } | |
1318 | ||
1319 | /************************ exported alb funcions ************************/ | |
1320 | ||
1321 | int bond_alb_initialize(struct bonding *bond, int rlb_enabled) | |
1322 | { | |
1323 | int res; | |
1324 | ||
1325 | res = tlb_initialize(bond); | |
1326 | if (res) { | |
1327 | return res; | |
1328 | } | |
1329 | ||
1330 | if (rlb_enabled) { | |
1331 | bond->alb_info.rlb_enabled = 1; | |
1332 | /* initialize rlb */ | |
1333 | res = rlb_initialize(bond); | |
1334 | if (res) { | |
1335 | tlb_deinitialize(bond); | |
1336 | return res; | |
1337 | } | |
b76850ab MW |
1338 | } else { |
1339 | bond->alb_info.rlb_enabled = 0; | |
1da177e4 LT |
1340 | } |
1341 | ||
1342 | return 0; | |
1343 | } | |
1344 | ||
1345 | void bond_alb_deinitialize(struct bonding *bond) | |
1346 | { | |
1347 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
1348 | ||
1349 | tlb_deinitialize(bond); | |
1350 | ||
1351 | if (bond_info->rlb_enabled) { | |
1352 | rlb_deinitialize(bond); | |
1353 | } | |
1354 | } | |
1355 | ||
1356 | int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev) | |
1357 | { | |
454d7c9b | 1358 | struct bonding *bond = netdev_priv(bond_dev); |
1da177e4 LT |
1359 | struct ethhdr *eth_data; |
1360 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
1361 | struct slave *tx_slave = NULL; | |
d3bb52b0 | 1362 | static const __be32 ip_bcast = htonl(0xffffffff); |
1da177e4 LT |
1363 | int hash_size = 0; |
1364 | int do_tx_balance = 1; | |
1365 | u32 hash_index = 0; | |
eddc9ec5 | 1366 | const u8 *hash_start = NULL; |
1da177e4 | 1367 | int res = 1; |
2d1ea19d | 1368 | struct ipv6hdr *ip6hdr; |
1da177e4 | 1369 | |
459a98ed | 1370 | skb_reset_mac_header(skb); |
1da177e4 LT |
1371 | eth_data = eth_hdr(skb); |
1372 | ||
0693e88e | 1373 | /* make sure that the curr_active_slave do not change during tx |
1da177e4 | 1374 | */ |
1da177e4 LT |
1375 | read_lock(&bond->curr_slave_lock); |
1376 | ||
1da177e4 | 1377 | switch (ntohs(skb->protocol)) { |
eddc9ec5 ACM |
1378 | case ETH_P_IP: { |
1379 | const struct iphdr *iph = ip_hdr(skb); | |
1380 | ||
a6700db1 | 1381 | if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast) || |
eddc9ec5 ACM |
1382 | (iph->daddr == ip_bcast) || |
1383 | (iph->protocol == IPPROTO_IGMP)) { | |
1da177e4 LT |
1384 | do_tx_balance = 0; |
1385 | break; | |
1386 | } | |
eddc9ec5 ACM |
1387 | hash_start = (char *)&(iph->daddr); |
1388 | hash_size = sizeof(iph->daddr); | |
1389 | } | |
1da177e4 LT |
1390 | break; |
1391 | case ETH_P_IPV6: | |
2d1ea19d VY |
1392 | /* IPv6 doesn't really use broadcast mac address, but leave |
1393 | * that here just in case. | |
1394 | */ | |
a6700db1 | 1395 | if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast)) { |
1da177e4 LT |
1396 | do_tx_balance = 0; |
1397 | break; | |
2d1ea19d VY |
1398 | } |
1399 | ||
1400 | /* IPv6 uses all-nodes multicast as an equivalent to | |
1401 | * broadcasts in IPv4. | |
1402 | */ | |
a6700db1 | 1403 | if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) { |
2d1ea19d VY |
1404 | do_tx_balance = 0; |
1405 | break; | |
1406 | } | |
1407 | ||
1408 | /* Additianally, DAD probes should not be tx-balanced as that | |
1409 | * will lead to false positives for duplicate addresses and | |
1410 | * prevent address configuration from working. | |
1411 | */ | |
1412 | ip6hdr = ipv6_hdr(skb); | |
1413 | if (ipv6_addr_any(&ip6hdr->saddr)) { | |
1414 | do_tx_balance = 0; | |
1415 | break; | |
1da177e4 LT |
1416 | } |
1417 | ||
0660e03f ACM |
1418 | hash_start = (char *)&(ipv6_hdr(skb)->daddr); |
1419 | hash_size = sizeof(ipv6_hdr(skb)->daddr); | |
1da177e4 LT |
1420 | break; |
1421 | case ETH_P_IPX: | |
d3bb52b0 | 1422 | if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) { |
1da177e4 LT |
1423 | /* something is wrong with this packet */ |
1424 | do_tx_balance = 0; | |
1425 | break; | |
1426 | } | |
1427 | ||
1428 | if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) { | |
1429 | /* The only protocol worth balancing in | |
1430 | * this family since it has an "ARP" like | |
1431 | * mechanism | |
1432 | */ | |
1433 | do_tx_balance = 0; | |
1434 | break; | |
1435 | } | |
1436 | ||
1437 | hash_start = (char*)eth_data->h_dest; | |
1438 | hash_size = ETH_ALEN; | |
1439 | break; | |
1440 | case ETH_P_ARP: | |
1441 | do_tx_balance = 0; | |
1442 | if (bond_info->rlb_enabled) { | |
1443 | tx_slave = rlb_arp_xmit(skb, bond); | |
1444 | } | |
1445 | break; | |
1446 | default: | |
1447 | do_tx_balance = 0; | |
1448 | break; | |
1449 | } | |
1450 | ||
1451 | if (do_tx_balance) { | |
1452 | hash_index = _simple_hash(hash_start, hash_size); | |
1453 | tx_slave = tlb_choose_channel(bond, hash_index, skb->len); | |
1454 | } | |
1455 | ||
1456 | if (!tx_slave) { | |
1457 | /* unbalanced or unassigned, send through primary */ | |
1458 | tx_slave = bond->curr_active_slave; | |
1459 | bond_info->unbalanced_load += skb->len; | |
1460 | } | |
1461 | ||
1462 | if (tx_slave && SLAVE_IS_OK(tx_slave)) { | |
1463 | if (tx_slave != bond->curr_active_slave) { | |
1464 | memcpy(eth_data->h_source, | |
1465 | tx_slave->dev->dev_addr, | |
1466 | ETH_ALEN); | |
1467 | } | |
1468 | ||
1469 | res = bond_dev_queue_xmit(bond, skb, tx_slave->dev); | |
1470 | } else { | |
1471 | if (tx_slave) { | |
f515e6b7 MU |
1472 | _lock_tx_hashtbl(bond); |
1473 | __tlb_clear_slave(bond, tx_slave, 0); | |
1474 | _unlock_tx_hashtbl(bond); | |
1da177e4 LT |
1475 | } |
1476 | } | |
1477 | ||
04502430 ED |
1478 | read_unlock(&bond->curr_slave_lock); |
1479 | ||
1da177e4 LT |
1480 | if (res) { |
1481 | /* no suitable interface, frame not sent */ | |
04502430 | 1482 | kfree_skb(skb); |
1da177e4 | 1483 | } |
ec634fe3 | 1484 | return NETDEV_TX_OK; |
1da177e4 LT |
1485 | } |
1486 | ||
1b76b316 | 1487 | void bond_alb_monitor(struct work_struct *work) |
1da177e4 | 1488 | { |
1b76b316 JV |
1489 | struct bonding *bond = container_of(work, struct bonding, |
1490 | alb_work.work); | |
1da177e4 LT |
1491 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); |
1492 | struct slave *slave; | |
1493 | int i; | |
1494 | ||
1495 | read_lock(&bond->lock); | |
1496 | ||
1da177e4 LT |
1497 | if (bond->slave_cnt == 0) { |
1498 | bond_info->tx_rebalance_counter = 0; | |
1499 | bond_info->lp_counter = 0; | |
1500 | goto re_arm; | |
1501 | } | |
1502 | ||
1503 | bond_info->tx_rebalance_counter++; | |
1504 | bond_info->lp_counter++; | |
1505 | ||
1506 | /* send learning packets */ | |
1507 | if (bond_info->lp_counter >= BOND_ALB_LP_TICKS) { | |
1508 | /* change of curr_active_slave involves swapping of mac addresses. | |
1509 | * in order to avoid this swapping from happening while | |
1510 | * sending the learning packets, the curr_slave_lock must be held for | |
1511 | * read. | |
1512 | */ | |
1513 | read_lock(&bond->curr_slave_lock); | |
1514 | ||
1515 | bond_for_each_slave(bond, slave, i) { | |
e944ef79 | 1516 | alb_send_learning_packets(slave, slave->dev->dev_addr); |
1da177e4 LT |
1517 | } |
1518 | ||
1519 | read_unlock(&bond->curr_slave_lock); | |
1520 | ||
1521 | bond_info->lp_counter = 0; | |
1522 | } | |
1523 | ||
1524 | /* rebalance tx traffic */ | |
1525 | if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) { | |
1526 | ||
1527 | read_lock(&bond->curr_slave_lock); | |
1528 | ||
1529 | bond_for_each_slave(bond, slave, i) { | |
1530 | tlb_clear_slave(bond, slave, 1); | |
1531 | if (slave == bond->curr_active_slave) { | |
1532 | SLAVE_TLB_INFO(slave).load = | |
1533 | bond_info->unbalanced_load / | |
1534 | BOND_TLB_REBALANCE_INTERVAL; | |
1535 | bond_info->unbalanced_load = 0; | |
1536 | } | |
1537 | } | |
1538 | ||
1539 | read_unlock(&bond->curr_slave_lock); | |
1540 | ||
1541 | bond_info->tx_rebalance_counter = 0; | |
1542 | } | |
1543 | ||
1544 | /* handle rlb stuff */ | |
1545 | if (bond_info->rlb_enabled) { | |
1da177e4 LT |
1546 | if (bond_info->primary_is_promisc && |
1547 | (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) { | |
1548 | ||
d0e81b7e JV |
1549 | /* |
1550 | * dev_set_promiscuity requires rtnl and | |
e6d265e8 | 1551 | * nothing else. Avoid race with bond_close. |
d0e81b7e JV |
1552 | */ |
1553 | read_unlock(&bond->lock); | |
e6d265e8 JV |
1554 | if (!rtnl_trylock()) { |
1555 | read_lock(&bond->lock); | |
1556 | goto re_arm; | |
1557 | } | |
d0e81b7e | 1558 | |
1da177e4 LT |
1559 | bond_info->rlb_promisc_timeout_counter = 0; |
1560 | ||
1561 | /* If the primary was set to promiscuous mode | |
1562 | * because a slave was disabled then | |
1563 | * it can now leave promiscuous mode. | |
1564 | */ | |
1565 | dev_set_promiscuity(bond->curr_active_slave->dev, -1); | |
1566 | bond_info->primary_is_promisc = 0; | |
1da177e4 | 1567 | |
d0e81b7e JV |
1568 | rtnl_unlock(); |
1569 | read_lock(&bond->lock); | |
1570 | } | |
1da177e4 LT |
1571 | |
1572 | if (bond_info->rlb_rebalance) { | |
1573 | bond_info->rlb_rebalance = 0; | |
1574 | rlb_rebalance(bond); | |
1575 | } | |
1576 | ||
1577 | /* check if clients need updating */ | |
1578 | if (bond_info->rx_ntt) { | |
1579 | if (bond_info->rlb_update_delay_counter) { | |
1580 | --bond_info->rlb_update_delay_counter; | |
1581 | } else { | |
1582 | rlb_update_rx_clients(bond); | |
1583 | if (bond_info->rlb_update_retry_counter) { | |
1584 | --bond_info->rlb_update_retry_counter; | |
1585 | } else { | |
1586 | bond_info->rx_ntt = 0; | |
1587 | } | |
1588 | } | |
1589 | } | |
1590 | } | |
1591 | ||
1592 | re_arm: | |
e6d265e8 JV |
1593 | queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks); |
1594 | ||
1da177e4 LT |
1595 | read_unlock(&bond->lock); |
1596 | } | |
1597 | ||
1598 | /* assumption: called before the slave is attached to the bond | |
1599 | * and not locked by the bond lock | |
1600 | */ | |
1601 | int bond_alb_init_slave(struct bonding *bond, struct slave *slave) | |
1602 | { | |
1603 | int res; | |
1604 | ||
b924551b | 1605 | res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr); |
1da177e4 LT |
1606 | if (res) { |
1607 | return res; | |
1608 | } | |
1609 | ||
1610 | /* caller must hold the bond lock for write since the mac addresses | |
1611 | * are compared and may be swapped. | |
1612 | */ | |
6603a6f2 | 1613 | read_lock(&bond->lock); |
1da177e4 LT |
1614 | |
1615 | res = alb_handle_addr_collision_on_attach(bond, slave); | |
1616 | ||
6603a6f2 | 1617 | read_unlock(&bond->lock); |
1da177e4 LT |
1618 | |
1619 | if (res) { | |
1620 | return res; | |
1621 | } | |
1622 | ||
1623 | tlb_init_slave(slave); | |
1624 | ||
1625 | /* order a rebalance ASAP */ | |
1626 | bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS; | |
1627 | ||
1628 | if (bond->alb_info.rlb_enabled) { | |
1629 | bond->alb_info.rlb_rebalance = 1; | |
1630 | } | |
1631 | ||
1632 | return 0; | |
1633 | } | |
1634 | ||
2543331d JV |
1635 | /* |
1636 | * Remove slave from tlb and rlb hash tables, and fix up MAC addresses | |
1637 | * if necessary. | |
1638 | * | |
1639 | * Caller must hold RTNL and no other locks | |
1640 | */ | |
1da177e4 LT |
1641 | void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave) |
1642 | { | |
1643 | if (bond->slave_cnt > 1) { | |
1644 | alb_change_hw_addr_on_detach(bond, slave); | |
1645 | } | |
1646 | ||
1647 | tlb_clear_slave(bond, slave, 0); | |
1648 | ||
1649 | if (bond->alb_info.rlb_enabled) { | |
1650 | bond->alb_info.next_rx_slave = NULL; | |
1651 | rlb_clear_slave(bond, slave); | |
1652 | } | |
1653 | } | |
1654 | ||
1655 | /* Caller must hold bond lock for read */ | |
1656 | void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link) | |
1657 | { | |
1658 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
1659 | ||
1660 | if (link == BOND_LINK_DOWN) { | |
1661 | tlb_clear_slave(bond, slave, 0); | |
1662 | if (bond->alb_info.rlb_enabled) { | |
1663 | rlb_clear_slave(bond, slave); | |
1664 | } | |
1665 | } else if (link == BOND_LINK_UP) { | |
1666 | /* order a rebalance ASAP */ | |
1667 | bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS; | |
1668 | if (bond->alb_info.rlb_enabled) { | |
1669 | bond->alb_info.rlb_rebalance = 1; | |
1670 | /* If the updelay module parameter is smaller than the | |
1671 | * forwarding delay of the switch the rebalance will | |
1672 | * not work because the rebalance arp replies will | |
1673 | * not be forwarded to the clients.. | |
1674 | */ | |
1675 | } | |
1676 | } | |
1677 | } | |
1678 | ||
1679 | /** | |
1680 | * bond_alb_handle_active_change - assign new curr_active_slave | |
1681 | * @bond: our bonding struct | |
1682 | * @new_slave: new slave to assign | |
1683 | * | |
1684 | * Set the bond->curr_active_slave to @new_slave and handle | |
1685 | * mac address swapping and promiscuity changes as needed. | |
1686 | * | |
059fe7a5 JV |
1687 | * If new_slave is NULL, caller must hold curr_slave_lock or |
1688 | * bond->lock for write. | |
1689 | * | |
1690 | * If new_slave is not NULL, caller must hold RTNL, bond->lock for | |
1691 | * read and curr_slave_lock for write. Processing here may sleep, so | |
1692 | * no other locks may be held. | |
1da177e4 LT |
1693 | */ |
1694 | void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave) | |
1f78d9f9 HE |
1695 | __releases(&bond->curr_slave_lock) |
1696 | __releases(&bond->lock) | |
1697 | __acquires(&bond->lock) | |
1698 | __acquires(&bond->curr_slave_lock) | |
1da177e4 LT |
1699 | { |
1700 | struct slave *swap_slave; | |
1701 | int i; | |
1702 | ||
1703 | if (bond->curr_active_slave == new_slave) { | |
1704 | return; | |
1705 | } | |
1706 | ||
1707 | if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) { | |
1708 | dev_set_promiscuity(bond->curr_active_slave->dev, -1); | |
1709 | bond->alb_info.primary_is_promisc = 0; | |
1710 | bond->alb_info.rlb_promisc_timeout_counter = 0; | |
1711 | } | |
1712 | ||
1713 | swap_slave = bond->curr_active_slave; | |
1714 | bond->curr_active_slave = new_slave; | |
1715 | ||
1716 | if (!new_slave || (bond->slave_cnt == 0)) { | |
1717 | return; | |
1718 | } | |
1719 | ||
1720 | /* set the new curr_active_slave to the bonds mac address | |
1721 | * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave | |
1722 | */ | |
1723 | if (!swap_slave) { | |
1724 | struct slave *tmp_slave; | |
1725 | /* find slave that is holding the bond's mac address */ | |
1726 | bond_for_each_slave(bond, tmp_slave, i) { | |
a6700db1 JP |
1727 | if (ether_addr_equal_64bits(tmp_slave->dev->dev_addr, |
1728 | bond->dev->dev_addr)) { | |
1da177e4 LT |
1729 | swap_slave = tmp_slave; |
1730 | break; | |
1731 | } | |
1732 | } | |
1733 | } | |
1734 | ||
059fe7a5 JV |
1735 | /* |
1736 | * Arrange for swap_slave and new_slave to temporarily be | |
1737 | * ignored so we can mess with their MAC addresses without | |
1738 | * fear of interference from transmit activity. | |
1739 | */ | |
1740 | if (swap_slave) { | |
1741 | tlb_clear_slave(bond, swap_slave, 1); | |
1742 | } | |
1743 | tlb_clear_slave(bond, new_slave, 1); | |
1744 | ||
1745 | write_unlock_bh(&bond->curr_slave_lock); | |
1746 | read_unlock(&bond->lock); | |
1747 | ||
e0138a66 JV |
1748 | ASSERT_RTNL(); |
1749 | ||
1da177e4 LT |
1750 | /* curr_active_slave must be set before calling alb_swap_mac_addr */ |
1751 | if (swap_slave) { | |
1752 | /* swap mac address */ | |
1753 | alb_swap_mac_addr(bond, swap_slave, new_slave); | |
1754 | } else { | |
1755 | /* set the new_slave to the bond mac address */ | |
b924551b | 1756 | alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr); |
059fe7a5 JV |
1757 | } |
1758 | ||
059fe7a5 JV |
1759 | if (swap_slave) { |
1760 | alb_fasten_mac_swap(bond, swap_slave, new_slave); | |
2543331d | 1761 | read_lock(&bond->lock); |
059fe7a5 | 1762 | } else { |
2543331d | 1763 | read_lock(&bond->lock); |
1da177e4 LT |
1764 | alb_send_learning_packets(new_slave, bond->dev->dev_addr); |
1765 | } | |
059fe7a5 JV |
1766 | |
1767 | write_lock_bh(&bond->curr_slave_lock); | |
1da177e4 LT |
1768 | } |
1769 | ||
059fe7a5 JV |
1770 | /* |
1771 | * Called with RTNL | |
1772 | */ | |
1da177e4 | 1773 | int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr) |
1f78d9f9 | 1774 | __acquires(&bond->lock) |
815bcc27 | 1775 | __releases(&bond->lock) |
1da177e4 | 1776 | { |
454d7c9b | 1777 | struct bonding *bond = netdev_priv(bond_dev); |
1da177e4 LT |
1778 | struct sockaddr *sa = addr; |
1779 | struct slave *slave, *swap_slave; | |
1780 | int res; | |
1781 | int i; | |
1782 | ||
1783 | if (!is_valid_ether_addr(sa->sa_data)) { | |
1784 | return -EADDRNOTAVAIL; | |
1785 | } | |
1786 | ||
1787 | res = alb_set_mac_address(bond, addr); | |
1788 | if (res) { | |
1789 | return res; | |
1790 | } | |
1791 | ||
1792 | memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len); | |
1793 | ||
1794 | /* If there is no curr_active_slave there is nothing else to do. | |
1795 | * Otherwise we'll need to pass the new address to it and handle | |
1796 | * duplications. | |
1797 | */ | |
1798 | if (!bond->curr_active_slave) { | |
1799 | return 0; | |
1800 | } | |
1801 | ||
1802 | swap_slave = NULL; | |
1803 | ||
1804 | bond_for_each_slave(bond, slave, i) { | |
a6700db1 JP |
1805 | if (ether_addr_equal_64bits(slave->dev->dev_addr, |
1806 | bond_dev->dev_addr)) { | |
1da177e4 LT |
1807 | swap_slave = slave; |
1808 | break; | |
1809 | } | |
1810 | } | |
1811 | ||
1812 | if (swap_slave) { | |
1813 | alb_swap_mac_addr(bond, swap_slave, bond->curr_active_slave); | |
059fe7a5 | 1814 | alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave); |
1da177e4 | 1815 | } else { |
b924551b | 1816 | alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr); |
1da177e4 | 1817 | |
815bcc27 | 1818 | read_lock(&bond->lock); |
1da177e4 LT |
1819 | alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr); |
1820 | if (bond->alb_info.rlb_enabled) { | |
1821 | /* inform clients mac address has changed */ | |
1822 | rlb_req_update_slave_clients(bond, bond->curr_active_slave); | |
1823 | } | |
815bcc27 | 1824 | read_unlock(&bond->lock); |
1da177e4 LT |
1825 | } |
1826 | ||
1827 | return 0; | |
1828 | } | |
1829 | ||
1830 | void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id) | |
1831 | { | |
1832 | if (bond->alb_info.current_alb_vlan && | |
1833 | (bond->alb_info.current_alb_vlan->vlan_id == vlan_id)) { | |
1834 | bond->alb_info.current_alb_vlan = NULL; | |
1835 | } | |
1836 | ||
1837 | if (bond->alb_info.rlb_enabled) { | |
1838 | rlb_clear_vlan(bond, vlan_id); | |
1839 | } | |
1840 | } | |
1841 |