Commit | Line | Data |
---|---|---|
0a65089e | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 LT |
2 | /* |
3 | * Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved. | |
1da177e4 LT |
4 | */ |
5 | ||
1da177e4 LT |
6 | #include <linux/skbuff.h> |
7 | #include <linux/netdevice.h> | |
8 | #include <linux/etherdevice.h> | |
9 | #include <linux/pkt_sched.h> | |
10 | #include <linux/spinlock.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/timer.h> | |
13 | #include <linux/ip.h> | |
14 | #include <linux/ipv6.h> | |
15 | #include <linux/if_arp.h> | |
16 | #include <linux/if_ether.h> | |
17 | #include <linux/if_bonding.h> | |
18 | #include <linux/if_vlan.h> | |
19 | #include <linux/in.h> | |
1da177e4 | 20 | #include <net/arp.h> |
2d1ea19d | 21 | #include <net/ipv6.h> |
c78b8b20 | 22 | #include <net/ndisc.h> |
1da177e4 | 23 | #include <asm/byteorder.h> |
1ef8019b DM |
24 | #include <net/bonding.h> |
25 | #include <net/bond_alb.h> | |
1da177e4 | 26 | |
f87fda00 | 27 | static const u8 mac_v6_allmcast[ETH_ALEN + 2] __long_aligned = { |
885a136c ED |
28 | 0x33, 0x33, 0x00, 0x00, 0x00, 0x01 |
29 | }; | |
1da177e4 LT |
30 | static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC; |
31 | ||
32 | #pragma pack(1) | |
33 | struct learning_pkt { | |
34 | u8 mac_dst[ETH_ALEN]; | |
35 | u8 mac_src[ETH_ALEN]; | |
d3bb52b0 | 36 | __be16 type; |
1da177e4 LT |
37 | u8 padding[ETH_ZLEN - ETH_HLEN]; |
38 | }; | |
39 | ||
40 | struct arp_pkt { | |
d3bb52b0 AV |
41 | __be16 hw_addr_space; |
42 | __be16 prot_addr_space; | |
1da177e4 LT |
43 | u8 hw_addr_len; |
44 | u8 prot_addr_len; | |
d3bb52b0 | 45 | __be16 op_code; |
1da177e4 | 46 | u8 mac_src[ETH_ALEN]; /* sender hardware address */ |
d3bb52b0 | 47 | __be32 ip_src; /* sender IP address */ |
1da177e4 | 48 | u8 mac_dst[ETH_ALEN]; /* target hardware address */ |
d3bb52b0 | 49 | __be32 ip_dst; /* target IP address */ |
1da177e4 LT |
50 | }; |
51 | #pragma pack() | |
52 | ||
53 | /* Forward declaration */ | |
6f238100 | 54 | static void alb_send_learning_packets(struct slave *slave, const u8 mac_addr[], |
d0c21d43 | 55 | bool strict_match); |
e53665c6 JB |
56 | static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp); |
57 | static void rlb_src_unlink(struct bonding *bond, u32 index); | |
58 | static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, | |
59 | u32 ip_dst_hash); | |
1da177e4 | 60 | |
eddc9ec5 | 61 | static inline u8 _simple_hash(const u8 *hash_start, int hash_size) |
1da177e4 LT |
62 | { |
63 | int i; | |
64 | u8 hash = 0; | |
65 | ||
fdb89d75 | 66 | for (i = 0; i < hash_size; i++) |
1da177e4 | 67 | hash ^= hash_start[i]; |
1da177e4 LT |
68 | |
69 | return hash; | |
70 | } | |
71 | ||
72 | /*********************** tlb specific functions ***************************/ | |
73 | ||
1da177e4 LT |
74 | static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load) |
75 | { | |
76 | if (save_load) { | |
77 | entry->load_history = 1 + entry->tx_bytes / | |
78 | BOND_TLB_REBALANCE_INTERVAL; | |
79 | entry->tx_bytes = 0; | |
80 | } | |
81 | ||
82 | entry->tx_slave = NULL; | |
83 | entry->next = TLB_NULL_INDEX; | |
84 | entry->prev = TLB_NULL_INDEX; | |
85 | } | |
86 | ||
87 | static inline void tlb_init_slave(struct slave *slave) | |
88 | { | |
89 | SLAVE_TLB_INFO(slave).load = 0; | |
90 | SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX; | |
91 | } | |
92 | ||
f515e6b7 MU |
93 | static void __tlb_clear_slave(struct bonding *bond, struct slave *slave, |
94 | int save_load) | |
1da177e4 LT |
95 | { |
96 | struct tlb_client_info *tx_hash_table; | |
97 | u32 index; | |
98 | ||
1da177e4 LT |
99 | /* clear slave from tx_hashtbl */ |
100 | tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl; | |
101 | ||
ce39a800 AG |
102 | /* skip this if we've already freed the tx hash table */ |
103 | if (tx_hash_table) { | |
104 | index = SLAVE_TLB_INFO(slave).head; | |
105 | while (index != TLB_NULL_INDEX) { | |
106 | u32 next_index = tx_hash_table[index].next; | |
86a5ad0a | 107 | |
ce39a800 AG |
108 | tlb_init_table_entry(&tx_hash_table[index], save_load); |
109 | index = next_index; | |
110 | } | |
1da177e4 LT |
111 | } |
112 | ||
1da177e4 | 113 | tlb_init_slave(slave); |
f515e6b7 | 114 | } |
5af47b2f | 115 | |
f515e6b7 MU |
116 | static void tlb_clear_slave(struct bonding *bond, struct slave *slave, |
117 | int save_load) | |
118 | { | |
4bab16d7 | 119 | spin_lock_bh(&bond->mode_lock); |
f515e6b7 | 120 | __tlb_clear_slave(bond, slave, save_load); |
4bab16d7 | 121 | spin_unlock_bh(&bond->mode_lock); |
1da177e4 LT |
122 | } |
123 | ||
124 | /* Must be called before starting the monitor timer */ | |
125 | static int tlb_initialize(struct bonding *bond) | |
126 | { | |
127 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
128 | int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info); | |
0d206a3a | 129 | struct tlb_client_info *new_hashtbl; |
1da177e4 LT |
130 | int i; |
131 | ||
243cb4e5 | 132 | new_hashtbl = kzalloc(size, GFP_KERNEL); |
e404decb | 133 | if (!new_hashtbl) |
6d9b6f42 | 134 | return -ENOMEM; |
e404decb | 135 | |
4bab16d7 | 136 | spin_lock_bh(&bond->mode_lock); |
0d206a3a MW |
137 | |
138 | bond_info->tx_hashtbl = new_hashtbl; | |
1da177e4 | 139 | |
fdb89d75 | 140 | for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) |
38dbaf0a | 141 | tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0); |
1da177e4 | 142 | |
4bab16d7 | 143 | spin_unlock_bh(&bond->mode_lock); |
1da177e4 LT |
144 | |
145 | return 0; | |
146 | } | |
147 | ||
148 | /* Must be called only after all slaves have been released */ | |
149 | static void tlb_deinitialize(struct bonding *bond) | |
150 | { | |
151 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
152 | ||
4bab16d7 | 153 | spin_lock_bh(&bond->mode_lock); |
1da177e4 LT |
154 | |
155 | kfree(bond_info->tx_hashtbl); | |
156 | bond_info->tx_hashtbl = NULL; | |
157 | ||
4bab16d7 | 158 | spin_unlock_bh(&bond->mode_lock); |
1da177e4 LT |
159 | } |
160 | ||
097811bb JP |
161 | static long long compute_gap(struct slave *slave) |
162 | { | |
163 | return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */ | |
164 | (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */ | |
165 | } | |
166 | ||
1da177e4 LT |
167 | static struct slave *tlb_get_least_loaded_slave(struct bonding *bond) |
168 | { | |
169 | struct slave *slave, *least_loaded; | |
9caff1e7 | 170 | struct list_head *iter; |
097811bb | 171 | long long max_gap; |
1da177e4 | 172 | |
097811bb JP |
173 | least_loaded = NULL; |
174 | max_gap = LLONG_MIN; | |
1da177e4 LT |
175 | |
176 | /* Find the slave with the largest gap */ | |
28c71926 | 177 | bond_for_each_slave_rcu(bond, slave, iter) { |
8557cd74 | 178 | if (bond_slave_can_tx(slave)) { |
097811bb JP |
179 | long long gap = compute_gap(slave); |
180 | ||
1da177e4 LT |
181 | if (max_gap < gap) { |
182 | least_loaded = slave; | |
183 | max_gap = gap; | |
184 | } | |
185 | } | |
186 | } | |
187 | ||
188 | return least_loaded; | |
189 | } | |
190 | ||
f515e6b7 MU |
191 | static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index, |
192 | u32 skb_len) | |
1da177e4 LT |
193 | { |
194 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
195 | struct tlb_client_info *hash_table; | |
196 | struct slave *assigned_slave; | |
197 | ||
1da177e4 LT |
198 | hash_table = bond_info->tx_hashtbl; |
199 | assigned_slave = hash_table[hash_index].tx_slave; | |
200 | if (!assigned_slave) { | |
201 | assigned_slave = tlb_get_least_loaded_slave(bond); | |
202 | ||
203 | if (assigned_slave) { | |
204 | struct tlb_slave_info *slave_info = | |
205 | &(SLAVE_TLB_INFO(assigned_slave)); | |
206 | u32 next_index = slave_info->head; | |
207 | ||
208 | hash_table[hash_index].tx_slave = assigned_slave; | |
209 | hash_table[hash_index].next = next_index; | |
210 | hash_table[hash_index].prev = TLB_NULL_INDEX; | |
211 | ||
fdb89d75 | 212 | if (next_index != TLB_NULL_INDEX) |
1da177e4 | 213 | hash_table[next_index].prev = hash_index; |
1da177e4 LT |
214 | |
215 | slave_info->head = hash_index; | |
216 | slave_info->load += | |
217 | hash_table[hash_index].load_history; | |
218 | } | |
219 | } | |
220 | ||
fdb89d75 | 221 | if (assigned_slave) |
1da177e4 | 222 | hash_table[hash_index].tx_bytes += skb_len; |
1da177e4 | 223 | |
1da177e4 LT |
224 | return assigned_slave; |
225 | } | |
226 | ||
f515e6b7 MU |
227 | static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index, |
228 | u32 skb_len) | |
229 | { | |
230 | struct slave *tx_slave; | |
547942ca | 231 | |
4057c58d | 232 | /* We don't need to disable softirq here, because |
f515e6b7 MU |
233 | * tlb_choose_channel() is only called by bond_alb_xmit() |
234 | * which already has softirq disabled. | |
235 | */ | |
4bab16d7 | 236 | spin_lock(&bond->mode_lock); |
f515e6b7 | 237 | tx_slave = __tlb_choose_channel(bond, hash_index, skb_len); |
4bab16d7 | 238 | spin_unlock(&bond->mode_lock); |
547942ca | 239 | |
f515e6b7 MU |
240 | return tx_slave; |
241 | } | |
242 | ||
1da177e4 | 243 | /*********************** rlb specific functions ***************************/ |
f515e6b7 | 244 | |
1da177e4 LT |
245 | /* when an ARP REPLY is received from a client update its info |
246 | * in the rx_hashtbl | |
247 | */ | |
248 | static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp) | |
249 | { | |
250 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
251 | struct rlb_client_info *client_info; | |
252 | u32 hash_index; | |
253 | ||
4bab16d7 | 254 | spin_lock_bh(&bond->mode_lock); |
1da177e4 | 255 | |
b85b6fb1 | 256 | hash_index = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src)); |
1da177e4 LT |
257 | client_info = &(bond_info->rx_hashtbl[hash_index]); |
258 | ||
259 | if ((client_info->assigned) && | |
260 | (client_info->ip_src == arp->ip_dst) && | |
42d782ac | 261 | (client_info->ip_dst == arp->ip_src) && |
a6700db1 | 262 | (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) { |
1da177e4 | 263 | /* update the clients MAC address */ |
ada0f863 | 264 | ether_addr_copy(client_info->mac_dst, arp->mac_src); |
1da177e4 LT |
265 | client_info->ntt = 1; |
266 | bond_info->rx_ntt = 1; | |
267 | } | |
268 | ||
4bab16d7 | 269 | spin_unlock_bh(&bond->mode_lock); |
1da177e4 LT |
270 | } |
271 | ||
de063b70 ED |
272 | static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond, |
273 | struct slave *slave) | |
1da177e4 | 274 | { |
de063b70 | 275 | struct arp_pkt *arp, _arp; |
1da177e4 | 276 | |
3aba891d | 277 | if (skb->protocol != cpu_to_be16(ETH_P_ARP)) |
b99215cd | 278 | goto out; |
1da177e4 | 279 | |
de063b70 ED |
280 | arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp); |
281 | if (!arp) | |
b99215cd | 282 | goto out; |
1da177e4 | 283 | |
e53665c6 JB |
284 | /* We received an ARP from arp->ip_src. |
285 | * We might have used this IP address previously (on the bonding host | |
286 | * itself or on a system that is bridged together with the bond). | |
287 | * However, if arp->mac_src is different than what is stored in | |
288 | * rx_hashtbl, some other host is now using the IP and we must prevent | |
289 | * sending out client updates with this IP address and the old MAC | |
290 | * address. | |
291 | * Clean up all hash table entries that have this address as ip_src but | |
292 | * have a different mac_src. | |
293 | */ | |
294 | rlb_purge_src_ip(bond, arp); | |
295 | ||
1da177e4 LT |
296 | if (arp->op_code == htons(ARPOP_REPLY)) { |
297 | /* update rx hash table for this ARP */ | |
298 | rlb_update_entry_from_arp(bond, arp); | |
7ea2e423 | 299 | slave_dbg(bond->dev, slave->dev, "Server received an ARP Reply from client\n"); |
1da177e4 | 300 | } |
b99215cd DM |
301 | out: |
302 | return RX_HANDLER_ANOTHER; | |
1da177e4 LT |
303 | } |
304 | ||
56924c38 NA |
305 | /* Caller must hold rcu_read_lock() */ |
306 | static struct slave *__rlb_next_rx_slave(struct bonding *bond) | |
1da177e4 LT |
307 | { |
308 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
6475ae4c VF |
309 | struct slave *before = NULL, *rx_slave = NULL, *slave; |
310 | struct list_head *iter; | |
311 | bool found = false; | |
1da177e4 | 312 | |
56924c38 | 313 | bond_for_each_slave_rcu(bond, slave, iter) { |
8557cd74 | 314 | if (!bond_slave_can_tx(slave)) |
6475ae4c VF |
315 | continue; |
316 | if (!found) { | |
317 | if (!before || before->speed < slave->speed) | |
318 | before = slave; | |
319 | } else { | |
320 | if (!rx_slave || rx_slave->speed < slave->speed) | |
1da177e4 | 321 | rx_slave = slave; |
1da177e4 | 322 | } |
6475ae4c VF |
323 | if (slave == bond_info->rx_slave) |
324 | found = true; | |
1da177e4 | 325 | } |
6475ae4c VF |
326 | /* we didn't find anything after the current or we have something |
327 | * better before and up to the current slave | |
328 | */ | |
329 | if (!rx_slave || (before && rx_slave->speed < before->speed)) | |
330 | rx_slave = before; | |
1da177e4 | 331 | |
6475ae4c VF |
332 | if (rx_slave) |
333 | bond_info->rx_slave = rx_slave; | |
1da177e4 LT |
334 | |
335 | return rx_slave; | |
336 | } | |
337 | ||
56924c38 NA |
338 | /* Caller must hold RTNL, rcu_read_lock is obtained only to silence checkers */ |
339 | static struct slave *rlb_next_rx_slave(struct bonding *bond) | |
28c71926 | 340 | { |
56924c38 | 341 | struct slave *rx_slave; |
28c71926 | 342 | |
56924c38 | 343 | ASSERT_RTNL(); |
28c71926 | 344 | |
56924c38 NA |
345 | rcu_read_lock(); |
346 | rx_slave = __rlb_next_rx_slave(bond); | |
347 | rcu_read_unlock(); | |
28c71926 | 348 | |
349 | return rx_slave; | |
350 | } | |
351 | ||
1da177e4 LT |
352 | /* teach the switch the mac of a disabled slave |
353 | * on the primary for fault tolerance | |
354 | * | |
62c5f518 | 355 | * Caller must hold RTNL |
1da177e4 | 356 | */ |
6f238100 JK |
357 | static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, |
358 | const u8 addr[]) | |
1da177e4 | 359 | { |
1c72cfdc | 360 | struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); |
4740d638 ED |
361 | |
362 | if (!curr_active) | |
1da177e4 | 363 | return; |
1da177e4 LT |
364 | |
365 | if (!bond->alb_info.primary_is_promisc) { | |
4740d638 | 366 | if (!dev_set_promiscuity(curr_active->dev, 1)) |
7e1a1ac1 WC |
367 | bond->alb_info.primary_is_promisc = 1; |
368 | else | |
369 | bond->alb_info.primary_is_promisc = 0; | |
1da177e4 LT |
370 | } |
371 | ||
372 | bond->alb_info.rlb_promisc_timeout_counter = 0; | |
373 | ||
4740d638 | 374 | alb_send_learning_packets(curr_active, addr, true); |
1da177e4 LT |
375 | } |
376 | ||
377 | /* slave being removed should not be active at this point | |
378 | * | |
b2e7aceb | 379 | * Caller must hold rtnl. |
1da177e4 LT |
380 | */ |
381 | static void rlb_clear_slave(struct bonding *bond, struct slave *slave) | |
382 | { | |
383 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
384 | struct rlb_client_info *rx_hash_table; | |
385 | u32 index, next_index; | |
386 | ||
387 | /* clear slave from rx_hashtbl */ | |
4bab16d7 | 388 | spin_lock_bh(&bond->mode_lock); |
1da177e4 LT |
389 | |
390 | rx_hash_table = bond_info->rx_hashtbl; | |
e53665c6 | 391 | index = bond_info->rx_hashtbl_used_head; |
1da177e4 | 392 | for (; index != RLB_NULL_INDEX; index = next_index) { |
e53665c6 | 393 | next_index = rx_hash_table[index].used_next; |
1da177e4 LT |
394 | if (rx_hash_table[index].slave == slave) { |
395 | struct slave *assigned_slave = rlb_next_rx_slave(bond); | |
396 | ||
397 | if (assigned_slave) { | |
398 | rx_hash_table[index].slave = assigned_slave; | |
cbeeea70 | 399 | if (is_valid_ether_addr(rx_hash_table[index].mac_dst)) { |
1da177e4 LT |
400 | bond_info->rx_hashtbl[index].ntt = 1; |
401 | bond_info->rx_ntt = 1; | |
402 | /* A slave has been removed from the | |
403 | * table because it is either disabled | |
404 | * or being released. We must retry the | |
405 | * update to avoid clients from not | |
406 | * being updated & disconnecting when | |
407 | * there is stress | |
408 | */ | |
409 | bond_info->rlb_update_retry_counter = | |
410 | RLB_UPDATE_RETRY; | |
411 | } | |
412 | } else { /* there is no active slave */ | |
413 | rx_hash_table[index].slave = NULL; | |
414 | } | |
415 | } | |
416 | } | |
417 | ||
4bab16d7 | 418 | spin_unlock_bh(&bond->mode_lock); |
1da177e4 | 419 | |
1c72cfdc | 420 | if (slave != rtnl_dereference(bond->curr_active_slave)) |
1da177e4 | 421 | rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr); |
1da177e4 LT |
422 | } |
423 | ||
424 | static void rlb_update_client(struct rlb_client_info *client_info) | |
425 | { | |
426 | int i; | |
427 | ||
4fa8667c | 428 | if (!client_info->slave || !is_valid_ether_addr(client_info->mac_dst)) |
1da177e4 | 429 | return; |
1da177e4 LT |
430 | |
431 | for (i = 0; i < RLB_ARP_BURST_SIZE; i++) { | |
432 | struct sk_buff *skb; | |
433 | ||
434 | skb = arp_create(ARPOP_REPLY, ETH_P_ARP, | |
435 | client_info->ip_dst, | |
436 | client_info->slave->dev, | |
437 | client_info->ip_src, | |
438 | client_info->mac_dst, | |
439 | client_info->slave->dev->dev_addr, | |
440 | client_info->mac_dst); | |
441 | if (!skb) { | |
7ea2e423 JW |
442 | slave_err(client_info->slave->bond->dev, |
443 | client_info->slave->dev, | |
444 | "failed to create an ARP packet\n"); | |
1da177e4 LT |
445 | continue; |
446 | } | |
447 | ||
448 | skb->dev = client_info->slave->dev; | |
449 | ||
d3ab3ffd | 450 | if (client_info->vlan_id) { |
b4bef1b5 JP |
451 | __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), |
452 | client_info->vlan_id); | |
1da177e4 LT |
453 | } |
454 | ||
455 | arp_xmit(skb); | |
456 | } | |
457 | } | |
458 | ||
459 | /* sends ARP REPLIES that update the clients that need updating */ | |
460 | static void rlb_update_rx_clients(struct bonding *bond) | |
461 | { | |
462 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
463 | struct rlb_client_info *client_info; | |
464 | u32 hash_index; | |
465 | ||
4bab16d7 | 466 | spin_lock_bh(&bond->mode_lock); |
1da177e4 | 467 | |
e53665c6 JB |
468 | hash_index = bond_info->rx_hashtbl_used_head; |
469 | for (; hash_index != RLB_NULL_INDEX; | |
470 | hash_index = client_info->used_next) { | |
1da177e4 LT |
471 | client_info = &(bond_info->rx_hashtbl[hash_index]); |
472 | if (client_info->ntt) { | |
473 | rlb_update_client(client_info); | |
35d75ee4 | 474 | if (bond_info->rlb_update_retry_counter == 0) |
1da177e4 | 475 | client_info->ntt = 0; |
1da177e4 LT |
476 | } |
477 | } | |
478 | ||
94e2bd68 | 479 | /* do not update the entries again until this counter is zero so that |
1da177e4 LT |
480 | * not to confuse the clients. |
481 | */ | |
482 | bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY; | |
483 | ||
4bab16d7 | 484 | spin_unlock_bh(&bond->mode_lock); |
1da177e4 LT |
485 | } |
486 | ||
487 | /* The slave was assigned a new mac address - update the clients */ | |
488 | static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave) | |
489 | { | |
490 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
491 | struct rlb_client_info *client_info; | |
492 | int ntt = 0; | |
493 | u32 hash_index; | |
494 | ||
4bab16d7 | 495 | spin_lock_bh(&bond->mode_lock); |
1da177e4 | 496 | |
e53665c6 JB |
497 | hash_index = bond_info->rx_hashtbl_used_head; |
498 | for (; hash_index != RLB_NULL_INDEX; | |
499 | hash_index = client_info->used_next) { | |
1da177e4 LT |
500 | client_info = &(bond_info->rx_hashtbl[hash_index]); |
501 | ||
502 | if ((client_info->slave == slave) && | |
cbeeea70 | 503 | is_valid_ether_addr(client_info->mac_dst)) { |
1da177e4 LT |
504 | client_info->ntt = 1; |
505 | ntt = 1; | |
506 | } | |
507 | } | |
508 | ||
4708a1b1 | 509 | /* update the team's flag only after the whole iteration */ |
1da177e4 LT |
510 | if (ntt) { |
511 | bond_info->rx_ntt = 1; | |
4708a1b1 | 512 | /* fasten the change */ |
1da177e4 LT |
513 | bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY; |
514 | } | |
515 | ||
4bab16d7 | 516 | spin_unlock_bh(&bond->mode_lock); |
1da177e4 LT |
517 | } |
518 | ||
519 | /* mark all clients using src_ip to be updated */ | |
d3bb52b0 | 520 | static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip) |
1da177e4 LT |
521 | { |
522 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
523 | struct rlb_client_info *client_info; | |
524 | u32 hash_index; | |
525 | ||
4bab16d7 | 526 | spin_lock(&bond->mode_lock); |
1da177e4 | 527 | |
e53665c6 JB |
528 | hash_index = bond_info->rx_hashtbl_used_head; |
529 | for (; hash_index != RLB_NULL_INDEX; | |
530 | hash_index = client_info->used_next) { | |
1da177e4 LT |
531 | client_info = &(bond_info->rx_hashtbl[hash_index]); |
532 | ||
533 | if (!client_info->slave) { | |
0a111a03 | 534 | netdev_err(bond->dev, "found a client with no channel in the client's hash table\n"); |
1da177e4 LT |
535 | continue; |
536 | } | |
547942ca | 537 | /* update all clients using this src_ip, that are not assigned |
1da177e4 LT |
538 | * to the team's address (curr_active_slave) and have a known |
539 | * unicast mac address. | |
540 | */ | |
541 | if ((client_info->ip_src == src_ip) && | |
a6700db1 JP |
542 | !ether_addr_equal_64bits(client_info->slave->dev->dev_addr, |
543 | bond->dev->dev_addr) && | |
cbeeea70 | 544 | is_valid_ether_addr(client_info->mac_dst)) { |
1da177e4 LT |
545 | client_info->ntt = 1; |
546 | bond_info->rx_ntt = 1; | |
547 | } | |
548 | } | |
549 | ||
4bab16d7 | 550 | spin_unlock(&bond->mode_lock); |
1da177e4 LT |
551 | } |
552 | ||
b7469e83 ED |
553 | static struct slave *rlb_choose_channel(struct sk_buff *skb, |
554 | struct bonding *bond, | |
555 | const struct arp_pkt *arp) | |
1da177e4 LT |
556 | { |
557 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
28c71926 | 558 | struct slave *assigned_slave, *curr_active_slave; |
1da177e4 LT |
559 | struct rlb_client_info *client_info; |
560 | u32 hash_index = 0; | |
561 | ||
4bab16d7 | 562 | spin_lock(&bond->mode_lock); |
1da177e4 | 563 | |
28c71926 | 564 | curr_active_slave = rcu_dereference(bond->curr_active_slave); |
565 | ||
e364a341 | 566 | hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst)); |
1da177e4 LT |
567 | client_info = &(bond_info->rx_hashtbl[hash_index]); |
568 | ||
569 | if (client_info->assigned) { | |
570 | if ((client_info->ip_src == arp->ip_src) && | |
571 | (client_info->ip_dst == arp->ip_dst)) { | |
572 | /* the entry is already assigned to this client */ | |
cbeeea70 | 573 | if (!is_broadcast_ether_addr(arp->mac_dst)) { |
1da177e4 | 574 | /* update mac address from arp */ |
ada0f863 | 575 | ether_addr_copy(client_info->mac_dst, arp->mac_dst); |
1da177e4 | 576 | } |
ada0f863 | 577 | ether_addr_copy(client_info->mac_src, arp->mac_src); |
1da177e4 LT |
578 | |
579 | assigned_slave = client_info->slave; | |
580 | if (assigned_slave) { | |
4bab16d7 | 581 | spin_unlock(&bond->mode_lock); |
1da177e4 LT |
582 | return assigned_slave; |
583 | } | |
584 | } else { | |
585 | /* the entry is already assigned to some other client, | |
586 | * move the old client to primary (curr_active_slave) so | |
587 | * that the new client can be assigned to this entry. | |
588 | */ | |
4740d638 | 589 | if (curr_active_slave && |
28c71926 | 590 | client_info->slave != curr_active_slave) { |
591 | client_info->slave = curr_active_slave; | |
1da177e4 LT |
592 | rlb_update_client(client_info); |
593 | } | |
594 | } | |
595 | } | |
596 | /* assign a new slave */ | |
28c71926 | 597 | assigned_slave = __rlb_next_rx_slave(bond); |
1da177e4 LT |
598 | |
599 | if (assigned_slave) { | |
e53665c6 JB |
600 | if (!(client_info->assigned && |
601 | client_info->ip_src == arp->ip_src)) { | |
602 | /* ip_src is going to be updated, | |
603 | * fix the src hash list | |
604 | */ | |
605 | u32 hash_src = _simple_hash((u8 *)&arp->ip_src, | |
606 | sizeof(arp->ip_src)); | |
607 | rlb_src_unlink(bond, hash_index); | |
608 | rlb_src_link(bond, hash_src, hash_index); | |
609 | } | |
610 | ||
1da177e4 LT |
611 | client_info->ip_src = arp->ip_src; |
612 | client_info->ip_dst = arp->ip_dst; | |
4057c58d | 613 | /* arp->mac_dst is broadcast for arp requests. |
1da177e4 LT |
614 | * will be updated with clients actual unicast mac address |
615 | * upon receiving an arp reply. | |
616 | */ | |
ada0f863 JP |
617 | ether_addr_copy(client_info->mac_dst, arp->mac_dst); |
618 | ether_addr_copy(client_info->mac_src, arp->mac_src); | |
1da177e4 LT |
619 | client_info->slave = assigned_slave; |
620 | ||
cbeeea70 | 621 | if (is_valid_ether_addr(client_info->mac_dst)) { |
1da177e4 LT |
622 | client_info->ntt = 1; |
623 | bond->alb_info.rx_ntt = 1; | |
624 | } else { | |
625 | client_info->ntt = 0; | |
626 | } | |
627 | ||
fb00bc2e | 628 | if (vlan_get_tag(skb, &client_info->vlan_id)) |
d3ab3ffd | 629 | client_info->vlan_id = 0; |
1da177e4 LT |
630 | |
631 | if (!client_info->assigned) { | |
e53665c6 | 632 | u32 prev_tbl_head = bond_info->rx_hashtbl_used_head; |
86a5ad0a | 633 | |
e53665c6 JB |
634 | bond_info->rx_hashtbl_used_head = hash_index; |
635 | client_info->used_next = prev_tbl_head; | |
1da177e4 | 636 | if (prev_tbl_head != RLB_NULL_INDEX) { |
e53665c6 | 637 | bond_info->rx_hashtbl[prev_tbl_head].used_prev = |
1da177e4 LT |
638 | hash_index; |
639 | } | |
640 | client_info->assigned = 1; | |
641 | } | |
642 | } | |
643 | ||
4bab16d7 | 644 | spin_unlock(&bond->mode_lock); |
1da177e4 LT |
645 | |
646 | return assigned_slave; | |
647 | } | |
648 | ||
649 | /* chooses (and returns) transmit channel for arp reply | |
650 | * does not choose channel for other arp types since they are | |
651 | * sent on the curr_active_slave | |
652 | */ | |
653 | static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond) | |
654 | { | |
1da177e4 | 655 | struct slave *tx_slave = NULL; |
d5410ac7 | 656 | struct net_device *dev; |
b7469e83 ED |
657 | struct arp_pkt *arp; |
658 | ||
659 | if (!pskb_network_may_pull(skb, sizeof(*arp))) | |
660 | return NULL; | |
661 | arp = (struct arp_pkt *)skb_network_header(skb); | |
1da177e4 | 662 | |
e74216b8 HL |
663 | /* Don't modify or load balance ARPs that do not originate |
664 | * from the bond itself or a VLAN directly above the bond. | |
567b871e | 665 | */ |
e74216b8 | 666 | if (!bond_slave_has_mac_rcu(bond, arp->mac_src)) |
567b871e | 667 | return NULL; |
668 | ||
d5410ac7 SS |
669 | dev = ip_dev_find(dev_net(bond->dev), arp->ip_src); |
670 | if (dev) { | |
f11e5bd1 | 671 | if (netif_is_any_bridge_master(dev)) { |
4f5d33f4 | 672 | dev_put(dev); |
d5410ac7 | 673 | return NULL; |
4f5d33f4 JV |
674 | } |
675 | dev_put(dev); | |
d5410ac7 SS |
676 | } |
677 | ||
f14c4e4e | 678 | if (arp->op_code == htons(ARPOP_REPLY)) { |
547942ca | 679 | /* the arp must be sent on the selected rx channel */ |
b7469e83 | 680 | tx_slave = rlb_choose_channel(skb, bond, arp); |
35d75ee4 | 681 | if (tx_slave) |
faeeb317 JW |
682 | bond_hw_addr_copy(arp->mac_src, tx_slave->dev->dev_addr, |
683 | tx_slave->dev->addr_len); | |
7ea2e423 JW |
684 | netdev_dbg(bond->dev, "(slave %s): Server sent ARP Reply packet\n", |
685 | tx_slave ? tx_slave->dev->name : "NULL"); | |
f14c4e4e | 686 | } else if (arp->op_code == htons(ARPOP_REQUEST)) { |
1da177e4 LT |
687 | /* Create an entry in the rx_hashtbl for this client as a |
688 | * place holder. | |
689 | * When the arp reply is received the entry will be updated | |
690 | * with the correct unicast address of the client. | |
691 | */ | |
b7469e83 | 692 | tx_slave = rlb_choose_channel(skb, bond, arp); |
1da177e4 | 693 | |
77c8e2c0 | 694 | /* The ARP reply packets must be delayed so that |
1da177e4 LT |
695 | * they can cancel out the influence of the ARP request. |
696 | */ | |
697 | bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY; | |
698 | ||
699 | /* arp requests are broadcast and are sent on the primary | |
700 | * the arp request will collapse all clients on the subnet to | |
701 | * the primary slave. We must register these clients to be | |
702 | * updated with their assigned mac. | |
703 | */ | |
704 | rlb_req_update_subnet_clients(bond, arp->ip_src); | |
7ea2e423 JW |
705 | netdev_dbg(bond->dev, "(slave %s): Server sent ARP Request packet\n", |
706 | tx_slave ? tx_slave->dev->name : "NULL"); | |
1da177e4 LT |
707 | } |
708 | ||
709 | return tx_slave; | |
710 | } | |
711 | ||
1da177e4 LT |
712 | static void rlb_rebalance(struct bonding *bond) |
713 | { | |
714 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
715 | struct slave *assigned_slave; | |
716 | struct rlb_client_info *client_info; | |
717 | int ntt; | |
718 | u32 hash_index; | |
719 | ||
4bab16d7 | 720 | spin_lock_bh(&bond->mode_lock); |
1da177e4 LT |
721 | |
722 | ntt = 0; | |
e53665c6 JB |
723 | hash_index = bond_info->rx_hashtbl_used_head; |
724 | for (; hash_index != RLB_NULL_INDEX; | |
725 | hash_index = client_info->used_next) { | |
1da177e4 | 726 | client_info = &(bond_info->rx_hashtbl[hash_index]); |
733ab639 | 727 | assigned_slave = __rlb_next_rx_slave(bond); |
1da177e4 LT |
728 | if (assigned_slave && (client_info->slave != assigned_slave)) { |
729 | client_info->slave = assigned_slave; | |
25780410 DB |
730 | if (!is_zero_ether_addr(client_info->mac_dst)) { |
731 | client_info->ntt = 1; | |
732 | ntt = 1; | |
733 | } | |
1da177e4 LT |
734 | } |
735 | } | |
736 | ||
737 | /* update the team's flag only after the whole iteration */ | |
35d75ee4 | 738 | if (ntt) |
1da177e4 | 739 | bond_info->rx_ntt = 1; |
4bab16d7 | 740 | spin_unlock_bh(&bond->mode_lock); |
1da177e4 LT |
741 | } |
742 | ||
547942ca | 743 | /* Caller must hold mode_lock */ |
e53665c6 JB |
744 | static void rlb_init_table_entry_dst(struct rlb_client_info *entry) |
745 | { | |
746 | entry->used_next = RLB_NULL_INDEX; | |
747 | entry->used_prev = RLB_NULL_INDEX; | |
748 | entry->assigned = 0; | |
749 | entry->slave = NULL; | |
d3ab3ffd | 750 | entry->vlan_id = 0; |
e53665c6 JB |
751 | } |
752 | static void rlb_init_table_entry_src(struct rlb_client_info *entry) | |
753 | { | |
754 | entry->src_first = RLB_NULL_INDEX; | |
755 | entry->src_prev = RLB_NULL_INDEX; | |
756 | entry->src_next = RLB_NULL_INDEX; | |
757 | } | |
758 | ||
1da177e4 LT |
759 | static void rlb_init_table_entry(struct rlb_client_info *entry) |
760 | { | |
761 | memset(entry, 0, sizeof(struct rlb_client_info)); | |
e53665c6 JB |
762 | rlb_init_table_entry_dst(entry); |
763 | rlb_init_table_entry_src(entry); | |
764 | } | |
765 | ||
766 | static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index) | |
767 | { | |
768 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
769 | u32 next_index = bond_info->rx_hashtbl[index].used_next; | |
770 | u32 prev_index = bond_info->rx_hashtbl[index].used_prev; | |
771 | ||
772 | if (index == bond_info->rx_hashtbl_used_head) | |
773 | bond_info->rx_hashtbl_used_head = next_index; | |
774 | if (prev_index != RLB_NULL_INDEX) | |
775 | bond_info->rx_hashtbl[prev_index].used_next = next_index; | |
776 | if (next_index != RLB_NULL_INDEX) | |
777 | bond_info->rx_hashtbl[next_index].used_prev = prev_index; | |
778 | } | |
779 | ||
780 | /* unlink a rlb hash table entry from the src list */ | |
781 | static void rlb_src_unlink(struct bonding *bond, u32 index) | |
782 | { | |
783 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
784 | u32 next_index = bond_info->rx_hashtbl[index].src_next; | |
785 | u32 prev_index = bond_info->rx_hashtbl[index].src_prev; | |
786 | ||
787 | bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX; | |
788 | bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX; | |
789 | ||
790 | if (next_index != RLB_NULL_INDEX) | |
791 | bond_info->rx_hashtbl[next_index].src_prev = prev_index; | |
792 | ||
793 | if (prev_index == RLB_NULL_INDEX) | |
794 | return; | |
795 | ||
796 | /* is prev_index pointing to the head of this list? */ | |
797 | if (bond_info->rx_hashtbl[prev_index].src_first == index) | |
798 | bond_info->rx_hashtbl[prev_index].src_first = next_index; | |
799 | else | |
800 | bond_info->rx_hashtbl[prev_index].src_next = next_index; | |
801 | ||
802 | } | |
803 | ||
804 | static void rlb_delete_table_entry(struct bonding *bond, u32 index) | |
805 | { | |
806 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
807 | struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]); | |
808 | ||
809 | rlb_delete_table_entry_dst(bond, index); | |
810 | rlb_init_table_entry_dst(entry); | |
811 | ||
812 | rlb_src_unlink(bond, index); | |
813 | } | |
814 | ||
815 | /* add the rx_hashtbl[ip_dst_hash] entry to the list | |
816 | * of entries with identical ip_src_hash | |
817 | */ | |
818 | static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash) | |
819 | { | |
820 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
821 | u32 next; | |
822 | ||
823 | bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash; | |
824 | next = bond_info->rx_hashtbl[ip_src_hash].src_first; | |
825 | bond_info->rx_hashtbl[ip_dst_hash].src_next = next; | |
826 | if (next != RLB_NULL_INDEX) | |
827 | bond_info->rx_hashtbl[next].src_prev = ip_dst_hash; | |
828 | bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash; | |
829 | } | |
830 | ||
547942ca NA |
831 | /* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does |
832 | * not match arp->mac_src | |
833 | */ | |
e53665c6 JB |
834 | static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp) |
835 | { | |
836 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
b85b6fb1 | 837 | u32 ip_src_hash = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src)); |
e53665c6 JB |
838 | u32 index; |
839 | ||
4bab16d7 | 840 | spin_lock_bh(&bond->mode_lock); |
e53665c6 JB |
841 | |
842 | index = bond_info->rx_hashtbl[ip_src_hash].src_first; | |
843 | while (index != RLB_NULL_INDEX) { | |
844 | struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]); | |
845 | u32 next_index = entry->src_next; | |
86a5ad0a | 846 | |
e53665c6 JB |
847 | if (entry->ip_src == arp->ip_src && |
848 | !ether_addr_equal_64bits(arp->mac_src, entry->mac_src)) | |
8ce390bb | 849 | rlb_delete_table_entry(bond, index); |
e53665c6 JB |
850 | index = next_index; |
851 | } | |
4bab16d7 | 852 | spin_unlock_bh(&bond->mode_lock); |
1da177e4 LT |
853 | } |
854 | ||
855 | static int rlb_initialize(struct bonding *bond) | |
856 | { | |
857 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
0d206a3a | 858 | struct rlb_client_info *new_hashtbl; |
1da177e4 LT |
859 | int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info); |
860 | int i; | |
861 | ||
0d206a3a | 862 | new_hashtbl = kmalloc(size, GFP_KERNEL); |
e404decb | 863 | if (!new_hashtbl) |
1da177e4 | 864 | return -1; |
e404decb | 865 | |
4bab16d7 | 866 | spin_lock_bh(&bond->mode_lock); |
0d206a3a MW |
867 | |
868 | bond_info->rx_hashtbl = new_hashtbl; | |
1da177e4 | 869 | |
e53665c6 | 870 | bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX; |
1da177e4 | 871 | |
35d75ee4 | 872 | for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) |
1da177e4 | 873 | rlb_init_table_entry(bond_info->rx_hashtbl + i); |
1da177e4 | 874 | |
4bab16d7 | 875 | spin_unlock_bh(&bond->mode_lock); |
1da177e4 | 876 | |
1da177e4 | 877 | /* register to receive ARPs */ |
3aba891d | 878 | bond->recv_probe = rlb_arp_recv; |
1da177e4 LT |
879 | |
880 | return 0; | |
881 | } | |
882 | ||
883 | static void rlb_deinitialize(struct bonding *bond) | |
884 | { | |
885 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
886 | ||
4bab16d7 | 887 | spin_lock_bh(&bond->mode_lock); |
1da177e4 LT |
888 | |
889 | kfree(bond_info->rx_hashtbl); | |
890 | bond_info->rx_hashtbl = NULL; | |
e53665c6 | 891 | bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX; |
1da177e4 | 892 | |
4bab16d7 | 893 | spin_unlock_bh(&bond->mode_lock); |
1da177e4 LT |
894 | } |
895 | ||
896 | static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id) | |
897 | { | |
898 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
899 | u32 curr_index; | |
900 | ||
4bab16d7 | 901 | spin_lock_bh(&bond->mode_lock); |
1da177e4 | 902 | |
e53665c6 | 903 | curr_index = bond_info->rx_hashtbl_used_head; |
1da177e4 LT |
904 | while (curr_index != RLB_NULL_INDEX) { |
905 | struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]); | |
e53665c6 | 906 | u32 next_index = bond_info->rx_hashtbl[curr_index].used_next; |
1da177e4 | 907 | |
d3ab3ffd | 908 | if (curr->vlan_id == vlan_id) |
e53665c6 | 909 | rlb_delete_table_entry(bond, curr_index); |
1da177e4 LT |
910 | |
911 | curr_index = next_index; | |
912 | } | |
913 | ||
4bab16d7 | 914 | spin_unlock_bh(&bond->mode_lock); |
1da177e4 LT |
915 | } |
916 | ||
917 | /*********************** tlb/rlb shared functions *********************/ | |
918 | ||
6f238100 | 919 | static void alb_send_lp_vid(struct slave *slave, const u8 mac_addr[], |
d6b694c0 | 920 | __be16 vlan_proto, u16 vid) |
1da177e4 | 921 | { |
1da177e4 | 922 | struct learning_pkt pkt; |
7aa64981 | 923 | struct sk_buff *skb; |
1da177e4 | 924 | int size = sizeof(struct learning_pkt); |
1da177e4 LT |
925 | |
926 | memset(&pkt, 0, size); | |
ada0f863 JP |
927 | ether_addr_copy(pkt.mac_dst, mac_addr); |
928 | ether_addr_copy(pkt.mac_src, mac_addr); | |
96a0922c | 929 | pkt.type = cpu_to_be16(ETH_P_LOOPBACK); |
1da177e4 | 930 | |
7aa64981 VF |
931 | skb = dev_alloc_skb(size); |
932 | if (!skb) | |
933 | return; | |
1da177e4 | 934 | |
b952f4df | 935 | skb_put_data(skb, &pkt, size); |
7aa64981 VF |
936 | |
937 | skb_reset_mac_header(skb); | |
938 | skb->network_header = skb->mac_header + ETH_HLEN; | |
939 | skb->protocol = pkt.type; | |
940 | skb->priority = TC_PRIO_CONTROL; | |
941 | skb->dev = slave->dev; | |
942 | ||
7ea2e423 JW |
943 | slave_dbg(slave->bond->dev, slave->dev, |
944 | "Send learning packet: mac %pM vlan %d\n", mac_addr, vid); | |
21706ee8 | 945 | |
b4bef1b5 JP |
946 | if (vid) |
947 | __vlan_hwaccel_put_tag(skb, vlan_proto, vid); | |
7aa64981 VF |
948 | |
949 | dev_queue_xmit(skb); | |
950 | } | |
1da177e4 | 951 | |
b3208b20 DA |
952 | struct alb_walk_data { |
953 | struct bonding *bond; | |
954 | struct slave *slave; | |
6f238100 | 955 | const u8 *mac_addr; |
b3208b20 DA |
956 | bool strict_match; |
957 | }; | |
958 | ||
eff74233 TY |
959 | static int alb_upper_dev_walk(struct net_device *upper, |
960 | struct netdev_nested_priv *priv) | |
b3208b20 | 961 | { |
eff74233 | 962 | struct alb_walk_data *data = (struct alb_walk_data *)priv->data; |
b3208b20 | 963 | bool strict_match = data->strict_match; |
6f238100 | 964 | const u8 *mac_addr = data->mac_addr; |
b3208b20 DA |
965 | struct bonding *bond = data->bond; |
966 | struct slave *slave = data->slave; | |
b3208b20 DA |
967 | struct bond_vlan_tag *tags; |
968 | ||
21706ee8 | 969 | if (is_vlan_dev(upper) && |
f3b0a18b | 970 | bond->dev->lower_level == upper->lower_level - 1) { |
21706ee8 | 971 | if (upper->addr_assign_type == NET_ADDR_STOLEN) { |
b3208b20 DA |
972 | alb_send_lp_vid(slave, mac_addr, |
973 | vlan_dev_vlan_proto(upper), | |
974 | vlan_dev_vlan_id(upper)); | |
21706ee8 | 975 | } else { |
b3208b20 DA |
976 | alb_send_lp_vid(slave, upper->dev_addr, |
977 | vlan_dev_vlan_proto(upper), | |
978 | vlan_dev_vlan_id(upper)); | |
979 | } | |
980 | } | |
981 | ||
982 | /* If this is a macvlan device, then only send updates | |
983 | * when strict_match is turned off. | |
984 | */ | |
985 | if (netif_is_macvlan(upper) && !strict_match) { | |
986 | tags = bond_verify_device_path(bond->dev, upper, 0); | |
987 | if (IS_ERR_OR_NULL(tags)) | |
d6b83f1e ZS |
988 | return -ENOMEM; |
989 | ||
b3208b20 DA |
990 | alb_send_lp_vid(slave, upper->dev_addr, |
991 | tags[0].vlan_proto, tags[0].vlan_id); | |
992 | kfree(tags); | |
993 | } | |
994 | ||
995 | return 0; | |
996 | } | |
997 | ||
6f238100 | 998 | static void alb_send_learning_packets(struct slave *slave, const u8 mac_addr[], |
d0c21d43 | 999 | bool strict_match) |
7aa64981 VF |
1000 | { |
1001 | struct bonding *bond = bond_get_bond_by_slave(slave); | |
eff74233 | 1002 | struct netdev_nested_priv priv; |
b3208b20 DA |
1003 | struct alb_walk_data data = { |
1004 | .strict_match = strict_match, | |
1005 | .mac_addr = mac_addr, | |
1006 | .slave = slave, | |
1007 | .bond = bond, | |
1008 | }; | |
5bf94b83 | 1009 | |
eff74233 | 1010 | priv.data = (void *)&data; |
5bf94b83 | 1011 | /* send untagged */ |
d6b694c0 | 1012 | alb_send_lp_vid(slave, mac_addr, 0, 0); |
5bf94b83 | 1013 | |
14af9963 VY |
1014 | /* loop through all devices and see if we need to send a packet |
1015 | * for that device. | |
1016 | */ | |
5bf94b83 | 1017 | rcu_read_lock(); |
eff74233 | 1018 | netdev_walk_all_upper_dev_rcu(bond->dev, alb_upper_dev_walk, &priv); |
5bf94b83 | 1019 | rcu_read_unlock(); |
1da177e4 LT |
1020 | } |
1021 | ||
6f238100 | 1022 | static int alb_set_slave_mac_addr(struct slave *slave, const u8 addr[], |
faeeb317 | 1023 | unsigned int len) |
1da177e4 LT |
1024 | { |
1025 | struct net_device *dev = slave->dev; | |
faeeb317 | 1026 | struct sockaddr_storage ss; |
1da177e4 | 1027 | |
01844098 | 1028 | if (BOND_MODE(slave->bond) == BOND_MODE_TLB) { |
6f238100 | 1029 | __dev_addr_set(dev, addr, len); |
1da177e4 LT |
1030 | return 0; |
1031 | } | |
1032 | ||
547942ca NA |
1033 | /* for rlb each slave must have a unique hw mac addresses so that |
1034 | * each slave will receive packets destined to a different mac | |
1035 | */ | |
faeeb317 JW |
1036 | memcpy(ss.__data, addr, len); |
1037 | ss.ss_family = dev->type; | |
3a37a963 | 1038 | if (dev_set_mac_address(dev, (struct sockaddr *)&ss, NULL)) { |
7ea2e423 | 1039 | slave_err(slave->bond->dev, dev, "dev_set_mac_address on slave failed! ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n"); |
1da177e4 LT |
1040 | return -EOPNOTSUPP; |
1041 | } | |
1042 | return 0; | |
1043 | } | |
1044 | ||
547942ca | 1045 | /* Swap MAC addresses between two slaves. |
059fe7a5 JV |
1046 | * |
1047 | * Called with RTNL held, and no other locks. | |
059fe7a5 | 1048 | */ |
43547ea6 | 1049 | static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2) |
1da177e4 | 1050 | { |
faeeb317 | 1051 | u8 tmp_mac_addr[MAX_ADDR_LEN]; |
1da177e4 | 1052 | |
faeeb317 JW |
1053 | bond_hw_addr_copy(tmp_mac_addr, slave1->dev->dev_addr, |
1054 | slave1->dev->addr_len); | |
1055 | alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr, | |
1056 | slave2->dev->addr_len); | |
1057 | alb_set_slave_mac_addr(slave2, tmp_mac_addr, | |
1058 | slave1->dev->addr_len); | |
1da177e4 | 1059 | |
059fe7a5 JV |
1060 | } |
1061 | ||
547942ca | 1062 | /* Send learning packets after MAC address swap. |
059fe7a5 | 1063 | * |
2543331d | 1064 | * Called with RTNL and no other locks |
059fe7a5 JV |
1065 | */ |
1066 | static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1, | |
1067 | struct slave *slave2) | |
1068 | { | |
8557cd74 | 1069 | int slaves_state_differ = (bond_slave_can_tx(slave1) != bond_slave_can_tx(slave2)); |
059fe7a5 JV |
1070 | struct slave *disabled_slave = NULL; |
1071 | ||
2543331d JV |
1072 | ASSERT_RTNL(); |
1073 | ||
1da177e4 | 1074 | /* fasten the change in the switch */ |
8557cd74 | 1075 | if (bond_slave_can_tx(slave1)) { |
d0c21d43 | 1076 | alb_send_learning_packets(slave1, slave1->dev->dev_addr, false); |
1da177e4 LT |
1077 | if (bond->alb_info.rlb_enabled) { |
1078 | /* inform the clients that the mac address | |
1079 | * has changed | |
1080 | */ | |
1081 | rlb_req_update_slave_clients(bond, slave1); | |
1082 | } | |
1083 | } else { | |
1084 | disabled_slave = slave1; | |
1085 | } | |
1086 | ||
8557cd74 | 1087 | if (bond_slave_can_tx(slave2)) { |
d0c21d43 | 1088 | alb_send_learning_packets(slave2, slave2->dev->dev_addr, false); |
1da177e4 LT |
1089 | if (bond->alb_info.rlb_enabled) { |
1090 | /* inform the clients that the mac address | |
1091 | * has changed | |
1092 | */ | |
1093 | rlb_req_update_slave_clients(bond, slave2); | |
1094 | } | |
1095 | } else { | |
1096 | disabled_slave = slave2; | |
1097 | } | |
1098 | ||
1099 | if (bond->alb_info.rlb_enabled && slaves_state_differ) { | |
1100 | /* A disabled slave was assigned an active mac addr */ | |
1101 | rlb_teach_disabled_mac_on_primary(bond, | |
1102 | disabled_slave->dev->dev_addr); | |
1103 | } | |
1104 | } | |
1105 | ||
1106 | /** | |
1107 | * alb_change_hw_addr_on_detach | |
1108 | * @bond: bonding we're working on | |
1109 | * @slave: the slave that was just detached | |
1110 | * | |
1111 | * We assume that @slave was already detached from the slave list. | |
1112 | * | |
1113 | * If @slave's permanent hw address is different both from its current | |
1114 | * address and from @bond's address, then somewhere in the bond there's | |
1115 | * a slave that has @slave's permanet address as its current address. | |
252b5d37 | 1116 | * We'll make sure that slave no longer uses @slave's permanent address. |
1da177e4 | 1117 | * |
2543331d | 1118 | * Caller must hold RTNL and no other locks |
1da177e4 LT |
1119 | */ |
1120 | static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave) | |
1121 | { | |
1122 | int perm_curr_diff; | |
1123 | int perm_bond_diff; | |
b88ec38d | 1124 | struct slave *found_slave; |
1da177e4 | 1125 | |
a6700db1 JP |
1126 | perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr, |
1127 | slave->dev->dev_addr); | |
1128 | perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr, | |
1129 | bond->dev->dev_addr); | |
1da177e4 LT |
1130 | |
1131 | if (perm_curr_diff && perm_bond_diff) { | |
b88ec38d | 1132 | found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr); |
1da177e4 | 1133 | |
b88ec38d | 1134 | if (found_slave) { |
b88ec38d VF |
1135 | alb_swap_mac_addr(slave, found_slave); |
1136 | alb_fasten_mac_swap(bond, slave, found_slave); | |
1da177e4 LT |
1137 | } |
1138 | } | |
1139 | } | |
1140 | ||
1141 | /** | |
1142 | * alb_handle_addr_collision_on_attach | |
1143 | * @bond: bonding we're working on | |
1144 | * @slave: the slave that was just attached | |
1145 | * | |
1146 | * checks uniqueness of slave's mac address and handles the case the | |
1147 | * new slave uses the bonds mac address. | |
1148 | * | |
1149 | * If the permanent hw address of @slave is @bond's hw address, we need to | |
1150 | * find a different hw address to give @slave, that isn't in use by any other | |
77c8e2c0 | 1151 | * slave in the bond. This address must be, of course, one of the permanent |
1da177e4 LT |
1152 | * addresses of the other slaves. |
1153 | * | |
1154 | * We go over the slave list, and for each slave there we compare its | |
1155 | * permanent hw address with the current address of all the other slaves. | |
1156 | * If no match was found, then we've found a slave with a permanent address | |
1157 | * that isn't used by any other slave in the bond, so we can assign it to | |
1158 | * @slave. | |
1159 | * | |
1160 | * assumption: this function is called before @slave is attached to the | |
cedb743f | 1161 | * bond slave list. |
1da177e4 LT |
1162 | */ |
1163 | static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave) | |
1164 | { | |
4740d638 | 1165 | struct slave *has_bond_addr = rcu_access_pointer(bond->curr_active_slave); |
9caff1e7 VF |
1166 | struct slave *tmp_slave1, *free_mac_slave = NULL; |
1167 | struct list_head *iter; | |
1da177e4 | 1168 | |
0965a1f3 | 1169 | if (!bond_has_slaves(bond)) { |
1da177e4 LT |
1170 | /* this is the first slave */ |
1171 | return 0; | |
1172 | } | |
1173 | ||
1174 | /* if slave's mac address differs from bond's mac address | |
1175 | * check uniqueness of slave's mac address against the other | |
1176 | * slaves in the bond. | |
1177 | */ | |
a6700db1 | 1178 | if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) { |
cedb743f | 1179 | if (!bond_slave_has_mac(bond, slave->dev->dev_addr)) |
6b38aefe | 1180 | return 0; |
1da177e4 | 1181 | |
6b38aefe | 1182 | /* Try setting slave mac to bond address and fall-through |
547942ca NA |
1183 | * to code handling that situation below... |
1184 | */ | |
faeeb317 JW |
1185 | alb_set_slave_mac_addr(slave, bond->dev->dev_addr, |
1186 | bond->dev->addr_len); | |
1da177e4 LT |
1187 | } |
1188 | ||
1189 | /* The slave's address is equal to the address of the bond. | |
1190 | * Search for a spare address in the bond for this slave. | |
1191 | */ | |
9caff1e7 | 1192 | bond_for_each_slave(bond, tmp_slave1, iter) { |
cedb743f | 1193 | if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) { |
1da177e4 LT |
1194 | /* no slave has tmp_slave1's perm addr |
1195 | * as its curr addr | |
1196 | */ | |
1197 | free_mac_slave = tmp_slave1; | |
1198 | break; | |
1199 | } | |
1200 | ||
1201 | if (!has_bond_addr) { | |
a6700db1 JP |
1202 | if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr, |
1203 | bond->dev->dev_addr)) { | |
1da177e4 LT |
1204 | |
1205 | has_bond_addr = tmp_slave1; | |
1206 | } | |
1207 | } | |
1208 | } | |
1209 | ||
1210 | if (free_mac_slave) { | |
faeeb317 JW |
1211 | alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr, |
1212 | free_mac_slave->dev->addr_len); | |
1da177e4 | 1213 | |
7ea2e423 JW |
1214 | slave_warn(bond->dev, slave->dev, "the slave hw address is in use by the bond; giving it the hw address of %s\n", |
1215 | free_mac_slave->dev->name); | |
1da177e4 LT |
1216 | |
1217 | } else if (has_bond_addr) { | |
7ea2e423 | 1218 | slave_err(bond->dev, slave->dev, "the slave hw address 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"); |
1da177e4 LT |
1219 | return -EFAULT; |
1220 | } | |
1221 | ||
1222 | return 0; | |
1223 | } | |
1224 | ||
1225 | /** | |
1226 | * alb_set_mac_address | |
f6e81b89 LJ |
1227 | * @bond: bonding we're working on |
1228 | * @addr: MAC address to set | |
1da177e4 LT |
1229 | * |
1230 | * In TLB mode all slaves are configured to the bond's hw address, but set | |
1231 | * their dev_addr field to different addresses (based on their permanent hw | |
1232 | * addresses). | |
1233 | * | |
1234 | * For each slave, this function sets the interface to the new address and then | |
1235 | * changes its dev_addr field to its previous value. | |
1236 | * | |
1237 | * Unwinding assumes bond's mac address has not yet changed. | |
1238 | */ | |
1239 | static int alb_set_mac_address(struct bonding *bond, void *addr) | |
1240 | { | |
81f23b13 | 1241 | struct slave *slave, *rollback_slave; |
9caff1e7 | 1242 | struct list_head *iter; |
faeeb317 JW |
1243 | struct sockaddr_storage ss; |
1244 | char tmp_addr[MAX_ADDR_LEN]; | |
1da177e4 | 1245 | int res; |
1da177e4 | 1246 | |
dec1e90e | 1247 | if (bond->alb_info.rlb_enabled) |
1da177e4 | 1248 | return 0; |
1da177e4 | 1249 | |
9caff1e7 | 1250 | bond_for_each_slave(bond, slave, iter) { |
1da177e4 | 1251 | /* save net_device's current hw address */ |
faeeb317 JW |
1252 | bond_hw_addr_copy(tmp_addr, slave->dev->dev_addr, |
1253 | slave->dev->addr_len); | |
1da177e4 | 1254 | |
3a37a963 | 1255 | res = dev_set_mac_address(slave->dev, addr, NULL); |
1da177e4 LT |
1256 | |
1257 | /* restore net_device's hw address */ | |
6f238100 | 1258 | dev_addr_set(slave->dev, tmp_addr); |
1da177e4 | 1259 | |
eb7cc59a | 1260 | if (res) |
1da177e4 | 1261 | goto unwind; |
1da177e4 LT |
1262 | } |
1263 | ||
1264 | return 0; | |
1265 | ||
1266 | unwind: | |
faeeb317 JW |
1267 | memcpy(ss.__data, bond->dev->dev_addr, bond->dev->addr_len); |
1268 | ss.ss_family = bond->dev->type; | |
1da177e4 LT |
1269 | |
1270 | /* unwind from head to the slave that failed */ | |
9caff1e7 | 1271 | bond_for_each_slave(bond, rollback_slave, iter) { |
81f23b13 VF |
1272 | if (rollback_slave == slave) |
1273 | break; | |
faeeb317 JW |
1274 | bond_hw_addr_copy(tmp_addr, rollback_slave->dev->dev_addr, |
1275 | rollback_slave->dev->addr_len); | |
1276 | dev_set_mac_address(rollback_slave->dev, | |
3a37a963 | 1277 | (struct sockaddr *)&ss, NULL); |
6f238100 | 1278 | dev_addr_set(rollback_slave->dev, tmp_addr); |
1da177e4 LT |
1279 | } |
1280 | ||
1281 | return res; | |
1282 | } | |
1283 | ||
0da8aa00 SS |
1284 | /* determine if the packet is NA or NS */ |
1285 | static bool alb_determine_nd(struct sk_buff *skb, struct bonding *bond) | |
1286 | { | |
1287 | struct ipv6hdr *ip6hdr; | |
1288 | struct icmp6hdr *hdr; | |
1289 | ||
1290 | if (!pskb_network_may_pull(skb, sizeof(*ip6hdr))) | |
1291 | return true; | |
1292 | ||
1293 | ip6hdr = ipv6_hdr(skb); | |
1294 | if (ip6hdr->nexthdr != IPPROTO_ICMPV6) | |
1295 | return false; | |
1296 | ||
1297 | if (!pskb_network_may_pull(skb, sizeof(*ip6hdr) + sizeof(*hdr))) | |
1298 | return true; | |
1299 | ||
1300 | hdr = icmp6_hdr(skb); | |
1301 | return hdr->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT || | |
1302 | hdr->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION; | |
1303 | } | |
1304 | ||
4057c58d | 1305 | /************************ exported alb functions ************************/ |
1da177e4 LT |
1306 | |
1307 | int bond_alb_initialize(struct bonding *bond, int rlb_enabled) | |
1308 | { | |
1309 | int res; | |
1310 | ||
1311 | res = tlb_initialize(bond); | |
35d75ee4 | 1312 | if (res) |
1da177e4 | 1313 | return res; |
1da177e4 LT |
1314 | |
1315 | if (rlb_enabled) { | |
1da177e4 LT |
1316 | res = rlb_initialize(bond); |
1317 | if (res) { | |
1318 | tlb_deinitialize(bond); | |
1319 | return res; | |
1320 | } | |
ab84db25 | 1321 | bond->alb_info.rlb_enabled = 1; |
b76850ab MW |
1322 | } else { |
1323 | bond->alb_info.rlb_enabled = 0; | |
1da177e4 LT |
1324 | } |
1325 | ||
1326 | return 0; | |
1327 | } | |
1328 | ||
1329 | void bond_alb_deinitialize(struct bonding *bond) | |
1330 | { | |
1331 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
1332 | ||
1333 | tlb_deinitialize(bond); | |
1334 | ||
dda0fd5c | 1335 | if (bond_info->rlb_enabled) |
1da177e4 | 1336 | rlb_deinitialize(bond); |
1da177e4 LT |
1337 | } |
1338 | ||
dbdc8a21 TZ |
1339 | static netdev_tx_t bond_do_alb_xmit(struct sk_buff *skb, struct bonding *bond, |
1340 | struct slave *tx_slave) | |
9a49aba1 MB |
1341 | { |
1342 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
1343 | struct ethhdr *eth_data = eth_hdr(skb); | |
1344 | ||
1345 | if (!tx_slave) { | |
1346 | /* unbalanced or unassigned, send through primary */ | |
1347 | tx_slave = rcu_dereference(bond->curr_active_slave); | |
e9f0fb88 MB |
1348 | if (bond->params.tlb_dynamic_lb) |
1349 | bond_info->unbalanced_load += skb->len; | |
9a49aba1 MB |
1350 | } |
1351 | ||
8557cd74 | 1352 | if (tx_slave && bond_slave_can_tx(tx_slave)) { |
b5091b55 | 1353 | if (tx_slave != rcu_access_pointer(bond->curr_active_slave)) { |
9a49aba1 MB |
1354 | ether_addr_copy(eth_data->h_source, |
1355 | tx_slave->dev->dev_addr); | |
1356 | } | |
1357 | ||
ae46f184 | 1358 | return bond_dev_queue_xmit(bond, skb, tx_slave->dev); |
9a49aba1 MB |
1359 | } |
1360 | ||
e9f0fb88 | 1361 | if (tx_slave && bond->params.tlb_dynamic_lb) { |
4bab16d7 | 1362 | spin_lock(&bond->mode_lock); |
9a49aba1 | 1363 | __tlb_clear_slave(bond, tx_slave, 0); |
4bab16d7 | 1364 | spin_unlock(&bond->mode_lock); |
9a49aba1 MB |
1365 | } |
1366 | ||
1367 | /* no suitable interface, frame not sent */ | |
ae46f184 | 1368 | return bond_tx_drop(bond->dev, skb); |
9a49aba1 MB |
1369 | } |
1370 | ||
34b37e20 MG |
1371 | struct slave *bond_xmit_tlb_slave_get(struct bonding *bond, |
1372 | struct sk_buff *skb) | |
f05b42ea | 1373 | { |
f05b42ea | 1374 | struct slave *tx_slave = NULL; |
34b37e20 | 1375 | struct ethhdr *eth_data; |
f05b42ea MB |
1376 | u32 hash_index; |
1377 | ||
1378 | skb_reset_mac_header(skb); | |
1379 | eth_data = eth_hdr(skb); | |
1380 | ||
1381 | /* Do not TX balance any multicast or broadcast */ | |
1382 | if (!is_multicast_ether_addr(eth_data->h_dest)) { | |
1383 | switch (skb->protocol) { | |
f05b42ea | 1384 | case htons(ETH_P_IPV6): |
0da8aa00 SS |
1385 | if (alb_determine_nd(skb, bond)) |
1386 | break; | |
1387 | fallthrough; | |
1388 | case htons(ETH_P_IP): | |
f05b42ea | 1389 | hash_index = bond_xmit_hash(bond, skb); |
e9f0fb88 MB |
1390 | if (bond->params.tlb_dynamic_lb) { |
1391 | tx_slave = tlb_choose_channel(bond, | |
1392 | hash_index & 0xFF, | |
1393 | skb->len); | |
1394 | } else { | |
ee637714 MB |
1395 | struct bond_up_slave *slaves; |
1396 | unsigned int count; | |
e9f0fb88 | 1397 | |
ed7d4f02 | 1398 | slaves = rcu_dereference(bond->usable_slaves); |
6aa7de05 | 1399 | count = slaves ? READ_ONCE(slaves->count) : 0; |
ee637714 | 1400 | if (likely(count)) |
6b794c1c | 1401 | tx_slave = slaves->arr[hash_index % |
ee637714 | 1402 | count]; |
e9f0fb88 | 1403 | } |
f05b42ea MB |
1404 | break; |
1405 | } | |
1406 | } | |
34b37e20 | 1407 | return tx_slave; |
f05b42ea MB |
1408 | } |
1409 | ||
34b37e20 | 1410 | netdev_tx_t bond_tlb_xmit(struct sk_buff *skb, struct net_device *bond_dev) |
1da177e4 | 1411 | { |
454d7c9b | 1412 | struct bonding *bond = netdev_priv(bond_dev); |
34b37e20 MG |
1413 | struct slave *tx_slave; |
1414 | ||
1415 | tx_slave = bond_xmit_tlb_slave_get(bond, skb); | |
1416 | return bond_do_alb_xmit(skb, bond, tx_slave); | |
1417 | } | |
1418 | ||
1419 | struct slave *bond_xmit_alb_slave_get(struct bonding *bond, | |
1420 | struct sk_buff *skb) | |
1421 | { | |
1da177e4 | 1422 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); |
d3bb52b0 | 1423 | static const __be32 ip_bcast = htonl(0xffffffff); |
34b37e20 MG |
1424 | struct slave *tx_slave = NULL; |
1425 | const u8 *hash_start = NULL; | |
9a49aba1 | 1426 | bool do_tx_balance = true; |
34b37e20 | 1427 | struct ethhdr *eth_data; |
1da177e4 | 1428 | u32 hash_index = 0; |
34b37e20 | 1429 | int hash_size = 0; |
1da177e4 | 1430 | |
459a98ed | 1431 | skb_reset_mac_header(skb); |
1da177e4 LT |
1432 | eth_data = eth_hdr(skb); |
1433 | ||
1da177e4 | 1434 | switch (ntohs(skb->protocol)) { |
eddc9ec5 | 1435 | case ETH_P_IP: { |
38f88c45 | 1436 | const struct iphdr *iph; |
eddc9ec5 | 1437 | |
cbeeea70 | 1438 | if (is_broadcast_ether_addr(eth_data->h_dest) || |
38f88c45 ED |
1439 | !pskb_network_may_pull(skb, sizeof(*iph))) { |
1440 | do_tx_balance = false; | |
1441 | break; | |
1442 | } | |
1443 | iph = ip_hdr(skb); | |
1444 | if (iph->daddr == ip_bcast || iph->protocol == IPPROTO_IGMP) { | |
9a49aba1 | 1445 | do_tx_balance = false; |
1da177e4 LT |
1446 | break; |
1447 | } | |
eddc9ec5 ACM |
1448 | hash_start = (char *)&(iph->daddr); |
1449 | hash_size = sizeof(iph->daddr); | |
1da177e4 | 1450 | break; |
38f88c45 ED |
1451 | } |
1452 | case ETH_P_IPV6: { | |
1453 | const struct ipv6hdr *ip6hdr; | |
1454 | ||
2d1ea19d VY |
1455 | /* IPv6 doesn't really use broadcast mac address, but leave |
1456 | * that here just in case. | |
1457 | */ | |
cbeeea70 | 1458 | if (is_broadcast_ether_addr(eth_data->h_dest)) { |
9a49aba1 | 1459 | do_tx_balance = false; |
1da177e4 | 1460 | break; |
2d1ea19d VY |
1461 | } |
1462 | ||
1463 | /* IPv6 uses all-nodes multicast as an equivalent to | |
1464 | * broadcasts in IPv4. | |
1465 | */ | |
a6700db1 | 1466 | if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) { |
9a49aba1 | 1467 | do_tx_balance = false; |
2d1ea19d VY |
1468 | break; |
1469 | } | |
1470 | ||
0da8aa00 | 1471 | if (alb_determine_nd(skb, bond)) { |
38f88c45 ED |
1472 | do_tx_balance = false; |
1473 | break; | |
1474 | } | |
0da8aa00 SS |
1475 | |
1476 | /* The IPv6 header is pulled by alb_determine_nd */ | |
38f88c45 | 1477 | /* Additionally, DAD probes should not be tx-balanced as that |
2d1ea19d VY |
1478 | * will lead to false positives for duplicate addresses and |
1479 | * prevent address configuration from working. | |
1480 | */ | |
1481 | ip6hdr = ipv6_hdr(skb); | |
1482 | if (ipv6_addr_any(&ip6hdr->saddr)) { | |
9a49aba1 | 1483 | do_tx_balance = false; |
2d1ea19d | 1484 | break; |
1da177e4 LT |
1485 | } |
1486 | ||
38f88c45 ED |
1487 | hash_start = (char *)&ip6hdr->daddr; |
1488 | hash_size = sizeof(ip6hdr->daddr); | |
1da177e4 | 1489 | break; |
38f88c45 | 1490 | } |
1da177e4 | 1491 | case ETH_P_ARP: |
9a49aba1 | 1492 | do_tx_balance = false; |
dda0fd5c | 1493 | if (bond_info->rlb_enabled) |
1da177e4 | 1494 | tx_slave = rlb_arp_xmit(skb, bond); |
1da177e4 LT |
1495 | break; |
1496 | default: | |
9a49aba1 | 1497 | do_tx_balance = false; |
1da177e4 LT |
1498 | break; |
1499 | } | |
1500 | ||
1501 | if (do_tx_balance) { | |
e79c1055 DB |
1502 | if (bond->params.tlb_dynamic_lb) { |
1503 | hash_index = _simple_hash(hash_start, hash_size); | |
1504 | tx_slave = tlb_choose_channel(bond, hash_index, skb->len); | |
1505 | } else { | |
1506 | /* | |
1507 | * do_tx_balance means we are free to select the tx_slave | |
1508 | * So we do exactly what tlb would do for hash selection | |
1509 | */ | |
1510 | ||
1511 | struct bond_up_slave *slaves; | |
1512 | unsigned int count; | |
1513 | ||
ed7d4f02 | 1514 | slaves = rcu_dereference(bond->usable_slaves); |
e79c1055 DB |
1515 | count = slaves ? READ_ONCE(slaves->count) : 0; |
1516 | if (likely(count)) | |
1517 | tx_slave = slaves->arr[bond_xmit_hash(bond, skb) % | |
1518 | count]; | |
1519 | } | |
1da177e4 | 1520 | } |
34b37e20 MG |
1521 | return tx_slave; |
1522 | } | |
1523 | ||
1524 | netdev_tx_t bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev) | |
1525 | { | |
1526 | struct bonding *bond = netdev_priv(bond_dev); | |
1527 | struct slave *tx_slave = NULL; | |
1da177e4 | 1528 | |
34b37e20 | 1529 | tx_slave = bond_xmit_alb_slave_get(bond, skb); |
9a49aba1 | 1530 | return bond_do_alb_xmit(skb, bond, tx_slave); |
1da177e4 LT |
1531 | } |
1532 | ||
1b76b316 | 1533 | void bond_alb_monitor(struct work_struct *work) |
1da177e4 | 1534 | { |
1b76b316 JV |
1535 | struct bonding *bond = container_of(work, struct bonding, |
1536 | alb_work.work); | |
1da177e4 | 1537 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); |
9caff1e7 | 1538 | struct list_head *iter; |
1da177e4 | 1539 | struct slave *slave; |
1da177e4 | 1540 | |
0965a1f3 | 1541 | if (!bond_has_slaves(bond)) { |
dac8e00f | 1542 | atomic_set(&bond_info->tx_rebalance_counter, 0); |
1da177e4 LT |
1543 | bond_info->lp_counter = 0; |
1544 | goto re_arm; | |
1545 | } | |
1546 | ||
733ab639 | 1547 | rcu_read_lock(); |
1548 | ||
dac8e00f | 1549 | atomic_inc(&bond_info->tx_rebalance_counter); |
1da177e4 LT |
1550 | bond_info->lp_counter++; |
1551 | ||
1552 | /* send learning packets */ | |
7eacd038 | 1553 | if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) { |
d0c21d43 VY |
1554 | bool strict_match; |
1555 | ||
d0c21d43 VY |
1556 | bond_for_each_slave_rcu(bond, slave, iter) { |
1557 | /* If updating current_active, use all currently | |
4057c58d | 1558 | * user mac addresses (!strict_match). Otherwise, only |
d0c21d43 | 1559 | * use mac of the slave device. |
14af9963 | 1560 | * In RLB mode, we always use strict matches. |
d0c21d43 | 1561 | */ |
4740d638 | 1562 | strict_match = (slave != rcu_access_pointer(bond->curr_active_slave) || |
14af9963 | 1563 | bond_info->rlb_enabled); |
d0c21d43 VY |
1564 | alb_send_learning_packets(slave, slave->dev->dev_addr, |
1565 | strict_match); | |
1566 | } | |
1da177e4 LT |
1567 | bond_info->lp_counter = 0; |
1568 | } | |
1569 | ||
1570 | /* rebalance tx traffic */ | |
dac8e00f | 1571 | if (atomic_read(&bond_info->tx_rebalance_counter) >= BOND_TLB_REBALANCE_TICKS) { |
733ab639 | 1572 | bond_for_each_slave_rcu(bond, slave, iter) { |
1da177e4 | 1573 | tlb_clear_slave(bond, slave, 1); |
4740d638 | 1574 | if (slave == rcu_access_pointer(bond->curr_active_slave)) { |
1da177e4 LT |
1575 | SLAVE_TLB_INFO(slave).load = |
1576 | bond_info->unbalanced_load / | |
1577 | BOND_TLB_REBALANCE_INTERVAL; | |
1578 | bond_info->unbalanced_load = 0; | |
1579 | } | |
1580 | } | |
dac8e00f | 1581 | atomic_set(&bond_info->tx_rebalance_counter, 0); |
1da177e4 LT |
1582 | } |
1583 | ||
1da177e4 | 1584 | if (bond_info->rlb_enabled) { |
1da177e4 LT |
1585 | if (bond_info->primary_is_promisc && |
1586 | (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) { | |
1587 | ||
547942ca | 1588 | /* dev_set_promiscuity requires rtnl and |
1f2cd845 DM |
1589 | * nothing else. Avoid race with bond_close. |
1590 | */ | |
733ab639 | 1591 | rcu_read_unlock(); |
1592 | if (!rtnl_trylock()) | |
1f2cd845 | 1593 | goto re_arm; |
1f2cd845 | 1594 | |
1da177e4 LT |
1595 | bond_info->rlb_promisc_timeout_counter = 0; |
1596 | ||
1597 | /* If the primary was set to promiscuous mode | |
1598 | * because a slave was disabled then | |
1599 | * it can now leave promiscuous mode. | |
1600 | */ | |
4740d638 ED |
1601 | dev_set_promiscuity(rtnl_dereference(bond->curr_active_slave)->dev, |
1602 | -1); | |
1da177e4 | 1603 | bond_info->primary_is_promisc = 0; |
1f2cd845 DM |
1604 | |
1605 | rtnl_unlock(); | |
733ab639 | 1606 | rcu_read_lock(); |
d0e81b7e | 1607 | } |
1da177e4 LT |
1608 | |
1609 | if (bond_info->rlb_rebalance) { | |
1610 | bond_info->rlb_rebalance = 0; | |
1611 | rlb_rebalance(bond); | |
1612 | } | |
1613 | ||
1614 | /* check if clients need updating */ | |
1615 | if (bond_info->rx_ntt) { | |
1616 | if (bond_info->rlb_update_delay_counter) { | |
1617 | --bond_info->rlb_update_delay_counter; | |
1618 | } else { | |
1619 | rlb_update_rx_clients(bond); | |
dda0fd5c | 1620 | if (bond_info->rlb_update_retry_counter) |
1da177e4 | 1621 | --bond_info->rlb_update_retry_counter; |
dda0fd5c | 1622 | else |
1da177e4 | 1623 | bond_info->rx_ntt = 0; |
1da177e4 LT |
1624 | } |
1625 | } | |
1626 | } | |
733ab639 | 1627 | rcu_read_unlock(); |
1da177e4 | 1628 | re_arm: |
e6d265e8 | 1629 | queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks); |
1da177e4 LT |
1630 | } |
1631 | ||
1632 | /* assumption: called before the slave is attached to the bond | |
1633 | * and not locked by the bond lock | |
1634 | */ | |
1635 | int bond_alb_init_slave(struct bonding *bond, struct slave *slave) | |
1636 | { | |
1637 | int res; | |
1638 | ||
faeeb317 JW |
1639 | res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr, |
1640 | slave->dev->addr_len); | |
dda0fd5c | 1641 | if (res) |
1da177e4 | 1642 | return res; |
1da177e4 | 1643 | |
1da177e4 | 1644 | res = alb_handle_addr_collision_on_attach(bond, slave); |
dda0fd5c | 1645 | if (res) |
1da177e4 | 1646 | return res; |
1da177e4 LT |
1647 | |
1648 | tlb_init_slave(slave); | |
1649 | ||
1650 | /* order a rebalance ASAP */ | |
dac8e00f ED |
1651 | atomic_set(&bond->alb_info.tx_rebalance_counter, |
1652 | BOND_TLB_REBALANCE_TICKS); | |
1da177e4 | 1653 | |
dda0fd5c | 1654 | if (bond->alb_info.rlb_enabled) |
1da177e4 | 1655 | bond->alb_info.rlb_rebalance = 1; |
1da177e4 LT |
1656 | |
1657 | return 0; | |
1658 | } | |
1659 | ||
547942ca | 1660 | /* Remove slave from tlb and rlb hash tables, and fix up MAC addresses |
2543331d JV |
1661 | * if necessary. |
1662 | * | |
1663 | * Caller must hold RTNL and no other locks | |
1664 | */ | |
1da177e4 LT |
1665 | void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave) |
1666 | { | |
0965a1f3 | 1667 | if (bond_has_slaves(bond)) |
1da177e4 | 1668 | alb_change_hw_addr_on_detach(bond, slave); |
1da177e4 LT |
1669 | |
1670 | tlb_clear_slave(bond, slave, 0); | |
1671 | ||
1672 | if (bond->alb_info.rlb_enabled) { | |
6475ae4c | 1673 | bond->alb_info.rx_slave = NULL; |
1da177e4 LT |
1674 | rlb_clear_slave(bond, slave); |
1675 | } | |
6b794c1c | 1676 | |
1da177e4 LT |
1677 | } |
1678 | ||
1da177e4 LT |
1679 | void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link) |
1680 | { | |
1681 | struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond)); | |
1682 | ||
1683 | if (link == BOND_LINK_DOWN) { | |
1684 | tlb_clear_slave(bond, slave, 0); | |
73ac0cd4 | 1685 | if (bond->alb_info.rlb_enabled) |
1da177e4 | 1686 | rlb_clear_slave(bond, slave); |
1da177e4 LT |
1687 | } else if (link == BOND_LINK_UP) { |
1688 | /* order a rebalance ASAP */ | |
dac8e00f ED |
1689 | atomic_set(&bond_info->tx_rebalance_counter, |
1690 | BOND_TLB_REBALANCE_TICKS); | |
1da177e4 LT |
1691 | if (bond->alb_info.rlb_enabled) { |
1692 | bond->alb_info.rlb_rebalance = 1; | |
1693 | /* If the updelay module parameter is smaller than the | |
1694 | * forwarding delay of the switch the rebalance will | |
1695 | * not work because the rebalance arp replies will | |
1696 | * not be forwarded to the clients.. | |
1697 | */ | |
1698 | } | |
1699 | } | |
6b794c1c MB |
1700 | |
1701 | if (bond_is_nondyn_tlb(bond)) { | |
ee637714 | 1702 | if (bond_update_slave_arr(bond, NULL)) |
6b794c1c MB |
1703 | pr_err("Failed to build slave-array for TLB mode.\n"); |
1704 | } | |
1da177e4 LT |
1705 | } |
1706 | ||
1707 | /** | |
1708 | * bond_alb_handle_active_change - assign new curr_active_slave | |
1709 | * @bond: our bonding struct | |
1710 | * @new_slave: new slave to assign | |
1711 | * | |
1712 | * Set the bond->curr_active_slave to @new_slave and handle | |
1713 | * mac address swapping and promiscuity changes as needed. | |
1714 | * | |
62c5f518 | 1715 | * Caller must hold RTNL |
1da177e4 LT |
1716 | */ |
1717 | void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave) | |
1718 | { | |
1719 | struct slave *swap_slave; | |
4740d638 | 1720 | struct slave *curr_active; |
1da177e4 | 1721 | |
62c5f518 | 1722 | curr_active = rtnl_dereference(bond->curr_active_slave); |
4740d638 | 1723 | if (curr_active == new_slave) |
1da177e4 | 1724 | return; |
1da177e4 | 1725 | |
4740d638 ED |
1726 | if (curr_active && bond->alb_info.primary_is_promisc) { |
1727 | dev_set_promiscuity(curr_active->dev, -1); | |
1da177e4 LT |
1728 | bond->alb_info.primary_is_promisc = 0; |
1729 | bond->alb_info.rlb_promisc_timeout_counter = 0; | |
1730 | } | |
1731 | ||
4740d638 | 1732 | swap_slave = curr_active; |
278b2083 | 1733 | rcu_assign_pointer(bond->curr_active_slave, new_slave); |
1da177e4 | 1734 | |
0965a1f3 | 1735 | if (!new_slave || !bond_has_slaves(bond)) |
1da177e4 | 1736 | return; |
1da177e4 LT |
1737 | |
1738 | /* set the new curr_active_slave to the bonds mac address | |
1739 | * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave | |
1740 | */ | |
b88ec38d VF |
1741 | if (!swap_slave) |
1742 | swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr); | |
1da177e4 | 1743 | |
547942ca | 1744 | /* Arrange for swap_slave and new_slave to temporarily be |
059fe7a5 JV |
1745 | * ignored so we can mess with their MAC addresses without |
1746 | * fear of interference from transmit activity. | |
1747 | */ | |
dec1e90e | 1748 | if (swap_slave) |
059fe7a5 | 1749 | tlb_clear_slave(bond, swap_slave, 1); |
059fe7a5 JV |
1750 | tlb_clear_slave(bond, new_slave, 1); |
1751 | ||
4996b909 VF |
1752 | /* in TLB mode, the slave might flip down/up with the old dev_addr, |
1753 | * and thus filter bond->dev_addr's packets, so force bond's mac | |
1754 | */ | |
01844098 | 1755 | if (BOND_MODE(bond) == BOND_MODE_TLB) { |
faeeb317 JW |
1756 | struct sockaddr_storage ss; |
1757 | u8 tmp_addr[MAX_ADDR_LEN]; | |
4996b909 | 1758 | |
faeeb317 JW |
1759 | bond_hw_addr_copy(tmp_addr, new_slave->dev->dev_addr, |
1760 | new_slave->dev->addr_len); | |
4996b909 | 1761 | |
faeeb317 JW |
1762 | bond_hw_addr_copy(ss.__data, bond->dev->dev_addr, |
1763 | bond->dev->addr_len); | |
1764 | ss.ss_family = bond->dev->type; | |
4996b909 | 1765 | /* we don't care if it can't change its mac, best effort */ |
3a37a963 PM |
1766 | dev_set_mac_address(new_slave->dev, (struct sockaddr *)&ss, |
1767 | NULL); | |
4996b909 | 1768 | |
6f238100 | 1769 | dev_addr_set(new_slave->dev, tmp_addr); |
4996b909 VF |
1770 | } |
1771 | ||
1da177e4 LT |
1772 | /* curr_active_slave must be set before calling alb_swap_mac_addr */ |
1773 | if (swap_slave) { | |
1774 | /* swap mac address */ | |
43547ea6 | 1775 | alb_swap_mac_addr(swap_slave, new_slave); |
059fe7a5 JV |
1776 | alb_fasten_mac_swap(bond, swap_slave, new_slave); |
1777 | } else { | |
b88ec38d | 1778 | /* set the new_slave to the bond mac address */ |
faeeb317 JW |
1779 | alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr, |
1780 | bond->dev->addr_len); | |
d0c21d43 VY |
1781 | alb_send_learning_packets(new_slave, bond->dev->dev_addr, |
1782 | false); | |
1da177e4 LT |
1783 | } |
1784 | } | |
1785 | ||
ecfede42 | 1786 | /* Called with RTNL */ |
1da177e4 LT |
1787 | int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr) |
1788 | { | |
454d7c9b | 1789 | struct bonding *bond = netdev_priv(bond_dev); |
faeeb317 | 1790 | struct sockaddr_storage *ss = addr; |
4740d638 | 1791 | struct slave *curr_active; |
b88ec38d | 1792 | struct slave *swap_slave; |
1da177e4 | 1793 | int res; |
1da177e4 | 1794 | |
faeeb317 | 1795 | if (!is_valid_ether_addr(ss->__data)) |
1da177e4 | 1796 | return -EADDRNOTAVAIL; |
1da177e4 LT |
1797 | |
1798 | res = alb_set_mac_address(bond, addr); | |
73ac0cd4 | 1799 | if (res) |
1da177e4 | 1800 | return res; |
1da177e4 | 1801 | |
6f238100 | 1802 | dev_addr_set(bond_dev, ss->__data); |
1da177e4 LT |
1803 | |
1804 | /* If there is no curr_active_slave there is nothing else to do. | |
1805 | * Otherwise we'll need to pass the new address to it and handle | |
1806 | * duplications. | |
1807 | */ | |
4740d638 ED |
1808 | curr_active = rtnl_dereference(bond->curr_active_slave); |
1809 | if (!curr_active) | |
1da177e4 | 1810 | return 0; |
1da177e4 | 1811 | |
b88ec38d | 1812 | swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr); |
1da177e4 LT |
1813 | |
1814 | if (swap_slave) { | |
4740d638 ED |
1815 | alb_swap_mac_addr(swap_slave, curr_active); |
1816 | alb_fasten_mac_swap(bond, swap_slave, curr_active); | |
1da177e4 | 1817 | } else { |
faeeb317 JW |
1818 | alb_set_slave_mac_addr(curr_active, bond_dev->dev_addr, |
1819 | bond_dev->addr_len); | |
1da177e4 | 1820 | |
4740d638 | 1821 | alb_send_learning_packets(curr_active, |
d0c21d43 | 1822 | bond_dev->dev_addr, false); |
1da177e4 LT |
1823 | if (bond->alb_info.rlb_enabled) { |
1824 | /* inform clients mac address has changed */ | |
4740d638 | 1825 | rlb_req_update_slave_clients(bond, curr_active); |
1da177e4 LT |
1826 | } |
1827 | } | |
1828 | ||
1829 | return 0; | |
1830 | } | |
1831 | ||
1832 | void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id) | |
1833 | { | |
73ac0cd4 | 1834 | if (bond->alb_info.rlb_enabled) |
1da177e4 | 1835 | rlb_clear_vlan(bond, vlan_id); |
1da177e4 LT |
1836 | } |
1837 |