projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf
[deliverable/linux.git] / drivers / net / bonding / bond_alb.c
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 #ifndef __long_aligned
46 #define __long_aligned __attribute__((aligned((sizeof(long)))))
47 #endif
48 static const u8 mac_bcast[ETH_ALEN] __long_aligned = {
49 0xff, 0xff, 0xff, 0xff, 0xff, 0xff
50 };
51 static const u8 mac_v6_allmcast[ETH_ALEN] __long_aligned = {
52 0x33, 0x33, 0x00, 0x00, 0x00, 0x01
53 };
54 static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
55
56 #pragma pack(1)
57 struct learning_pkt {
58 u8 mac_dst[ETH_ALEN];
59 u8 mac_src[ETH_ALEN];
60 __be16 type;
61 u8 padding[ETH_ZLEN - ETH_HLEN];
62 };
63
64 struct arp_pkt {
65 __be16 hw_addr_space;
66 __be16 prot_addr_space;
67 u8 hw_addr_len;
68 u8 prot_addr_len;
69 __be16 op_code;
70 u8 mac_src[ETH_ALEN]; /* sender hardware address */
71 __be32 ip_src; /* sender IP address */
72 u8 mac_dst[ETH_ALEN]; /* target hardware address */
73 __be32 ip_dst; /* target IP address */
74 };
75 #pragma pack()
76
77 static inline struct arp_pkt *arp_pkt(const struct sk_buff *skb)
78 {
79 return (struct arp_pkt *)skb_network_header(skb);
80 }
81
82 /* Forward declaration */
83 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
84 bool strict_match);
85 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp);
86 static void rlb_src_unlink(struct bonding *bond, u32 index);
87 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash,
88 u32 ip_dst_hash);
89
90 static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
91 {
92 int i;
93 u8 hash = 0;
94
95 for (i = 0; i < hash_size; i++)
96 hash ^= hash_start[i];
97
98 return hash;
99 }
100
101 /*********************** tlb specific functions ***************************/
102
103 static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
104 {
105 if (save_load) {
106 entry->load_history = 1 + entry->tx_bytes /
107 BOND_TLB_REBALANCE_INTERVAL;
108 entry->tx_bytes = 0;
109 }
110
111 entry->tx_slave = NULL;
112 entry->next = TLB_NULL_INDEX;
113 entry->prev = TLB_NULL_INDEX;
114 }
115
116 static inline void tlb_init_slave(struct slave *slave)
117 {
118 SLAVE_TLB_INFO(slave).load = 0;
119 SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
120 }
121
122 static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
123 int save_load)
124 {
125 struct tlb_client_info *tx_hash_table;
126 u32 index;
127
128 /* clear slave from tx_hashtbl */
129 tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
130
131 /* skip this if we've already freed the tx hash table */
132 if (tx_hash_table) {
133 index = SLAVE_TLB_INFO(slave).head;
134 while (index != TLB_NULL_INDEX) {
135 u32 next_index = tx_hash_table[index].next;
136 tlb_init_table_entry(&tx_hash_table[index], save_load);
137 index = next_index;
138 }
139 }
140
141 tlb_init_slave(slave);
142 }
143
144 static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
145 int save_load)
146 {
147 spin_lock_bh(&bond->mode_lock);
148 __tlb_clear_slave(bond, slave, save_load);
149 spin_unlock_bh(&bond->mode_lock);
150 }
151
152 /* Must be called before starting the monitor timer */
153 static int tlb_initialize(struct bonding *bond)
154 {
155 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
156 int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
157 struct tlb_client_info *new_hashtbl;
158 int i;
159
160 new_hashtbl = kzalloc(size, GFP_KERNEL);
161 if (!new_hashtbl)
162 return -1;
163
164 spin_lock_bh(&bond->mode_lock);
165
166 bond_info->tx_hashtbl = new_hashtbl;
167
168 for (i = 0; i < TLB_HASH_TABLE_SIZE; i++)
169 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 0);
170
171 spin_unlock_bh(&bond->mode_lock);
172
173 return 0;
174 }
175
176 /* Must be called only after all slaves have been released */
177 static void tlb_deinitialize(struct bonding *bond)
178 {
179 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
180
181 spin_lock_bh(&bond->mode_lock);
182
183 kfree(bond_info->tx_hashtbl);
184 bond_info->tx_hashtbl = NULL;
185
186 spin_unlock_bh(&bond->mode_lock);
187 }
188
189 static long long compute_gap(struct slave *slave)
190 {
191 return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
192 (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
193 }
194
195 static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
196 {
197 struct slave *slave, *least_loaded;
198 struct list_head *iter;
199 long long max_gap;
200
201 least_loaded = NULL;
202 max_gap = LLONG_MIN;
203
204 /* Find the slave with the largest gap */
205 bond_for_each_slave_rcu(bond, slave, iter) {
206 if (bond_slave_can_tx(slave)) {
207 long long gap = compute_gap(slave);
208
209 if (max_gap < gap) {
210 least_loaded = slave;
211 max_gap = gap;
212 }
213 }
214 }
215
216 return least_loaded;
217 }
218
219 static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
220 u32 skb_len)
221 {
222 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
223 struct tlb_client_info *hash_table;
224 struct slave *assigned_slave;
225
226 hash_table = bond_info->tx_hashtbl;
227 assigned_slave = hash_table[hash_index].tx_slave;
228 if (!assigned_slave) {
229 assigned_slave = tlb_get_least_loaded_slave(bond);
230
231 if (assigned_slave) {
232 struct tlb_slave_info *slave_info =
233 &(SLAVE_TLB_INFO(assigned_slave));
234 u32 next_index = slave_info->head;
235
236 hash_table[hash_index].tx_slave = assigned_slave;
237 hash_table[hash_index].next = next_index;
238 hash_table[hash_index].prev = TLB_NULL_INDEX;
239
240 if (next_index != TLB_NULL_INDEX)
241 hash_table[next_index].prev = hash_index;
242
243 slave_info->head = hash_index;
244 slave_info->load +=
245 hash_table[hash_index].load_history;
246 }
247 }
248
249 if (assigned_slave)
250 hash_table[hash_index].tx_bytes += skb_len;
251
252 return assigned_slave;
253 }
254
255 static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
256 u32 skb_len)
257 {
258 struct slave *tx_slave;
259
260 /* We don't need to disable softirq here, becase
261 * tlb_choose_channel() is only called by bond_alb_xmit()
262 * which already has softirq disabled.
263 */
264 spin_lock(&bond->mode_lock);
265 tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
266 spin_unlock(&bond->mode_lock);
267
268 return tx_slave;
269 }
270
271 /*********************** rlb specific functions ***************************/
272
273 /* when an ARP REPLY is received from a client update its info
274 * in the rx_hashtbl
275 */
276 static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
277 {
278 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
279 struct rlb_client_info *client_info;
280 u32 hash_index;
281
282 spin_lock_bh(&bond->mode_lock);
283
284 hash_index = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
285 client_info = &(bond_info->rx_hashtbl[hash_index]);
286
287 if ((client_info->assigned) &&
288 (client_info->ip_src == arp->ip_dst) &&
289 (client_info->ip_dst == arp->ip_src) &&
290 (!ether_addr_equal_64bits(client_info->mac_dst, arp->mac_src))) {
291 /* update the clients MAC address */
292 ether_addr_copy(client_info->mac_dst, arp->mac_src);
293 client_info->ntt = 1;
294 bond_info->rx_ntt = 1;
295 }
296
297 spin_unlock_bh(&bond->mode_lock);
298 }
299
300 static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
301 struct slave *slave)
302 {
303 struct arp_pkt *arp, _arp;
304
305 if (skb->protocol != cpu_to_be16(ETH_P_ARP))
306 goto out;
307
308 arp = skb_header_pointer(skb, 0, sizeof(_arp), &_arp);
309 if (!arp)
310 goto out;
311
312 /* We received an ARP from arp->ip_src.
313 * We might have used this IP address previously (on the bonding host
314 * itself or on a system that is bridged together with the bond).
315 * However, if arp->mac_src is different than what is stored in
316 * rx_hashtbl, some other host is now using the IP and we must prevent
317 * sending out client updates with this IP address and the old MAC
318 * address.
319 * Clean up all hash table entries that have this address as ip_src but
320 * have a different mac_src.
321 */
322 rlb_purge_src_ip(bond, arp);
323
324 if (arp->op_code == htons(ARPOP_REPLY)) {
325 /* update rx hash table for this ARP */
326 rlb_update_entry_from_arp(bond, arp);
327 netdev_dbg(bond->dev, "Server received an ARP Reply from client\n");
328 }
329 out:
330 return RX_HANDLER_ANOTHER;
331 }
332
333 /* Caller must hold rcu_read_lock() */
334 static struct slave *__rlb_next_rx_slave(struct bonding *bond)
335 {
336 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
337 struct slave *before = NULL, *rx_slave = NULL, *slave;
338 struct list_head *iter;
339 bool found = false;
340
341 bond_for_each_slave_rcu(bond, slave, iter) {
342 if (!bond_slave_can_tx(slave))
343 continue;
344 if (!found) {
345 if (!before || before->speed < slave->speed)
346 before = slave;
347 } else {
348 if (!rx_slave || rx_slave->speed < slave->speed)
349 rx_slave = slave;
350 }
351 if (slave == bond_info->rx_slave)
352 found = true;
353 }
354 /* we didn't find anything after the current or we have something
355 * better before and up to the current slave
356 */
357 if (!rx_slave || (before && rx_slave->speed < before->speed))
358 rx_slave = before;
359
360 if (rx_slave)
361 bond_info->rx_slave = rx_slave;
362
363 return rx_slave;
364 }
365
366 /* Caller must hold RTNL, rcu_read_lock is obtained only to silence checkers */
367 static struct slave *rlb_next_rx_slave(struct bonding *bond)
368 {
369 struct slave *rx_slave;
370
371 ASSERT_RTNL();
372
373 rcu_read_lock();
374 rx_slave = __rlb_next_rx_slave(bond);
375 rcu_read_unlock();
376
377 return rx_slave;
378 }
379
380 /* teach the switch the mac of a disabled slave
381 * on the primary for fault tolerance
382 *
383 * Caller must hold RTNL
384 */
385 static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
386 {
387 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
388
389 if (!curr_active)
390 return;
391
392 if (!bond->alb_info.primary_is_promisc) {
393 if (!dev_set_promiscuity(curr_active->dev, 1))
394 bond->alb_info.primary_is_promisc = 1;
395 else
396 bond->alb_info.primary_is_promisc = 0;
397 }
398
399 bond->alb_info.rlb_promisc_timeout_counter = 0;
400
401 alb_send_learning_packets(curr_active, addr, true);
402 }
403
404 /* slave being removed should not be active at this point
405 *
406 * Caller must hold rtnl.
407 */
408 static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
409 {
410 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
411 struct rlb_client_info *rx_hash_table;
412 u32 index, next_index;
413
414 /* clear slave from rx_hashtbl */
415 spin_lock_bh(&bond->mode_lock);
416
417 rx_hash_table = bond_info->rx_hashtbl;
418 index = bond_info->rx_hashtbl_used_head;
419 for (; index != RLB_NULL_INDEX; index = next_index) {
420 next_index = rx_hash_table[index].used_next;
421 if (rx_hash_table[index].slave == slave) {
422 struct slave *assigned_slave = rlb_next_rx_slave(bond);
423
424 if (assigned_slave) {
425 rx_hash_table[index].slave = assigned_slave;
426 if (!ether_addr_equal_64bits(rx_hash_table[index].mac_dst,
427 mac_bcast)) {
428 bond_info->rx_hashtbl[index].ntt = 1;
429 bond_info->rx_ntt = 1;
430 /* A slave has been removed from the
431 * table because it is either disabled
432 * or being released. We must retry the
433 * update to avoid clients from not
434 * being updated & disconnecting when
435 * there is stress
436 */
437 bond_info->rlb_update_retry_counter =
438 RLB_UPDATE_RETRY;
439 }
440 } else { /* there is no active slave */
441 rx_hash_table[index].slave = NULL;
442 }
443 }
444 }
445
446 spin_unlock_bh(&bond->mode_lock);
447
448 if (slave != rtnl_dereference(bond->curr_active_slave))
449 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
450 }
451
452 static void rlb_update_client(struct rlb_client_info *client_info)
453 {
454 int i;
455
456 if (!client_info->slave)
457 return;
458
459 for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
460 struct sk_buff *skb;
461
462 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
463 client_info->ip_dst,
464 client_info->slave->dev,
465 client_info->ip_src,
466 client_info->mac_dst,
467 client_info->slave->dev->dev_addr,
468 client_info->mac_dst);
469 if (!skb) {
470 netdev_err(client_info->slave->bond->dev,
471 "failed to create an ARP packet\n");
472 continue;
473 }
474
475 skb->dev = client_info->slave->dev;
476
477 if (client_info->vlan_id) {
478 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
479 client_info->vlan_id);
480 }
481
482 arp_xmit(skb);
483 }
484 }
485
486 /* sends ARP REPLIES that update the clients that need updating */
487 static void rlb_update_rx_clients(struct bonding *bond)
488 {
489 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
490 struct rlb_client_info *client_info;
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 if (client_info->ntt) {
500 rlb_update_client(client_info);
501 if (bond_info->rlb_update_retry_counter == 0)
502 client_info->ntt = 0;
503 }
504 }
505
506 /* do not update the entries again until this counter is zero so that
507 * not to confuse the clients.
508 */
509 bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
510
511 spin_unlock_bh(&bond->mode_lock);
512 }
513
514 /* The slave was assigned a new mac address - update the clients */
515 static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
516 {
517 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
518 struct rlb_client_info *client_info;
519 int ntt = 0;
520 u32 hash_index;
521
522 spin_lock_bh(&bond->mode_lock);
523
524 hash_index = bond_info->rx_hashtbl_used_head;
525 for (; hash_index != RLB_NULL_INDEX;
526 hash_index = client_info->used_next) {
527 client_info = &(bond_info->rx_hashtbl[hash_index]);
528
529 if ((client_info->slave == slave) &&
530 !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
531 client_info->ntt = 1;
532 ntt = 1;
533 }
534 }
535
536 /* update the team's flag only after the whole iteration */
537 if (ntt) {
538 bond_info->rx_ntt = 1;
539 /* fasten the change */
540 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
541 }
542
543 spin_unlock_bh(&bond->mode_lock);
544 }
545
546 /* mark all clients using src_ip to be updated */
547 static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
548 {
549 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
550 struct rlb_client_info *client_info;
551 u32 hash_index;
552
553 spin_lock(&bond->mode_lock);
554
555 hash_index = bond_info->rx_hashtbl_used_head;
556 for (; hash_index != RLB_NULL_INDEX;
557 hash_index = client_info->used_next) {
558 client_info = &(bond_info->rx_hashtbl[hash_index]);
559
560 if (!client_info->slave) {
561 netdev_err(bond->dev, "found a client with no channel in the client's hash table\n");
562 continue;
563 }
564 /* update all clients using this src_ip, that are not assigned
565 * to the team's address (curr_active_slave) and have a known
566 * unicast mac address.
567 */
568 if ((client_info->ip_src == src_ip) &&
569 !ether_addr_equal_64bits(client_info->slave->dev->dev_addr,
570 bond->dev->dev_addr) &&
571 !ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
572 client_info->ntt = 1;
573 bond_info->rx_ntt = 1;
574 }
575 }
576
577 spin_unlock(&bond->mode_lock);
578 }
579
580 static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
581 {
582 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
583 struct arp_pkt *arp = arp_pkt(skb);
584 struct slave *assigned_slave, *curr_active_slave;
585 struct rlb_client_info *client_info;
586 u32 hash_index = 0;
587
588 spin_lock(&bond->mode_lock);
589
590 curr_active_slave = rcu_dereference(bond->curr_active_slave);
591
592 hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_dst));
593 client_info = &(bond_info->rx_hashtbl[hash_index]);
594
595 if (client_info->assigned) {
596 if ((client_info->ip_src == arp->ip_src) &&
597 (client_info->ip_dst == arp->ip_dst)) {
598 /* the entry is already assigned to this client */
599 if (!ether_addr_equal_64bits(arp->mac_dst, mac_bcast)) {
600 /* update mac address from arp */
601 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
602 }
603 ether_addr_copy(client_info->mac_src, arp->mac_src);
604
605 assigned_slave = client_info->slave;
606 if (assigned_slave) {
607 spin_unlock(&bond->mode_lock);
608 return assigned_slave;
609 }
610 } else {
611 /* the entry is already assigned to some other client,
612 * move the old client to primary (curr_active_slave) so
613 * that the new client can be assigned to this entry.
614 */
615 if (curr_active_slave &&
616 client_info->slave != curr_active_slave) {
617 client_info->slave = curr_active_slave;
618 rlb_update_client(client_info);
619 }
620 }
621 }
622 /* assign a new slave */
623 assigned_slave = __rlb_next_rx_slave(bond);
624
625 if (assigned_slave) {
626 if (!(client_info->assigned &&
627 client_info->ip_src == arp->ip_src)) {
628 /* ip_src is going to be updated,
629 * fix the src hash list
630 */
631 u32 hash_src = _simple_hash((u8 *)&arp->ip_src,
632 sizeof(arp->ip_src));
633 rlb_src_unlink(bond, hash_index);
634 rlb_src_link(bond, hash_src, hash_index);
635 }
636
637 client_info->ip_src = arp->ip_src;
638 client_info->ip_dst = arp->ip_dst;
639 /* arp->mac_dst is broadcast for arp reqeusts.
640 * will be updated with clients actual unicast mac address
641 * upon receiving an arp reply.
642 */
643 ether_addr_copy(client_info->mac_dst, arp->mac_dst);
644 ether_addr_copy(client_info->mac_src, arp->mac_src);
645 client_info->slave = assigned_slave;
646
647 if (!ether_addr_equal_64bits(client_info->mac_dst, mac_bcast)) {
648 client_info->ntt = 1;
649 bond->alb_info.rx_ntt = 1;
650 } else {
651 client_info->ntt = 0;
652 }
653
654 if (vlan_get_tag(skb, &client_info->vlan_id))
655 client_info->vlan_id = 0;
656
657 if (!client_info->assigned) {
658 u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
659 bond_info->rx_hashtbl_used_head = hash_index;
660 client_info->used_next = prev_tbl_head;
661 if (prev_tbl_head != RLB_NULL_INDEX) {
662 bond_info->rx_hashtbl[prev_tbl_head].used_prev =
663 hash_index;
664 }
665 client_info->assigned = 1;
666 }
667 }
668
669 spin_unlock(&bond->mode_lock);
670
671 return assigned_slave;
672 }
673
674 /* chooses (and returns) transmit channel for arp reply
675 * does not choose channel for other arp types since they are
676 * sent on the curr_active_slave
677 */
678 static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
679 {
680 struct arp_pkt *arp = arp_pkt(skb);
681 struct slave *tx_slave = NULL;
682
683 /* Don't modify or load balance ARPs that do not originate locally
684 * (e.g.,arrive via a bridge).
685 */
686 if (!bond_slave_has_mac_rx(bond, arp->mac_src))
687 return NULL;
688
689 if (arp->op_code == htons(ARPOP_REPLY)) {
690 /* the arp must be sent on the selected rx channel */
691 tx_slave = rlb_choose_channel(skb, bond);
692 if (tx_slave)
693 ether_addr_copy(arp->mac_src, tx_slave->dev->dev_addr);
694 netdev_dbg(bond->dev, "Server sent ARP Reply packet\n");
695 } else if (arp->op_code == htons(ARPOP_REQUEST)) {
696 /* Create an entry in the rx_hashtbl for this client as a
697 * place holder.
698 * When the arp reply is received the entry will be updated
699 * with the correct unicast address of the client.
700 */
701 rlb_choose_channel(skb, bond);
702
703 /* The ARP reply packets must be delayed so that
704 * they can cancel out the influence of the ARP request.
705 */
706 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
707
708 /* arp requests are broadcast and are sent on the primary
709 * the arp request will collapse all clients on the subnet to
710 * the primary slave. We must register these clients to be
711 * updated with their assigned mac.
712 */
713 rlb_req_update_subnet_clients(bond, arp->ip_src);
714 netdev_dbg(bond->dev, "Server sent ARP Request packet\n");
715 }
716
717 return tx_slave;
718 }
719
720 static void rlb_rebalance(struct bonding *bond)
721 {
722 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
723 struct slave *assigned_slave;
724 struct rlb_client_info *client_info;
725 int ntt;
726 u32 hash_index;
727
728 spin_lock_bh(&bond->mode_lock);
729
730 ntt = 0;
731 hash_index = bond_info->rx_hashtbl_used_head;
732 for (; hash_index != RLB_NULL_INDEX;
733 hash_index = client_info->used_next) {
734 client_info = &(bond_info->rx_hashtbl[hash_index]);
735 assigned_slave = __rlb_next_rx_slave(bond);
736 if (assigned_slave && (client_info->slave != assigned_slave)) {
737 client_info->slave = assigned_slave;
738 client_info->ntt = 1;
739 ntt = 1;
740 }
741 }
742
743 /* update the team's flag only after the whole iteration */
744 if (ntt)
745 bond_info->rx_ntt = 1;
746 spin_unlock_bh(&bond->mode_lock);
747 }
748
749 /* Caller must hold mode_lock */
750 static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
751 {
752 entry->used_next = RLB_NULL_INDEX;
753 entry->used_prev = RLB_NULL_INDEX;
754 entry->assigned = 0;
755 entry->slave = NULL;
756 entry->vlan_id = 0;
757 }
758 static void rlb_init_table_entry_src(struct rlb_client_info *entry)
759 {
760 entry->src_first = RLB_NULL_INDEX;
761 entry->src_prev = RLB_NULL_INDEX;
762 entry->src_next = RLB_NULL_INDEX;
763 }
764
765 static void rlb_init_table_entry(struct rlb_client_info *entry)
766 {
767 memset(entry, 0, sizeof(struct rlb_client_info));
768 rlb_init_table_entry_dst(entry);
769 rlb_init_table_entry_src(entry);
770 }
771
772 static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
773 {
774 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
775 u32 next_index = bond_info->rx_hashtbl[index].used_next;
776 u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
777
778 if (index == bond_info->rx_hashtbl_used_head)
779 bond_info->rx_hashtbl_used_head = next_index;
780 if (prev_index != RLB_NULL_INDEX)
781 bond_info->rx_hashtbl[prev_index].used_next = next_index;
782 if (next_index != RLB_NULL_INDEX)
783 bond_info->rx_hashtbl[next_index].used_prev = prev_index;
784 }
785
786 /* unlink a rlb hash table entry from the src list */
787 static void rlb_src_unlink(struct bonding *bond, u32 index)
788 {
789 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
790 u32 next_index = bond_info->rx_hashtbl[index].src_next;
791 u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
792
793 bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
794 bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
795
796 if (next_index != RLB_NULL_INDEX)
797 bond_info->rx_hashtbl[next_index].src_prev = prev_index;
798
799 if (prev_index == RLB_NULL_INDEX)
800 return;
801
802 /* is prev_index pointing to the head of this list? */
803 if (bond_info->rx_hashtbl[prev_index].src_first == index)
804 bond_info->rx_hashtbl[prev_index].src_first = next_index;
805 else
806 bond_info->rx_hashtbl[prev_index].src_next = next_index;
807
808 }
809
810 static void rlb_delete_table_entry(struct bonding *bond, u32 index)
811 {
812 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
813 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
814
815 rlb_delete_table_entry_dst(bond, index);
816 rlb_init_table_entry_dst(entry);
817
818 rlb_src_unlink(bond, index);
819 }
820
821 /* add the rx_hashtbl[ip_dst_hash] entry to the list
822 * of entries with identical ip_src_hash
823 */
824 static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
825 {
826 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
827 u32 next;
828
829 bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
830 next = bond_info->rx_hashtbl[ip_src_hash].src_first;
831 bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
832 if (next != RLB_NULL_INDEX)
833 bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
834 bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
835 }
836
837 /* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does
838 * not match arp->mac_src
839 */
840 static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
841 {
842 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
843 u32 ip_src_hash = _simple_hash((u8 *)&(arp->ip_src), sizeof(arp->ip_src));
844 u32 index;
845
846 spin_lock_bh(&bond->mode_lock);
847
848 index = bond_info->rx_hashtbl[ip_src_hash].src_first;
849 while (index != RLB_NULL_INDEX) {
850 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
851 u32 next_index = entry->src_next;
852 if (entry->ip_src == arp->ip_src &&
853 !ether_addr_equal_64bits(arp->mac_src, entry->mac_src))
854 rlb_delete_table_entry(bond, index);
855 index = next_index;
856 }
857 spin_unlock_bh(&bond->mode_lock);
858 }
859
860 static int rlb_initialize(struct bonding *bond)
861 {
862 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
863 struct rlb_client_info *new_hashtbl;
864 int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
865 int i;
866
867 new_hashtbl = kmalloc(size, GFP_KERNEL);
868 if (!new_hashtbl)
869 return -1;
870
871 spin_lock_bh(&bond->mode_lock);
872
873 bond_info->rx_hashtbl = new_hashtbl;
874
875 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
876
877 for (i = 0; i < RLB_HASH_TABLE_SIZE; i++)
878 rlb_init_table_entry(bond_info->rx_hashtbl + i);
879
880 spin_unlock_bh(&bond->mode_lock);
881
882 /* register to receive ARPs */
883 bond->recv_probe = rlb_arp_recv;
884
885 return 0;
886 }
887
888 static void rlb_deinitialize(struct bonding *bond)
889 {
890 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
891
892 spin_lock_bh(&bond->mode_lock);
893
894 kfree(bond_info->rx_hashtbl);
895 bond_info->rx_hashtbl = NULL;
896 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
897
898 spin_unlock_bh(&bond->mode_lock);
899 }
900
901 static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
902 {
903 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
904 u32 curr_index;
905
906 spin_lock_bh(&bond->mode_lock);
907
908 curr_index = bond_info->rx_hashtbl_used_head;
909 while (curr_index != RLB_NULL_INDEX) {
910 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
911 u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
912
913 if (curr->vlan_id == vlan_id)
914 rlb_delete_table_entry(bond, curr_index);
915
916 curr_index = next_index;
917 }
918
919 spin_unlock_bh(&bond->mode_lock);
920 }
921
922 /*********************** tlb/rlb shared functions *********************/
923
924 static void alb_send_lp_vid(struct slave *slave, u8 mac_addr[],
925 __be16 vlan_proto, u16 vid)
926 {
927 struct learning_pkt pkt;
928 struct sk_buff *skb;
929 int size = sizeof(struct learning_pkt);
930 char *data;
931
932 memset(&pkt, 0, size);
933 ether_addr_copy(pkt.mac_dst, mac_addr);
934 ether_addr_copy(pkt.mac_src, mac_addr);
935 pkt.type = cpu_to_be16(ETH_P_LOOPBACK);
936
937 skb = dev_alloc_skb(size);
938 if (!skb)
939 return;
940
941 data = skb_put(skb, size);
942 memcpy(data, &pkt, size);
943
944 skb_reset_mac_header(skb);
945 skb->network_header = skb->mac_header + ETH_HLEN;
946 skb->protocol = pkt.type;
947 skb->priority = TC_PRIO_CONTROL;
948 skb->dev = slave->dev;
949
950 if (vid)
951 __vlan_hwaccel_put_tag(skb, vlan_proto, vid);
952
953 dev_queue_xmit(skb);
954 }
955
956 static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[],
957 bool strict_match)
958 {
959 struct bonding *bond = bond_get_bond_by_slave(slave);
960 struct net_device *upper;
961 struct list_head *iter;
962 struct bond_vlan_tag *tags;
963
964 /* send untagged */
965 alb_send_lp_vid(slave, mac_addr, 0, 0);
966
967 /* loop through all devices and see if we need to send a packet
968 * for that device.
969 */
970 rcu_read_lock();
971 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
972 if (is_vlan_dev(upper) && vlan_get_encap_level(upper) == 0) {
973 if (strict_match &&
974 ether_addr_equal_64bits(mac_addr,
975 upper->dev_addr)) {
976 alb_send_lp_vid(slave, mac_addr,
977 vlan_dev_vlan_proto(upper),
978 vlan_dev_vlan_id(upper));
979 } else if (!strict_match) {
980 alb_send_lp_vid(slave, upper->dev_addr,
981 vlan_dev_vlan_proto(upper),
982 vlan_dev_vlan_id(upper));
983 }
984 }
985
986 /* If this is a macvlan device, then only send updates
987 * when strict_match is turned off.
988 */
989 if (netif_is_macvlan(upper) && !strict_match) {
990 tags = bond_verify_device_path(bond->dev, upper, 0);
991 if (IS_ERR_OR_NULL(tags))
992 BUG();
993 alb_send_lp_vid(slave, upper->dev_addr,
994 tags[0].vlan_proto, tags[0].vlan_id);
995 kfree(tags);
996 }
997 }
998 rcu_read_unlock();
999 }
1000
1001 static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[])
1002 {
1003 struct net_device *dev = slave->dev;
1004 struct sockaddr s_addr;
1005
1006 if (BOND_MODE(slave->bond) == BOND_MODE_TLB) {
1007 memcpy(dev->dev_addr, addr, dev->addr_len);
1008 return 0;
1009 }
1010
1011 /* for rlb each slave must have a unique hw mac addresses so that
1012 * each slave will receive packets destined to a different mac
1013 */
1014 memcpy(s_addr.sa_data, addr, dev->addr_len);
1015 s_addr.sa_family = dev->type;
1016 if (dev_set_mac_address(dev, &s_addr)) {
1017 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",
1018 dev->name);
1019 return -EOPNOTSUPP;
1020 }
1021 return 0;
1022 }
1023
1024 /* Swap MAC addresses between two slaves.
1025 *
1026 * Called with RTNL held, and no other locks.
1027 */
1028 static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1029 {
1030 u8 tmp_mac_addr[ETH_ALEN];
1031
1032 ether_addr_copy(tmp_mac_addr, slave1->dev->dev_addr);
1033 alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr);
1034 alb_set_slave_mac_addr(slave2, tmp_mac_addr);
1035
1036 }
1037
1038 /* Send learning packets after MAC address swap.
1039 *
1040 * Called with RTNL and no other locks
1041 */
1042 static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1043 struct slave *slave2)
1044 {
1045 int slaves_state_differ = (bond_slave_can_tx(slave1) != bond_slave_can_tx(slave2));
1046 struct slave *disabled_slave = NULL;
1047
1048 ASSERT_RTNL();
1049
1050 /* fasten the change in the switch */
1051 if (bond_slave_can_tx(slave1)) {
1052 alb_send_learning_packets(slave1, slave1->dev->dev_addr, false);
1053 if (bond->alb_info.rlb_enabled) {
1054 /* inform the clients that the mac address
1055 * has changed
1056 */
1057 rlb_req_update_slave_clients(bond, slave1);
1058 }
1059 } else {
1060 disabled_slave = slave1;
1061 }
1062
1063 if (bond_slave_can_tx(slave2)) {
1064 alb_send_learning_packets(slave2, slave2->dev->dev_addr, false);
1065 if (bond->alb_info.rlb_enabled) {
1066 /* inform the clients that the mac address
1067 * has changed
1068 */
1069 rlb_req_update_slave_clients(bond, slave2);
1070 }
1071 } else {
1072 disabled_slave = slave2;
1073 }
1074
1075 if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1076 /* A disabled slave was assigned an active mac addr */
1077 rlb_teach_disabled_mac_on_primary(bond,
1078 disabled_slave->dev->dev_addr);
1079 }
1080 }
1081
1082 /**
1083 * alb_change_hw_addr_on_detach
1084 * @bond: bonding we're working on
1085 * @slave: the slave that was just detached
1086 *
1087 * We assume that @slave was already detached from the slave list.
1088 *
1089 * If @slave's permanent hw address is different both from its current
1090 * address and from @bond's address, then somewhere in the bond there's
1091 * a slave that has @slave's permanet address as its current address.
1092 * We'll make sure that that slave no longer uses @slave's permanent address.
1093 *
1094 * Caller must hold RTNL and no other locks
1095 */
1096 static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1097 {
1098 int perm_curr_diff;
1099 int perm_bond_diff;
1100 struct slave *found_slave;
1101
1102 perm_curr_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1103 slave->dev->dev_addr);
1104 perm_bond_diff = !ether_addr_equal_64bits(slave->perm_hwaddr,
1105 bond->dev->dev_addr);
1106
1107 if (perm_curr_diff && perm_bond_diff) {
1108 found_slave = bond_slave_has_mac(bond, slave->perm_hwaddr);
1109
1110 if (found_slave) {
1111 alb_swap_mac_addr(slave, found_slave);
1112 alb_fasten_mac_swap(bond, slave, found_slave);
1113 }
1114 }
1115 }
1116
1117 /**
1118 * alb_handle_addr_collision_on_attach
1119 * @bond: bonding we're working on
1120 * @slave: the slave that was just attached
1121 *
1122 * checks uniqueness of slave's mac address and handles the case the
1123 * new slave uses the bonds mac address.
1124 *
1125 * If the permanent hw address of @slave is @bond's hw address, we need to
1126 * find a different hw address to give @slave, that isn't in use by any other
1127 * slave in the bond. This address must be, of course, one of the permanent
1128 * addresses of the other slaves.
1129 *
1130 * We go over the slave list, and for each slave there we compare its
1131 * permanent hw address with the current address of all the other slaves.
1132 * If no match was found, then we've found a slave with a permanent address
1133 * that isn't used by any other slave in the bond, so we can assign it to
1134 * @slave.
1135 *
1136 * assumption: this function is called before @slave is attached to the
1137 * bond slave list.
1138 */
1139 static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1140 {
1141 struct slave *has_bond_addr = rcu_access_pointer(bond->curr_active_slave);
1142 struct slave *tmp_slave1, *free_mac_slave = NULL;
1143 struct list_head *iter;
1144
1145 if (!bond_has_slaves(bond)) {
1146 /* this is the first slave */
1147 return 0;
1148 }
1149
1150 /* if slave's mac address differs from bond's mac address
1151 * check uniqueness of slave's mac address against the other
1152 * slaves in the bond.
1153 */
1154 if (!ether_addr_equal_64bits(slave->perm_hwaddr, bond->dev->dev_addr)) {
1155 if (!bond_slave_has_mac(bond, slave->dev->dev_addr))
1156 return 0;
1157
1158 /* Try setting slave mac to bond address and fall-through
1159 * to code handling that situation below...
1160 */
1161 alb_set_slave_mac_addr(slave, bond->dev->dev_addr);
1162 }
1163
1164 /* The slave's address is equal to the address of the bond.
1165 * Search for a spare address in the bond for this slave.
1166 */
1167 bond_for_each_slave(bond, tmp_slave1, iter) {
1168 if (!bond_slave_has_mac(bond, tmp_slave1->perm_hwaddr)) {
1169 /* no slave has tmp_slave1's perm addr
1170 * as its curr addr
1171 */
1172 free_mac_slave = tmp_slave1;
1173 break;
1174 }
1175
1176 if (!has_bond_addr) {
1177 if (ether_addr_equal_64bits(tmp_slave1->dev->dev_addr,
1178 bond->dev->dev_addr)) {
1179
1180 has_bond_addr = tmp_slave1;
1181 }
1182 }
1183 }
1184
1185 if (free_mac_slave) {
1186 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr);
1187
1188 netdev_warn(bond->dev, "the hw address of slave %s is in use by the bond; giving it the hw address of %s\n",
1189 slave->dev->name, free_mac_slave->dev->name);
1190
1191 } else if (has_bond_addr) {
1192 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",
1193 slave->dev->name);
1194 return -EFAULT;
1195 }
1196
1197 return 0;
1198 }
1199
1200 /**
1201 * alb_set_mac_address
1202 * @bond:
1203 * @addr:
1204 *
1205 * In TLB mode all slaves are configured to the bond's hw address, but set
1206 * their dev_addr field to different addresses (based on their permanent hw
1207 * addresses).
1208 *
1209 * For each slave, this function sets the interface to the new address and then
1210 * changes its dev_addr field to its previous value.
1211 *
1212 * Unwinding assumes bond's mac address has not yet changed.
1213 */
1214 static int alb_set_mac_address(struct bonding *bond, void *addr)
1215 {
1216 struct slave *slave, *rollback_slave;
1217 struct list_head *iter;
1218 struct sockaddr sa;
1219 char tmp_addr[ETH_ALEN];
1220 int res;
1221
1222 if (bond->alb_info.rlb_enabled)
1223 return 0;
1224
1225 bond_for_each_slave(bond, slave, iter) {
1226 /* save net_device's current hw address */
1227 ether_addr_copy(tmp_addr, slave->dev->dev_addr);
1228
1229 res = dev_set_mac_address(slave->dev, addr);
1230
1231 /* restore net_device's hw address */
1232 ether_addr_copy(slave->dev->dev_addr, tmp_addr);
1233
1234 if (res)
1235 goto unwind;
1236 }
1237
1238 return 0;
1239
1240 unwind:
1241 memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1242 sa.sa_family = bond->dev->type;
1243
1244 /* unwind from head to the slave that failed */
1245 bond_for_each_slave(bond, rollback_slave, iter) {
1246 if (rollback_slave == slave)
1247 break;
1248 ether_addr_copy(tmp_addr, rollback_slave->dev->dev_addr);
1249 dev_set_mac_address(rollback_slave->dev, &sa);
1250 ether_addr_copy(rollback_slave->dev->dev_addr, tmp_addr);
1251 }
1252
1253 return res;
1254 }
1255
1256 /************************ exported alb funcions ************************/
1257
1258 int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1259 {
1260 int res;
1261
1262 res = tlb_initialize(bond);
1263 if (res)
1264 return res;
1265
1266 if (rlb_enabled) {
1267 bond->alb_info.rlb_enabled = 1;
1268 res = rlb_initialize(bond);
1269 if (res) {
1270 tlb_deinitialize(bond);
1271 return res;
1272 }
1273 } else {
1274 bond->alb_info.rlb_enabled = 0;
1275 }
1276
1277 return 0;
1278 }
1279
1280 void bond_alb_deinitialize(struct bonding *bond)
1281 {
1282 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1283
1284 tlb_deinitialize(bond);
1285
1286 if (bond_info->rlb_enabled)
1287 rlb_deinitialize(bond);
1288 }
1289
1290 static int bond_do_alb_xmit(struct sk_buff *skb, struct bonding *bond,
1291 struct slave *tx_slave)
1292 {
1293 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1294 struct ethhdr *eth_data = eth_hdr(skb);
1295
1296 if (!tx_slave) {
1297 /* unbalanced or unassigned, send through primary */
1298 tx_slave = rcu_dereference(bond->curr_active_slave);
1299 if (bond->params.tlb_dynamic_lb)
1300 bond_info->unbalanced_load += skb->len;
1301 }
1302
1303 if (tx_slave && bond_slave_can_tx(tx_slave)) {
1304 if (tx_slave != rcu_access_pointer(bond->curr_active_slave)) {
1305 ether_addr_copy(eth_data->h_source,
1306 tx_slave->dev->dev_addr);
1307 }
1308
1309 bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1310 goto out;
1311 }
1312
1313 if (tx_slave && bond->params.tlb_dynamic_lb) {
1314 spin_lock(&bond->mode_lock);
1315 __tlb_clear_slave(bond, tx_slave, 0);
1316 spin_unlock(&bond->mode_lock);
1317 }
1318
1319 /* no suitable interface, frame not sent */
1320 bond_tx_drop(bond->dev, skb);
1321 out:
1322 return NETDEV_TX_OK;
1323 }
1324
1325 int bond_tlb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1326 {
1327 struct bonding *bond = netdev_priv(bond_dev);
1328 struct ethhdr *eth_data;
1329 struct slave *tx_slave = NULL;
1330 u32 hash_index;
1331
1332 skb_reset_mac_header(skb);
1333 eth_data = eth_hdr(skb);
1334
1335 /* Do not TX balance any multicast or broadcast */
1336 if (!is_multicast_ether_addr(eth_data->h_dest)) {
1337 switch (skb->protocol) {
1338 case htons(ETH_P_IP):
1339 case htons(ETH_P_IPX):
1340 /* In case of IPX, it will falback to L2 hash */
1341 case htons(ETH_P_IPV6):
1342 hash_index = bond_xmit_hash(bond, skb);
1343 if (bond->params.tlb_dynamic_lb) {
1344 tx_slave = tlb_choose_channel(bond,
1345 hash_index & 0xFF,
1346 skb->len);
1347 } else {
1348 struct bond_up_slave *slaves;
1349 unsigned int count;
1350
1351 slaves = rcu_dereference(bond->slave_arr);
1352 count = slaves ? ACCESS_ONCE(slaves->count) : 0;
1353 if (likely(count))
1354 tx_slave = slaves->arr[hash_index %
1355 count];
1356 }
1357 break;
1358 }
1359 }
1360 return bond_do_alb_xmit(skb, bond, tx_slave);
1361 }
1362
1363 int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1364 {
1365 struct bonding *bond = netdev_priv(bond_dev);
1366 struct ethhdr *eth_data;
1367 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1368 struct slave *tx_slave = NULL;
1369 static const __be32 ip_bcast = htonl(0xffffffff);
1370 int hash_size = 0;
1371 bool do_tx_balance = true;
1372 u32 hash_index = 0;
1373 const u8 *hash_start = NULL;
1374 struct ipv6hdr *ip6hdr;
1375
1376 skb_reset_mac_header(skb);
1377 eth_data = eth_hdr(skb);
1378
1379 switch (ntohs(skb->protocol)) {
1380 case ETH_P_IP: {
1381 const struct iphdr *iph = ip_hdr(skb);
1382
1383 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast) ||
1384 (iph->daddr == ip_bcast) ||
1385 (iph->protocol == IPPROTO_IGMP)) {
1386 do_tx_balance = false;
1387 break;
1388 }
1389 hash_start = (char *)&(iph->daddr);
1390 hash_size = sizeof(iph->daddr);
1391 }
1392 break;
1393 case ETH_P_IPV6:
1394 /* IPv6 doesn't really use broadcast mac address, but leave
1395 * that here just in case.
1396 */
1397 if (ether_addr_equal_64bits(eth_data->h_dest, mac_bcast)) {
1398 do_tx_balance = false;
1399 break;
1400 }
1401
1402 /* IPv6 uses all-nodes multicast as an equivalent to
1403 * broadcasts in IPv4.
1404 */
1405 if (ether_addr_equal_64bits(eth_data->h_dest, mac_v6_allmcast)) {
1406 do_tx_balance = false;
1407 break;
1408 }
1409
1410 /* Additianally, DAD probes should not be tx-balanced as that
1411 * will lead to false positives for duplicate addresses and
1412 * prevent address configuration from working.
1413 */
1414 ip6hdr = ipv6_hdr(skb);
1415 if (ipv6_addr_any(&ip6hdr->saddr)) {
1416 do_tx_balance = false;
1417 break;
1418 }
1419
1420 hash_start = (char *)&(ipv6_hdr(skb)->daddr);
1421 hash_size = sizeof(ipv6_hdr(skb)->daddr);
1422 break;
1423 case ETH_P_IPX:
1424 if (ipx_hdr(skb)->ipx_checksum != IPX_NO_CHECKSUM) {
1425 /* something is wrong with this packet */
1426 do_tx_balance = false;
1427 break;
1428 }
1429
1430 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1431 /* The only protocol worth balancing in
1432 * this family since it has an "ARP" like
1433 * mechanism
1434 */
1435 do_tx_balance = false;
1436 break;
1437 }
1438
1439 hash_start = (char *)eth_data->h_dest;
1440 hash_size = ETH_ALEN;
1441 break;
1442 case ETH_P_ARP:
1443 do_tx_balance = false;
1444 if (bond_info->rlb_enabled)
1445 tx_slave = rlb_arp_xmit(skb, bond);
1446 break;
1447 default:
1448 do_tx_balance = false;
1449 break;
1450 }
1451
1452 if (do_tx_balance) {
1453 hash_index = _simple_hash(hash_start, hash_size);
1454 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1455 }
1456
1457 return bond_do_alb_xmit(skb, bond, tx_slave);
1458 }
1459
1460 void bond_alb_monitor(struct work_struct *work)
1461 {
1462 struct bonding *bond = container_of(work, struct bonding,
1463 alb_work.work);
1464 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1465 struct list_head *iter;
1466 struct slave *slave;
1467
1468 if (!bond_has_slaves(bond)) {
1469 bond_info->tx_rebalance_counter = 0;
1470 bond_info->lp_counter = 0;
1471 goto re_arm;
1472 }
1473
1474 rcu_read_lock();
1475
1476 bond_info->tx_rebalance_counter++;
1477 bond_info->lp_counter++;
1478
1479 /* send learning packets */
1480 if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1481 bool strict_match;
1482
1483 bond_for_each_slave_rcu(bond, slave, iter) {
1484 /* If updating current_active, use all currently
1485 * user mac addreses (!strict_match). Otherwise, only
1486 * use mac of the slave device.
1487 * In RLB mode, we always use strict matches.
1488 */
1489 strict_match = (slave != rcu_access_pointer(bond->curr_active_slave) ||
1490 bond_info->rlb_enabled);
1491 alb_send_learning_packets(slave, slave->dev->dev_addr,
1492 strict_match);
1493 }
1494 bond_info->lp_counter = 0;
1495 }
1496
1497 /* rebalance tx traffic */
1498 if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1499 bond_for_each_slave_rcu(bond, slave, iter) {
1500 tlb_clear_slave(bond, slave, 1);
1501 if (slave == rcu_access_pointer(bond->curr_active_slave)) {
1502 SLAVE_TLB_INFO(slave).load =
1503 bond_info->unbalanced_load /
1504 BOND_TLB_REBALANCE_INTERVAL;
1505 bond_info->unbalanced_load = 0;
1506 }
1507 }
1508 bond_info->tx_rebalance_counter = 0;
1509 }
1510
1511 if (bond_info->rlb_enabled) {
1512 if (bond_info->primary_is_promisc &&
1513 (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1514
1515 /* dev_set_promiscuity requires rtnl and
1516 * nothing else. Avoid race with bond_close.
1517 */
1518 rcu_read_unlock();
1519 if (!rtnl_trylock())
1520 goto re_arm;
1521
1522 bond_info->rlb_promisc_timeout_counter = 0;
1523
1524 /* If the primary was set to promiscuous mode
1525 * because a slave was disabled then
1526 * it can now leave promiscuous mode.
1527 */
1528 dev_set_promiscuity(rtnl_dereference(bond->curr_active_slave)->dev,
1529 -1);
1530 bond_info->primary_is_promisc = 0;
1531
1532 rtnl_unlock();
1533 rcu_read_lock();
1534 }
1535
1536 if (bond_info->rlb_rebalance) {
1537 bond_info->rlb_rebalance = 0;
1538 rlb_rebalance(bond);
1539 }
1540
1541 /* check if clients need updating */
1542 if (bond_info->rx_ntt) {
1543 if (bond_info->rlb_update_delay_counter) {
1544 --bond_info->rlb_update_delay_counter;
1545 } else {
1546 rlb_update_rx_clients(bond);
1547 if (bond_info->rlb_update_retry_counter)
1548 --bond_info->rlb_update_retry_counter;
1549 else
1550 bond_info->rx_ntt = 0;
1551 }
1552 }
1553 }
1554 rcu_read_unlock();
1555 re_arm:
1556 queue_delayed_work(bond->wq, &bond->alb_work, alb_delta_in_ticks);
1557 }
1558
1559 /* assumption: called before the slave is attached to the bond
1560 * and not locked by the bond lock
1561 */
1562 int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1563 {
1564 int res;
1565
1566 res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr);
1567 if (res)
1568 return res;
1569
1570 res = alb_handle_addr_collision_on_attach(bond, slave);
1571 if (res)
1572 return res;
1573
1574 tlb_init_slave(slave);
1575
1576 /* order a rebalance ASAP */
1577 bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1578
1579 if (bond->alb_info.rlb_enabled)
1580 bond->alb_info.rlb_rebalance = 1;
1581
1582 return 0;
1583 }
1584
1585 /* Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1586 * if necessary.
1587 *
1588 * Caller must hold RTNL and no other locks
1589 */
1590 void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1591 {
1592 if (bond_has_slaves(bond))
1593 alb_change_hw_addr_on_detach(bond, slave);
1594
1595 tlb_clear_slave(bond, slave, 0);
1596
1597 if (bond->alb_info.rlb_enabled) {
1598 bond->alb_info.rx_slave = NULL;
1599 rlb_clear_slave(bond, slave);
1600 }
1601
1602 }
1603
1604 void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1605 {
1606 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1607
1608 if (link == BOND_LINK_DOWN) {
1609 tlb_clear_slave(bond, slave, 0);
1610 if (bond->alb_info.rlb_enabled)
1611 rlb_clear_slave(bond, slave);
1612 } else if (link == BOND_LINK_UP) {
1613 /* order a rebalance ASAP */
1614 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1615 if (bond->alb_info.rlb_enabled) {
1616 bond->alb_info.rlb_rebalance = 1;
1617 /* If the updelay module parameter is smaller than the
1618 * forwarding delay of the switch the rebalance will
1619 * not work because the rebalance arp replies will
1620 * not be forwarded to the clients..
1621 */
1622 }
1623 }
1624
1625 if (bond_is_nondyn_tlb(bond)) {
1626 if (bond_update_slave_arr(bond, NULL))
1627 pr_err("Failed to build slave-array for TLB mode.\n");
1628 }
1629 }
1630
1631 /**
1632 * bond_alb_handle_active_change - assign new curr_active_slave
1633 * @bond: our bonding struct
1634 * @new_slave: new slave to assign
1635 *
1636 * Set the bond->curr_active_slave to @new_slave and handle
1637 * mac address swapping and promiscuity changes as needed.
1638 *
1639 * Caller must hold RTNL
1640 */
1641 void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1642 {
1643 struct slave *swap_slave;
1644 struct slave *curr_active;
1645
1646 curr_active = rtnl_dereference(bond->curr_active_slave);
1647 if (curr_active == new_slave)
1648 return;
1649
1650 if (curr_active && bond->alb_info.primary_is_promisc) {
1651 dev_set_promiscuity(curr_active->dev, -1);
1652 bond->alb_info.primary_is_promisc = 0;
1653 bond->alb_info.rlb_promisc_timeout_counter = 0;
1654 }
1655
1656 swap_slave = curr_active;
1657 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1658
1659 if (!new_slave || !bond_has_slaves(bond))
1660 return;
1661
1662 /* set the new curr_active_slave to the bonds mac address
1663 * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1664 */
1665 if (!swap_slave)
1666 swap_slave = bond_slave_has_mac(bond, bond->dev->dev_addr);
1667
1668 /* Arrange for swap_slave and new_slave to temporarily be
1669 * ignored so we can mess with their MAC addresses without
1670 * fear of interference from transmit activity.
1671 */
1672 if (swap_slave)
1673 tlb_clear_slave(bond, swap_slave, 1);
1674 tlb_clear_slave(bond, new_slave, 1);
1675
1676 /* in TLB mode, the slave might flip down/up with the old dev_addr,
1677 * and thus filter bond->dev_addr's packets, so force bond's mac
1678 */
1679 if (BOND_MODE(bond) == BOND_MODE_TLB) {
1680 struct sockaddr sa;
1681 u8 tmp_addr[ETH_ALEN];
1682
1683 ether_addr_copy(tmp_addr, new_slave->dev->dev_addr);
1684
1685 memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1686 sa.sa_family = bond->dev->type;
1687 /* we don't care if it can't change its mac, best effort */
1688 dev_set_mac_address(new_slave->dev, &sa);
1689
1690 ether_addr_copy(new_slave->dev->dev_addr, tmp_addr);
1691 }
1692
1693 /* curr_active_slave must be set before calling alb_swap_mac_addr */
1694 if (swap_slave) {
1695 /* swap mac address */
1696 alb_swap_mac_addr(swap_slave, new_slave);
1697 alb_fasten_mac_swap(bond, swap_slave, new_slave);
1698 } else {
1699 /* set the new_slave to the bond mac address */
1700 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr);
1701 alb_send_learning_packets(new_slave, bond->dev->dev_addr,
1702 false);
1703 }
1704 }
1705
1706 /* Called with RTNL */
1707 int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1708 {
1709 struct bonding *bond = netdev_priv(bond_dev);
1710 struct sockaddr *sa = addr;
1711 struct slave *curr_active;
1712 struct slave *swap_slave;
1713 int res;
1714
1715 if (!is_valid_ether_addr(sa->sa_data))
1716 return -EADDRNOTAVAIL;
1717
1718 res = alb_set_mac_address(bond, addr);
1719 if (res)
1720 return res;
1721
1722 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1723
1724 /* If there is no curr_active_slave there is nothing else to do.
1725 * Otherwise we'll need to pass the new address to it and handle
1726 * duplications.
1727 */
1728 curr_active = rtnl_dereference(bond->curr_active_slave);
1729 if (!curr_active)
1730 return 0;
1731
1732 swap_slave = bond_slave_has_mac(bond, bond_dev->dev_addr);
1733
1734 if (swap_slave) {
1735 alb_swap_mac_addr(swap_slave, curr_active);
1736 alb_fasten_mac_swap(bond, swap_slave, curr_active);
1737 } else {
1738 alb_set_slave_mac_addr(curr_active, bond_dev->dev_addr);
1739
1740 alb_send_learning_packets(curr_active,
1741 bond_dev->dev_addr, false);
1742 if (bond->alb_info.rlb_enabled) {
1743 /* inform clients mac address has changed */
1744 rlb_req_update_slave_clients(bond, curr_active);
1745 }
1746 }
1747
1748 return 0;
1749 }
1750
1751 void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1752 {
1753 if (bond->alb_info.rlb_enabled)
1754 rlb_clear_vlan(bond, vlan_id);
1755 }
1756
Linux kernel - Deliverables
This page took 0.068175 seconds and 5 git commands to generate.