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 | |
adf8d3ff | 15 | * with this program; if not, see <http://www.gnu.org/licenses/>. |
1da177e4 LT |
16 | * |
17 | * The full GNU General Public License is included in this distribution in the | |
18 | * file called LICENSE. | |
19 | * | |
1da177e4 LT |
20 | */ |
21 | ||
a4aee5c8 JP |
22 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
23 | ||
1da177e4 LT |
24 | #include <linux/skbuff.h> |
25 | #include <linux/netdevice.h> | |
26 | #include <linux/etherdevice.h> | |
27 | #include <linux/pkt_sched.h> | |
28 | #include <linux/spinlock.h> | |
29 | #include <linux/slab.h> | |
30 | #include <linux/timer.h> | |
31 | #include <linux/ip.h> | |
32 | #include <linux/ipv6.h> | |
33 | #include <linux/if_arp.h> | |
34 | #include <linux/if_ether.h> | |
35 | #include <linux/if_bonding.h> | |
36 | #include <linux/if_vlan.h> | |
37 | #include <linux/in.h> | |
38 | #include <net/ipx.h> | |
39 | #include <net/arp.h> | |
2d1ea19d | 40 | #include <net/ipv6.h> |
1da177e4 LT |
41 | #include <asm/byteorder.h> |
42 | #include "bonding.h" | |
43 | #include "bond_alb.h" | |
44 | ||
45 | ||
1da177e4 | 46 | |
885a136c ED |
47 | #ifndef __long_aligned |
48 | #define __long_aligned __attribute__((aligned((sizeof(long))))) | |
49 | #endif | |
50 | static const u8 mac_bcast[ETH_ALEN] __long_aligned = { | |
51 | 0xff, 0xff, 0xff, 0xff, 0xff, 0xff | |
52 | }; | |
53 | static const u8 mac_v6_allmcast[ETH_ALEN] __long_aligned = { | |
54 | 0x33, 0x33, 0x00, 0x00, 0x00, 0x01 | |
55 | }; | |
1da177e4 LT |
56 | static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC; |
57 | ||
58 | #pragma pack(1) | |
59 | struct learning_pkt { | |
60 | u8 mac_dst[ETH_ALEN]; | |
61 | u8 mac_src[ETH_ALEN]; | |
d3bb52b0 | 62 | __be16 type; |
1da177e4 LT |
63 | u8 padding[ETH_ZLEN - ETH_HLEN]; |
64 | }; | |
65 | ||
66 | struct arp_pkt { | |
d3bb52b0 AV |
67 | __be16 hw_addr_space; |
68 | __be16 prot_addr_space; | |
1da177e4 LT |
69 | u8 hw_addr_len; |
70 | u8 prot_addr_len; | |
d3bb52b0 | 71 | __be16 op_code; |
1da177e4 | 72 | u8 mac_src[ETH_ALEN]; /* sender hardware address */ |
d3bb52b0 | 73 | __be32 ip_src; /* sender IP address */ |
1da177e4 | 74 | u8 mac_dst[ETH_ALEN]; /* target hardware address */ |
d3bb52b0 | 75 | __be32 ip_dst; /* target IP address */ |
1da177e4 LT |
76 | }; |
77 | #pragma pack() | |
78 | ||
a16aeb36 ACM |
79 | static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb) |
80 | { | |
d56f90a7 | 81 | return (struct arp_pkt *)skb_network_header(skb); |
a16aeb36 ACM |
82 | } |
83 | ||
1da177e4 LT |
84 | /* Forward declaration */ |
85 | static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[]); | |
e53665c6 JB |
86 | static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp); |
87 | static void rlb_src_unlink(struct bonding *bond, u32 index); | |
88 | static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, | |
89 | u32 ip_dst_hash); | |
1da177e4 | 90 | |
eddc9ec5 | 91 | static inline u8 _simple_hash(const u8 *hash_start, int hash_size) |
1da177e4 LT |
92 | { |
93 | int i; | |
94 | u8 hash = 0; | |
95 | ||
fdb89d75 | 96 | for (i = 0; i < hash_size; i++) |
1da177e4 | 97 | hash ^= hash_start[i]; |
1da177e4 LT |
98 | |
99 | return hash; | |
100 | } | |
101 | ||
102 | /*********************** tlb specific functions ***************************/ | |
103 | ||
f515e6b7 | 104 | static inline void _lock_tx_hashtbl_bh(struct bonding *bond) |
1da177e4 | 105 | { |
6603a6f2 | 106 | spin_lock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock)); |
1da177e4 LT |
107 | } |
108 | ||
f515e6b7 | 109 | static inline void _unlock_tx_hashtbl_bh(struct bonding *bond) |
1da177e4 | 110 | { |
6603a6f2 | 111 | spin_unlock_bh(&(BOND_ALB_INFO(bond).tx_hashtbl_lock)); |
1da177e4 LT |
112 | } |
113 | ||
f515e6b7 MU |
114 | static inline void _lock_tx_hashtbl(struct bonding *bond) |
115 | { | |
116 | spin_lock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock)); | |
117 | } | |
118 | ||
119 | static inline void _unlock_tx_hashtbl(struct bonding *bond) | |
120 | { | |
121 | spin_unlock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock)); | |
122 | } | |
123 | ||
1da177e4 LT |
124 | /* Caller must hold tx_hashtbl lock */ |
125 | static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load) | |
126 | { | |
127 | if (save_load) { | |
128 | entry->load_history = 1 + entry->tx_bytes / | |
129 | BOND_TLB_REBALANCE_INTERVAL; | |
130 | entry->tx_bytes = 0; | |
131 | } | |
132 | ||
133 | entry->tx_slave = NULL; | |
134 | entry->next = TLB_NULL_INDEX; | |
135 | entry->prev = TLB_NULL_INDEX; | |
136 | } | |
137 | ||
138 | static inline void tlb_init_slave(struct slave *slave) | |
139 | { | |
140 | SLAVE_TLB_INFO(slave).load = 0; | |
141 | SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX; | |
142 | } | |
143 | ||
f515e6b7 MU |
144 | /* Caller must hold bond lock for read, BH disabled */ |
145 | static void __tlb_clear_slave(struct bonding *bond, struct slave *slave, | |
146 | int save_load) | |
1da177e4 LT |
147 | { |
148 | struct tlb_client_info *tx_hash_table; | |
149 | u32 index; | |
150 | ||
1da177e4 LT |
151 | /* clear slave from tx_hashtbl */ |
152 | tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl; | |
153 | ||
ce39a800 AG |
154 | /* skip this if we've already freed the tx hash table */ |
155 | if (tx_hash_table) { | |
156 | index = SLAVE_TLB_INFO(slave).head; | |
157 | while (index != TLB_NULL_INDEX) { | |
158 | u32 next_index = tx_hash_table[index].next; | |
159 | tlb_init_table_entry(&tx_hash_table[index], save_load); | |
160 | index = next_index; | |
161 | } | |
1da177e4 LT |
162 | } |
163 | ||
1da177e4 | 164 | tlb_init_slave(slave); |
f515e6b7 | 165 | } |
5af47b2f | 166 | |
f515e6b7 MU |
167 | /* Caller must hold bond lock for read */ |
168 | static void tlb_clear_slave(struct bonding *bond, struct slave *slave, | |
169 | int save_load) | |
170 | { | |
171 | _lock_tx_hashtbl_bh(bond); | |
172 | __tlb_clear_slave(bond, slave, save_load); | |
173 | _unlock_tx_hashtbl_bh(bond); | |
1da177e4 LT |
174 | } |
175 | ||
176 | /* Must be called before starting the monitor timer */ | |
177 | static int tlb_initialize(struct bonding *bond) | |
178 | { | |
179 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
180 | int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info); | |
0d206a3a | 181 | struct tlb_client_info *new_hashtbl; |
1da177e4 LT |
182 | int i; |
183 | ||
243cb4e5 | 184 | new_hashtbl = kzalloc(size, GFP_KERNEL); |
e404decb | 185 | if (!new_hashtbl) |
1da177e4 | 186 | return -1; |
e404decb | 187 | |
f515e6b7 | 188 | _lock_tx_hashtbl_bh(bond); |
0d206a3a MW |
189 | |
190 | bond_info->tx_hashtbl = new_hashtbl; | |
1da177e4 | 191 | |
fdb89d75 | 192 | for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) |
38dbaf0a | 193 | tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0); |
1da177e4 | 194 | |
f515e6b7 | 195 | _unlock_tx_hashtbl_bh(bond); |
1da177e4 LT |
196 | |
197 | return 0; | |
198 | } | |
199 | ||
200 | /* Must be called only after all slaves have been released */ | |
201 | static void tlb_deinitialize(struct bonding *bond) | |
202 | { | |
203 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
204 | ||
f515e6b7 | 205 | _lock_tx_hashtbl_bh(bond); |
1da177e4 LT |
206 | |
207 | kfree(bond_info->tx_hashtbl); | |
208 | bond_info->tx_hashtbl = NULL; | |
209 | ||
f515e6b7 | 210 | _unlock_tx_hashtbl_bh(bond); |
1da177e4 LT |
211 | } |
212 | ||
097811bb JP |
213 | static long long compute_gap(struct slave *slave) |
214 | { | |
215 | return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */ | |
216 | (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */ | |
217 | } | |
218 | ||
1da177e4 LT |
219 | /* Caller must hold bond lock for read */ |
220 | static struct slave *tlb_get_least_loaded_slave(struct bonding *bond) | |
221 | { | |
222 | struct slave *slave, *least_loaded; | |
9caff1e7 | 223 | struct list_head *iter; |
097811bb | 224 | long long max_gap; |
1da177e4 | 225 | |
097811bb JP |
226 | least_loaded = NULL; |
227 | max_gap = LLONG_MIN; | |
1da177e4 LT |
228 | |
229 | /* Find the slave with the largest gap */ | |
28c71926 | 230 | bond_for_each_slave_rcu(bond, slave, iter) { |
1da177e4 | 231 | if (SLAVE_IS_OK(slave)) { |
097811bb JP |
232 | long long gap = compute_gap(slave); |
233 | ||
1da177e4 LT |
234 | if (max_gap < gap) { |
235 | least_loaded = slave; | |
236 | max_gap = gap; | |
237 | } | |
238 | } | |
239 | } | |
240 | ||
241 | return least_loaded; | |
242 | } | |
243 | ||
f515e6b7 MU |
244 | static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index, |
245 | u32 skb_len) | |
1da177e4 LT |
246 | { |
247 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
248 | struct tlb_client_info *hash_table; | |
249 | struct slave *assigned_slave; | |
250 | ||
1da177e4 LT |
251 | hash_table = bond_info->tx_hashtbl; |
252 | assigned_slave = hash_table[hash_index].tx_slave; | |
253 | if (!assigned_slave) { | |
254 | assigned_slave = tlb_get_least_loaded_slave(bond); | |
255 | ||
256 | if (assigned_slave) { | |
257 | struct tlb_slave_info *slave_info = | |
258 | &(SLAVE_TLB_INFO(assigned_slave)); | |
259 | u32 next_index = slave_info->head; | |
260 | ||
261 | hash_table[hash_index].tx_slave = assigned_slave; | |
262 | hash_table[hash_index].next = next_index; | |
263 | hash_table[hash_index].prev = TLB_NULL_INDEX; | |
264 | ||
fdb89d75 | 265 | if (next_index != TLB_NULL_INDEX) |
1da177e4 | 266 | hash_table[next_index].prev = hash_index; |
1da177e4 LT |
267 | |
268 | slave_info->head = hash_index; | |
269 | slave_info->load += | |
270 | hash_table[hash_index].load_history; | |
271 | } | |
272 | } | |
273 | ||
fdb89d75 | 274 | if (assigned_slave) |
1da177e4 | 275 | hash_table[hash_index].tx_bytes += skb_len; |
1da177e4 | 276 | |
1da177e4 LT |
277 | return assigned_slave; |
278 | } | |
279 | ||
f515e6b7 MU |
280 | /* Caller must hold bond lock for read */ |
281 | static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index, | |
282 | u32 skb_len) | |
283 | { | |
284 | struct slave *tx_slave; | |
285 | /* | |
286 | * We don't need to disable softirq here, becase | |
287 | * tlb_choose_channel() is only called by bond_alb_xmit() | |
288 | * which already has softirq disabled. | |
289 | */ | |
290 | _lock_tx_hashtbl(bond); | |
291 | tx_slave = __tlb_choose_channel(bond, hash_index, skb_len); | |
292 | _unlock_tx_hashtbl(bond); | |
293 | return tx_slave; | |
294 | } | |
295 | ||
1da177e4 | 296 | /*********************** rlb specific functions ***************************/ |
f515e6b7 | 297 | static inline void _lock_rx_hashtbl_bh(struct bonding *bond) |
1da177e4 | 298 | { |
6603a6f2 | 299 | spin_lock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock)); |
1da177e4 LT |
300 | } |
301 | ||
f515e6b7 | 302 | static inline void _unlock_rx_hashtbl_bh(struct bonding *bond) |
1da177e4 | 303 | { |
6603a6f2 | 304 | spin_unlock_bh(&(BOND_ALB_INFO(bond).rx_hashtbl_lock)); |
1da177e4 LT |
305 | } |
306 | ||
f515e6b7 MU |
307 | static inline void _lock_rx_hashtbl(struct bonding *bond) |
308 | { | |
309 | spin_lock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock)); | |
310 | } | |
311 | ||
312 | static inline void _unlock_rx_hashtbl(struct bonding *bond) | |
313 | { | |
314 | spin_unlock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock)); | |
315 | } | |
316 | ||
1da177e4 LT |
317 | /* when an ARP REPLY is received from a client update its info |
318 | * in the rx_hashtbl | |
319 | */ | |
320 | static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp) | |
321 | { | |
322 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
323 | struct rlb_client_info *client_info; | |
324 | u32 hash_index; | |
325 | ||
f515e6b7 | 326 | _lock_rx_hashtbl_bh(bond); |
1da177e4 | 327 | |
b85b6fb1 | 328 | hash_index = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src)); |
1da177e4 LT |
329 | client_info = &(bond_info->rx_hashtbl[hash_index]); |
330 | ||
331 | if ((client_info->assigned) && | |
332 | (client_info->ip_src == arp->ip_dst) && | |
42d782ac | 333 | (client_info->ip_dst == arp->ip_src) && |
a6700db1 | 334 | (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) { |
1da177e4 | 335 | /* update the clients MAC address */ |
ada0f863 | 336 | ether_addr_copy(client_info->mac_dst, arp->mac_src); |
1da177e4 LT |
337 | client_info->ntt = 1; |
338 | bond_info->rx_ntt = 1; | |
339 | } | |
340 | ||
f515e6b7 | 341 | _unlock_rx_hashtbl_bh(bond); |
1da177e4 LT |
342 | } |
343 | ||
de063b70 ED |
344 | static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond, |
345 | struct slave *slave) | |
1da177e4 | 346 | { |
de063b70 | 347 | struct arp_pkt *arp, _arp; |
1da177e4 | 348 | |
3aba891d | 349 | if (skb->protocol != cpu_to_be16(ETH_P_ARP)) |
b99215cd | 350 | goto out; |
1da177e4 | 351 | |
de063b70 ED |
352 | arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp); |
353 | if (!arp) | |
b99215cd | 354 | goto out; |
1da177e4 | 355 | |
e53665c6 JB |
356 | /* We received an ARP from arp->ip_src. |
357 | * We might have used this IP address previously (on the bonding host | |
358 | * itself or on a system that is bridged together with the bond). | |
359 | * However, if arp->mac_src is different than what is stored in | |
360 | * rx_hashtbl, some other host is now using the IP and we must prevent | |
361 | * sending out client updates with this IP address and the old MAC | |
362 | * address. | |
363 | * Clean up all hash table entries that have this address as ip_src but | |
364 | * have a different mac_src. | |
365 | */ | |
366 | rlb_purge_src_ip(bond, arp); | |
367 | ||
1da177e4 LT |
368 | if (arp->op_code == htons(ARPOP_REPLY)) { |
369 | /* update rx hash table for this ARP */ | |
370 | rlb_update_entry_from_arp(bond, arp); | |
5a03cdb7 | 371 | pr_debug("Server received an ARP Reply from client\n"); |
1da177e4 | 372 | } |
b99215cd DM |
373 | out: |
374 | return RX_HANDLER_ANOTHER; | |
1da177e4 LT |
375 | } |
376 | ||
377 | /* Caller must hold bond lock for read */ | |
378 | static struct slave *rlb_next_rx_slave(struct bonding *bond) | |
379 | { | |
380 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
6475ae4c VF |
381 | struct slave *before = NULL, *rx_slave = NULL, *slave; |
382 | struct list_head *iter; | |
383 | bool found = false; | |
1da177e4 | 384 | |
6475ae4c VF |
385 | bond_for_each_slave(bond, slave, iter) { |
386 | if (!SLAVE_IS_OK(slave)) | |
387 | continue; | |
388 | if (!found) { | |
389 | if (!before || before->speed < slave->speed) | |
390 | before = slave; | |
391 | } else { | |
392 | if (!rx_slave || rx_slave->speed < slave->speed) | |
1da177e4 | 393 | rx_slave = slave; |
1da177e4 | 394 | } |
6475ae4c VF |
395 | if (slave == bond_info->rx_slave) |
396 | found = true; | |
1da177e4 | 397 | } |
6475ae4c VF |
398 | /* we didn't find anything after the current or we have something |
399 | * better before and up to the current slave | |
400 | */ | |
401 | if (!rx_slave || (before && rx_slave->speed < before->speed)) | |
402 | rx_slave = before; | |
1da177e4 | 403 | |
6475ae4c VF |
404 | if (rx_slave) |
405 | bond_info->rx_slave = rx_slave; | |
1da177e4 LT |
406 | |
407 | return rx_slave; | |
408 | } | |
409 | ||
28c71926 | 410 | /* Caller must hold rcu_read_lock() for read */ |
411 | static struct slave *__rlb_next_rx_slave(struct bonding *bond) | |
412 | { | |
413 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
414 | struct slave *before = NULL, *rx_slave = NULL, *slave; | |
415 | struct list_head *iter; | |
416 | bool found = false; | |
417 | ||
418 | bond_for_each_slave_rcu(bond, slave, iter) { | |
419 | if (!SLAVE_IS_OK(slave)) | |
420 | continue; | |
421 | if (!found) { | |
422 | if (!before || before->speed < slave->speed) | |
423 | before = slave; | |
424 | } else { | |
425 | if (!rx_slave || rx_slave->speed < slave->speed) | |
426 | rx_slave = slave; | |
427 | } | |
428 | if (slave == bond_info->rx_slave) | |
429 | found = true; | |
430 | } | |
431 | /* we didn't find anything after the current or we have something | |
432 | * better before and up to the current slave | |
433 | */ | |
434 | if (!rx_slave || (before && rx_slave->speed < before->speed)) | |
435 | rx_slave = before; | |
436 | ||
437 | if (rx_slave) | |
438 | bond_info->rx_slave = rx_slave; | |
439 | ||
440 | return rx_slave; | |
441 | } | |
442 | ||
1da177e4 LT |
443 | /* teach the switch the mac of a disabled slave |
444 | * on the primary for fault tolerance | |
445 | * | |
446 | * Caller must hold bond->curr_slave_lock for write or bond lock for write | |
447 | */ | |
448 | static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[]) | |
449 | { | |
fdb89d75 | 450 | if (!bond->curr_active_slave) |
1da177e4 | 451 | return; |
1da177e4 LT |
452 | |
453 | if (!bond->alb_info.primary_is_promisc) { | |
7e1a1ac1 WC |
454 | if (!dev_set_promiscuity(bond->curr_active_slave->dev, 1)) |
455 | bond->alb_info.primary_is_promisc = 1; | |
456 | else | |
457 | bond->alb_info.primary_is_promisc = 0; | |
1da177e4 LT |
458 | } |
459 | ||
460 | bond->alb_info.rlb_promisc_timeout_counter = 0; | |
461 | ||
462 | alb_send_learning_packets(bond->curr_active_slave, addr); | |
463 | } | |
464 | ||
465 | /* slave being removed should not be active at this point | |
466 | * | |
b2e7aceb | 467 | * Caller must hold rtnl. |
1da177e4 LT |
468 | */ |
469 | static void rlb_clear_slave(struct bonding *bond, struct slave *slave) | |
470 | { | |
471 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
472 | struct rlb_client_info *rx_hash_table; | |
473 | u32 index, next_index; | |
474 | ||
475 | /* clear slave from rx_hashtbl */ | |
f515e6b7 | 476 | _lock_rx_hashtbl_bh(bond); |
1da177e4 LT |
477 | |
478 | rx_hash_table = bond_info->rx_hashtbl; | |
e53665c6 | 479 | index = bond_info->rx_hashtbl_used_head; |
1da177e4 | 480 | for (; index != RLB_NULL_INDEX; index = next_index) { |
e53665c6 | 481 | next_index = rx_hash_table[index].used_next; |
1da177e4 LT |
482 | if (rx_hash_table[index].slave == slave) { |
483 | struct slave *assigned_slave = rlb_next_rx_slave(bond); | |
484 | ||
485 | if (assigned_slave) { | |
486 | rx_hash_table[index].slave = assigned_slave; | |
a6700db1 JP |
487 | if (!ether_addr_equal_64bits(rx_hash_table[index].mac_dst, |
488 | mac_bcast)) { | |
1da177e4 LT |
489 | bond_info->rx_hashtbl[index].ntt = 1; |
490 | bond_info->rx_ntt = 1; | |
491 | /* A slave has been removed from the | |
492 | * table because it is either disabled | |
493 | * or being released. We must retry the | |
494 | * update to avoid clients from not | |
495 | * being updated & disconnecting when | |
496 | * there is stress | |
497 | */ | |
498 | bond_info->rlb_update_retry_counter = | |
499 | RLB_UPDATE_RETRY; | |
500 | } | |
501 | } else { /* there is no active slave */ | |
502 | rx_hash_table[index].slave = NULL; | |
503 | } | |
504 | } | |
505 | } | |
506 | ||
f515e6b7 | 507 | _unlock_rx_hashtbl_bh(bond); |
1da177e4 | 508 | |
6603a6f2 | 509 | write_lock_bh(&bond->curr_slave_lock); |
1da177e4 | 510 | |
fdb89d75 | 511 | if (slave != bond->curr_active_slave) |
1da177e4 | 512 | rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr); |
1da177e4 | 513 | |
6603a6f2 | 514 | write_unlock_bh(&bond->curr_slave_lock); |
1da177e4 LT |
515 | } |
516 | ||
517 | static void rlb_update_client(struct rlb_client_info *client_info) | |
518 | { | |
519 | int i; | |
520 | ||
35d75ee4 | 521 | if (!client_info->slave) |
1da177e4 | 522 | return; |
1da177e4 LT |
523 | |
524 | for (i = 0; i < RLB_ARP_BURST_SIZE; i++) { | |
525 | struct sk_buff *skb; | |
526 | ||
527 | skb = arp_create(ARPOP_REPLY, ETH_P_ARP, | |
528 | client_info->ip_dst, | |
529 | client_info->slave->dev, | |
530 | client_info->ip_src, | |
531 | client_info->mac_dst, | |
532 | client_info->slave->dev->dev_addr, | |
533 | client_info->mac_dst); | |
534 | if (!skb) { | |
a4aee5c8 | 535 | pr_err("%s: Error: failed to create an ARP packet\n", |
471cb5a3 | 536 | client_info->slave->bond->dev->name); |
1da177e4 LT |
537 | continue; |
538 | } | |
539 | ||
540 | skb->dev = client_info->slave->dev; | |
541 | ||
d3ab3ffd | 542 | if (client_info->vlan_id) { |
86a9bad3 | 543 | skb = vlan_put_tag(skb, htons(ETH_P_8021Q), client_info->vlan_id); |
1da177e4 | 544 | if (!skb) { |
a4aee5c8 | 545 | pr_err("%s: Error: failed to insert VLAN tag\n", |
471cb5a3 | 546 | client_info->slave->bond->dev->name); |
1da177e4 LT |
547 | continue; |
548 | } | |
549 | } | |
550 | ||
551 | arp_xmit(skb); | |
552 | } | |
553 | } | |
554 | ||
555 | /* sends ARP REPLIES that update the clients that need updating */ | |
556 | static void rlb_update_rx_clients(struct bonding *bond) | |
557 | { | |
558 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
559 | struct rlb_client_info *client_info; | |
560 | u32 hash_index; | |
561 | ||
f515e6b7 | 562 | _lock_rx_hashtbl_bh(bond); |
1da177e4 | 563 | |
e53665c6 JB |
564 | hash_index = bond_info->rx_hashtbl_used_head; |
565 | for (; hash_index != RLB_NULL_INDEX; | |
566 | hash_index = client_info->used_next) { | |
1da177e4 LT |
567 | client_info = &(bond_info->rx_hashtbl[hash_index]); |
568 | if (client_info->ntt) { | |
569 | rlb_update_client(client_info); | |
35d75ee4 | 570 | if (bond_info->rlb_update_retry_counter == 0) |
1da177e4 | 571 | client_info->ntt = 0; |
1da177e4 LT |
572 | } |
573 | } | |
574 | ||
94e2bd68 | 575 | /* do not update the entries again until this counter is zero so that |
1da177e4 LT |
576 | * not to confuse the clients. |
577 | */ | |
578 | bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY; | |
579 | ||
f515e6b7 | 580 | _unlock_rx_hashtbl_bh(bond); |
1da177e4 LT |
581 | } |
582 | ||
583 | /* The slave was assigned a new mac address - update the clients */ | |
584 | static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave) | |
585 | { | |
586 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
587 | struct rlb_client_info *client_info; | |
588 | int ntt = 0; | |
589 | u32 hash_index; | |
590 | ||
f515e6b7 | 591 | _lock_rx_hashtbl_bh(bond); |
1da177e4 | 592 | |
e53665c6 JB |
593 | hash_index = bond_info->rx_hashtbl_used_head; |
594 | for (; hash_index != RLB_NULL_INDEX; | |
595 | hash_index = client_info->used_next) { | |
1da177e4 LT |
596 | client_info = &(bond_info->rx_hashtbl[hash_index]); |
597 | ||
598 | if ((client_info->slave == slave) && | |
a6700db1 | 599 | !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) { |
1da177e4 LT |
600 | client_info->ntt = 1; |
601 | ntt = 1; | |
602 | } | |
603 | } | |
604 | ||
4708a1b1 | 605 | /* update the team's flag only after the whole iteration */ |
1da177e4 LT |
606 | if (ntt) { |
607 | bond_info->rx_ntt = 1; | |
4708a1b1 | 608 | /* fasten the change */ |
1da177e4 LT |
609 | bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY; |
610 | } | |
611 | ||
f515e6b7 | 612 | _unlock_rx_hashtbl_bh(bond); |
1da177e4 LT |
613 | } |
614 | ||
615 | /* mark all clients using src_ip to be updated */ | |
d3bb52b0 | 616 | static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip) |
1da177e4 LT |
617 | { |
618 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
619 | struct rlb_client_info *client_info; | |
620 | u32 hash_index; | |
621 | ||
622 | _lock_rx_hashtbl(bond); | |
623 | ||
e53665c6 JB |
624 | hash_index = bond_info->rx_hashtbl_used_head; |
625 | for (; hash_index != RLB_NULL_INDEX; | |
626 | hash_index = client_info->used_next) { | |
1da177e4 LT |
627 | client_info = &(bond_info->rx_hashtbl[hash_index]); |
628 | ||
629 | if (!client_info->slave) { | |
a4aee5c8 | 630 | pr_err("%s: Error: found a client with no channel in the client's hash table\n", |
4e0952c7 | 631 | bond->dev->name); |
1da177e4 LT |
632 | continue; |
633 | } | |
634 | /*update all clients using this src_ip, that are not assigned | |
635 | * to the team's address (curr_active_slave) and have a known | |
636 | * unicast mac address. | |
637 | */ | |
638 | if ((client_info->ip_src == src_ip) && | |
a6700db1 JP |
639 | !ether_addr_equal_64bits(client_info->slave->dev->dev_addr, |
640 | bond->dev->dev_addr) && | |
641 | !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) { | |
1da177e4 LT |
642 | client_info->ntt = 1; |
643 | bond_info->rx_ntt = 1; | |
644 | } | |
645 | } | |
646 | ||
647 | _unlock_rx_hashtbl(bond); | |
648 | } | |
649 | ||
650 | /* Caller must hold both bond and ptr locks for read */ | |
651 | static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond) | |
652 | { | |
653 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
a16aeb36 | 654 | struct arp_pkt *arp = arp_pkt(skb); |
28c71926 | 655 | struct slave *assigned_slave, *curr_active_slave; |
1da177e4 LT |
656 | struct rlb_client_info *client_info; |
657 | u32 hash_index = 0; | |
658 | ||
659 | _lock_rx_hashtbl(bond); | |
660 | ||
28c71926 | 661 | curr_active_slave = rcu_dereference(bond->curr_active_slave); |
662 | ||
e364a341 | 663 | hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst)); |
1da177e4 LT |
664 | client_info = &(bond_info->rx_hashtbl[hash_index]); |
665 | ||
666 | if (client_info->assigned) { | |
667 | if ((client_info->ip_src == arp->ip_src) && | |
668 | (client_info->ip_dst == arp->ip_dst)) { | |
669 | /* the entry is already assigned to this client */ | |
a6700db1 | 670 | if (!ether_addr_equal_64bits(arp->mac_dst, mac_bcast)) { |
1da177e4 | 671 | /* update mac address from arp */ |
ada0f863 | 672 | ether_addr_copy(client_info->mac_dst, arp->mac_dst); |
1da177e4 | 673 | } |
ada0f863 | 674 | ether_addr_copy(client_info->mac_src, arp->mac_src); |
1da177e4 LT |
675 | |
676 | assigned_slave = client_info->slave; | |
677 | if (assigned_slave) { | |
678 | _unlock_rx_hashtbl(bond); | |
679 | return assigned_slave; | |
680 | } | |
681 | } else { | |
682 | /* the entry is already assigned to some other client, | |
683 | * move the old client to primary (curr_active_slave) so | |
684 | * that the new client can be assigned to this entry. | |
685 | */ | |
686 | if (bond->curr_active_slave && | |
28c71926 | 687 | client_info->slave != curr_active_slave) { |
688 | client_info->slave = curr_active_slave; | |
1da177e4 LT |
689 | rlb_update_client(client_info); |
690 | } | |
691 | } | |
692 | } | |
693 | /* assign a new slave */ | |
28c71926 | 694 | assigned_slave = __rlb_next_rx_slave(bond); |
1da177e4 LT |
695 | |
696 | if (assigned_slave) { | |
e53665c6 JB |
697 | if (!(client_info->assigned && |
698 | client_info->ip_src == arp->ip_src)) { | |
699 | /* ip_src is going to be updated, | |
700 | * fix the src hash list | |
701 | */ | |
702 | u32 hash_src = _simple_hash((u8 *)&arp->ip_src, | |
703 | sizeof(arp->ip_src)); | |
704 | rlb_src_unlink(bond, hash_index); | |
705 | rlb_src_link(bond, hash_src, hash_index); | |
706 | } | |
707 | ||
1da177e4 LT |
708 | client_info->ip_src = arp->ip_src; |
709 | client_info->ip_dst = arp->ip_dst; | |
710 | /* arp->mac_dst is broadcast for arp reqeusts. | |
711 | * will be updated with clients actual unicast mac address | |
712 | * upon receiving an arp reply. | |
713 | */ | |
ada0f863 JP |
714 | ether_addr_copy(client_info->mac_dst, arp->mac_dst); |
715 | ether_addr_copy(client_info->mac_src, arp->mac_src); | |
1da177e4 LT |
716 | client_info->slave = assigned_slave; |
717 | ||
a6700db1 | 718 | if (!ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) { |
1da177e4 LT |
719 | client_info->ntt = 1; |
720 | bond->alb_info.rx_ntt = 1; | |
721 | } else { | |
722 | client_info->ntt = 0; | |
723 | } | |
724 | ||
fb00bc2e | 725 | if (vlan_get_tag(skb, &client_info->vlan_id)) |
d3ab3ffd | 726 | client_info->vlan_id = 0; |
1da177e4 LT |
727 | |
728 | if (!client_info->assigned) { | |
e53665c6 JB |
729 | u32 prev_tbl_head = bond_info->rx_hashtbl_used_head; |
730 | bond_info->rx_hashtbl_used_head = hash_index; | |
731 | client_info->used_next = prev_tbl_head; | |
1da177e4 | 732 | if (prev_tbl_head != RLB_NULL_INDEX) { |
e53665c6 | 733 | bond_info->rx_hashtbl[prev_tbl_head].used_prev = |
1da177e4 LT |
734 | hash_index; |
735 | } | |
736 | client_info->assigned = 1; | |
737 | } | |
738 | } | |
739 | ||
740 | _unlock_rx_hashtbl(bond); | |
741 | ||
742 | return assigned_slave; | |
743 | } | |
744 | ||
745 | /* chooses (and returns) transmit channel for arp reply | |
746 | * does not choose channel for other arp types since they are | |
747 | * sent on the curr_active_slave | |
748 | */ | |
749 | static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond) | |
750 | { | |
a16aeb36 | 751 | struct arp_pkt *arp = arp_pkt(skb); |
1da177e4 LT |
752 | struct slave *tx_slave = NULL; |
753 | ||
567b871e | 754 | /* Don't modify or load balance ARPs that do not originate locally |
755 | * (e.g.,arrive via a bridge). | |
756 | */ | |
28c71926 | 757 | if (!bond_slave_has_mac_rcu(bond, arp->mac_src)) |
567b871e | 758 | return NULL; |
759 | ||
f14c4e4e | 760 | if (arp->op_code == htons(ARPOP_REPLY)) { |
1da177e4 LT |
761 | /* the arp must be sent on the selected |
762 | * rx channel | |
763 | */ | |
764 | tx_slave = rlb_choose_channel(skb, bond); | |
35d75ee4 | 765 | if (tx_slave) |
ada0f863 | 766 | ether_addr_copy(arp->mac_src, tx_slave->dev->dev_addr); |
5a03cdb7 | 767 | pr_debug("Server sent ARP Reply packet\n"); |
f14c4e4e | 768 | } else if (arp->op_code == htons(ARPOP_REQUEST)) { |
1da177e4 LT |
769 | /* Create an entry in the rx_hashtbl for this client as a |
770 | * place holder. | |
771 | * When the arp reply is received the entry will be updated | |
772 | * with the correct unicast address of the client. | |
773 | */ | |
774 | rlb_choose_channel(skb, bond); | |
775 | ||
77c8e2c0 | 776 | /* The ARP reply packets must be delayed so that |
1da177e4 LT |
777 | * they can cancel out the influence of the ARP request. |
778 | */ | |
779 | bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY; | |
780 | ||
781 | /* arp requests are broadcast and are sent on the primary | |
782 | * the arp request will collapse all clients on the subnet to | |
783 | * the primary slave. We must register these clients to be | |
784 | * updated with their assigned mac. | |
785 | */ | |
786 | rlb_req_update_subnet_clients(bond, arp->ip_src); | |
5a03cdb7 | 787 | pr_debug("Server sent ARP Request packet\n"); |
1da177e4 LT |
788 | } |
789 | ||
790 | return tx_slave; | |
791 | } | |
792 | ||
793 | /* Caller must hold bond lock for read */ | |
794 | static void rlb_rebalance(struct bonding *bond) | |
795 | { | |
796 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
797 | struct slave *assigned_slave; | |
798 | struct rlb_client_info *client_info; | |
799 | int ntt; | |
800 | u32 hash_index; | |
801 | ||
f515e6b7 | 802 | _lock_rx_hashtbl_bh(bond); |
1da177e4 LT |
803 | |
804 | ntt = 0; | |
e53665c6 JB |
805 | hash_index = bond_info->rx_hashtbl_used_head; |
806 | for (; hash_index != RLB_NULL_INDEX; | |
807 | hash_index = client_info->used_next) { | |
1da177e4 | 808 | client_info = &(bond_info->rx_hashtbl[hash_index]); |
733ab639 | 809 | assigned_slave = __rlb_next_rx_slave(bond); |
1da177e4 LT |
810 | if (assigned_slave && (client_info->slave != assigned_slave)) { |
811 | client_info->slave = assigned_slave; | |
812 | client_info->ntt = 1; | |
813 | ntt = 1; | |
814 | } | |
815 | } | |
816 | ||
817 | /* update the team's flag only after the whole iteration */ | |
35d75ee4 | 818 | if (ntt) |
1da177e4 | 819 | bond_info->rx_ntt = 1; |
f515e6b7 | 820 | _unlock_rx_hashtbl_bh(bond); |
1da177e4 LT |
821 | } |
822 | ||
823 | /* Caller must hold rx_hashtbl lock */ | |
e53665c6 JB |
824 | static void rlb_init_table_entry_dst(struct rlb_client_info *entry) |
825 | { | |
826 | entry->used_next = RLB_NULL_INDEX; | |
827 | entry->used_prev = RLB_NULL_INDEX; | |
828 | entry->assigned = 0; | |
829 | entry->slave = NULL; | |
d3ab3ffd | 830 | entry->vlan_id = 0; |
e53665c6 JB |
831 | } |
832 | static void rlb_init_table_entry_src(struct rlb_client_info *entry) | |
833 | { | |
834 | entry->src_first = RLB_NULL_INDEX; | |
835 | entry->src_prev = RLB_NULL_INDEX; | |
836 | entry->src_next = RLB_NULL_INDEX; | |
837 | } | |
838 | ||
1da177e4 LT |
839 | static void rlb_init_table_entry(struct rlb_client_info *entry) |
840 | { | |
841 | memset(entry, 0, sizeof(struct rlb_client_info)); | |
e53665c6 JB |
842 | rlb_init_table_entry_dst(entry); |
843 | rlb_init_table_entry_src(entry); | |
844 | } | |
845 | ||
846 | static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index) | |
847 | { | |
848 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
849 | u32 next_index = bond_info->rx_hashtbl[index].used_next; | |
850 | u32 prev_index = bond_info->rx_hashtbl[index].used_prev; | |
851 | ||
852 | if (index == bond_info->rx_hashtbl_used_head) | |
853 | bond_info->rx_hashtbl_used_head = next_index; | |
854 | if (prev_index != RLB_NULL_INDEX) | |
855 | bond_info->rx_hashtbl[prev_index].used_next = next_index; | |
856 | if (next_index != RLB_NULL_INDEX) | |
857 | bond_info->rx_hashtbl[next_index].used_prev = prev_index; | |
858 | } | |
859 | ||
860 | /* unlink a rlb hash table entry from the src list */ | |
861 | static void rlb_src_unlink(struct bonding *bond, u32 index) | |
862 | { | |
863 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
864 | u32 next_index = bond_info->rx_hashtbl[index].src_next; | |
865 | u32 prev_index = bond_info->rx_hashtbl[index].src_prev; | |
866 | ||
867 | bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX; | |
868 | bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX; | |
869 | ||
870 | if (next_index != RLB_NULL_INDEX) | |
871 | bond_info->rx_hashtbl[next_index].src_prev = prev_index; | |
872 | ||
873 | if (prev_index == RLB_NULL_INDEX) | |
874 | return; | |
875 | ||
876 | /* is prev_index pointing to the head of this list? */ | |
877 | if (bond_info->rx_hashtbl[prev_index].src_first == index) | |
878 | bond_info->rx_hashtbl[prev_index].src_first = next_index; | |
879 | else | |
880 | bond_info->rx_hashtbl[prev_index].src_next = next_index; | |
881 | ||
882 | } | |
883 | ||
884 | static void rlb_delete_table_entry(struct bonding *bond, u32 index) | |
885 | { | |
886 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
887 | struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]); | |
888 | ||
889 | rlb_delete_table_entry_dst(bond, index); | |
890 | rlb_init_table_entry_dst(entry); | |
891 | ||
892 | rlb_src_unlink(bond, index); | |
893 | } | |
894 | ||
895 | /* add the rx_hashtbl[ip_dst_hash] entry to the list | |
896 | * of entries with identical ip_src_hash | |
897 | */ | |
898 | static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash) | |
899 | { | |
900 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
901 | u32 next; | |
902 | ||
903 | bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash; | |
904 | next = bond_info->rx_hashtbl[ip_src_hash].src_first; | |
905 | bond_info->rx_hashtbl[ip_dst_hash].src_next = next; | |
906 | if (next != RLB_NULL_INDEX) | |
907 | bond_info->rx_hashtbl[next].src_prev = ip_dst_hash; | |
908 | bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash; | |
909 | } | |
910 | ||
911 | /* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does | |
912 | * not match arp->mac_src */ | |
913 | static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp) | |
914 | { | |
915 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
b85b6fb1 | 916 | u32 ip_src_hash = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src)); |
e53665c6 JB |
917 | u32 index; |
918 | ||
919 | _lock_rx_hashtbl_bh(bond); | |
920 | ||
921 | index = bond_info->rx_hashtbl[ip_src_hash].src_first; | |
922 | while (index != RLB_NULL_INDEX) { | |
923 | struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]); | |
924 | u32 next_index = entry->src_next; | |
925 | if (entry->ip_src == arp->ip_src && | |
926 | !ether_addr_equal_64bits(arp->mac_src, entry->mac_src)) | |
927 | rlb_delete_table_entry(bond, index); | |
928 | index = next_index; | |
929 | } | |
930 | _unlock_rx_hashtbl_bh(bond); | |
1da177e4 LT |
931 | } |
932 | ||
933 | static int rlb_initialize(struct bonding *bond) | |
934 | { | |
935 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
0d206a3a | 936 | struct rlb_client_info *new_hashtbl; |
1da177e4 LT |
937 | int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info); |
938 | int i; | |
939 | ||
0d206a3a | 940 | new_hashtbl = kmalloc(size, GFP_KERNEL); |
e404decb | 941 | if (!new_hashtbl) |
1da177e4 | 942 | return -1; |
e404decb | 943 | |
f515e6b7 | 944 | _lock_rx_hashtbl_bh(bond); |
0d206a3a MW |
945 | |
946 | bond_info->rx_hashtbl = new_hashtbl; | |
1da177e4 | 947 | |
e53665c6 | 948 | bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX; |
1da177e4 | 949 | |
35d75ee4 | 950 | for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) |
1da177e4 | 951 | rlb_init_table_entry(bond_info->rx_hashtbl + i); |
1da177e4 | 952 | |
f515e6b7 | 953 | _unlock_rx_hashtbl_bh(bond); |
1da177e4 | 954 | |
1da177e4 | 955 | /* register to receive ARPs */ |
3aba891d | 956 | bond->recv_probe = rlb_arp_recv; |
1da177e4 LT |
957 | |
958 | return 0; | |
959 | } | |
960 | ||
961 | static void rlb_deinitialize(struct bonding *bond) | |
962 | { | |
963 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
964 | ||
f515e6b7 | 965 | _lock_rx_hashtbl_bh(bond); |
1da177e4 LT |
966 | |
967 | kfree(bond_info->rx_hashtbl); | |
968 | bond_info->rx_hashtbl = NULL; | |
e53665c6 | 969 | bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX; |
1da177e4 | 970 | |
f515e6b7 | 971 | _unlock_rx_hashtbl_bh(bond); |
1da177e4 LT |
972 | } |
973 | ||
974 | static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id) | |
975 | { | |
976 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
977 | u32 curr_index; | |
978 | ||
f515e6b7 | 979 | _lock_rx_hashtbl_bh(bond); |
1da177e4 | 980 | |
e53665c6 | 981 | curr_index = bond_info->rx_hashtbl_used_head; |
1da177e4 LT |
982 | while (curr_index != RLB_NULL_INDEX) { |
983 | struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]); | |
e53665c6 | 984 | u32 next_index = bond_info->rx_hashtbl[curr_index].used_next; |
1da177e4 | 985 | |
d3ab3ffd | 986 | if (curr->vlan_id == vlan_id) |
e53665c6 | 987 | rlb_delete_table_entry(bond, curr_index); |
1da177e4 LT |
988 | |
989 | curr_index = next_index; | |
990 | } | |
991 | ||
f515e6b7 | 992 | _unlock_rx_hashtbl_bh(bond); |
1da177e4 LT |
993 | } |
994 | ||
995 | /*********************** tlb/rlb shared functions *********************/ | |
996 | ||
7aa64981 VF |
997 | static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[], |
998 | u16 vid) | |
1da177e4 | 999 | { |
1da177e4 | 1000 | struct learning_pkt pkt; |
7aa64981 | 1001 | struct sk_buff *skb; |
1da177e4 | 1002 | int size = sizeof(struct learning_pkt); |
7aa64981 | 1003 | char *data; |
1da177e4 LT |
1004 | |
1005 | memset(&pkt, 0, size); | |
ada0f863 JP |
1006 | ether_addr_copy(pkt.mac_dst, mac_addr); |
1007 | ether_addr_copy(pkt.mac_src, mac_addr); | |
96a0922c | 1008 | pkt.type = cpu_to_be16(ETH_P_LOOPBACK); |
1da177e4 | 1009 | |
7aa64981 VF |
1010 | skb = dev_alloc_skb(size); |
1011 | if (!skb) | |
1012 | return; | |
1da177e4 | 1013 | |
7aa64981 VF |
1014 | data = skb_put(skb, size); |
1015 | memcpy(data, &pkt, size); | |
1016 | ||
1017 | skb_reset_mac_header(skb); | |
1018 | skb->network_header = skb->mac_header + ETH_HLEN; | |
1019 | skb->protocol = pkt.type; | |
1020 | skb->priority = TC_PRIO_CONTROL; | |
1021 | skb->dev = slave->dev; | |
1022 | ||
1023 | if (vid) { | |
1024 | skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vid); | |
1da177e4 | 1025 | if (!skb) { |
7aa64981 VF |
1026 | pr_err("%s: Error: failed to insert VLAN tag\n", |
1027 | slave->bond->dev->name); | |
1da177e4 LT |
1028 | return; |
1029 | } | |
7aa64981 VF |
1030 | } |
1031 | ||
1032 | dev_queue_xmit(skb); | |
1033 | } | |
1da177e4 | 1034 | |
1da177e4 | 1035 | |
7aa64981 VF |
1036 | static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[]) |
1037 | { | |
1038 | struct bonding *bond = bond_get_bond_by_slave(slave); | |
5bf94b83 VF |
1039 | struct net_device *upper; |
1040 | struct list_head *iter; | |
1041 | ||
1042 | /* send untagged */ | |
1043 | alb_send_lp_vid(slave, mac_addr, 0); | |
1044 | ||
1045 | /* loop through vlans and send one packet for each */ | |
1046 | rcu_read_lock(); | |
2f268f12 | 1047 | netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) { |
5bf94b83 VF |
1048 | if (upper->priv_flags & IFF_802_1Q_VLAN) |
1049 | alb_send_lp_vid(slave, mac_addr, | |
1050 | vlan_dev_vlan_id(upper)); | |
1da177e4 | 1051 | } |
5bf94b83 | 1052 | rcu_read_unlock(); |
1da177e4 LT |
1053 | } |
1054 | ||
b924551b | 1055 | static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[]) |
1da177e4 LT |
1056 | { |
1057 | struct net_device *dev = slave->dev; | |
1058 | struct sockaddr s_addr; | |
1059 | ||
b924551b | 1060 | if (slave->bond->params.mode == BOND_MODE_TLB) { |
1da177e4 LT |
1061 | memcpy(dev->dev_addr, addr, dev->addr_len); |
1062 | return 0; | |
1063 | } | |
1064 | ||
1065 | /* for rlb each slave must have a unique hw mac addresses so that */ | |
1066 | /* each slave will receive packets destined to a different mac */ | |
1067 | memcpy(s_addr.sa_data, addr, dev->addr_len); | |
1068 | s_addr.sa_family = dev->type; | |
1069 | if (dev_set_mac_address(dev, &s_addr)) { | |
a4aee5c8 JP |
1070 | pr_err("%s: Error: dev_set_mac_address of dev %s failed!\n" |
1071 | "ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n", | |
471cb5a3 | 1072 | slave->bond->dev->name, dev->name); |
1da177e4 LT |
1073 | return -EOPNOTSUPP; |
1074 | } | |
1075 | return 0; | |
1076 | } | |
1077 | ||
059fe7a5 JV |
1078 | /* |
1079 | * Swap MAC addresses between two slaves. | |
1080 | * | |
1081 | * Called with RTNL held, and no other locks. | |
1082 | * | |
1083 | */ | |
1084 | ||
43547ea6 | 1085 | static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2) |
1da177e4 | 1086 | { |
1da177e4 | 1087 | u8 tmp_mac_addr[ETH_ALEN]; |
1da177e4 | 1088 | |
ada0f863 | 1089 | ether_addr_copy(tmp_mac_addr, slave1->dev->dev_addr); |
b924551b JB |
1090 | alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr); |
1091 | alb_set_slave_mac_addr(slave2, tmp_mac_addr); | |
1da177e4 | 1092 | |
059fe7a5 JV |
1093 | } |
1094 | ||
1095 | /* | |
1096 | * Send learning packets after MAC address swap. | |
1097 | * | |
2543331d | 1098 | * Called with RTNL and no other locks |
059fe7a5 JV |
1099 | */ |
1100 | static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1, | |
1101 | struct slave *slave2) | |
1102 | { | |
1103 | int slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2)); | |
1104 | struct slave *disabled_slave = NULL; | |
1105 | ||
2543331d JV |
1106 | ASSERT_RTNL(); |
1107 | ||
1da177e4 LT |
1108 | /* fasten the change in the switch */ |
1109 | if (SLAVE_IS_OK(slave1)) { | |
1110 | alb_send_learning_packets(slave1, slave1->dev->dev_addr); | |
1111 | if (bond->alb_info.rlb_enabled) { | |
1112 | /* inform the clients that the mac address | |
1113 | * has changed | |
1114 | */ | |
1115 | rlb_req_update_slave_clients(bond, slave1); | |
1116 | } | |
1117 | } else { | |
1118 | disabled_slave = slave1; | |
1119 | } | |
1120 | ||
1121 | if (SLAVE_IS_OK(slave2)) { | |
1122 | alb_send_learning_packets(slave2, slave2->dev->dev_addr); | |
1123 | if (bond->alb_info.rlb_enabled) { | |
1124 | /* inform the clients that the mac address | |
1125 | * has changed | |
1126 | */ | |
1127 | rlb_req_update_slave_clients(bond, slave2); | |
1128 | } | |
1129 | } else { | |
1130 | disabled_slave = slave2; | |
1131 | } | |
1132 | ||
1133 | if (bond->alb_info.rlb_enabled && slaves_state_differ) { | |
1134 | /* A disabled slave was assigned an active mac addr */ | |
1135 | rlb_teach_disabled_mac_on_primary(bond, | |
1136 | disabled_slave->dev->dev_addr); | |
1137 | } | |
1138 | } | |
1139 | ||
1140 | /** | |
1141 | * alb_change_hw_addr_on_detach | |
1142 | * @bond: bonding we're working on | |
1143 | * @slave: the slave that was just detached | |
1144 | * | |
1145 | * We assume that @slave was already detached from the slave list. | |
1146 | * | |
1147 | * If @slave's permanent hw address is different both from its current | |
1148 | * address and from @bond's address, then somewhere in the bond there's | |
1149 | * a slave that has @slave's permanet address as its current address. | |
1150 | * We'll make sure that that slave no longer uses @slave's permanent address. | |
1151 | * | |
2543331d | 1152 | * Caller must hold RTNL and no other locks |
1da177e4 LT |
1153 | */ |
1154 | static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave) | |
1155 | { | |
1156 | int perm_curr_diff; | |
1157 | int perm_bond_diff; | |
b88ec38d | 1158 | struct slave *found_slave; |
1da177e4 | 1159 | |
a6700db1 JP |
1160 | perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr, |
1161 | slave->dev->dev_addr); | |
1162 | perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr, | |
1163 | bond->dev->dev_addr); | |
1da177e4 LT |
1164 | |
1165 | if (perm_curr_diff && perm_bond_diff) { | |
b88ec38d | 1166 | found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr); |
1da177e4 | 1167 | |
b88ec38d | 1168 | if (found_slave) { |
059fe7a5 | 1169 | /* locking: needs RTNL and nothing else */ |
b88ec38d VF |
1170 | alb_swap_mac_addr(slave, found_slave); |
1171 | alb_fasten_mac_swap(bond, slave, found_slave); | |
1da177e4 LT |
1172 | } |
1173 | } | |
1174 | } | |
1175 | ||
1176 | /** | |
1177 | * alb_handle_addr_collision_on_attach | |
1178 | * @bond: bonding we're working on | |
1179 | * @slave: the slave that was just attached | |
1180 | * | |
1181 | * checks uniqueness of slave's mac address and handles the case the | |
1182 | * new slave uses the bonds mac address. | |
1183 | * | |
1184 | * If the permanent hw address of @slave is @bond's hw address, we need to | |
1185 | * find a different hw address to give @slave, that isn't in use by any other | |
77c8e2c0 | 1186 | * slave in the bond. This address must be, of course, one of the permanent |
1da177e4 LT |
1187 | * addresses of the other slaves. |
1188 | * | |
1189 | * We go over the slave list, and for each slave there we compare its | |
1190 | * permanent hw address with the current address of all the other slaves. | |
1191 | * If no match was found, then we've found a slave with a permanent address | |
1192 | * that isn't used by any other slave in the bond, so we can assign it to | |
1193 | * @slave. | |
1194 | * | |
1195 | * assumption: this function is called before @slave is attached to the | |
cedb743f | 1196 | * bond slave list. |
1da177e4 LT |
1197 | */ |
1198 | static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave) | |
1199 | { | |
1da177e4 | 1200 | struct slave *has_bond_addr = bond->curr_active_slave; |
9caff1e7 VF |
1201 | struct slave *tmp_slave1, *free_mac_slave = NULL; |
1202 | struct list_head *iter; | |
1da177e4 | 1203 | |
0965a1f3 | 1204 | if (!bond_has_slaves(bond)) { |
1da177e4 LT |
1205 | /* this is the first slave */ |
1206 | return 0; | |
1207 | } | |
1208 | ||
1209 | /* if slave's mac address differs from bond's mac address | |
1210 | * check uniqueness of slave's mac address against the other | |
1211 | * slaves in the bond. | |
1212 | */ | |
a6700db1 | 1213 | if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) { |
cedb743f | 1214 | if (!bond_slave_has_mac(bond, slave->dev->dev_addr)) |
6b38aefe | 1215 | return 0; |
1da177e4 | 1216 | |
6b38aefe JL |
1217 | /* Try setting slave mac to bond address and fall-through |
1218 | to code handling that situation below... */ | |
b924551b | 1219 | alb_set_slave_mac_addr(slave, bond->dev->dev_addr); |
1da177e4 LT |
1220 | } |
1221 | ||
1222 | /* The slave's address is equal to the address of the bond. | |
1223 | * Search for a spare address in the bond for this slave. | |
1224 | */ | |
9caff1e7 | 1225 | bond_for_each_slave(bond, tmp_slave1, iter) { |
cedb743f | 1226 | if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) { |
1da177e4 LT |
1227 | /* no slave has tmp_slave1's perm addr |
1228 | * as its curr addr | |
1229 | */ | |
1230 | free_mac_slave = tmp_slave1; | |
1231 | break; | |
1232 | } | |
1233 | ||
1234 | if (!has_bond_addr) { | |
a6700db1 JP |
1235 | if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr, |
1236 | bond->dev->dev_addr)) { | |
1da177e4 LT |
1237 | |
1238 | has_bond_addr = tmp_slave1; | |
1239 | } | |
1240 | } | |
1241 | } | |
1242 | ||
1243 | if (free_mac_slave) { | |
b924551b | 1244 | alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr); |
1da177e4 | 1245 | |
91565ebb JP |
1246 | pr_warn("%s: Warning: the hw address of slave %s is in use by the bond; giving it the hw address of %s\n", |
1247 | bond->dev->name, slave->dev->name, | |
1248 | free_mac_slave->dev->name); | |
1da177e4 LT |
1249 | |
1250 | } else if (has_bond_addr) { | |
a4aee5c8 | 1251 | 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 | 1252 | bond->dev->name, slave->dev->name); |
1da177e4 LT |
1253 | return -EFAULT; |
1254 | } | |
1255 | ||
1256 | return 0; | |
1257 | } | |
1258 | ||
1259 | /** | |
1260 | * alb_set_mac_address | |
1261 | * @bond: | |
1262 | * @addr: | |
1263 | * | |
1264 | * In TLB mode all slaves are configured to the bond's hw address, but set | |
1265 | * their dev_addr field to different addresses (based on their permanent hw | |
1266 | * addresses). | |
1267 | * | |
1268 | * For each slave, this function sets the interface to the new address and then | |
1269 | * changes its dev_addr field to its previous value. | |
1270 | * | |
1271 | * Unwinding assumes bond's mac address has not yet changed. | |
1272 | */ | |
1273 | static int alb_set_mac_address(struct bonding *bond, void *addr) | |
1274 | { | |
81f23b13 | 1275 | struct slave *slave, *rollback_slave; |
9caff1e7 | 1276 | struct list_head *iter; |
dec1e90e | 1277 | struct sockaddr sa; |
81f23b13 | 1278 | char tmp_addr[ETH_ALEN]; |
1da177e4 | 1279 | int res; |
1da177e4 | 1280 | |
dec1e90e | 1281 | if (bond->alb_info.rlb_enabled) |
1da177e4 | 1282 | return 0; |
1da177e4 | 1283 | |
9caff1e7 | 1284 | bond_for_each_slave(bond, slave, iter) { |
1da177e4 | 1285 | /* save net_device's current hw address */ |
ada0f863 | 1286 | ether_addr_copy(tmp_addr, slave->dev->dev_addr); |
1da177e4 LT |
1287 | |
1288 | res = dev_set_mac_address(slave->dev, addr); | |
1289 | ||
1290 | /* restore net_device's hw address */ | |
ada0f863 | 1291 | ether_addr_copy(slave->dev->dev_addr, tmp_addr); |
1da177e4 | 1292 | |
eb7cc59a | 1293 | if (res) |
1da177e4 | 1294 | goto unwind; |
1da177e4 LT |
1295 | } |
1296 | ||
1297 | return 0; | |
1298 | ||
1299 | unwind: | |
1300 | memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len); | |
1301 | sa.sa_family = bond->dev->type; | |
1302 | ||
1303 | /* unwind from head to the slave that failed */ | |
9caff1e7 | 1304 | bond_for_each_slave(bond, rollback_slave, iter) { |
81f23b13 VF |
1305 | if (rollback_slave == slave) |
1306 | break; | |
ada0f863 | 1307 | ether_addr_copy(tmp_addr, rollback_slave->dev->dev_addr); |
81f23b13 | 1308 | dev_set_mac_address(rollback_slave->dev, &sa); |
ada0f863 | 1309 | ether_addr_copy(rollback_slave->dev->dev_addr, tmp_addr); |
1da177e4 LT |
1310 | } |
1311 | ||
1312 | return res; | |
1313 | } | |
1314 | ||
1315 | /************************ exported alb funcions ************************/ | |
1316 | ||
1317 | int bond_alb_initialize(struct bonding *bond, int rlb_enabled) | |
1318 | { | |
1319 | int res; | |
1320 | ||
1321 | res = tlb_initialize(bond); | |
35d75ee4 | 1322 | if (res) |
1da177e4 | 1323 | return res; |
1da177e4 LT |
1324 | |
1325 | if (rlb_enabled) { | |
1326 | bond->alb_info.rlb_enabled = 1; | |
1327 | /* initialize rlb */ | |
1328 | res = rlb_initialize(bond); | |
1329 | if (res) { | |
1330 | tlb_deinitialize(bond); | |
1331 | return res; | |
1332 | } | |
b76850ab MW |
1333 | } else { |
1334 | bond->alb_info.rlb_enabled = 0; | |
1da177e4 LT |
1335 | } |
1336 | ||
1337 | return 0; | |
1338 | } | |
1339 | ||
1340 | void bond_alb_deinitialize(struct bonding *bond) | |
1341 | { | |
1342 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
1343 | ||
1344 | tlb_deinitialize(bond); | |
1345 | ||
dda0fd5c | 1346 | if (bond_info->rlb_enabled) |
1da177e4 | 1347 | rlb_deinitialize(bond); |
1da177e4 LT |
1348 | } |
1349 | ||
9a49aba1 MB |
1350 | static int bond_do_alb_xmit(struct sk_buff *skb, struct bonding *bond, |
1351 | struct slave *tx_slave) | |
1352 | { | |
1353 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
1354 | struct ethhdr *eth_data = eth_hdr(skb); | |
1355 | ||
1356 | if (!tx_slave) { | |
1357 | /* unbalanced or unassigned, send through primary */ | |
1358 | tx_slave = rcu_dereference(bond->curr_active_slave); | |
e9f0fb88 MB |
1359 | if (bond->params.tlb_dynamic_lb) |
1360 | bond_info->unbalanced_load += skb->len; | |
9a49aba1 MB |
1361 | } |
1362 | ||
1363 | if (tx_slave && SLAVE_IS_OK(tx_slave)) { | |
1364 | if (tx_slave != rcu_dereference(bond->curr_active_slave)) { | |
1365 | ether_addr_copy(eth_data->h_source, | |
1366 | tx_slave->dev->dev_addr); | |
1367 | } | |
1368 | ||
1369 | bond_dev_queue_xmit(bond, skb, tx_slave->dev); | |
1370 | goto out; | |
1371 | } | |
1372 | ||
e9f0fb88 | 1373 | if (tx_slave && bond->params.tlb_dynamic_lb) { |
9a49aba1 MB |
1374 | _lock_tx_hashtbl(bond); |
1375 | __tlb_clear_slave(bond, tx_slave, 0); | |
1376 | _unlock_tx_hashtbl(bond); | |
1377 | } | |
1378 | ||
1379 | /* no suitable interface, frame not sent */ | |
1380 | dev_kfree_skb_any(skb); | |
1381 | out: | |
1382 | return NETDEV_TX_OK; | |
1383 | } | |
1384 | ||
f05b42ea MB |
1385 | int bond_tlb_xmit(struct sk_buff *skb, struct net_device *bond_dev) |
1386 | { | |
1387 | struct bonding *bond = netdev_priv(bond_dev); | |
1388 | struct ethhdr *eth_data; | |
1389 | struct slave *tx_slave = NULL; | |
1390 | u32 hash_index; | |
1391 | ||
1392 | skb_reset_mac_header(skb); | |
1393 | eth_data = eth_hdr(skb); | |
1394 | ||
1395 | /* Do not TX balance any multicast or broadcast */ | |
1396 | if (!is_multicast_ether_addr(eth_data->h_dest)) { | |
1397 | switch (skb->protocol) { | |
1398 | case htons(ETH_P_IP): | |
1399 | case htons(ETH_P_IPX): | |
1400 | /* In case of IPX, it will falback to L2 hash */ | |
1401 | case htons(ETH_P_IPV6): | |
1402 | hash_index = bond_xmit_hash(bond, skb); | |
e9f0fb88 MB |
1403 | if (bond->params.tlb_dynamic_lb) { |
1404 | tx_slave = tlb_choose_channel(bond, | |
1405 | hash_index & 0xFF, | |
1406 | skb->len); | |
1407 | } else { | |
1408 | struct list_head *iter; | |
1409 | int idx = hash_index % bond->slave_cnt; | |
1410 | ||
1411 | bond_for_each_slave_rcu(bond, tx_slave, iter) | |
1412 | if (--idx < 0) | |
1413 | break; | |
1414 | } | |
f05b42ea MB |
1415 | break; |
1416 | } | |
1417 | } | |
f05b42ea MB |
1418 | return bond_do_alb_xmit(skb, bond, tx_slave); |
1419 | } | |
1420 | ||
1da177e4 LT |
1421 | int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev) |
1422 | { | |
454d7c9b | 1423 | struct bonding *bond = netdev_priv(bond_dev); |
1da177e4 LT |
1424 | struct ethhdr *eth_data; |
1425 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
1426 | struct slave *tx_slave = NULL; | |
d3bb52b0 | 1427 | static const __be32 ip_bcast = htonl(0xffffffff); |
1da177e4 | 1428 | int hash_size = 0; |
9a49aba1 | 1429 | bool do_tx_balance = true; |
1da177e4 | 1430 | u32 hash_index = 0; |
eddc9ec5 | 1431 | const u8 *hash_start = NULL; |
2d1ea19d | 1432 | struct ipv6hdr *ip6hdr; |
1da177e4 | 1433 | |
459a98ed | 1434 | skb_reset_mac_header(skb); |
1da177e4 LT |
1435 | eth_data = eth_hdr(skb); |
1436 | ||
1da177e4 | 1437 | switch (ntohs(skb->protocol)) { |
eddc9ec5 ACM |
1438 | case ETH_P_IP: { |
1439 | const struct iphdr *iph = ip_hdr(skb); | |
1440 | ||
a6700db1 | 1441 | if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast) || |
eddc9ec5 ACM |
1442 | (iph->daddr == ip_bcast) || |
1443 | (iph->protocol == IPPROTO_IGMP)) { | |
9a49aba1 | 1444 | do_tx_balance = false; |
1da177e4 LT |
1445 | break; |
1446 | } | |
eddc9ec5 ACM |
1447 | hash_start = (char *)&(iph->daddr); |
1448 | hash_size = sizeof(iph->daddr); | |
1449 | } | |
1da177e4 LT |
1450 | break; |
1451 | case ETH_P_IPV6: | |
2d1ea19d VY |
1452 | /* IPv6 doesn't really use broadcast mac address, but leave |
1453 | * that here just in case. | |
1454 | */ | |
a6700db1 | 1455 | if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast)) { |
9a49aba1 | 1456 | do_tx_balance = false; |
1da177e4 | 1457 | break; |
2d1ea19d VY |
1458 | } |
1459 | ||
1460 | /* IPv6 uses all-nodes multicast as an equivalent to | |
1461 | * broadcasts in IPv4. | |
1462 | */ | |
a6700db1 | 1463 | if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) { |
9a49aba1 | 1464 | do_tx_balance = false; |
2d1ea19d VY |
1465 | break; |
1466 | } | |
1467 | ||
1468 | /* Additianally, DAD probes should not be tx-balanced as that | |
1469 | * will lead to false positives for duplicate addresses and | |
1470 | * prevent address configuration from working. | |
1471 | */ | |
1472 | ip6hdr = ipv6_hdr(skb); | |
1473 | if (ipv6_addr_any(&ip6hdr->saddr)) { | |
9a49aba1 | 1474 | do_tx_balance = false; |
2d1ea19d | 1475 | break; |
1da177e4 LT |
1476 | } |
1477 | ||
0660e03f ACM |
1478 | hash_start = (char *)&(ipv6_hdr(skb)->daddr); |
1479 | hash_size = sizeof(ipv6_hdr(skb)->daddr); | |
1da177e4 LT |
1480 | break; |
1481 | case ETH_P_IPX: | |
d3bb52b0 | 1482 | if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) { |
1da177e4 | 1483 | /* something is wrong with this packet */ |
9a49aba1 | 1484 | do_tx_balance = false; |
1da177e4 LT |
1485 | break; |
1486 | } | |
1487 | ||
1488 | if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) { | |
1489 | /* The only protocol worth balancing in | |
1490 | * this family since it has an "ARP" like | |
1491 | * mechanism | |
1492 | */ | |
9a49aba1 | 1493 | do_tx_balance = false; |
1da177e4 LT |
1494 | break; |
1495 | } | |
1496 | ||
b85b6fb1 | 1497 | hash_start = (char *)eth_data->h_dest; |
1da177e4 LT |
1498 | hash_size = ETH_ALEN; |
1499 | break; | |
1500 | case ETH_P_ARP: | |
9a49aba1 | 1501 | do_tx_balance = false; |
dda0fd5c | 1502 | if (bond_info->rlb_enabled) |
1da177e4 | 1503 | tx_slave = rlb_arp_xmit(skb, bond); |
1da177e4 LT |
1504 | break; |
1505 | default: | |
9a49aba1 | 1506 | do_tx_balance = false; |
1da177e4 LT |
1507 | break; |
1508 | } | |
1509 | ||
1510 | if (do_tx_balance) { | |
1511 | hash_index = _simple_hash(hash_start, hash_size); | |
1512 | tx_slave = tlb_choose_channel(bond, hash_index, skb->len); | |
1513 | } | |
1514 | ||
9a49aba1 | 1515 | return bond_do_alb_xmit(skb, bond, tx_slave); |
1da177e4 LT |
1516 | } |
1517 | ||
1b76b316 | 1518 | void bond_alb_monitor(struct work_struct *work) |
1da177e4 | 1519 | { |
1b76b316 JV |
1520 | struct bonding *bond = container_of(work, struct bonding, |
1521 | alb_work.work); | |
1da177e4 | 1522 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); |
9caff1e7 | 1523 | struct list_head *iter; |
1da177e4 | 1524 | struct slave *slave; |
1da177e4 | 1525 | |
0965a1f3 | 1526 | if (!bond_has_slaves(bond)) { |
1da177e4 LT |
1527 | bond_info->tx_rebalance_counter = 0; |
1528 | bond_info->lp_counter = 0; | |
1529 | goto re_arm; | |
1530 | } | |
1531 | ||
733ab639 | 1532 | rcu_read_lock(); |
1533 | ||
1da177e4 LT |
1534 | bond_info->tx_rebalance_counter++; |
1535 | bond_info->lp_counter++; | |
1536 | ||
1537 | /* send learning packets */ | |
7eacd038 | 1538 | if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) { |
1da177e4 LT |
1539 | /* change of curr_active_slave involves swapping of mac addresses. |
1540 | * in order to avoid this swapping from happening while | |
1541 | * sending the learning packets, the curr_slave_lock must be held for | |
1542 | * read. | |
1543 | */ | |
1544 | read_lock(&bond->curr_slave_lock); | |
1545 | ||
733ab639 | 1546 | bond_for_each_slave_rcu(bond, slave, iter) |
e944ef79 | 1547 | alb_send_learning_packets(slave, slave->dev->dev_addr); |
1da177e4 LT |
1548 | |
1549 | read_unlock(&bond->curr_slave_lock); | |
1550 | ||
1551 | bond_info->lp_counter = 0; | |
1552 | } | |
1553 | ||
1554 | /* rebalance tx traffic */ | |
1555 | if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) { | |
1556 | ||
1557 | read_lock(&bond->curr_slave_lock); | |
1558 | ||
733ab639 | 1559 | bond_for_each_slave_rcu(bond, slave, iter) { |
1da177e4 LT |
1560 | tlb_clear_slave(bond, slave, 1); |
1561 | if (slave == bond->curr_active_slave) { | |
1562 | SLAVE_TLB_INFO(slave).load = | |
1563 | bond_info->unbalanced_load / | |
1564 | BOND_TLB_REBALANCE_INTERVAL; | |
1565 | bond_info->unbalanced_load = 0; | |
1566 | } | |
1567 | } | |
1568 | ||
1569 | read_unlock(&bond->curr_slave_lock); | |
1570 | ||
1571 | bond_info->tx_rebalance_counter = 0; | |
1572 | } | |
1573 | ||
1574 | /* handle rlb stuff */ | |
1575 | if (bond_info->rlb_enabled) { | |
1da177e4 LT |
1576 | if (bond_info->primary_is_promisc && |
1577 | (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) { | |
1578 | ||
1f2cd845 DM |
1579 | /* |
1580 | * dev_set_promiscuity requires rtnl and | |
1581 | * nothing else. Avoid race with bond_close. | |
1582 | */ | |
733ab639 | 1583 | rcu_read_unlock(); |
1584 | if (!rtnl_trylock()) | |
1f2cd845 | 1585 | goto re_arm; |
1f2cd845 | 1586 | |
1da177e4 LT |
1587 | bond_info->rlb_promisc_timeout_counter = 0; |
1588 | ||
1589 | /* If the primary was set to promiscuous mode | |
1590 | * because a slave was disabled then | |
1591 | * it can now leave promiscuous mode. | |
1592 | */ | |
1593 | dev_set_promiscuity(bond->curr_active_slave->dev, -1); | |
1594 | bond_info->primary_is_promisc = 0; | |
1f2cd845 DM |
1595 | |
1596 | rtnl_unlock(); | |
733ab639 | 1597 | rcu_read_lock(); |
d0e81b7e | 1598 | } |
1da177e4 LT |
1599 | |
1600 | if (bond_info->rlb_rebalance) { | |
1601 | bond_info->rlb_rebalance = 0; | |
1602 | rlb_rebalance(bond); | |
1603 | } | |
1604 | ||
1605 | /* check if clients need updating */ | |
1606 | if (bond_info->rx_ntt) { | |
1607 | if (bond_info->rlb_update_delay_counter) { | |
1608 | --bond_info->rlb_update_delay_counter; | |
1609 | } else { | |
1610 | rlb_update_rx_clients(bond); | |
dda0fd5c | 1611 | if (bond_info->rlb_update_retry_counter) |
1da177e4 | 1612 | --bond_info->rlb_update_retry_counter; |
dda0fd5c | 1613 | else |
1da177e4 | 1614 | bond_info->rx_ntt = 0; |
1da177e4 LT |
1615 | } |
1616 | } | |
1617 | } | |
733ab639 | 1618 | rcu_read_unlock(); |
1da177e4 | 1619 | re_arm: |
e6d265e8 | 1620 | queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks); |
1da177e4 LT |
1621 | } |
1622 | ||
1623 | /* assumption: called before the slave is attached to the bond | |
1624 | * and not locked by the bond lock | |
1625 | */ | |
1626 | int bond_alb_init_slave(struct bonding *bond, struct slave *slave) | |
1627 | { | |
1628 | int res; | |
1629 | ||
b924551b | 1630 | res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr); |
dda0fd5c | 1631 | if (res) |
1da177e4 | 1632 | return res; |
1da177e4 | 1633 | |
1da177e4 | 1634 | res = alb_handle_addr_collision_on_attach(bond, slave); |
dda0fd5c | 1635 | if (res) |
1da177e4 | 1636 | return res; |
1da177e4 LT |
1637 | |
1638 | tlb_init_slave(slave); | |
1639 | ||
1640 | /* order a rebalance ASAP */ | |
1641 | bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS; | |
1642 | ||
dda0fd5c | 1643 | if (bond->alb_info.rlb_enabled) |
1da177e4 | 1644 | bond->alb_info.rlb_rebalance = 1; |
1da177e4 LT |
1645 | |
1646 | return 0; | |
1647 | } | |
1648 | ||
2543331d JV |
1649 | /* |
1650 | * Remove slave from tlb and rlb hash tables, and fix up MAC addresses | |
1651 | * if necessary. | |
1652 | * | |
1653 | * Caller must hold RTNL and no other locks | |
1654 | */ | |
1da177e4 LT |
1655 | void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave) |
1656 | { | |
0965a1f3 | 1657 | if (bond_has_slaves(bond)) |
1da177e4 | 1658 | alb_change_hw_addr_on_detach(bond, slave); |
1da177e4 LT |
1659 | |
1660 | tlb_clear_slave(bond, slave, 0); | |
1661 | ||
1662 | if (bond->alb_info.rlb_enabled) { | |
6475ae4c | 1663 | bond->alb_info.rx_slave = NULL; |
1da177e4 LT |
1664 | rlb_clear_slave(bond, slave); |
1665 | } | |
1666 | } | |
1667 | ||
1668 | /* Caller must hold bond lock for read */ | |
1669 | void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link) | |
1670 | { | |
1671 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
1672 | ||
1673 | if (link == BOND_LINK_DOWN) { | |
1674 | tlb_clear_slave(bond, slave, 0); | |
73ac0cd4 | 1675 | if (bond->alb_info.rlb_enabled) |
1da177e4 | 1676 | rlb_clear_slave(bond, slave); |
1da177e4 LT |
1677 | } else if (link == BOND_LINK_UP) { |
1678 | /* order a rebalance ASAP */ | |
1679 | bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS; | |
1680 | if (bond->alb_info.rlb_enabled) { | |
1681 | bond->alb_info.rlb_rebalance = 1; | |
1682 | /* If the updelay module parameter is smaller than the | |
1683 | * forwarding delay of the switch the rebalance will | |
1684 | * not work because the rebalance arp replies will | |
1685 | * not be forwarded to the clients.. | |
1686 | */ | |
1687 | } | |
1688 | } | |
1689 | } | |
1690 | ||
1691 | /** | |
1692 | * bond_alb_handle_active_change - assign new curr_active_slave | |
1693 | * @bond: our bonding struct | |
1694 | * @new_slave: new slave to assign | |
1695 | * | |
1696 | * Set the bond->curr_active_slave to @new_slave and handle | |
1697 | * mac address swapping and promiscuity changes as needed. | |
1698 | * | |
059fe7a5 JV |
1699 | * If new_slave is NULL, caller must hold curr_slave_lock or |
1700 | * bond->lock for write. | |
1701 | * | |
b2e7aceb | 1702 | * If new_slave is not NULL, caller must hold RTNL, curr_slave_lock |
1703 | * for write. Processing here may sleep, so no other locks may be held. | |
1da177e4 LT |
1704 | */ |
1705 | void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave) | |
1f78d9f9 | 1706 | __releases(&bond->curr_slave_lock) |
1f78d9f9 | 1707 | __acquires(&bond->curr_slave_lock) |
1da177e4 LT |
1708 | { |
1709 | struct slave *swap_slave; | |
1da177e4 | 1710 | |
dec1e90e | 1711 | if (bond->curr_active_slave == new_slave) |
1da177e4 | 1712 | return; |
1da177e4 LT |
1713 | |
1714 | if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) { | |
1715 | dev_set_promiscuity(bond->curr_active_slave->dev, -1); | |
1716 | bond->alb_info.primary_is_promisc = 0; | |
1717 | bond->alb_info.rlb_promisc_timeout_counter = 0; | |
1718 | } | |
1719 | ||
1720 | swap_slave = bond->curr_active_slave; | |
278b2083 | 1721 | rcu_assign_pointer(bond->curr_active_slave, new_slave); |
1da177e4 | 1722 | |
0965a1f3 | 1723 | if (!new_slave || !bond_has_slaves(bond)) |
1da177e4 | 1724 | return; |
1da177e4 LT |
1725 | |
1726 | /* set the new curr_active_slave to the bonds mac address | |
1727 | * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave | |
1728 | */ | |
b88ec38d VF |
1729 | if (!swap_slave) |
1730 | swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr); | |
1da177e4 | 1731 | |
059fe7a5 JV |
1732 | /* |
1733 | * Arrange for swap_slave and new_slave to temporarily be | |
1734 | * ignored so we can mess with their MAC addresses without | |
1735 | * fear of interference from transmit activity. | |
1736 | */ | |
dec1e90e | 1737 | if (swap_slave) |
059fe7a5 | 1738 | tlb_clear_slave(bond, swap_slave, 1); |
059fe7a5 JV |
1739 | tlb_clear_slave(bond, new_slave, 1); |
1740 | ||
1741 | write_unlock_bh(&bond->curr_slave_lock); | |
059fe7a5 | 1742 | |
e0138a66 JV |
1743 | ASSERT_RTNL(); |
1744 | ||
4996b909 VF |
1745 | /* in TLB mode, the slave might flip down/up with the old dev_addr, |
1746 | * and thus filter bond->dev_addr's packets, so force bond's mac | |
1747 | */ | |
1748 | if (bond->params.mode == BOND_MODE_TLB) { | |
1749 | struct sockaddr sa; | |
1750 | u8 tmp_addr[ETH_ALEN]; | |
1751 | ||
ada0f863 | 1752 | ether_addr_copy(tmp_addr, new_slave->dev->dev_addr); |
4996b909 VF |
1753 | |
1754 | memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len); | |
1755 | sa.sa_family = bond->dev->type; | |
1756 | /* we don't care if it can't change its mac, best effort */ | |
1757 | dev_set_mac_address(new_slave->dev, &sa); | |
1758 | ||
ada0f863 | 1759 | ether_addr_copy(new_slave->dev->dev_addr, tmp_addr); |
4996b909 VF |
1760 | } |
1761 | ||
1da177e4 LT |
1762 | /* curr_active_slave must be set before calling alb_swap_mac_addr */ |
1763 | if (swap_slave) { | |
1764 | /* swap mac address */ | |
43547ea6 | 1765 | alb_swap_mac_addr(swap_slave, new_slave); |
059fe7a5 JV |
1766 | alb_fasten_mac_swap(bond, swap_slave, new_slave); |
1767 | } else { | |
b88ec38d VF |
1768 | /* set the new_slave to the bond mac address */ |
1769 | alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr); | |
1da177e4 LT |
1770 | alb_send_learning_packets(new_slave, bond->dev->dev_addr); |
1771 | } | |
059fe7a5 JV |
1772 | |
1773 | write_lock_bh(&bond->curr_slave_lock); | |
1da177e4 LT |
1774 | } |
1775 | ||
059fe7a5 JV |
1776 | /* |
1777 | * Called with RTNL | |
1778 | */ | |
1da177e4 | 1779 | int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr) |
1f78d9f9 | 1780 | __acquires(&bond->lock) |
815bcc27 | 1781 | __releases(&bond->lock) |
1da177e4 | 1782 | { |
454d7c9b | 1783 | struct bonding *bond = netdev_priv(bond_dev); |
1da177e4 | 1784 | struct sockaddr *sa = addr; |
b88ec38d | 1785 | struct slave *swap_slave; |
1da177e4 | 1786 | int res; |
1da177e4 | 1787 | |
73ac0cd4 | 1788 | if (!is_valid_ether_addr(sa->sa_data)) |
1da177e4 | 1789 | return -EADDRNOTAVAIL; |
1da177e4 LT |
1790 | |
1791 | res = alb_set_mac_address(bond, addr); | |
73ac0cd4 | 1792 | if (res) |
1da177e4 | 1793 | return res; |
1da177e4 LT |
1794 | |
1795 | memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len); | |
1796 | ||
1797 | /* If there is no curr_active_slave there is nothing else to do. | |
1798 | * Otherwise we'll need to pass the new address to it and handle | |
1799 | * duplications. | |
1800 | */ | |
73ac0cd4 | 1801 | if (!bond->curr_active_slave) |
1da177e4 | 1802 | return 0; |
1da177e4 | 1803 | |
b88ec38d | 1804 | swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr); |
1da177e4 LT |
1805 | |
1806 | if (swap_slave) { | |
43547ea6 | 1807 | alb_swap_mac_addr(swap_slave, bond->curr_active_slave); |
059fe7a5 | 1808 | alb_fasten_mac_swap(bond, swap_slave, bond->curr_active_slave); |
1da177e4 | 1809 | } else { |
b924551b | 1810 | alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr); |
1da177e4 | 1811 | |
815bcc27 | 1812 | read_lock(&bond->lock); |
1da177e4 LT |
1813 | alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr); |
1814 | if (bond->alb_info.rlb_enabled) { | |
1815 | /* inform clients mac address has changed */ | |
1816 | rlb_req_update_slave_clients(bond, bond->curr_active_slave); | |
1817 | } | |
815bcc27 | 1818 | read_unlock(&bond->lock); |
1da177e4 LT |
1819 | } |
1820 | ||
1821 | return 0; | |
1822 | } | |
1823 | ||
1824 | void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id) | |
1825 | { | |
73ac0cd4 | 1826 | if (bond->alb_info.rlb_enabled) |
1da177e4 | 1827 | rlb_clear_vlan(bond, vlan_id); |
1da177e4 LT |
1828 | } |
1829 |