?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'block-5.11-2021-01-10' of git://git.kernel.dk/linux-block
[linux/fpc-iii.git] / drivers / net / bonding / bond_alb.c
blobc3091e00dd5fb295315c550509050ac16849387d
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved.
4 */
5
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>
20 #include <net/ipx.h>
21 #include <net/arp.h>
22 #include <net/ipv6.h>
23 #include <asm/byteorder.h>
24 #include <net/bonding.h>
25 #include <net/bond_alb.h>
26
27 static const u8 mac_v6_allmcast[ETH_ALEN + 2] __long_aligned = {
28 0x33, 0x33, 0x00, 0x00, 0x00, 0x01
29 };
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];
36 __be16 type;
37 u8 padding[ETH_ZLEN - ETH_HLEN];
38 };
39
40 struct arp_pkt {
41 __be16 hw_addr_space;
42 __be16 prot_addr_space;
43 u8 hw_addr_len;
44 u8 prot_addr_len;
45 __be16 op_code;
46 u8 mac_src[ETH_ALEN]; /* sender hardware address */
47 __be32 ip_src; /* sender IP address */
48 u8 mac_dst[ETH_ALEN]; /* target hardware address */
49 __be32 ip_dst; /* target IP address */
50 };
51 #pragma pack()
52
53 /* Forward declaration */
54 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
55 bool strict_match);
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);
60
61 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
62 {
63 int i;
64 u8 hash = 0;
65
66 for (i = 0; i < hash_size; i++)
67 hash ^= hash_start[i];
68
69 return hash;
70 }
71
72 /*********************** tlb specific functions ***************************/
73
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
93 static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
94 int save_load)
95 {
96 struct tlb_client_info *tx_hash_table;
97 u32 index;
98
99 /* clear slave from tx_hashtbl */
100 tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
101
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;
107 tlb_init_table_entry(&tx_hash_table[index], save_load);
108 index = next_index;
109 }
110 }
111
112 tlb_init_slave(slave);
113 }
114
115 static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
116 int save_load)
117 {
118 spin_lock_bh(&bond->mode_lock);
119 __tlb_clear_slave(bond, slave, save_load);
120 spin_unlock_bh(&bond->mode_lock);
121 }
122
123 /* Must be called before starting the monitor timer */
124 static int tlb_initialize(struct bonding *bond)
125 {
126 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
127 int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
128 struct tlb_client_info *new_hashtbl;
129 int i;
130
131 new_hashtbl = kzalloc(size, GFP_KERNEL);
132 if (!new_hashtbl)
133 return -ENOMEM;
134
135 spin_lock_bh(&bond->mode_lock);
136
137 bond_info->tx_hashtbl = new_hashtbl;
138
139 for (i = 0; i < TLB_HASH_TABLE_SIZE; i++)
140 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
141
142 spin_unlock_bh(&bond->mode_lock);
143
144 return 0;
145 }
146
147 /* Must be called only after all slaves have been released */
148 static void tlb_deinitialize(struct bonding *bond)
149 {
150 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
151
152 spin_lock_bh(&bond->mode_lock);
153
154 kfree(bond_info->tx_hashtbl);
155 bond_info->tx_hashtbl = NULL;
156
157 spin_unlock_bh(&bond->mode_lock);
158 }
159
160 static long long compute_gap(struct slave *slave)
161 {
162 return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
163 (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
164 }
165
166 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
167 {
168 struct slave *slave, *least_loaded;
169 struct list_head *iter;
170 long long max_gap;
171
172 least_loaded = NULL;
173 max_gap = LLONG_MIN;
174
175 /* Find the slave with the largest gap */
176 bond_for_each_slave_rcu(bond, slave, iter) {
177 if (bond_slave_can_tx(slave)) {
178 long long gap = compute_gap(slave);
179
180 if (max_gap < gap) {
181 least_loaded = slave;
182 max_gap = gap;
183 }
184 }
185 }
186
187 return least_loaded;
188 }
189
190 static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
191 u32 skb_len)
192 {
193 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
194 struct tlb_client_info *hash_table;
195 struct slave *assigned_slave;
196
197 hash_table = bond_info->tx_hashtbl;
198 assigned_slave = hash_table[hash_index].tx_slave;
199 if (!assigned_slave) {
200 assigned_slave = tlb_get_least_loaded_slave(bond);
201
202 if (assigned_slave) {
203 struct tlb_slave_info *slave_info =
204 &(SLAVE_TLB_INFO(assigned_slave));
205 u32 next_index = slave_info->head;
206
207 hash_table[hash_index].tx_slave = assigned_slave;
208 hash_table[hash_index].next = next_index;
209 hash_table[hash_index].prev = TLB_NULL_INDEX;
210
211 if (next_index != TLB_NULL_INDEX)
212 hash_table[next_index].prev = hash_index;
213
214 slave_info->head = hash_index;
215 slave_info->load +=
216 hash_table[hash_index].load_history;
217 }
218 }
219
220 if (assigned_slave)
221 hash_table[hash_index].tx_bytes += skb_len;
222
223 return assigned_slave;
224 }
225
226 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
227 u32 skb_len)
228 {
229 struct slave *tx_slave;
230
231 /* We don't need to disable softirq here, becase
232 * tlb_choose_channel() is only called by bond_alb_xmit()
233 * which already has softirq disabled.
234 */
235 spin_lock(&bond->mode_lock);
236 tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
237 spin_unlock(&bond->mode_lock);
238
239 return tx_slave;
240 }
241
242 /*********************** rlb specific functions ***************************/
243
244 /* when an ARP REPLY is received from a client update its info
245 * in the rx_hashtbl
246 */
247 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
248 {
249 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
250 struct rlb_client_info *client_info;
251 u32 hash_index;
252
253 spin_lock_bh(&bond->mode_lock);
254
255 hash_index = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
256 client_info = &(bond_info->rx_hashtbl[hash_index]);
257
258 if ((client_info->assigned) &&
259 (client_info->ip_src == arp->ip_dst) &&
260 (client_info->ip_dst == arp->ip_src) &&
261 (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) {
262 /* update the clients MAC address */
263 ether_addr_copy(client_info->mac_dst, arp->mac_src);
264 client_info->ntt = 1;
265 bond_info->rx_ntt = 1;
266 }
267
268 spin_unlock_bh(&bond->mode_lock);
269 }
270
271 static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
272 struct slave *slave)
273 {
274 struct arp_pkt *arp, _arp;
275
276 if (skb->protocol != cpu_to_be16(ETH_P_ARP))
277 goto out;
278
279 arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp);
280 if (!arp)
281 goto out;
282
283 /* We received an ARP from arp->ip_src.
284 * We might have used this IP address previously (on the bonding host
285 * itself or on a system that is bridged together with the bond).
286 * However, if arp->mac_src is different than what is stored in
287 * rx_hashtbl, some other host is now using the IP and we must prevent
288 * sending out client updates with this IP address and the old MAC
289 * address.
290 * Clean up all hash table entries that have this address as ip_src but
291 * have a different mac_src.
292 */
293 rlb_purge_src_ip(bond, arp);
294
295 if (arp->op_code == htons(ARPOP_REPLY)) {
296 /* update rx hash table for this ARP */
297 rlb_update_entry_from_arp(bond, arp);
298 slave_dbg(bond->dev, slave->dev, "Server received an ARP Reply from client\n");
299 }
300 out:
301 return RX_HANDLER_ANOTHER;
302 }
303
304 /* Caller must hold rcu_read_lock() */
305 static struct slave *__rlb_next_rx_slave(struct bonding *bond)
306 {
307 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
308 struct slave *before = NULL, *rx_slave = NULL, *slave;
309 struct list_head *iter;
310 bool found = false;
311
312 bond_for_each_slave_rcu(bond, slave, iter) {
313 if (!bond_slave_can_tx(slave))
314 continue;
315 if (!found) {
316 if (!before || before->speed < slave->speed)
317 before = slave;
318 } else {
319 if (!rx_slave || rx_slave->speed < slave->speed)
320 rx_slave = slave;
321 }
322 if (slave == bond_info->rx_slave)
323 found = true;
324 }
325 /* we didn't find anything after the current or we have something
326 * better before and up to the current slave
327 */
328 if (!rx_slave || (before && rx_slave->speed < before->speed))
329 rx_slave = before;
330
331 if (rx_slave)
332 bond_info->rx_slave = rx_slave;
333
334 return rx_slave;
335 }
336
337 /* Caller must hold RTNL, rcu_read_lock is obtained only to silence checkers */
338 static struct slave *rlb_next_rx_slave(struct bonding *bond)
339 {
340 struct slave *rx_slave;
341
342 ASSERT_RTNL();
343
344 rcu_read_lock();
345 rx_slave = __rlb_next_rx_slave(bond);
346 rcu_read_unlock();
347
348 return rx_slave;
349 }
350
351 /* teach the switch the mac of a disabled slave
352 * on the primary for fault tolerance
353 *
354 * Caller must hold RTNL
355 */
356 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
357 {
358 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
359
360 if (!curr_active)
361 return;
362
363 if (!bond->alb_info.primary_is_promisc) {
364 if (!dev_set_promiscuity(curr_active->dev, 1))
365 bond->alb_info.primary_is_promisc = 1;
366 else
367 bond->alb_info.primary_is_promisc = 0;
368 }
369
370 bond->alb_info.rlb_promisc_timeout_counter = 0;
371
372 alb_send_learning_packets(curr_active, addr, true);
373 }
374
375 /* slave being removed should not be active at this point
376 *
377 * Caller must hold rtnl.
378 */
379 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
380 {
381 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
382 struct rlb_client_info *rx_hash_table;
383 u32 index, next_index;
384
385 /* clear slave from rx_hashtbl */
386 spin_lock_bh(&bond->mode_lock);
387
388 rx_hash_table = bond_info->rx_hashtbl;
389 index = bond_info->rx_hashtbl_used_head;
390 for (; index != RLB_NULL_INDEX; index = next_index) {
391 next_index = rx_hash_table[index].used_next;
392 if (rx_hash_table[index].slave == slave) {
393 struct slave *assigned_slave = rlb_next_rx_slave(bond);
394
395 if (assigned_slave) {
396 rx_hash_table[index].slave = assigned_slave;
397 if (is_valid_ether_addr(rx_hash_table[index].mac_dst)) {
398 bond_info->rx_hashtbl[index].ntt = 1;
399 bond_info->rx_ntt = 1;
400 /* A slave has been removed from the
401 * table because it is either disabled
402 * or being released. We must retry the
403 * update to avoid clients from not
404 * being updated & disconnecting when
405 * there is stress
406 */
407 bond_info->rlb_update_retry_counter =
408 RLB_UPDATE_RETRY;
409 }
410 } else { /* there is no active slave */
411 rx_hash_table[index].slave = NULL;
412 }
413 }
414 }
415
416 spin_unlock_bh(&bond->mode_lock);
417
418 if (slave != rtnl_dereference(bond->curr_active_slave))
419 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
420 }
421
422 static void rlb_update_client(struct rlb_client_info *client_info)
423 {
424 int i;
425
426 if (!client_info->slave || !is_valid_ether_addr(client_info->mac_dst))
427 return;
428
429 for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
430 struct sk_buff *skb;
431
432 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
433 client_info->ip_dst,
434 client_info->slave->dev,
435 client_info->ip_src,
436 client_info->mac_dst,
437 client_info->slave->dev->dev_addr,
438 client_info->mac_dst);
439 if (!skb) {
440 slave_err(client_info->slave->bond->dev,
441 client_info->slave->dev,
442 "failed to create an ARP packet\n");
443 continue;
444 }
445
446 skb->dev = client_info->slave->dev;
447
448 if (client_info->vlan_id) {
449 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
450 client_info->vlan_id);
451 }
452
453 arp_xmit(skb);
454 }
455 }
456
457 /* sends ARP REPLIES that update the clients that need updating */
458 static void rlb_update_rx_clients(struct bonding *bond)
459 {
460 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
461 struct rlb_client_info *client_info;
462 u32 hash_index;
463
464 spin_lock_bh(&bond->mode_lock);
465
466 hash_index = bond_info->rx_hashtbl_used_head;
467 for (; hash_index != RLB_NULL_INDEX;
468 hash_index = client_info->used_next) {
469 client_info = &(bond_info->rx_hashtbl[hash_index]);
470 if (client_info->ntt) {
471 rlb_update_client(client_info);
472 if (bond_info->rlb_update_retry_counter == 0)
473 client_info->ntt = 0;
474 }
475 }
476
477 /* do not update the entries again until this counter is zero so that
478 * not to confuse the clients.
479 */
480 bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
481
482 spin_unlock_bh(&bond->mode_lock);
483 }
484
485 /* The slave was assigned a new mac address - update the clients */
486 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
487 {
488 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
489 struct rlb_client_info *client_info;
490 int ntt = 0;
491 u32 hash_index;
492
493 spin_lock_bh(&bond->mode_lock);
494
495 hash_index = bond_info->rx_hashtbl_used_head;
496 for (; hash_index != RLB_NULL_INDEX;
497 hash_index = client_info->used_next) {
498 client_info = &(bond_info->rx_hashtbl[hash_index]);
499
500 if ((client_info->slave == slave) &&
501 is_valid_ether_addr(client_info->mac_dst)) {
502 client_info->ntt = 1;
503 ntt = 1;
504 }
505 }
506
507 /* update the team's flag only after the whole iteration */
508 if (ntt) {
509 bond_info->rx_ntt = 1;
510 /* fasten the change */
511 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
512 }
513
514 spin_unlock_bh(&bond->mode_lock);
515 }
516
517 /* mark all clients using src_ip to be updated */
518 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
519 {
520 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
521 struct rlb_client_info *client_info;
522 u32 hash_index;
523
524 spin_lock(&bond->mode_lock);
525
526 hash_index = bond_info->rx_hashtbl_used_head;
527 for (; hash_index != RLB_NULL_INDEX;
528 hash_index = client_info->used_next) {
529 client_info = &(bond_info->rx_hashtbl[hash_index]);
530
531 if (!client_info->slave) {
532 netdev_err(bond->dev, "found a client with no channel in the client's hash table\n");
533 continue;
534 }
535 /* update all clients using this src_ip, that are not assigned
536 * to the team's address (curr_active_slave) and have a known
537 * unicast mac address.
538 */
539 if ((client_info->ip_src == src_ip) &&
540 !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
541 bond->dev->dev_addr) &&
542 is_valid_ether_addr(client_info->mac_dst)) {
543 client_info->ntt = 1;
544 bond_info->rx_ntt = 1;
545 }
546 }
547
548 spin_unlock(&bond->mode_lock);
549 }
550
551 static struct slave *rlb_choose_channel(struct sk_buff *skb,
552 struct bonding *bond,
553 const struct arp_pkt *arp)
554 {
555 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
556 struct slave *assigned_slave, *curr_active_slave;
557 struct rlb_client_info *client_info;
558 u32 hash_index = 0;
559
560 spin_lock(&bond->mode_lock);
561
562 curr_active_slave = rcu_dereference(bond->curr_active_slave);
563
564 hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
565 client_info = &(bond_info->rx_hashtbl[hash_index]);
566
567 if (client_info->assigned) {
568 if ((client_info->ip_src == arp->ip_src) &&
569 (client_info->ip_dst == arp->ip_dst)) {
570 /* the entry is already assigned to this client */
571 if (!is_broadcast_ether_addr(arp->mac_dst)) {
572 /* update mac address from arp */
573 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
574 }
575 ether_addr_copy(client_info->mac_src, arp->mac_src);
576
577 assigned_slave = client_info->slave;
578 if (assigned_slave) {
579 spin_unlock(&bond->mode_lock);
580 return assigned_slave;
581 }
582 } else {
583 /* the entry is already assigned to some other client,
584 * move the old client to primary (curr_active_slave) so
585 * that the new client can be assigned to this entry.
586 */
587 if (curr_active_slave &&
588 client_info->slave != curr_active_slave) {
589 client_info->slave = curr_active_slave;
590 rlb_update_client(client_info);
591 }
592 }
593 }
594 /* assign a new slave */
595 assigned_slave = __rlb_next_rx_slave(bond);
596
597 if (assigned_slave) {
598 if (!(client_info->assigned &&
599 client_info->ip_src == arp->ip_src)) {
600 /* ip_src is going to be updated,
601 * fix the src hash list
602 */
603 u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
604 sizeof(arp->ip_src));
605 rlb_src_unlink(bond, hash_index);
606 rlb_src_link(bond, hash_src, hash_index);
607 }
608
609 client_info->ip_src = arp->ip_src;
610 client_info->ip_dst = arp->ip_dst;
611 /* arp->mac_dst is broadcast for arp reqeusts.
612 * will be updated with clients actual unicast mac address
613 * upon receiving an arp reply.
614 */
615 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
616 ether_addr_copy(client_info->mac_src, arp->mac_src);
617 client_info->slave = assigned_slave;
618
619 if (is_valid_ether_addr(client_info->mac_dst)) {
620 client_info->ntt = 1;
621 bond->alb_info.rx_ntt = 1;
622 } else {
623 client_info->ntt = 0;
624 }
625
626 if (vlan_get_tag(skb, &client_info->vlan_id))
627 client_info->vlan_id = 0;
628
629 if (!client_info->assigned) {
630 u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
631 bond_info->rx_hashtbl_used_head = hash_index;
632 client_info->used_next = prev_tbl_head;
633 if (prev_tbl_head != RLB_NULL_INDEX) {
634 bond_info->rx_hashtbl[prev_tbl_head].used_prev =
635 hash_index;
636 }
637 client_info->assigned = 1;
638 }
639 }
640
641 spin_unlock(&bond->mode_lock);
642
643 return assigned_slave;
644 }
645
646 /* chooses (and returns) transmit channel for arp reply
647 * does not choose channel for other arp types since they are
648 * sent on the curr_active_slave
649 */
650 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
651 {
652 struct slave *tx_slave = NULL;
653 struct arp_pkt *arp;
654
655 if (!pskb_network_may_pull(skb, sizeof(*arp)))
656 return NULL;
657 arp = (struct arp_pkt *)skb_network_header(skb);
658
659 /* Don't modify or load balance ARPs that do not originate locally
660 * (e.g.,arrive via a bridge).
661 */
662 if (!bond_slave_has_mac_rx(bond, arp->mac_src))
663 return NULL;
664
665 if (arp->op_code == htons(ARPOP_REPLY)) {
666 /* the arp must be sent on the selected rx channel */
667 tx_slave = rlb_choose_channel(skb, bond, arp);
668 if (tx_slave)
669 bond_hw_addr_copy(arp->mac_src, tx_slave->dev->dev_addr,
670 tx_slave->dev->addr_len);
671 netdev_dbg(bond->dev, "(slave %s): Server sent ARP Reply packet\n",
672 tx_slave ? tx_slave->dev->name : "NULL");
673 } else if (arp->op_code == htons(ARPOP_REQUEST)) {
674 /* Create an entry in the rx_hashtbl for this client as a
675 * place holder.
676 * When the arp reply is received the entry will be updated
677 * with the correct unicast address of the client.
678 */
679 tx_slave = rlb_choose_channel(skb, bond, arp);
680
681 /* The ARP reply packets must be delayed so that
682 * they can cancel out the influence of the ARP request.
683 */
684 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
685
686 /* arp requests are broadcast and are sent on the primary
687 * the arp request will collapse all clients on the subnet to
688 * the primary slave. We must register these clients to be
689 * updated with their assigned mac.
690 */
691 rlb_req_update_subnet_clients(bond, arp->ip_src);
692 netdev_dbg(bond->dev, "(slave %s): Server sent ARP Request packet\n",
693 tx_slave ? tx_slave->dev->name : "NULL");
694 }
695
696 return tx_slave;
697 }
698
699 static void rlb_rebalance(struct bonding *bond)
700 {
701 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
702 struct slave *assigned_slave;
703 struct rlb_client_info *client_info;
704 int ntt;
705 u32 hash_index;
706
707 spin_lock_bh(&bond->mode_lock);
708
709 ntt = 0;
710 hash_index = bond_info->rx_hashtbl_used_head;
711 for (; hash_index != RLB_NULL_INDEX;
712 hash_index = client_info->used_next) {
713 client_info = &(bond_info->rx_hashtbl[hash_index]);
714 assigned_slave = __rlb_next_rx_slave(bond);
715 if (assigned_slave && (client_info->slave != assigned_slave)) {
716 client_info->slave = assigned_slave;
717 if (!is_zero_ether_addr(client_info->mac_dst)) {
718 client_info->ntt = 1;
719 ntt = 1;
720 }
721 }
722 }
723
724 /* update the team's flag only after the whole iteration */
725 if (ntt)
726 bond_info->rx_ntt = 1;
727 spin_unlock_bh(&bond->mode_lock);
728 }
729
730 /* Caller must hold mode_lock */
731 static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
732 {
733 entry->used_next = RLB_NULL_INDEX;
734 entry->used_prev = RLB_NULL_INDEX;
735 entry->assigned = 0;
736 entry->slave = NULL;
737 entry->vlan_id = 0;
738 }
739 static void rlb_init_table_entry_src(struct rlb_client_info *entry)
740 {
741 entry->src_first = RLB_NULL_INDEX;
742 entry->src_prev = RLB_NULL_INDEX;
743 entry->src_next = RLB_NULL_INDEX;
744 }
745
746 static void rlb_init_table_entry(struct rlb_client_info *entry)
747 {
748 memset(entry, 0, sizeof(struct rlb_client_info));
749 rlb_init_table_entry_dst(entry);
750 rlb_init_table_entry_src(entry);
751 }
752
753 static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
754 {
755 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
756 u32 next_index = bond_info->rx_hashtbl[index].used_next;
757 u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
758
759 if (index == bond_info->rx_hashtbl_used_head)
760 bond_info->rx_hashtbl_used_head = next_index;
761 if (prev_index != RLB_NULL_INDEX)
762 bond_info->rx_hashtbl[prev_index].used_next = next_index;
763 if (next_index != RLB_NULL_INDEX)
764 bond_info->rx_hashtbl[next_index].used_prev = prev_index;
765 }
766
767 /* unlink a rlb hash table entry from the src list */
768 static void rlb_src_unlink(struct bonding *bond, u32 index)
769 {
770 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
771 u32 next_index = bond_info->rx_hashtbl[index].src_next;
772 u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
773
774 bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
775 bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
776
777 if (next_index != RLB_NULL_INDEX)
778 bond_info->rx_hashtbl[next_index].src_prev = prev_index;
779
780 if (prev_index == RLB_NULL_INDEX)
781 return;
782
783 /* is prev_index pointing to the head of this list? */
784 if (bond_info->rx_hashtbl[prev_index].src_first == index)
785 bond_info->rx_hashtbl[prev_index].src_first = next_index;
786 else
787 bond_info->rx_hashtbl[prev_index].src_next = next_index;
788
789 }
790
791 static void rlb_delete_table_entry(struct bonding *bond, u32 index)
792 {
793 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
794 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
795
796 rlb_delete_table_entry_dst(bond, index);
797 rlb_init_table_entry_dst(entry);
798
799 rlb_src_unlink(bond, index);
800 }
801
802 /* add the rx_hashtbl[ip_dst_hash] entry to the list
803 * of entries with identical ip_src_hash
804 */
805 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
806 {
807 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
808 u32 next;
809
810 bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
811 next = bond_info->rx_hashtbl[ip_src_hash].src_first;
812 bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
813 if (next != RLB_NULL_INDEX)
814 bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
815 bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
816 }
817
818 /* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does
819 * not match arp->mac_src
820 */
821 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
822 {
823 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
824 u32 ip_src_hash = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
825 u32 index;
826
827 spin_lock_bh(&bond->mode_lock);
828
829 index = bond_info->rx_hashtbl[ip_src_hash].src_first;
830 while (index != RLB_NULL_INDEX) {
831 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
832 u32 next_index = entry->src_next;
833 if (entry->ip_src == arp->ip_src &&
834 !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
835 rlb_delete_table_entry(bond, index);
836 index = next_index;
837 }
838 spin_unlock_bh(&bond->mode_lock);
839 }
840
841 static int rlb_initialize(struct bonding *bond)
842 {
843 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
844 struct rlb_client_info *new_hashtbl;
845 int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
846 int i;
847
848 new_hashtbl = kmalloc(size, GFP_KERNEL);
849 if (!new_hashtbl)
850 return -1;
851
852 spin_lock_bh(&bond->mode_lock);
853
854 bond_info->rx_hashtbl = new_hashtbl;
855
856 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
857
858 for (i = 0; i < RLB_HASH_TABLE_SIZE; i++)
859 rlb_init_table_entry(bond_info->rx_hashtbl + i);
860
861 spin_unlock_bh(&bond->mode_lock);
862
863 /* register to receive ARPs */
864 bond->recv_probe = rlb_arp_recv;
865
866 return 0;
867 }
868
869 static void rlb_deinitialize(struct bonding *bond)
870 {
871 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
872
873 spin_lock_bh(&bond->mode_lock);
874
875 kfree(bond_info->rx_hashtbl);
876 bond_info->rx_hashtbl = NULL;
877 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
878
879 spin_unlock_bh(&bond->mode_lock);
880 }
881
882 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
883 {
884 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
885 u32 curr_index;
886
887 spin_lock_bh(&bond->mode_lock);
888
889 curr_index = bond_info->rx_hashtbl_used_head;
890 while (curr_index != RLB_NULL_INDEX) {
891 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
892 u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
893
894 if (curr->vlan_id == vlan_id)
895 rlb_delete_table_entry(bond, curr_index);
896
897 curr_index = next_index;
898 }
899
900 spin_unlock_bh(&bond->mode_lock);
901 }
902
903 /*********************** tlb/rlb shared functions *********************/
904
905 static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
906 __be16 vlan_proto, u16 vid)
907 {
908 struct learning_pkt pkt;
909 struct sk_buff *skb;
910 int size = sizeof(struct learning_pkt);
911
912 memset(&pkt, 0, size);
913 ether_addr_copy(pkt.mac_dst, mac_addr);
914 ether_addr_copy(pkt.mac_src, mac_addr);
915 pkt.type = cpu_to_be16(ETH_P_LOOPBACK);
916
917 skb = dev_alloc_skb(size);
918 if (!skb)
919 return;
920
921 skb_put_data(skb, &pkt, size);
922
923 skb_reset_mac_header(skb);
924 skb->network_header = skb->mac_header + ETH_HLEN;
925 skb->protocol = pkt.type;
926 skb->priority = TC_PRIO_CONTROL;
927 skb->dev = slave->dev;
928
929 slave_dbg(slave->bond->dev, slave->dev,
930 "Send learning packet: mac %pM vlan %d\n", mac_addr, vid);
931
932 if (vid)
933 __vlan_hwaccel_put_tag(skb, vlan_proto, vid);
934
935 dev_queue_xmit(skb);
936 }
937
938 struct alb_walk_data {
939 struct bonding *bond;
940 struct slave *slave;
941 u8 *mac_addr;
942 bool strict_match;
943 };
944
945 static int alb_upper_dev_walk(struct net_device *upper,
946 struct netdev_nested_priv *priv)
947 {
948 struct alb_walk_data *data = (struct alb_walk_data *)priv->data;
949 bool strict_match = data->strict_match;
950 struct bonding *bond = data->bond;
951 struct slave *slave = data->slave;
952 u8 *mac_addr = data->mac_addr;
953 struct bond_vlan_tag *tags;
954
955 if (is_vlan_dev(upper) &&
956 bond->dev->lower_level == upper->lower_level - 1) {
957 if (upper->addr_assign_type == NET_ADDR_STOLEN) {
958 alb_send_lp_vid(slave, mac_addr,
959 vlan_dev_vlan_proto(upper),
960 vlan_dev_vlan_id(upper));
961 } else {
962 alb_send_lp_vid(slave, upper->dev_addr,
963 vlan_dev_vlan_proto(upper),
964 vlan_dev_vlan_id(upper));
965 }
966 }
967
968 /* If this is a macvlan device, then only send updates
969 * when strict_match is turned off.
970 */
971 if (netif_is_macvlan(upper) && !strict_match) {
972 tags = bond_verify_device_path(bond->dev, upper, 0);
973 if (IS_ERR_OR_NULL(tags))
974 BUG();
975 alb_send_lp_vid(slave, upper->dev_addr,
976 tags[0].vlan_proto, tags[0].vlan_id);
977 kfree(tags);
978 }
979
980 return 0;
981 }
982
983 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
984 bool strict_match)
985 {
986 struct bonding *bond = bond_get_bond_by_slave(slave);
987 struct netdev_nested_priv priv;
988 struct alb_walk_data data = {
989 .strict_match = strict_match,
990 .mac_addr = mac_addr,
991 .slave = slave,
992 .bond = bond,
993 };
994
995 priv.data = (void *)&data;
996 /* send untagged */
997 alb_send_lp_vid(slave, mac_addr, 0, 0);
998
999 /* loop through all devices and see if we need to send a packet
1000 * for that device.
1001 */
1002 rcu_read_lock();
1003 netdev_walk_all_upper_dev_rcu(bond->dev, alb_upper_dev_walk, &priv);
1004 rcu_read_unlock();
1005 }
1006
1007 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[],
1008 unsigned int len)
1009 {
1010 struct net_device *dev = slave->dev;
1011 struct sockaddr_storage ss;
1012
1013 if (BOND_MODE(slave->bond) == BOND_MODE_TLB) {
1014 memcpy(dev->dev_addr, addr, len);
1015 return 0;
1016 }
1017
1018 /* for rlb each slave must have a unique hw mac addresses so that
1019 * each slave will receive packets destined to a different mac
1020 */
1021 memcpy(ss.__data, addr, len);
1022 ss.ss_family = dev->type;
1023 if (dev_set_mac_address(dev, (struct sockaddr *)&ss, NULL)) {
1024 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");
1025 return -EOPNOTSUPP;
1026 }
1027 return 0;
1028 }
1029
1030 /* Swap MAC addresses between two slaves.
1031 *
1032 * Called with RTNL held, and no other locks.
1033 */
1034 static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1035 {
1036 u8 tmp_mac_addr[MAX_ADDR_LEN];
1037
1038 bond_hw_addr_copy(tmp_mac_addr, slave1->dev->dev_addr,
1039 slave1->dev->addr_len);
1040 alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr,
1041 slave2->dev->addr_len);
1042 alb_set_slave_mac_addr(slave2, tmp_mac_addr,
1043 slave1->dev->addr_len);
1044
1045 }
1046
1047 /* Send learning packets after MAC address swap.
1048 *
1049 * Called with RTNL and no other locks
1050 */
1051 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1052 struct slave *slave2)
1053 {
1054 int slaves_state_differ = (bond_slave_can_tx(slave1) != bond_slave_can_tx(slave2));
1055 struct slave *disabled_slave = NULL;
1056
1057 ASSERT_RTNL();
1058
1059 /* fasten the change in the switch */
1060 if (bond_slave_can_tx(slave1)) {
1061 alb_send_learning_packets(slave1, slave1->dev->dev_addr, false);
1062 if (bond->alb_info.rlb_enabled) {
1063 /* inform the clients that the mac address
1064 * has changed
1065 */
1066 rlb_req_update_slave_clients(bond, slave1);
1067 }
1068 } else {
1069 disabled_slave = slave1;
1070 }
1071
1072 if (bond_slave_can_tx(slave2)) {
1073 alb_send_learning_packets(slave2, slave2->dev->dev_addr, false);
1074 if (bond->alb_info.rlb_enabled) {
1075 /* inform the clients that the mac address
1076 * has changed
1077 */
1078 rlb_req_update_slave_clients(bond, slave2);
1079 }
1080 } else {
1081 disabled_slave = slave2;
1082 }
1083
1084 if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1085 /* A disabled slave was assigned an active mac addr */
1086 rlb_teach_disabled_mac_on_primary(bond,
1087 disabled_slave->dev->dev_addr);
1088 }
1089 }
1090
1091 /**
1092 * alb_change_hw_addr_on_detach
1093 * @bond: bonding we're working on
1094 * @slave: the slave that was just detached
1095 *
1096 * We assume that @slave was already detached from the slave list.
1097 *
1098 * If @slave's permanent hw address is different both from its current
1099 * address and from @bond's address, then somewhere in the bond there's
1100 * a slave that has @slave's permanet address as its current address.
1101 * We'll make sure that that slave no longer uses @slave's permanent address.
1102 *
1103 * Caller must hold RTNL and no other locks
1104 */
1105 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1106 {
1107 int perm_curr_diff;
1108 int perm_bond_diff;
1109 struct slave *found_slave;
1110
1111 perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1112 slave->dev->dev_addr);
1113 perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1114 bond->dev->dev_addr);
1115
1116 if (perm_curr_diff && perm_bond_diff) {
1117 found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1118
1119 if (found_slave) {
1120 alb_swap_mac_addr(slave, found_slave);
1121 alb_fasten_mac_swap(bond, slave, found_slave);
1122 }
1123 }
1124 }
1125
1126 /**
1127 * alb_handle_addr_collision_on_attach
1128 * @bond: bonding we're working on
1129 * @slave: the slave that was just attached
1130 *
1131 * checks uniqueness of slave's mac address and handles the case the
1132 * new slave uses the bonds mac address.
1133 *
1134 * If the permanent hw address of @slave is @bond's hw address, we need to
1135 * find a different hw address to give @slave, that isn't in use by any other
1136 * slave in the bond. This address must be, of course, one of the permanent
1137 * addresses of the other slaves.
1138 *
1139 * We go over the slave list, and for each slave there we compare its
1140 * permanent hw address with the current address of all the other slaves.
1141 * If no match was found, then we've found a slave with a permanent address
1142 * that isn't used by any other slave in the bond, so we can assign it to
1143 * @slave.
1144 *
1145 * assumption: this function is called before @slave is attached to the
1146 * bond slave list.
1147 */
1148 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1149 {
1150 struct slave *has_bond_addr = rcu_access_pointer(bond->curr_active_slave);
1151 struct slave *tmp_slave1, *free_mac_slave = NULL;
1152 struct list_head *iter;
1153
1154 if (!bond_has_slaves(bond)) {
1155 /* this is the first slave */
1156 return 0;
1157 }
1158
1159 /* if slave's mac address differs from bond's mac address
1160 * check uniqueness of slave's mac address against the other
1161 * slaves in the bond.
1162 */
1163 if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1164 if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1165 return 0;
1166
1167 /* Try setting slave mac to bond address and fall-through
1168 * to code handling that situation below...
1169 */
1170 alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1171 bond->dev->addr_len);
1172 }
1173
1174 /* The slave's address is equal to the address of the bond.
1175 * Search for a spare address in the bond for this slave.
1176 */
1177 bond_for_each_slave(bond, tmp_slave1, iter) {
1178 if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1179 /* no slave has tmp_slave1's perm addr
1180 * as its curr addr
1181 */
1182 free_mac_slave = tmp_slave1;
1183 break;
1184 }
1185
1186 if (!has_bond_addr) {
1187 if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1188 bond->dev->dev_addr)) {
1189
1190 has_bond_addr = tmp_slave1;
1191 }
1192 }
1193 }
1194
1195 if (free_mac_slave) {
1196 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1197 free_mac_slave->dev->addr_len);
1198
1199 slave_warn(bond->dev, slave->dev, "the slave hw address is in use by the bond; giving it the hw address of %s\n",
1200 free_mac_slave->dev->name);
1201
1202 } else if (has_bond_addr) {
1203 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");
1204 return -EFAULT;
1205 }
1206
1207 return 0;
1208 }
1209
1210 /**
1211 * alb_set_mac_address
1212 * @bond: bonding we're working on
1213 * @addr: MAC address to set
1214 *
1215 * In TLB mode all slaves are configured to the bond's hw address, but set
1216 * their dev_addr field to different addresses (based on their permanent hw
1217 * addresses).
1218 *
1219 * For each slave, this function sets the interface to the new address and then
1220 * changes its dev_addr field to its previous value.
1221 *
1222 * Unwinding assumes bond's mac address has not yet changed.
1223 */
1224 static int alb_set_mac_address(struct bonding *bond, void *addr)
1225 {
1226 struct slave *slave, *rollback_slave;
1227 struct list_head *iter;
1228 struct sockaddr_storage ss;
1229 char tmp_addr[MAX_ADDR_LEN];
1230 int res;
1231
1232 if (bond->alb_info.rlb_enabled)
1233 return 0;
1234
1235 bond_for_each_slave(bond, slave, iter) {
1236 /* save net_device's current hw address */
1237 bond_hw_addr_copy(tmp_addr, slave->dev->dev_addr,
1238 slave->dev->addr_len);
1239
1240 res = dev_set_mac_address(slave->dev, addr, NULL);
1241
1242 /* restore net_device's hw address */
1243 bond_hw_addr_copy(slave->dev->dev_addr, tmp_addr,
1244 slave->dev->addr_len);
1245
1246 if (res)
1247 goto unwind;
1248 }
1249
1250 return 0;
1251
1252 unwind:
1253 memcpy(ss.__data, bond->dev->dev_addr, bond->dev->addr_len);
1254 ss.ss_family = bond->dev->type;
1255
1256 /* unwind from head to the slave that failed */
1257 bond_for_each_slave(bond, rollback_slave, iter) {
1258 if (rollback_slave == slave)
1259 break;
1260 bond_hw_addr_copy(tmp_addr, rollback_slave->dev->dev_addr,
1261 rollback_slave->dev->addr_len);
1262 dev_set_mac_address(rollback_slave->dev,
1263 (struct sockaddr *)&ss, NULL);
1264 bond_hw_addr_copy(rollback_slave->dev->dev_addr, tmp_addr,
1265 rollback_slave->dev->addr_len);
1266 }
1267
1268 return res;
1269 }
1270
1271 /************************ exported alb funcions ************************/
1272
1273 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1274 {
1275 int res;
1276
1277 res = tlb_initialize(bond);
1278 if (res)
1279 return res;
1280
1281 if (rlb_enabled) {
1282 bond->alb_info.rlb_enabled = 1;
1283 res = rlb_initialize(bond);
1284 if (res) {
1285 tlb_deinitialize(bond);
1286 return res;
1287 }
1288 } else {
1289 bond->alb_info.rlb_enabled = 0;
1290 }
1291
1292 return 0;
1293 }
1294
1295 void bond_alb_deinitialize(struct bonding *bond)
1296 {
1297 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1298
1299 tlb_deinitialize(bond);
1300
1301 if (bond_info->rlb_enabled)
1302 rlb_deinitialize(bond);
1303 }
1304
1305 static netdev_tx_t bond_do_alb_xmit(struct sk_buff *skb, struct bonding *bond,
1306 struct slave *tx_slave)
1307 {
1308 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1309 struct ethhdr *eth_data = eth_hdr(skb);
1310
1311 if (!tx_slave) {
1312 /* unbalanced or unassigned, send through primary */
1313 tx_slave = rcu_dereference(bond->curr_active_slave);
1314 if (bond->params.tlb_dynamic_lb)
1315 bond_info->unbalanced_load += skb->len;
1316 }
1317
1318 if (tx_slave && bond_slave_can_tx(tx_slave)) {
1319 if (tx_slave != rcu_access_pointer(bond->curr_active_slave)) {
1320 ether_addr_copy(eth_data->h_source,
1321 tx_slave->dev->dev_addr);
1322 }
1323
1324 return bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1325 }
1326
1327 if (tx_slave && bond->params.tlb_dynamic_lb) {
1328 spin_lock(&bond->mode_lock);
1329 __tlb_clear_slave(bond, tx_slave, 0);
1330 spin_unlock(&bond->mode_lock);
1331 }
1332
1333 /* no suitable interface, frame not sent */
1334 return bond_tx_drop(bond->dev, skb);
1335 }
1336
1337 struct slave *bond_xmit_tlb_slave_get(struct bonding *bond,
1338 struct sk_buff *skb)
1339 {
1340 struct slave *tx_slave = NULL;
1341 struct ethhdr *eth_data;
1342 u32 hash_index;
1343
1344 skb_reset_mac_header(skb);
1345 eth_data = eth_hdr(skb);
1346
1347 /* Do not TX balance any multicast or broadcast */
1348 if (!is_multicast_ether_addr(eth_data->h_dest)) {
1349 switch (skb->protocol) {
1350 case htons(ETH_P_IP):
1351 case htons(ETH_P_IPX):
1352 /* In case of IPX, it will falback to L2 hash */
1353 case htons(ETH_P_IPV6):
1354 hash_index = bond_xmit_hash(bond, skb);
1355 if (bond->params.tlb_dynamic_lb) {
1356 tx_slave = tlb_choose_channel(bond,
1357 hash_index & 0xFF,
1358 skb->len);
1359 } else {
1360 struct bond_up_slave *slaves;
1361 unsigned int count;
1362
1363 slaves = rcu_dereference(bond->usable_slaves);
1364 count = slaves ? READ_ONCE(slaves->count) : 0;
1365 if (likely(count))
1366 tx_slave = slaves->arr[hash_index %
1367 count];
1368 }
1369 break;
1370 }
1371 }
1372 return tx_slave;
1373 }
1374
1375 netdev_tx_t bond_tlb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1376 {
1377 struct bonding *bond = netdev_priv(bond_dev);
1378 struct slave *tx_slave;
1379
1380 tx_slave = bond_xmit_tlb_slave_get(bond, skb);
1381 return bond_do_alb_xmit(skb, bond, tx_slave);
1382 }
1383
1384 struct slave *bond_xmit_alb_slave_get(struct bonding *bond,
1385 struct sk_buff *skb)
1386 {
1387 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1388 static const __be32 ip_bcast = htonl(0xffffffff);
1389 struct slave *tx_slave = NULL;
1390 const u8 *hash_start = NULL;
1391 bool do_tx_balance = true;
1392 struct ethhdr *eth_data;
1393 u32 hash_index = 0;
1394 int hash_size = 0;
1395
1396 skb_reset_mac_header(skb);
1397 eth_data = eth_hdr(skb);
1398
1399 switch (ntohs(skb->protocol)) {
1400 case ETH_P_IP: {
1401 const struct iphdr *iph;
1402
1403 if (is_broadcast_ether_addr(eth_data->h_dest) ||
1404 !pskb_network_may_pull(skb, sizeof(*iph))) {
1405 do_tx_balance = false;
1406 break;
1407 }
1408 iph = ip_hdr(skb);
1409 if (iph->daddr == ip_bcast || iph->protocol == IPPROTO_IGMP) {
1410 do_tx_balance = false;
1411 break;
1412 }
1413 hash_start = (char *)&(iph->daddr);
1414 hash_size = sizeof(iph->daddr);
1415 break;
1416 }
1417 case ETH_P_IPV6: {
1418 const struct ipv6hdr *ip6hdr;
1419
1420 /* IPv6 doesn't really use broadcast mac address, but leave
1421 * that here just in case.
1422 */
1423 if (is_broadcast_ether_addr(eth_data->h_dest)) {
1424 do_tx_balance = false;
1425 break;
1426 }
1427
1428 /* IPv6 uses all-nodes multicast as an equivalent to
1429 * broadcasts in IPv4.
1430 */
1431 if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1432 do_tx_balance = false;
1433 break;
1434 }
1435
1436 if (!pskb_network_may_pull(skb, sizeof(*ip6hdr))) {
1437 do_tx_balance = false;
1438 break;
1439 }
1440 /* Additionally, DAD probes should not be tx-balanced as that
1441 * will lead to false positives for duplicate addresses and
1442 * prevent address configuration from working.
1443 */
1444 ip6hdr = ipv6_hdr(skb);
1445 if (ipv6_addr_any(&ip6hdr->saddr)) {
1446 do_tx_balance = false;
1447 break;
1448 }
1449
1450 hash_start = (char *)&ip6hdr->daddr;
1451 hash_size = sizeof(ip6hdr->daddr);
1452 break;
1453 }
1454 case ETH_P_IPX: {
1455 const struct ipxhdr *ipxhdr;
1456
1457 if (pskb_network_may_pull(skb, sizeof(*ipxhdr))) {
1458 do_tx_balance = false;
1459 break;
1460 }
1461 ipxhdr = (struct ipxhdr *)skb_network_header(skb);
1462
1463 if (ipxhdr->ipx_checksum != IPX_NO_CHECKSUM) {
1464 /* something is wrong with this packet */
1465 do_tx_balance = false;
1466 break;
1467 }
1468
1469 if (ipxhdr->ipx_type != IPX_TYPE_NCP) {
1470 /* The only protocol worth balancing in
1471 * this family since it has an "ARP" like
1472 * mechanism
1473 */
1474 do_tx_balance = false;
1475 break;
1476 }
1477
1478 eth_data = eth_hdr(skb);
1479 hash_start = (char *)eth_data->h_dest;
1480 hash_size = ETH_ALEN;
1481 break;
1482 }
1483 case ETH_P_ARP:
1484 do_tx_balance = false;
1485 if (bond_info->rlb_enabled)
1486 tx_slave = rlb_arp_xmit(skb, bond);
1487 break;
1488 default:
1489 do_tx_balance = false;
1490 break;
1491 }
1492
1493 if (do_tx_balance) {
1494 if (bond->params.tlb_dynamic_lb) {
1495 hash_index = _simple_hash(hash_start, hash_size);
1496 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1497 } else {
1498 /*
1499 * do_tx_balance means we are free to select the tx_slave
1500 * So we do exactly what tlb would do for hash selection
1501 */
1502
1503 struct bond_up_slave *slaves;
1504 unsigned int count;
1505
1506 slaves = rcu_dereference(bond->usable_slaves);
1507 count = slaves ? READ_ONCE(slaves->count) : 0;
1508 if (likely(count))
1509 tx_slave = slaves->arr[bond_xmit_hash(bond, skb) %
1510 count];
1511 }
1512 }
1513 return tx_slave;
1514 }
1515
1516 netdev_tx_t bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1517 {
1518 struct bonding *bond = netdev_priv(bond_dev);
1519 struct slave *tx_slave = NULL;
1520
1521 tx_slave = bond_xmit_alb_slave_get(bond, skb);
1522 return bond_do_alb_xmit(skb, bond, tx_slave);
1523 }
1524
1525 void bond_alb_monitor(struct work_struct *work)
1526 {
1527 struct bonding *bond = container_of(work, struct bonding,
1528 alb_work.work);
1529 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1530 struct list_head *iter;
1531 struct slave *slave;
1532
1533 if (!bond_has_slaves(bond)) {
1534 bond_info->tx_rebalance_counter = 0;
1535 bond_info->lp_counter = 0;
1536 goto re_arm;
1537 }
1538
1539 rcu_read_lock();
1540
1541 bond_info->tx_rebalance_counter++;
1542 bond_info->lp_counter++;
1543
1544 /* send learning packets */
1545 if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1546 bool strict_match;
1547
1548 bond_for_each_slave_rcu(bond, slave, iter) {
1549 /* If updating current_active, use all currently
1550 * user mac addreses (!strict_match). Otherwise, only
1551 * use mac of the slave device.
1552 * In RLB mode, we always use strict matches.
1553 */
1554 strict_match = (slave != rcu_access_pointer(bond->curr_active_slave) ||
1555 bond_info->rlb_enabled);
1556 alb_send_learning_packets(slave, slave->dev->dev_addr,
1557 strict_match);
1558 }
1559 bond_info->lp_counter = 0;
1560 }
1561
1562 /* rebalance tx traffic */
1563 if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1564 bond_for_each_slave_rcu(bond, slave, iter) {
1565 tlb_clear_slave(bond, slave, 1);
1566 if (slave == rcu_access_pointer(bond->curr_active_slave)) {
1567 SLAVE_TLB_INFO(slave).load =
1568 bond_info->unbalanced_load /
1569 BOND_TLB_REBALANCE_INTERVAL;
1570 bond_info->unbalanced_load = 0;
1571 }
1572 }
1573 bond_info->tx_rebalance_counter = 0;
1574 }
1575
1576 if (bond_info->rlb_enabled) {
1577 if (bond_info->primary_is_promisc &&
1578 (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1579
1580 /* dev_set_promiscuity requires rtnl and
1581 * nothing else. Avoid race with bond_close.
1582 */
1583 rcu_read_unlock();
1584 if (!rtnl_trylock())
1585 goto re_arm;
1586
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(rtnl_dereference(bond->curr_active_slave)->dev,
1594 -1);
1595 bond_info->primary_is_promisc = 0;
1596
1597 rtnl_unlock();
1598 rcu_read_lock();
1599 }
1600
1601 if (bond_info->rlb_rebalance) {
1602 bond_info->rlb_rebalance = 0;
1603 rlb_rebalance(bond);
1604 }
1605
1606 /* check if clients need updating */
1607 if (bond_info->rx_ntt) {
1608 if (bond_info->rlb_update_delay_counter) {
1609 --bond_info->rlb_update_delay_counter;
1610 } else {
1611 rlb_update_rx_clients(bond);
1612 if (bond_info->rlb_update_retry_counter)
1613 --bond_info->rlb_update_retry_counter;
1614 else
1615 bond_info->rx_ntt = 0;
1616 }
1617 }
1618 }
1619 rcu_read_unlock();
1620 re_arm:
1621 queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1622 }
1623
1624 /* assumption: called before the slave is attached to the bond
1625 * and not locked by the bond lock
1626 */
1627 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1628 {
1629 int res;
1630
1631 res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1632 slave->dev->addr_len);
1633 if (res)
1634 return res;
1635
1636 res = alb_handle_addr_collision_on_attach(bond, slave);
1637 if (res)
1638 return res;
1639
1640 tlb_init_slave(slave);
1641
1642 /* order a rebalance ASAP */
1643 bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1644
1645 if (bond->alb_info.rlb_enabled)
1646 bond->alb_info.rlb_rebalance = 1;
1647
1648 return 0;
1649 }
1650
1651 /* Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1652 * if necessary.
1653 *
1654 * Caller must hold RTNL and no other locks
1655 */
1656 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1657 {
1658 if (bond_has_slaves(bond))
1659 alb_change_hw_addr_on_detach(bond, slave);
1660
1661 tlb_clear_slave(bond, slave, 0);
1662
1663 if (bond->alb_info.rlb_enabled) {
1664 bond->alb_info.rx_slave = NULL;
1665 rlb_clear_slave(bond, slave);
1666 }
1667
1668 }
1669
1670 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1671 {
1672 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1673
1674 if (link == BOND_LINK_DOWN) {
1675 tlb_clear_slave(bond, slave, 0);
1676 if (bond->alb_info.rlb_enabled)
1677 rlb_clear_slave(bond, slave);
1678 } else if (link == BOND_LINK_UP) {
1679 /* order a rebalance ASAP */
1680 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1681 if (bond->alb_info.rlb_enabled) {
1682 bond->alb_info.rlb_rebalance = 1;
1683 /* If the updelay module parameter is smaller than the
1684 * forwarding delay of the switch the rebalance will
1685 * not work because the rebalance arp replies will
1686 * not be forwarded to the clients..
1687 */
1688 }
1689 }
1690
1691 if (bond_is_nondyn_tlb(bond)) {
1692 if (bond_update_slave_arr(bond, NULL))
1693 pr_err("Failed to build slave-array for TLB mode.\n");
1694 }
1695 }
1696
1697 /**
1698 * bond_alb_handle_active_change - assign new curr_active_slave
1699 * @bond: our bonding struct
1700 * @new_slave: new slave to assign
1701 *
1702 * Set the bond->curr_active_slave to @new_slave and handle
1703 * mac address swapping and promiscuity changes as needed.
1704 *
1705 * Caller must hold RTNL
1706 */
1707 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1708 {
1709 struct slave *swap_slave;
1710 struct slave *curr_active;
1711
1712 curr_active = rtnl_dereference(bond->curr_active_slave);
1713 if (curr_active == new_slave)
1714 return;
1715
1716 if (curr_active && bond->alb_info.primary_is_promisc) {
1717 dev_set_promiscuity(curr_active->dev, -1);
1718 bond->alb_info.primary_is_promisc = 0;
1719 bond->alb_info.rlb_promisc_timeout_counter = 0;
1720 }
1721
1722 swap_slave = curr_active;
1723 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1724
1725 if (!new_slave || !bond_has_slaves(bond))
1726 return;
1727
1728 /* set the new curr_active_slave to the bonds mac address
1729 * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1730 */
1731 if (!swap_slave)
1732 swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1733
1734 /* Arrange for swap_slave and new_slave to temporarily be
1735 * ignored so we can mess with their MAC addresses without
1736 * fear of interference from transmit activity.
1737 */
1738 if (swap_slave)
1739 tlb_clear_slave(bond, swap_slave, 1);
1740 tlb_clear_slave(bond, new_slave, 1);
1741
1742 /* in TLB mode, the slave might flip down/up with the old dev_addr,
1743 * and thus filter bond->dev_addr's packets, so force bond's mac
1744 */
1745 if (BOND_MODE(bond) == BOND_MODE_TLB) {
1746 struct sockaddr_storage ss;
1747 u8 tmp_addr[MAX_ADDR_LEN];
1748
1749 bond_hw_addr_copy(tmp_addr, new_slave->dev->dev_addr,
1750 new_slave->dev->addr_len);
1751
1752 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
1753 bond->dev->addr_len);
1754 ss.ss_family = bond->dev->type;
1755 /* we don't care if it can't change its mac, best effort */
1756 dev_set_mac_address(new_slave->dev, (struct sockaddr *)&ss,
1757 NULL);
1758
1759 bond_hw_addr_copy(new_slave->dev->dev_addr, tmp_addr,
1760 new_slave->dev->addr_len);
1761 }
1762
1763 /* curr_active_slave must be set before calling alb_swap_mac_addr */
1764 if (swap_slave) {
1765 /* swap mac address */
1766 alb_swap_mac_addr(swap_slave, new_slave);
1767 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1768 } else {
1769 /* set the new_slave to the bond mac address */
1770 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1771 bond->dev->addr_len);
1772 alb_send_learning_packets(new_slave, bond->dev->dev_addr,
1773 false);
1774 }
1775 }
1776
1777 /* Called with RTNL */
1778 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1779 {
1780 struct bonding *bond = netdev_priv(bond_dev);
1781 struct sockaddr_storage *ss = addr;
1782 struct slave *curr_active;
1783 struct slave *swap_slave;
1784 int res;
1785
1786 if (!is_valid_ether_addr(ss->__data))
1787 return -EADDRNOTAVAIL;
1788
1789 res = alb_set_mac_address(bond, addr);
1790 if (res)
1791 return res;
1792
1793 bond_hw_addr_copy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
1794
1795 /* If there is no curr_active_slave there is nothing else to do.
1796 * Otherwise we'll need to pass the new address to it and handle
1797 * duplications.
1798 */
1799 curr_active = rtnl_dereference(bond->curr_active_slave);
1800 if (!curr_active)
1801 return 0;
1802
1803 swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1804
1805 if (swap_slave) {
1806 alb_swap_mac_addr(swap_slave, curr_active);
1807 alb_fasten_mac_swap(bond, swap_slave, curr_active);
1808 } else {
1809 alb_set_slave_mac_addr(curr_active, bond_dev->dev_addr,
1810 bond_dev->addr_len);
1811
1812 alb_send_learning_packets(curr_active,
1813 bond_dev->dev_addr, false);
1814 if (bond->alb_info.rlb_enabled) {
1815 /* inform clients mac address has changed */
1816 rlb_req_update_slave_clients(bond, curr_active);
1817 }
1818 }
1819
1820 return 0;
1821 }
1822
1823 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1824 {
1825 if (bond->alb_info.rlb_enabled)
1826 rlb_clear_vlan(bond, vlan_id);
1827 }
1828
Linux with added FPC-III support