?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[cris-mirror.git] / drivers / net / bonding / bond_alb.c
blob1ed9529e7bd1de923697731dc0db0ceaa926ef1f
1 /*
2 * Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of the GNU General Public License as published by the
6 * Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 *
9 * This program is distributed in the hope that it will be useful, but
10 * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
11 * or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * for more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, see <http://www.gnu.org/licenses/>.
16 *
17 * The full GNU General Public License is included in this distribution in the
18 * file called LICENSE.
19 *
20 */
21
22 #include <linux/skbuff.h>
23 #include <linux/netdevice.h>
24 #include <linux/etherdevice.h>
25 #include <linux/pkt_sched.h>
26 #include <linux/spinlock.h>
27 #include <linux/slab.h>
28 #include <linux/timer.h>
29 #include <linux/ip.h>
30 #include <linux/ipv6.h>
31 #include <linux/if_arp.h>
32 #include <linux/if_ether.h>
33 #include <linux/if_bonding.h>
34 #include <linux/if_vlan.h>
35 #include <linux/in.h>
36 #include <net/ipx.h>
37 #include <net/arp.h>
38 #include <net/ipv6.h>
39 #include <asm/byteorder.h>
40 #include <net/bonding.h>
41 #include <net/bond_alb.h>
42
43
44
45 static const u8 mac_bcast[ETH_ALEN + 2] __long_aligned = {
46 0xff, 0xff, 0xff, 0xff, 0xff, 0xff
47 };
48 static const u8 mac_v6_allmcast[ETH_ALEN + 2] __long_aligned = {
49 0x33, 0x33, 0x00, 0x00, 0x00, 0x01
50 };
51 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
52
53 #pragma pack(1)
54 struct learning_pkt {
55 u8 mac_dst[ETH_ALEN];
56 u8 mac_src[ETH_ALEN];
57 __be16 type;
58 u8 padding[ETH_ZLEN - ETH_HLEN];
59 };
60
61 struct arp_pkt {
62 __be16 hw_addr_space;
63 __be16 prot_addr_space;
64 u8 hw_addr_len;
65 u8 prot_addr_len;
66 __be16 op_code;
67 u8 mac_src[ETH_ALEN]; /* sender hardware address */
68 __be32 ip_src; /* sender IP address */
69 u8 mac_dst[ETH_ALEN]; /* target hardware address */
70 __be32 ip_dst; /* target IP address */
71 };
72 #pragma pack()
73
74 static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb)
75 {
76 return (struct arp_pkt *)skb_network_header(skb);
77 }
78
79 /* Forward declaration */
80 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
81 bool strict_match);
82 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp);
83 static void rlb_src_unlink(struct bonding *bond, u32 index);
84 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash,
85 u32 ip_dst_hash);
86
87 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
88 {
89 int i;
90 u8 hash = 0;
91
92 for (i = 0; i < hash_size; i++)
93 hash ^= hash_start[i];
94
95 return hash;
96 }
97
98 /*********************** tlb specific functions ***************************/
99
100 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
101 {
102 if (save_load) {
103 entry->load_history = 1 + entry->tx_bytes /
104 BOND_TLB_REBALANCE_INTERVAL;
105 entry->tx_bytes = 0;
106 }
107
108 entry->tx_slave = NULL;
109 entry->next = TLB_NULL_INDEX;
110 entry->prev = TLB_NULL_INDEX;
111 }
112
113 static inline void tlb_init_slave(struct slave *slave)
114 {
115 SLAVE_TLB_INFO(slave).load = 0;
116 SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
117 }
118
119 static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
120 int save_load)
121 {
122 struct tlb_client_info *tx_hash_table;
123 u32 index;
124
125 /* clear slave from tx_hashtbl */
126 tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
127
128 /* skip this if we've already freed the tx hash table */
129 if (tx_hash_table) {
130 index = SLAVE_TLB_INFO(slave).head;
131 while (index != TLB_NULL_INDEX) {
132 u32 next_index = tx_hash_table[index].next;
133 tlb_init_table_entry(&tx_hash_table[index], save_load);
134 index = next_index;
135 }
136 }
137
138 tlb_init_slave(slave);
139 }
140
141 static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
142 int save_load)
143 {
144 spin_lock_bh(&bond->mode_lock);
145 __tlb_clear_slave(bond, slave, save_load);
146 spin_unlock_bh(&bond->mode_lock);
147 }
148
149 /* Must be called before starting the monitor timer */
150 static int tlb_initialize(struct bonding *bond)
151 {
152 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
153 int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
154 struct tlb_client_info *new_hashtbl;
155 int i;
156
157 new_hashtbl = kzalloc(size, GFP_KERNEL);
158 if (!new_hashtbl)
159 return -ENOMEM;
160
161 spin_lock_bh(&bond->mode_lock);
162
163 bond_info->tx_hashtbl = new_hashtbl;
164
165 for (i = 0; i < TLB_HASH_TABLE_SIZE; i++)
166 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
167
168 spin_unlock_bh(&bond->mode_lock);
169
170 return 0;
171 }
172
173 /* Must be called only after all slaves have been released */
174 static void tlb_deinitialize(struct bonding *bond)
175 {
176 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
177
178 spin_lock_bh(&bond->mode_lock);
179
180 kfree(bond_info->tx_hashtbl);
181 bond_info->tx_hashtbl = NULL;
182
183 spin_unlock_bh(&bond->mode_lock);
184 }
185
186 static long long compute_gap(struct slave *slave)
187 {
188 return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
189 (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
190 }
191
192 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
193 {
194 struct slave *slave, *least_loaded;
195 struct list_head *iter;
196 long long max_gap;
197
198 least_loaded = NULL;
199 max_gap = LLONG_MIN;
200
201 /* Find the slave with the largest gap */
202 bond_for_each_slave_rcu(bond, slave, iter) {
203 if (bond_slave_can_tx(slave)) {
204 long long gap = compute_gap(slave);
205
206 if (max_gap < gap) {
207 least_loaded = slave;
208 max_gap = gap;
209 }
210 }
211 }
212
213 return least_loaded;
214 }
215
216 static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
217 u32 skb_len)
218 {
219 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
220 struct tlb_client_info *hash_table;
221 struct slave *assigned_slave;
222
223 hash_table = bond_info->tx_hashtbl;
224 assigned_slave = hash_table[hash_index].tx_slave;
225 if (!assigned_slave) {
226 assigned_slave = tlb_get_least_loaded_slave(bond);
227
228 if (assigned_slave) {
229 struct tlb_slave_info *slave_info =
230 &(SLAVE_TLB_INFO(assigned_slave));
231 u32 next_index = slave_info->head;
232
233 hash_table[hash_index].tx_slave = assigned_slave;
234 hash_table[hash_index].next = next_index;
235 hash_table[hash_index].prev = TLB_NULL_INDEX;
236
237 if (next_index != TLB_NULL_INDEX)
238 hash_table[next_index].prev = hash_index;
239
240 slave_info->head = hash_index;
241 slave_info->load +=
242 hash_table[hash_index].load_history;
243 }
244 }
245
246 if (assigned_slave)
247 hash_table[hash_index].tx_bytes += skb_len;
248
249 return assigned_slave;
250 }
251
252 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
253 u32 skb_len)
254 {
255 struct slave *tx_slave;
256
257 /* We don't need to disable softirq here, becase
258 * tlb_choose_channel() is only called by bond_alb_xmit()
259 * which already has softirq disabled.
260 */
261 spin_lock(&bond->mode_lock);
262 tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
263 spin_unlock(&bond->mode_lock);
264
265 return tx_slave;
266 }
267
268 /*********************** rlb specific functions ***************************/
269
270 /* when an ARP REPLY is received from a client update its info
271 * in the rx_hashtbl
272 */
273 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
274 {
275 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
276 struct rlb_client_info *client_info;
277 u32 hash_index;
278
279 spin_lock_bh(&bond->mode_lock);
280
281 hash_index = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
282 client_info = &(bond_info->rx_hashtbl[hash_index]);
283
284 if ((client_info->assigned) &&
285 (client_info->ip_src == arp->ip_dst) &&
286 (client_info->ip_dst == arp->ip_src) &&
287 (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) {
288 /* update the clients MAC address */
289 ether_addr_copy(client_info->mac_dst, arp->mac_src);
290 client_info->ntt = 1;
291 bond_info->rx_ntt = 1;
292 }
293
294 spin_unlock_bh(&bond->mode_lock);
295 }
296
297 static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
298 struct slave *slave)
299 {
300 struct arp_pkt *arp, _arp;
301
302 if (skb->protocol != cpu_to_be16(ETH_P_ARP))
303 goto out;
304
305 arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp);
306 if (!arp)
307 goto out;
308
309 /* We received an ARP from arp->ip_src.
310 * We might have used this IP address previously (on the bonding host
311 * itself or on a system that is bridged together with the bond).
312 * However, if arp->mac_src is different than what is stored in
313 * rx_hashtbl, some other host is now using the IP and we must prevent
314 * sending out client updates with this IP address and the old MAC
315 * address.
316 * Clean up all hash table entries that have this address as ip_src but
317 * have a different mac_src.
318 */
319 rlb_purge_src_ip(bond, arp);
320
321 if (arp->op_code == htons(ARPOP_REPLY)) {
322 /* update rx hash table for this ARP */
323 rlb_update_entry_from_arp(bond, arp);
324 netdev_dbg(bond->dev, "Server received an ARP Reply from client\n");
325 }
326 out:
327 return RX_HANDLER_ANOTHER;
328 }
329
330 /* Caller must hold rcu_read_lock() */
331 static struct slave *__rlb_next_rx_slave(struct bonding *bond)
332 {
333 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
334 struct slave *before = NULL, *rx_slave = NULL, *slave;
335 struct list_head *iter;
336 bool found = false;
337
338 bond_for_each_slave_rcu(bond, slave, iter) {
339 if (!bond_slave_can_tx(slave))
340 continue;
341 if (!found) {
342 if (!before || before->speed < slave->speed)
343 before = slave;
344 } else {
345 if (!rx_slave || rx_slave->speed < slave->speed)
346 rx_slave = slave;
347 }
348 if (slave == bond_info->rx_slave)
349 found = true;
350 }
351 /* we didn't find anything after the current or we have something
352 * better before and up to the current slave
353 */
354 if (!rx_slave || (before && rx_slave->speed < before->speed))
355 rx_slave = before;
356
357 if (rx_slave)
358 bond_info->rx_slave = rx_slave;
359
360 return rx_slave;
361 }
362
363 /* Caller must hold RTNL, rcu_read_lock is obtained only to silence checkers */
364 static struct slave *rlb_next_rx_slave(struct bonding *bond)
365 {
366 struct slave *rx_slave;
367
368 ASSERT_RTNL();
369
370 rcu_read_lock();
371 rx_slave = __rlb_next_rx_slave(bond);
372 rcu_read_unlock();
373
374 return rx_slave;
375 }
376
377 /* teach the switch the mac of a disabled slave
378 * on the primary for fault tolerance
379 *
380 * Caller must hold RTNL
381 */
382 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
383 {
384 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
385
386 if (!curr_active)
387 return;
388
389 if (!bond->alb_info.primary_is_promisc) {
390 if (!dev_set_promiscuity(curr_active->dev, 1))
391 bond->alb_info.primary_is_promisc = 1;
392 else
393 bond->alb_info.primary_is_promisc = 0;
394 }
395
396 bond->alb_info.rlb_promisc_timeout_counter = 0;
397
398 alb_send_learning_packets(curr_active, addr, true);
399 }
400
401 /* slave being removed should not be active at this point
402 *
403 * Caller must hold rtnl.
404 */
405 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
406 {
407 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
408 struct rlb_client_info *rx_hash_table;
409 u32 index, next_index;
410
411 /* clear slave from rx_hashtbl */
412 spin_lock_bh(&bond->mode_lock);
413
414 rx_hash_table = bond_info->rx_hashtbl;
415 index = bond_info->rx_hashtbl_used_head;
416 for (; index != RLB_NULL_INDEX; index = next_index) {
417 next_index = rx_hash_table[index].used_next;
418 if (rx_hash_table[index].slave == slave) {
419 struct slave *assigned_slave = rlb_next_rx_slave(bond);
420
421 if (assigned_slave) {
422 rx_hash_table[index].slave = assigned_slave;
423 if (!ether_addr_equal_64bits(rx_hash_table[index].mac_dst,
424 mac_bcast)) {
425 bond_info->rx_hashtbl[index].ntt = 1;
426 bond_info->rx_ntt = 1;
427 /* A slave has been removed from the
428 * table because it is either disabled
429 * or being released. We must retry the
430 * update to avoid clients from not
431 * being updated & disconnecting when
432 * there is stress
433 */
434 bond_info->rlb_update_retry_counter =
435 RLB_UPDATE_RETRY;
436 }
437 } else { /* there is no active slave */
438 rx_hash_table[index].slave = NULL;
439 }
440 }
441 }
442
443 spin_unlock_bh(&bond->mode_lock);
444
445 if (slave != rtnl_dereference(bond->curr_active_slave))
446 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
447 }
448
449 static void rlb_update_client(struct rlb_client_info *client_info)
450 {
451 int i;
452
453 if (!client_info->slave)
454 return;
455
456 for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
457 struct sk_buff *skb;
458
459 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
460 client_info->ip_dst,
461 client_info->slave->dev,
462 client_info->ip_src,
463 client_info->mac_dst,
464 client_info->slave->dev->dev_addr,
465 client_info->mac_dst);
466 if (!skb) {
467 netdev_err(client_info->slave->bond->dev,
468 "failed to create an ARP packet\n");
469 continue;
470 }
471
472 skb->dev = client_info->slave->dev;
473
474 if (client_info->vlan_id) {
475 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
476 client_info->vlan_id);
477 }
478
479 arp_xmit(skb);
480 }
481 }
482
483 /* sends ARP REPLIES that update the clients that need updating */
484 static void rlb_update_rx_clients(struct bonding *bond)
485 {
486 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
487 struct rlb_client_info *client_info;
488 u32 hash_index;
489
490 spin_lock_bh(&bond->mode_lock);
491
492 hash_index = bond_info->rx_hashtbl_used_head;
493 for (; hash_index != RLB_NULL_INDEX;
494 hash_index = client_info->used_next) {
495 client_info = &(bond_info->rx_hashtbl[hash_index]);
496 if (client_info->ntt) {
497 rlb_update_client(client_info);
498 if (bond_info->rlb_update_retry_counter == 0)
499 client_info->ntt = 0;
500 }
501 }
502
503 /* do not update the entries again until this counter is zero so that
504 * not to confuse the clients.
505 */
506 bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
507
508 spin_unlock_bh(&bond->mode_lock);
509 }
510
511 /* The slave was assigned a new mac address - update the clients */
512 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
513 {
514 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
515 struct rlb_client_info *client_info;
516 int ntt = 0;
517 u32 hash_index;
518
519 spin_lock_bh(&bond->mode_lock);
520
521 hash_index = bond_info->rx_hashtbl_used_head;
522 for (; hash_index != RLB_NULL_INDEX;
523 hash_index = client_info->used_next) {
524 client_info = &(bond_info->rx_hashtbl[hash_index]);
525
526 if ((client_info->slave == slave) &&
527 !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
528 client_info->ntt = 1;
529 ntt = 1;
530 }
531 }
532
533 /* update the team's flag only after the whole iteration */
534 if (ntt) {
535 bond_info->rx_ntt = 1;
536 /* fasten the change */
537 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
538 }
539
540 spin_unlock_bh(&bond->mode_lock);
541 }
542
543 /* mark all clients using src_ip to be updated */
544 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
545 {
546 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
547 struct rlb_client_info *client_info;
548 u32 hash_index;
549
550 spin_lock(&bond->mode_lock);
551
552 hash_index = bond_info->rx_hashtbl_used_head;
553 for (; hash_index != RLB_NULL_INDEX;
554 hash_index = client_info->used_next) {
555 client_info = &(bond_info->rx_hashtbl[hash_index]);
556
557 if (!client_info->slave) {
558 netdev_err(bond->dev, "found a client with no channel in the client's hash table\n");
559 continue;
560 }
561 /* update all clients using this src_ip, that are not assigned
562 * to the team's address (curr_active_slave) and have a known
563 * unicast mac address.
564 */
565 if ((client_info->ip_src == src_ip) &&
566 !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
567 bond->dev->dev_addr) &&
568 !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
569 client_info->ntt = 1;
570 bond_info->rx_ntt = 1;
571 }
572 }
573
574 spin_unlock(&bond->mode_lock);
575 }
576
577 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
578 {
579 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
580 struct arp_pkt *arp = arp_pkt(skb);
581 struct slave *assigned_slave, *curr_active_slave;
582 struct rlb_client_info *client_info;
583 u32 hash_index = 0;
584
585 spin_lock(&bond->mode_lock);
586
587 curr_active_slave = rcu_dereference(bond->curr_active_slave);
588
589 hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
590 client_info = &(bond_info->rx_hashtbl[hash_index]);
591
592 if (client_info->assigned) {
593 if ((client_info->ip_src == arp->ip_src) &&
594 (client_info->ip_dst == arp->ip_dst)) {
595 /* the entry is already assigned to this client */
596 if (!ether_addr_equal_64bits(arp->mac_dst, mac_bcast)) {
597 /* update mac address from arp */
598 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
599 }
600 ether_addr_copy(client_info->mac_src, arp->mac_src);
601
602 assigned_slave = client_info->slave;
603 if (assigned_slave) {
604 spin_unlock(&bond->mode_lock);
605 return assigned_slave;
606 }
607 } else {
608 /* the entry is already assigned to some other client,
609 * move the old client to primary (curr_active_slave) so
610 * that the new client can be assigned to this entry.
611 */
612 if (curr_active_slave &&
613 client_info->slave != curr_active_slave) {
614 client_info->slave = curr_active_slave;
615 rlb_update_client(client_info);
616 }
617 }
618 }
619 /* assign a new slave */
620 assigned_slave = __rlb_next_rx_slave(bond);
621
622 if (assigned_slave) {
623 if (!(client_info->assigned &&
624 client_info->ip_src == arp->ip_src)) {
625 /* ip_src is going to be updated,
626 * fix the src hash list
627 */
628 u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
629 sizeof(arp->ip_src));
630 rlb_src_unlink(bond, hash_index);
631 rlb_src_link(bond, hash_src, hash_index);
632 }
633
634 client_info->ip_src = arp->ip_src;
635 client_info->ip_dst = arp->ip_dst;
636 /* arp->mac_dst is broadcast for arp reqeusts.
637 * will be updated with clients actual unicast mac address
638 * upon receiving an arp reply.
639 */
640 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
641 ether_addr_copy(client_info->mac_src, arp->mac_src);
642 client_info->slave = assigned_slave;
643
644 if (!ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
645 client_info->ntt = 1;
646 bond->alb_info.rx_ntt = 1;
647 } else {
648 client_info->ntt = 0;
649 }
650
651 if (vlan_get_tag(skb, &client_info->vlan_id))
652 client_info->vlan_id = 0;
653
654 if (!client_info->assigned) {
655 u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
656 bond_info->rx_hashtbl_used_head = hash_index;
657 client_info->used_next = prev_tbl_head;
658 if (prev_tbl_head != RLB_NULL_INDEX) {
659 bond_info->rx_hashtbl[prev_tbl_head].used_prev =
660 hash_index;
661 }
662 client_info->assigned = 1;
663 }
664 }
665
666 spin_unlock(&bond->mode_lock);
667
668 return assigned_slave;
669 }
670
671 /* chooses (and returns) transmit channel for arp reply
672 * does not choose channel for other arp types since they are
673 * sent on the curr_active_slave
674 */
675 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
676 {
677 struct arp_pkt *arp = arp_pkt(skb);
678 struct slave *tx_slave = NULL;
679
680 /* Don't modify or load balance ARPs that do not originate locally
681 * (e.g.,arrive via a bridge).
682 */
683 if (!bond_slave_has_mac_rx(bond, arp->mac_src))
684 return NULL;
685
686 if (arp->op_code == htons(ARPOP_REPLY)) {
687 /* the arp must be sent on the selected rx channel */
688 tx_slave = rlb_choose_channel(skb, bond);
689 if (tx_slave)
690 bond_hw_addr_copy(arp->mac_src, tx_slave->dev->dev_addr,
691 tx_slave->dev->addr_len);
692 netdev_dbg(bond->dev, "Server sent ARP Reply packet\n");
693 } else if (arp->op_code == htons(ARPOP_REQUEST)) {
694 /* Create an entry in the rx_hashtbl for this client as a
695 * place holder.
696 * When the arp reply is received the entry will be updated
697 * with the correct unicast address of the client.
698 */
699 rlb_choose_channel(skb, bond);
700
701 /* The ARP reply packets must be delayed so that
702 * they can cancel out the influence of the ARP request.
703 */
704 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
705
706 /* arp requests are broadcast and are sent on the primary
707 * the arp request will collapse all clients on the subnet to
708 * the primary slave. We must register these clients to be
709 * updated with their assigned mac.
710 */
711 rlb_req_update_subnet_clients(bond, arp->ip_src);
712 netdev_dbg(bond->dev, "Server sent ARP Request packet\n");
713 }
714
715 return tx_slave;
716 }
717
718 static void rlb_rebalance(struct bonding *bond)
719 {
720 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
721 struct slave *assigned_slave;
722 struct rlb_client_info *client_info;
723 int ntt;
724 u32 hash_index;
725
726 spin_lock_bh(&bond->mode_lock);
727
728 ntt = 0;
729 hash_index = bond_info->rx_hashtbl_used_head;
730 for (; hash_index != RLB_NULL_INDEX;
731 hash_index = client_info->used_next) {
732 client_info = &(bond_info->rx_hashtbl[hash_index]);
733 assigned_slave = __rlb_next_rx_slave(bond);
734 if (assigned_slave && (client_info->slave != assigned_slave)) {
735 client_info->slave = assigned_slave;
736 client_info->ntt = 1;
737 ntt = 1;
738 }
739 }
740
741 /* update the team's flag only after the whole iteration */
742 if (ntt)
743 bond_info->rx_ntt = 1;
744 spin_unlock_bh(&bond->mode_lock);
745 }
746
747 /* Caller must hold mode_lock */
748 static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
749 {
750 entry->used_next = RLB_NULL_INDEX;
751 entry->used_prev = RLB_NULL_INDEX;
752 entry->assigned = 0;
753 entry->slave = NULL;
754 entry->vlan_id = 0;
755 }
756 static void rlb_init_table_entry_src(struct rlb_client_info *entry)
757 {
758 entry->src_first = RLB_NULL_INDEX;
759 entry->src_prev = RLB_NULL_INDEX;
760 entry->src_next = RLB_NULL_INDEX;
761 }
762
763 static void rlb_init_table_entry(struct rlb_client_info *entry)
764 {
765 memset(entry, 0, sizeof(struct rlb_client_info));
766 rlb_init_table_entry_dst(entry);
767 rlb_init_table_entry_src(entry);
768 }
769
770 static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
771 {
772 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
773 u32 next_index = bond_info->rx_hashtbl[index].used_next;
774 u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
775
776 if (index == bond_info->rx_hashtbl_used_head)
777 bond_info->rx_hashtbl_used_head = next_index;
778 if (prev_index != RLB_NULL_INDEX)
779 bond_info->rx_hashtbl[prev_index].used_next = next_index;
780 if (next_index != RLB_NULL_INDEX)
781 bond_info->rx_hashtbl[next_index].used_prev = prev_index;
782 }
783
784 /* unlink a rlb hash table entry from the src list */
785 static void rlb_src_unlink(struct bonding *bond, u32 index)
786 {
787 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
788 u32 next_index = bond_info->rx_hashtbl[index].src_next;
789 u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
790
791 bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
792 bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
793
794 if (next_index != RLB_NULL_INDEX)
795 bond_info->rx_hashtbl[next_index].src_prev = prev_index;
796
797 if (prev_index == RLB_NULL_INDEX)
798 return;
799
800 /* is prev_index pointing to the head of this list? */
801 if (bond_info->rx_hashtbl[prev_index].src_first == index)
802 bond_info->rx_hashtbl[prev_index].src_first = next_index;
803 else
804 bond_info->rx_hashtbl[prev_index].src_next = next_index;
805
806 }
807
808 static void rlb_delete_table_entry(struct bonding *bond, u32 index)
809 {
810 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
811 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
812
813 rlb_delete_table_entry_dst(bond, index);
814 rlb_init_table_entry_dst(entry);
815
816 rlb_src_unlink(bond, index);
817 }
818
819 /* add the rx_hashtbl[ip_dst_hash] entry to the list
820 * of entries with identical ip_src_hash
821 */
822 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
823 {
824 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
825 u32 next;
826
827 bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
828 next = bond_info->rx_hashtbl[ip_src_hash].src_first;
829 bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
830 if (next != RLB_NULL_INDEX)
831 bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
832 bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
833 }
834
835 /* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does
836 * not match arp->mac_src
837 */
838 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
839 {
840 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
841 u32 ip_src_hash = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
842 u32 index;
843
844 spin_lock_bh(&bond->mode_lock);
845
846 index = bond_info->rx_hashtbl[ip_src_hash].src_first;
847 while (index != RLB_NULL_INDEX) {
848 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
849 u32 next_index = entry->src_next;
850 if (entry->ip_src == arp->ip_src &&
851 !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
852 rlb_delete_table_entry(bond, index);
853 index = next_index;
854 }
855 spin_unlock_bh(&bond->mode_lock);
856 }
857
858 static int rlb_initialize(struct bonding *bond)
859 {
860 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
861 struct rlb_client_info *new_hashtbl;
862 int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
863 int i;
864
865 new_hashtbl = kmalloc(size, GFP_KERNEL);
866 if (!new_hashtbl)
867 return -1;
868
869 spin_lock_bh(&bond->mode_lock);
870
871 bond_info->rx_hashtbl = new_hashtbl;
872
873 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
874
875 for (i = 0; i < RLB_HASH_TABLE_SIZE; i++)
876 rlb_init_table_entry(bond_info->rx_hashtbl + i);
877
878 spin_unlock_bh(&bond->mode_lock);
879
880 /* register to receive ARPs */
881 bond->recv_probe = rlb_arp_recv;
882
883 return 0;
884 }
885
886 static void rlb_deinitialize(struct bonding *bond)
887 {
888 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
889
890 spin_lock_bh(&bond->mode_lock);
891
892 kfree(bond_info->rx_hashtbl);
893 bond_info->rx_hashtbl = NULL;
894 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
895
896 spin_unlock_bh(&bond->mode_lock);
897 }
898
899 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
900 {
901 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
902 u32 curr_index;
903
904 spin_lock_bh(&bond->mode_lock);
905
906 curr_index = bond_info->rx_hashtbl_used_head;
907 while (curr_index != RLB_NULL_INDEX) {
908 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
909 u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
910
911 if (curr->vlan_id == vlan_id)
912 rlb_delete_table_entry(bond, curr_index);
913
914 curr_index = next_index;
915 }
916
917 spin_unlock_bh(&bond->mode_lock);
918 }
919
920 /*********************** tlb/rlb shared functions *********************/
921
922 static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
923 __be16 vlan_proto, u16 vid)
924 {
925 struct learning_pkt pkt;
926 struct sk_buff *skb;
927 int size = sizeof(struct learning_pkt);
928
929 memset(&pkt, 0, size);
930 ether_addr_copy(pkt.mac_dst, mac_addr);
931 ether_addr_copy(pkt.mac_src, mac_addr);
932 pkt.type = cpu_to_be16(ETH_P_LOOPBACK);
933
934 skb = dev_alloc_skb(size);
935 if (!skb)
936 return;
937
938 skb_put_data(skb, &pkt, size);
939
940 skb_reset_mac_header(skb);
941 skb->network_header = skb->mac_header + ETH_HLEN;
942 skb->protocol = pkt.type;
943 skb->priority = TC_PRIO_CONTROL;
944 skb->dev = slave->dev;
945
946 if (vid)
947 __vlan_hwaccel_put_tag(skb, vlan_proto, vid);
948
949 dev_queue_xmit(skb);
950 }
951
952 struct alb_walk_data {
953 struct bonding *bond;
954 struct slave *slave;
955 u8 *mac_addr;
956 bool strict_match;
957 };
958
959 static int alb_upper_dev_walk(struct net_device *upper, void *_data)
960 {
961 struct alb_walk_data *data = _data;
962 bool strict_match = data->strict_match;
963 struct bonding *bond = data->bond;
964 struct slave *slave = data->slave;
965 u8 *mac_addr = data->mac_addr;
966 struct bond_vlan_tag *tags;
967
968 if (is_vlan_dev(upper) && vlan_get_encap_level(upper) == 0) {
969 if (strict_match &&
970 ether_addr_equal_64bits(mac_addr,
971 upper->dev_addr)) {
972 alb_send_lp_vid(slave, mac_addr,
973 vlan_dev_vlan_proto(upper),
974 vlan_dev_vlan_id(upper));
975 } else if (!strict_match) {
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))
988 BUG();
989 alb_send_lp_vid(slave, upper->dev_addr,
990 tags[0].vlan_proto, tags[0].vlan_id);
991 kfree(tags);
992 }
993
994 return 0;
995 }
996
997 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
998 bool strict_match)
999 {
1000 struct bonding *bond = bond_get_bond_by_slave(slave);
1001 struct alb_walk_data data = {
1002 .strict_match = strict_match,
1003 .mac_addr = mac_addr,
1004 .slave = slave,
1005 .bond = bond,
1006 };
1007
1008 /* send untagged */
1009 alb_send_lp_vid(slave, mac_addr, 0, 0);
1010
1011 /* loop through all devices and see if we need to send a packet
1012 * for that device.
1013 */
1014 rcu_read_lock();
1015 netdev_walk_all_upper_dev_rcu(bond->dev, alb_upper_dev_walk, &data);
1016 rcu_read_unlock();
1017 }
1018
1019 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[],
1020 unsigned int len)
1021 {
1022 struct net_device *dev = slave->dev;
1023 struct sockaddr_storage ss;
1024
1025 if (BOND_MODE(slave->bond) == BOND_MODE_TLB) {
1026 memcpy(dev->dev_addr, addr, len);
1027 return 0;
1028 }
1029
1030 /* for rlb each slave must have a unique hw mac addresses so that
1031 * each slave will receive packets destined to a different mac
1032 */
1033 memcpy(ss.__data, addr, len);
1034 ss.ss_family = dev->type;
1035 if (dev_set_mac_address(dev, (struct sockaddr *)&ss)) {
1036 netdev_err(slave->bond->dev, "dev_set_mac_address of dev %s failed! ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n",
1037 dev->name);
1038 return -EOPNOTSUPP;
1039 }
1040 return 0;
1041 }
1042
1043 /* Swap MAC addresses between two slaves.
1044 *
1045 * Called with RTNL held, and no other locks.
1046 */
1047 static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1048 {
1049 u8 tmp_mac_addr[MAX_ADDR_LEN];
1050
1051 bond_hw_addr_copy(tmp_mac_addr, slave1->dev->dev_addr,
1052 slave1->dev->addr_len);
1053 alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr,
1054 slave2->dev->addr_len);
1055 alb_set_slave_mac_addr(slave2, tmp_mac_addr,
1056 slave1->dev->addr_len);
1057
1058 }
1059
1060 /* Send learning packets after MAC address swap.
1061 *
1062 * Called with RTNL and no other locks
1063 */
1064 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1065 struct slave *slave2)
1066 {
1067 int slaves_state_differ = (bond_slave_can_tx(slave1) != bond_slave_can_tx(slave2));
1068 struct slave *disabled_slave = NULL;
1069
1070 ASSERT_RTNL();
1071
1072 /* fasten the change in the switch */
1073 if (bond_slave_can_tx(slave1)) {
1074 alb_send_learning_packets(slave1, slave1->dev->dev_addr, false);
1075 if (bond->alb_info.rlb_enabled) {
1076 /* inform the clients that the mac address
1077 * has changed
1078 */
1079 rlb_req_update_slave_clients(bond, slave1);
1080 }
1081 } else {
1082 disabled_slave = slave1;
1083 }
1084
1085 if (bond_slave_can_tx(slave2)) {
1086 alb_send_learning_packets(slave2, slave2->dev->dev_addr, false);
1087 if (bond->alb_info.rlb_enabled) {
1088 /* inform the clients that the mac address
1089 * has changed
1090 */
1091 rlb_req_update_slave_clients(bond, slave2);
1092 }
1093 } else {
1094 disabled_slave = slave2;
1095 }
1096
1097 if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1098 /* A disabled slave was assigned an active mac addr */
1099 rlb_teach_disabled_mac_on_primary(bond,
1100 disabled_slave->dev->dev_addr);
1101 }
1102 }
1103
1104 /**
1105 * alb_change_hw_addr_on_detach
1106 * @bond: bonding we're working on
1107 * @slave: the slave that was just detached
1108 *
1109 * We assume that @slave was already detached from the slave list.
1110 *
1111 * If @slave's permanent hw address is different both from its current
1112 * address and from @bond's address, then somewhere in the bond there's
1113 * a slave that has @slave's permanet address as its current address.
1114 * We'll make sure that that slave no longer uses @slave's permanent address.
1115 *
1116 * Caller must hold RTNL and no other locks
1117 */
1118 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1119 {
1120 int perm_curr_diff;
1121 int perm_bond_diff;
1122 struct slave *found_slave;
1123
1124 perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1125 slave->dev->dev_addr);
1126 perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1127 bond->dev->dev_addr);
1128
1129 if (perm_curr_diff && perm_bond_diff) {
1130 found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1131
1132 if (found_slave) {
1133 alb_swap_mac_addr(slave, found_slave);
1134 alb_fasten_mac_swap(bond, slave, found_slave);
1135 }
1136 }
1137 }
1138
1139 /**
1140 * alb_handle_addr_collision_on_attach
1141 * @bond: bonding we're working on
1142 * @slave: the slave that was just attached
1143 *
1144 * checks uniqueness of slave's mac address and handles the case the
1145 * new slave uses the bonds mac address.
1146 *
1147 * If the permanent hw address of @slave is @bond's hw address, we need to
1148 * find a different hw address to give @slave, that isn't in use by any other
1149 * slave in the bond. This address must be, of course, one of the permanent
1150 * addresses of the other slaves.
1151 *
1152 * We go over the slave list, and for each slave there we compare its
1153 * permanent hw address with the current address of all the other slaves.
1154 * If no match was found, then we've found a slave with a permanent address
1155 * that isn't used by any other slave in the bond, so we can assign it to
1156 * @slave.
1157 *
1158 * assumption: this function is called before @slave is attached to the
1159 * bond slave list.
1160 */
1161 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1162 {
1163 struct slave *has_bond_addr = rcu_access_pointer(bond->curr_active_slave);
1164 struct slave *tmp_slave1, *free_mac_slave = NULL;
1165 struct list_head *iter;
1166
1167 if (!bond_has_slaves(bond)) {
1168 /* this is the first slave */
1169 return 0;
1170 }
1171
1172 /* if slave's mac address differs from bond's mac address
1173 * check uniqueness of slave's mac address against the other
1174 * slaves in the bond.
1175 */
1176 if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1177 if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1178 return 0;
1179
1180 /* Try setting slave mac to bond address and fall-through
1181 * to code handling that situation below...
1182 */
1183 alb_set_slave_mac_addr(slave, bond->dev->dev_addr,
1184 bond->dev->addr_len);
1185 }
1186
1187 /* The slave's address is equal to the address of the bond.
1188 * Search for a spare address in the bond for this slave.
1189 */
1190 bond_for_each_slave(bond, tmp_slave1, iter) {
1191 if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1192 /* no slave has tmp_slave1's perm addr
1193 * as its curr addr
1194 */
1195 free_mac_slave = tmp_slave1;
1196 break;
1197 }
1198
1199 if (!has_bond_addr) {
1200 if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1201 bond->dev->dev_addr)) {
1202
1203 has_bond_addr = tmp_slave1;
1204 }
1205 }
1206 }
1207
1208 if (free_mac_slave) {
1209 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1210 free_mac_slave->dev->addr_len);
1211
1212 netdev_warn(bond->dev, "the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1213 slave->dev->name, free_mac_slave->dev->name);
1214
1215 } else if (has_bond_addr) {
1216 netdev_err(bond->dev, "the hw address of slave %s is in use by the bond; couldn't find a slave with a free hw address to give it (this should not have happened)\n",
1217 slave->dev->name);
1218 return -EFAULT;
1219 }
1220
1221 return 0;
1222 }
1223
1224 /**
1225 * alb_set_mac_address
1226 * @bond:
1227 * @addr:
1228 *
1229 * In TLB mode all slaves are configured to the bond's hw address, but set
1230 * their dev_addr field to different addresses (based on their permanent hw
1231 * addresses).
1232 *
1233 * For each slave, this function sets the interface to the new address and then
1234 * changes its dev_addr field to its previous value.
1235 *
1236 * Unwinding assumes bond's mac address has not yet changed.
1237 */
1238 static int alb_set_mac_address(struct bonding *bond, void *addr)
1239 {
1240 struct slave *slave, *rollback_slave;
1241 struct list_head *iter;
1242 struct sockaddr_storage ss;
1243 char tmp_addr[MAX_ADDR_LEN];
1244 int res;
1245
1246 if (bond->alb_info.rlb_enabled)
1247 return 0;
1248
1249 bond_for_each_slave(bond, slave, iter) {
1250 /* save net_device's current hw address */
1251 bond_hw_addr_copy(tmp_addr, slave->dev->dev_addr,
1252 slave->dev->addr_len);
1253
1254 res = dev_set_mac_address(slave->dev, addr);
1255
1256 /* restore net_device's hw address */
1257 bond_hw_addr_copy(slave->dev->dev_addr, tmp_addr,
1258 slave->dev->addr_len);
1259
1260 if (res)
1261 goto unwind;
1262 }
1263
1264 return 0;
1265
1266 unwind:
1267 memcpy(ss.__data, bond->dev->dev_addr, bond->dev->addr_len);
1268 ss.ss_family = bond->dev->type;
1269
1270 /* unwind from head to the slave that failed */
1271 bond_for_each_slave(bond, rollback_slave, iter) {
1272 if (rollback_slave == slave)
1273 break;
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,
1277 (struct sockaddr *)&ss);
1278 bond_hw_addr_copy(rollback_slave->dev->dev_addr, tmp_addr,
1279 rollback_slave->dev->addr_len);
1280 }
1281
1282 return res;
1283 }
1284
1285 /************************ exported alb funcions ************************/
1286
1287 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1288 {
1289 int res;
1290
1291 res = tlb_initialize(bond);
1292 if (res)
1293 return res;
1294
1295 if (rlb_enabled) {
1296 bond->alb_info.rlb_enabled = 1;
1297 res = rlb_initialize(bond);
1298 if (res) {
1299 tlb_deinitialize(bond);
1300 return res;
1301 }
1302 } else {
1303 bond->alb_info.rlb_enabled = 0;
1304 }
1305
1306 return 0;
1307 }
1308
1309 void bond_alb_deinitialize(struct bonding *bond)
1310 {
1311 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1312
1313 tlb_deinitialize(bond);
1314
1315 if (bond_info->rlb_enabled)
1316 rlb_deinitialize(bond);
1317 }
1318
1319 static int bond_do_alb_xmit(struct sk_buff *skb, struct bonding *bond,
1320 struct slave *tx_slave)
1321 {
1322 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1323 struct ethhdr *eth_data = eth_hdr(skb);
1324
1325 if (!tx_slave) {
1326 /* unbalanced or unassigned, send through primary */
1327 tx_slave = rcu_dereference(bond->curr_active_slave);
1328 if (bond->params.tlb_dynamic_lb)
1329 bond_info->unbalanced_load += skb->len;
1330 }
1331
1332 if (tx_slave && bond_slave_can_tx(tx_slave)) {
1333 if (tx_slave != rcu_access_pointer(bond->curr_active_slave)) {
1334 ether_addr_copy(eth_data->h_source,
1335 tx_slave->dev->dev_addr);
1336 }
1337
1338 bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1339 goto out;
1340 }
1341
1342 if (tx_slave && bond->params.tlb_dynamic_lb) {
1343 spin_lock(&bond->mode_lock);
1344 __tlb_clear_slave(bond, tx_slave, 0);
1345 spin_unlock(&bond->mode_lock);
1346 }
1347
1348 /* no suitable interface, frame not sent */
1349 bond_tx_drop(bond->dev, skb);
1350 out:
1351 return NETDEV_TX_OK;
1352 }
1353
1354 int bond_tlb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1355 {
1356 struct bonding *bond = netdev_priv(bond_dev);
1357 struct ethhdr *eth_data;
1358 struct slave *tx_slave = NULL;
1359 u32 hash_index;
1360
1361 skb_reset_mac_header(skb);
1362 eth_data = eth_hdr(skb);
1363
1364 /* Do not TX balance any multicast or broadcast */
1365 if (!is_multicast_ether_addr(eth_data->h_dest)) {
1366 switch (skb->protocol) {
1367 case htons(ETH_P_IP):
1368 case htons(ETH_P_IPX):
1369 /* In case of IPX, it will falback to L2 hash */
1370 case htons(ETH_P_IPV6):
1371 hash_index = bond_xmit_hash(bond, skb);
1372 if (bond->params.tlb_dynamic_lb) {
1373 tx_slave = tlb_choose_channel(bond,
1374 hash_index & 0xFF,
1375 skb->len);
1376 } else {
1377 struct bond_up_slave *slaves;
1378 unsigned int count;
1379
1380 slaves = rcu_dereference(bond->slave_arr);
1381 count = slaves ? READ_ONCE(slaves->count) : 0;
1382 if (likely(count))
1383 tx_slave = slaves->arr[hash_index %
1384 count];
1385 }
1386 break;
1387 }
1388 }
1389 return bond_do_alb_xmit(skb, bond, tx_slave);
1390 }
1391
1392 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1393 {
1394 struct bonding *bond = netdev_priv(bond_dev);
1395 struct ethhdr *eth_data;
1396 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1397 struct slave *tx_slave = NULL;
1398 static const __be32 ip_bcast = htonl(0xffffffff);
1399 int hash_size = 0;
1400 bool do_tx_balance = true;
1401 u32 hash_index = 0;
1402 const u8 *hash_start = NULL;
1403 struct ipv6hdr *ip6hdr;
1404
1405 skb_reset_mac_header(skb);
1406 eth_data = eth_hdr(skb);
1407
1408 switch (ntohs(skb->protocol)) {
1409 case ETH_P_IP: {
1410 const struct iphdr *iph = ip_hdr(skb);
1411
1412 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast) ||
1413 (iph->daddr == ip_bcast) ||
1414 (iph->protocol == IPPROTO_IGMP)) {
1415 do_tx_balance = false;
1416 break;
1417 }
1418 hash_start = (char *)&(iph->daddr);
1419 hash_size = sizeof(iph->daddr);
1420 }
1421 break;
1422 case ETH_P_IPV6:
1423 /* IPv6 doesn't really use broadcast mac address, but leave
1424 * that here just in case.
1425 */
1426 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast)) {
1427 do_tx_balance = false;
1428 break;
1429 }
1430
1431 /* IPv6 uses all-nodes multicast as an equivalent to
1432 * broadcasts in IPv4.
1433 */
1434 if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1435 do_tx_balance = false;
1436 break;
1437 }
1438
1439 /* Additianally, DAD probes should not be tx-balanced as that
1440 * will lead to false positives for duplicate addresses and
1441 * prevent address configuration from working.
1442 */
1443 ip6hdr = ipv6_hdr(skb);
1444 if (ipv6_addr_any(&ip6hdr->saddr)) {
1445 do_tx_balance = false;
1446 break;
1447 }
1448
1449 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1450 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1451 break;
1452 case ETH_P_IPX:
1453 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1454 /* something is wrong with this packet */
1455 do_tx_balance = false;
1456 break;
1457 }
1458
1459 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1460 /* The only protocol worth balancing in
1461 * this family since it has an "ARP" like
1462 * mechanism
1463 */
1464 do_tx_balance = false;
1465 break;
1466 }
1467
1468 hash_start = (char *)eth_data->h_dest;
1469 hash_size = ETH_ALEN;
1470 break;
1471 case ETH_P_ARP:
1472 do_tx_balance = false;
1473 if (bond_info->rlb_enabled)
1474 tx_slave = rlb_arp_xmit(skb, bond);
1475 break;
1476 default:
1477 do_tx_balance = false;
1478 break;
1479 }
1480
1481 if (do_tx_balance) {
1482 hash_index = _simple_hash(hash_start, hash_size);
1483 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1484 }
1485
1486 return bond_do_alb_xmit(skb, bond, tx_slave);
1487 }
1488
1489 void bond_alb_monitor(struct work_struct *work)
1490 {
1491 struct bonding *bond = container_of(work, struct bonding,
1492 alb_work.work);
1493 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1494 struct list_head *iter;
1495 struct slave *slave;
1496
1497 if (!bond_has_slaves(bond)) {
1498 bond_info->tx_rebalance_counter = 0;
1499 bond_info->lp_counter = 0;
1500 goto re_arm;
1501 }
1502
1503 rcu_read_lock();
1504
1505 bond_info->tx_rebalance_counter++;
1506 bond_info->lp_counter++;
1507
1508 /* send learning packets */
1509 if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1510 bool strict_match;
1511
1512 bond_for_each_slave_rcu(bond, slave, iter) {
1513 /* If updating current_active, use all currently
1514 * user mac addreses (!strict_match). Otherwise, only
1515 * use mac of the slave device.
1516 * In RLB mode, we always use strict matches.
1517 */
1518 strict_match = (slave != rcu_access_pointer(bond->curr_active_slave) ||
1519 bond_info->rlb_enabled);
1520 alb_send_learning_packets(slave, slave->dev->dev_addr,
1521 strict_match);
1522 }
1523 bond_info->lp_counter = 0;
1524 }
1525
1526 /* rebalance tx traffic */
1527 if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1528 bond_for_each_slave_rcu(bond, slave, iter) {
1529 tlb_clear_slave(bond, slave, 1);
1530 if (slave == rcu_access_pointer(bond->curr_active_slave)) {
1531 SLAVE_TLB_INFO(slave).load =
1532 bond_info->unbalanced_load /
1533 BOND_TLB_REBALANCE_INTERVAL;
1534 bond_info->unbalanced_load = 0;
1535 }
1536 }
1537 bond_info->tx_rebalance_counter = 0;
1538 }
1539
1540 if (bond_info->rlb_enabled) {
1541 if (bond_info->primary_is_promisc &&
1542 (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1543
1544 /* dev_set_promiscuity requires rtnl and
1545 * nothing else. Avoid race with bond_close.
1546 */
1547 rcu_read_unlock();
1548 if (!rtnl_trylock())
1549 goto re_arm;
1550
1551 bond_info->rlb_promisc_timeout_counter = 0;
1552
1553 /* If the primary was set to promiscuous mode
1554 * because a slave was disabled then
1555 * it can now leave promiscuous mode.
1556 */
1557 dev_set_promiscuity(rtnl_dereference(bond->curr_active_slave)->dev,
1558 -1);
1559 bond_info->primary_is_promisc = 0;
1560
1561 rtnl_unlock();
1562 rcu_read_lock();
1563 }
1564
1565 if (bond_info->rlb_rebalance) {
1566 bond_info->rlb_rebalance = 0;
1567 rlb_rebalance(bond);
1568 }
1569
1570 /* check if clients need updating */
1571 if (bond_info->rx_ntt) {
1572 if (bond_info->rlb_update_delay_counter) {
1573 --bond_info->rlb_update_delay_counter;
1574 } else {
1575 rlb_update_rx_clients(bond);
1576 if (bond_info->rlb_update_retry_counter)
1577 --bond_info->rlb_update_retry_counter;
1578 else
1579 bond_info->rx_ntt = 0;
1580 }
1581 }
1582 }
1583 rcu_read_unlock();
1584 re_arm:
1585 queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1586 }
1587
1588 /* assumption: called before the slave is attached to the bond
1589 * and not locked by the bond lock
1590 */
1591 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1592 {
1593 int res;
1594
1595 res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1596 slave->dev->addr_len);
1597 if (res)
1598 return res;
1599
1600 res = alb_handle_addr_collision_on_attach(bond, slave);
1601 if (res)
1602 return res;
1603
1604 tlb_init_slave(slave);
1605
1606 /* order a rebalance ASAP */
1607 bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1608
1609 if (bond->alb_info.rlb_enabled)
1610 bond->alb_info.rlb_rebalance = 1;
1611
1612 return 0;
1613 }
1614
1615 /* Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1616 * if necessary.
1617 *
1618 * Caller must hold RTNL and no other locks
1619 */
1620 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1621 {
1622 if (bond_has_slaves(bond))
1623 alb_change_hw_addr_on_detach(bond, slave);
1624
1625 tlb_clear_slave(bond, slave, 0);
1626
1627 if (bond->alb_info.rlb_enabled) {
1628 bond->alb_info.rx_slave = NULL;
1629 rlb_clear_slave(bond, slave);
1630 }
1631
1632 }
1633
1634 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1635 {
1636 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1637
1638 if (link == BOND_LINK_DOWN) {
1639 tlb_clear_slave(bond, slave, 0);
1640 if (bond->alb_info.rlb_enabled)
1641 rlb_clear_slave(bond, slave);
1642 } else if (link == BOND_LINK_UP) {
1643 /* order a rebalance ASAP */
1644 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1645 if (bond->alb_info.rlb_enabled) {
1646 bond->alb_info.rlb_rebalance = 1;
1647 /* If the updelay module parameter is smaller than the
1648 * forwarding delay of the switch the rebalance will
1649 * not work because the rebalance arp replies will
1650 * not be forwarded to the clients..
1651 */
1652 }
1653 }
1654
1655 if (bond_is_nondyn_tlb(bond)) {
1656 if (bond_update_slave_arr(bond, NULL))
1657 pr_err("Failed to build slave-array for TLB mode.\n");
1658 }
1659 }
1660
1661 /**
1662 * bond_alb_handle_active_change - assign new curr_active_slave
1663 * @bond: our bonding struct
1664 * @new_slave: new slave to assign
1665 *
1666 * Set the bond->curr_active_slave to @new_slave and handle
1667 * mac address swapping and promiscuity changes as needed.
1668 *
1669 * Caller must hold RTNL
1670 */
1671 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1672 {
1673 struct slave *swap_slave;
1674 struct slave *curr_active;
1675
1676 curr_active = rtnl_dereference(bond->curr_active_slave);
1677 if (curr_active == new_slave)
1678 return;
1679
1680 if (curr_active && bond->alb_info.primary_is_promisc) {
1681 dev_set_promiscuity(curr_active->dev, -1);
1682 bond->alb_info.primary_is_promisc = 0;
1683 bond->alb_info.rlb_promisc_timeout_counter = 0;
1684 }
1685
1686 swap_slave = curr_active;
1687 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1688
1689 if (!new_slave || !bond_has_slaves(bond))
1690 return;
1691
1692 /* set the new curr_active_slave to the bonds mac address
1693 * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1694 */
1695 if (!swap_slave)
1696 swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1697
1698 /* Arrange for swap_slave and new_slave to temporarily be
1699 * ignored so we can mess with their MAC addresses without
1700 * fear of interference from transmit activity.
1701 */
1702 if (swap_slave)
1703 tlb_clear_slave(bond, swap_slave, 1);
1704 tlb_clear_slave(bond, new_slave, 1);
1705
1706 /* in TLB mode, the slave might flip down/up with the old dev_addr,
1707 * and thus filter bond->dev_addr's packets, so force bond's mac
1708 */
1709 if (BOND_MODE(bond) == BOND_MODE_TLB) {
1710 struct sockaddr_storage ss;
1711 u8 tmp_addr[MAX_ADDR_LEN];
1712
1713 bond_hw_addr_copy(tmp_addr, new_slave->dev->dev_addr,
1714 new_slave->dev->addr_len);
1715
1716 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
1717 bond->dev->addr_len);
1718 ss.ss_family = bond->dev->type;
1719 /* we don't care if it can't change its mac, best effort */
1720 dev_set_mac_address(new_slave->dev, (struct sockaddr *)&ss);
1721
1722 bond_hw_addr_copy(new_slave->dev->dev_addr, tmp_addr,
1723 new_slave->dev->addr_len);
1724 }
1725
1726 /* curr_active_slave must be set before calling alb_swap_mac_addr */
1727 if (swap_slave) {
1728 /* swap mac address */
1729 alb_swap_mac_addr(swap_slave, new_slave);
1730 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1731 } else {
1732 /* set the new_slave to the bond mac address */
1733 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1734 bond->dev->addr_len);
1735 alb_send_learning_packets(new_slave, bond->dev->dev_addr,
1736 false);
1737 }
1738 }
1739
1740 /* Called with RTNL */
1741 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1742 {
1743 struct bonding *bond = netdev_priv(bond_dev);
1744 struct sockaddr_storage *ss = addr;
1745 struct slave *curr_active;
1746 struct slave *swap_slave;
1747 int res;
1748
1749 if (!is_valid_ether_addr(ss->__data))
1750 return -EADDRNOTAVAIL;
1751
1752 res = alb_set_mac_address(bond, addr);
1753 if (res)
1754 return res;
1755
1756 bond_hw_addr_copy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
1757
1758 /* If there is no curr_active_slave there is nothing else to do.
1759 * Otherwise we'll need to pass the new address to it and handle
1760 * duplications.
1761 */
1762 curr_active = rtnl_dereference(bond->curr_active_slave);
1763 if (!curr_active)
1764 return 0;
1765
1766 swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1767
1768 if (swap_slave) {
1769 alb_swap_mac_addr(swap_slave, curr_active);
1770 alb_fasten_mac_swap(bond, swap_slave, curr_active);
1771 } else {
1772 alb_set_slave_mac_addr(curr_active, bond_dev->dev_addr,
1773 bond_dev->addr_len);
1774
1775 alb_send_learning_packets(curr_active,
1776 bond_dev->dev_addr, false);
1777 if (bond->alb_info.rlb_enabled) {
1778 /* inform clients mac address has changed */
1779 rlb_req_update_slave_clients(bond, curr_active);
1780 }
1781 }
1782
1783 return 0;
1784 }
1785
1786 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1787 {
1788 if (bond->alb_info.rlb_enabled)
1789 rlb_clear_vlan(bond, vlan_id);
1790 }
1791
CRIS linux kernel tree